Chemical Engineering and Processing

Copyright 2004 Elsevier B.V. All rights reserved

Volume 44, Issue 1, Pages 1-137 (January 2005)

1. A new method for producing anhydrous puffed borax
Pages 1-6
mer ahin, Nasrettin Genli and Mustafa zdemir

2. The influence of temperature and inlet velocity on cyclone pressure drop: a CFD study
Pages 7-12
Jolius Gimbun, T. G. Chuah, A. Fakhrul-Razi and Thomas S. Y. Choong

3. A pollution reduction methodology in reactor design

Pages 13-21
Qishi Chen and Xiao Feng

4. Effects of hydraulic residence time on metal uptake by activated sludge

Pages 23-32
Tlay A. zbelge, H. nder zbelge and Murat Tursun

5. Porous catalyst intraparticle status of parallel, equilibrium-restrained reactions

Pages 33-39
Guangsuo Yu, Fuchen Wang, Zhenghua Dai and Zunhong Yu

6. PID controller tuning using mathematical programming

Pages 41-49
George Syrcos and Ioannis K. Kookos

7. Separation of acetic acidwater mixtures through acrylonitrile grafted poly(vinyl

alcohol) membranes by pervaporation
Pages 51-58
N. Alghezawi, O. anli, L. Aras and G. Asman

8. Countercurrent flow distribution in structured packing via computed tomography

Pages 59-69
Shaibal Roy, A. Kemoun, M. H. Al-Dahhan, M. P. Dudukovic, Thomas B. Skourlis and Frits M.
Dautzenberg

9. Effect of cycling operations on an adsorbed natural gas storage

Pages 71-79
O. Pupier, V. Goetz and R. Fiscal

10. Effect of internal on the hydrodynamics in external-loop airlift reactors

Pages 81-87
Tongwang Zhang, Jinfu Wang, Tiefeng Wang, Jing Lin and Yong Jin

11. Multicriteria synthesis of flexible heat-exchanger networks with uncertain source-

stream temperatures
Pages 89-100
Cheng-Liang Chen and Ping-Sung Hung

12. On-line dynamic optimization and control strategy for improving the performance of
batch reactors
Pages 101-114
A. Arpornwichanop, P. Kittisupakorn and I. M. Mujtaba
13. Synthesis of nano-sized particles from metal carbonates by the method of reversed
mycelles
Pages 115-119
Christo Karagiozov and Dafina Momchilova

14. Mechanism of mass transfer from bubbles in dispersions: Part II: Mass transfer
coefficients in stirred gasliquid reactor and bubble column
Pages 121-130
V. Linek, M. Korda and T. Moucha

15. Separation of n-hexaneethyl acetate mixtures by azeotropic batch distillation with

heterogeneous entrainers
Pages 131-137
I. Rodriguez-Donis, J. Acosta-Esquijarosa, V. Gerbaud, E. Pardillo-Fondevila and X. Joulia

16. Inside front cover - Editorial Board EDITORIAL BOARD

Pages CO2-CO2

Volume 44, Issue 2, Pages 139-334 (February 2005)

Pneumatic Conveying and Handling of Particulate Solids
Edited by Janos Gyenis and Avi Levy

1. Special issue on conveying and handling of particulate solids

Pages 139-140
Janos Gyenis and Avi Levy

2. A review of the research work of Professor Predrag Marjanovi

Pages 141-151
David Mills

3. Influence of the particle diameter and density in the gas velocity in jet spouted beds
Pages 153-157
Mara J. San Jos, Sonia Alvarez, Alvaro Ortiz de Salazar, Martn Olazar and Javier Bilbao

4. Taking-off model of particles with a wide size distribution

Pages 159-166
Isabelle Descamps, Jean-Luc Harion and Bernard Baudoin

5. Solids deposition in low-velocity slug flow pneumatic conveying

Pages 167-173
J. Li, C. Webb, S. S. Pandiella, G. M. Campbell, T. Dyakowski, A. Cowell and D. McGlinchey

6. Identification of material specific attrition mechanisms for polymers in dilute phase

pneumatic conveying
Pages 175-185
Lars Frye and Wolfgang Peukert

7. Two-dimensional numerical simulations of the pneumatic drying in vertical pipes

Pages 187-192
I. Skuratovsky, A. Levy and I. Borde

8. The formation of fine particles by salting-out precipitation

Pages 193-200
Judit Tth, Andrea Kardos-Fodor and Susan Halsz-Pterfi

9. Micronized cocoa butter particles produced by a supercritical process

Pages 201-207
J. -J. Letourneau, S. Vigneau, P. Gonus and J. Fages
10. Food powder handling and processing: Industry problems, knowledge barriers and
research opportunities
Pages 209-214
John J. Fitzpatrick and Lilia Ahrn

11. Microencapsulation of particles using supercritical carbon dioxide

Pages 215-219
H. Krber and U. Teipel

12. Plasma spheroidization of ceramic particles

Pages 221-224
Z. Kroly and J. Szpvlgyi

13. Treatment of particulate metallurgical wastes in thermal plasmas

Pages 225-229
I. Mohai and J. Szpvlgyi
14. Defluidization modelling of pyrolysis of plastics in a conical spouted bed reactor
Pages 231-235
Roberto Aguado, Rubn Prieto, Mara J. San Jos, Sonia Alvarez, Martn Olazar and Javier
Bilbao

15. The energy of bed processing during drum granulation

Pages 237-243
Tadeusz Gluba

16. Control of aggregation in production and handling of nanoparticles

Pages 245-252
Wolfgang Peukert, Hans-Christoph Schwarzer and Frank Stenger

17. Effect of different types of impact surface on coal degradation

Pages 253-261
R. K. Sahoo and D. Roach

18. Using statistical moments to describe grinding in a ball mill for industrial-scale
process
Pages 263-266
Andrzej Heim, Tomasz P. Olejnik and Agnieszka Pawlak

19. On possible instability of throughputs in complex milling circuits

Pages 267-272
V. Mizonov, V. Zhukov, A. Korovkin and H. Berthiaux

20. Optimising design of continuous grinding mill-classifier systems

Pages 273-277
P. B. Kis, Cs. Mihlyk and B. G. Lakatos

21. Air classification of solid particles: a review

Pages 279-285
M. Shapiro and V. Galperin

22. Application of a vertical venturi separator for improved recycling of automotive tires
Pages 287-291
W. McBride and S. Keys

23. Particle movement during granular intermingling in a pulsated bottom mixer

Pages 293-296
Mikls Nemnyi and Attila J. Kovcs
24. Estimating the homogenization efficiency of mammoth silos by process dynamics
and simulations: Comparing the results of process dynamics with the simulations
Pages 297-302
D. L. Schott, L. A. van Wijk and W. J. Vlasblom

25. Assessing the homogeneity of powder mixtures by on-line electrical capacitance

Pages 303-313
N. Ehrhardt, M. Montagne, H. Berthiaux, B. Dalloz-Dubrujeaud and C. Gatumel

26. Solid transport in a pyrolysis pilot-scale rotary kiln: preliminary resultsstationary

and dynamic results
Pages 315-321
N. Descoins, J. -L. Dirion and T. Howes

27. Controlling dust emissions and explosion hazards in powder handling plants
Pages 323-326
Peter Wypych, Dave Cook and Paul Cooper

28. Photocatalytic degradation of trichloroethylene (TCE) over TiO2/silica gel in a

circulating fluidized bed (CFB) photoreactor
Pages 327-334
Tak Hyoung Lim and Sang Done Kim

29. Inside front cover - Editorial Board EDITORIAL BOARD

Pages CO2-CO2
 Chemical Engineering and Processing 44 (2005) 16

A new method for producing anhydrous puffed borax

mer Sahin , Nasrettin Genli, Mustafa zdemir
Department of Chemistry, Harran University, S. Urfa, Turkey

Received 30 June 2003; received in revised form 29 July 2003; accepted 8 March 2004
Available online 24 April 2004

Abstract

This paper describes the production of anhydrous puffed borax from borax pentahydrate (BPH) in a batch calcinator. The calcination of
BPH is incomplete since agglomeration starts at 300 C. In order to avoid agglomeration at temperatures higher than 300 C, the surface of
BPH particles have been covered with a CaO layer in aqueous media having a higher melting point than pure BPH. To investigate the effect
of this CaO layer on the calcination of BPH, the samples were fed into the batch calcinator. During the calcination process, quantities such as
the bulk density value, particle size distribution, sodium borate and calcium content of anhydrous borax have been determined as a function
of temperature. It was found that the anhydrous borax can be obtained by calcination of BPH particles covered with CaO in 50, 20 and 5 min
time intervals at a temperature range of 300, 400 and 500 C, respectively. As a result, puffed anhydrous borax of 99% purity with bulk density
of 0.082 g cm3 and containing 0.670% Ca2+ has been produced by this method in 5 min interval at 500 C.
2004 Elsevier B.V. All rights reserved.

Keywords: Borax pentahydrate; Anhydrous borax; Calcium oxide; Coating

1. Introduction With current technology, BA is produced from BPH or

Borax pentahydrate (BPH) is one of the most important
commercial boron compounds containing water of crystal-
lization. The structure formula of BPH can be best repre-
sented as Na2 B4 O5 (OH)4 2.67H2 O which means that there
are 2 mol of water in the molecular structure, remaining be-
ing real crystal water [1]. BPH is used in many areas, such as
perborate and boric acid production and detergent formula-
tions. However, its water content is not appropriate in some
applications such as the manufacture of high quality glass,
frit production ceramic and the refinement of precious met-
als. Thus, BPH should be dehydrated to anhydrous borax
(BA) state.
The production methods of BA from BPH or borax dec-
ahydrate may be categorized into three main groups. These
are: (a) azeotropic distillation, (b) melting and (c) dehydra-
tion in fluidized bed.
The method of azeotropic distillation is not yet suitable
for industrial purposes because of low yield and high heat
input needed. In addition, BPH cannot be converted by BA
using this method [2].
Corresponding author. Fax: +90-414-315-1998.
E-mail address: osahin@harran.edu.tr (. Sahin).
borax decahydrate (BDH) by the melting method which is a
two-stage process: first dehydration, and then fusion. BPH
or BDH is fused at a temperature higher than the melting
point of BA, ranging from 1000 to 1400 C at different parts
of firebox. The high temperature decreases the viscosity of
the molten mass, in order to produce proper fluidization
conditions for flow out from the firebox. The molten borax
is highly corrosive which can only be prevented by forming
a layer of solid calcined borax on the refractory material of
the furnace, thus the melting of borax is the most critical
an the most expensive step, which cause some difficulties in
operation and results in somewhat contaminated products.
Slow cooling gives crystal formation but immediate cooling
gives an amorphous glass [3]. By using this process, high
density of BA is produced, but there are many technical
difficulties in the process, such as corrosion and handling
problems. The other disadvantage of this method is that the
product needs crushing, grinding and homogenization before
being used.
Production of BA without fusion is very attractive with
respect to energy consumption and corrosion. Thus, the at-
tempts have mostly been in dehydration borax pentahydrate
in a fluidized bed via a stage-wise calcination [4,5]. The
main disadvantage of this process is the puffing of the par-

0255-2701/$ see front matter 2004 Elsevier B.V. All rights reserved.
doi:10.1016/j.cep.2004.03.004
2 . Sahin et al. / Chemical Engineering and Processing 44 (2005) 16
ticles during removal of water. In a fluidized bed, the calci-
nation of BPH cannot exceed 300 C at the earlier stage of
the process. At this temperature, it turns into an amorphous
form of BA and softens easily. Softening of the particles
cause agglomeration, which disturbs the fluidization condi-
tion in the fluidized bed calcinator [6].
BPH can be calcined to BA without puffing by at least
two-stage calcination which each needs at long time interval
in fluidized bed [7]. As a result, the calcination of BPH in
fluidized bed have some problem such as stage-wise, puffing,
agglomeration and needed long time.
Thus, a considerable number of attempts have been made
to develop a new method for production anhydrous borax
from BPH. As it is known the calcination of BPH cannot
be conducted above agglomeration temperature of 300 C in
fluidized bed. At temperatures higher than 300 C, the amor-
phous BA particles which soften readily adhere strongly to
each other since BPH dissolves in their water. This prob-
lem has been solved in our system by covering the particles
with a material that has a higher melting point than the fi-
nal agglomeration temperature. For this propose, the surface
of BPH was covered with Ca(OH)2 layer in aqueous me-
dia. The presence of Ca(OH)2 prevents the agglomeration of
BPH particles up to 500 C, since it controls the acceleration
of water vapor on the surface of BPH particles. Using this
method BPH can be converted to puffed BA containing 99%
Na2 B4 O7 in a 5-min interval at the temperature of 500 C.
The degree of puffing is very high, thus the calcination was
carried out in an oven.
The only disadvantage of this method is that the ob-
tained product (BA) has a lower bulk density. Such a puffed
product, with large quantities of liquids and gas, can be
loaded onto the expanded borax. For example, organic such
as trichloroethylene, cyclohexanone and pentachlorophenol
can be loaded at very high percentages onto the puffed bo-
rax, as can non-ionic and anionic surfactants, with the prod-
uct retaining its free-flowing characteristics. This renders
the puffed borax very useful in such diverse compositions
as dairy cleaners, fabric softeners and bath additives [8].
In addition, this puffed BA having large surface area can
be used for solidgas reactions. Recently, one popular appli-
Fig. 1. Schematic diagram of the experimental procedure.
cation of this BA reaction has been the production of sodium
borohydride. Sodium borohydride which is used as a hydro-
gen storage medium can also be synthesized by the reaction
of BA, with MgH2 through ball milling at room temperature
as shown the following reaction [9]:
8MgH2 +Na2 B4 O7 +Na2 CO3 4NaBH4 + 8MgO + CO2
The obtained anhydrous borax is very fragile, thus it can be
easily converted to powder form by grinding. This operation
increases the bulk density of BA.
2. Experiments
The calcination of BPH to BA was carried out in a batch
calcinator (thermolyne 6000 furnace) which is heated to a
predetermined calcination temperature (Fig. 1). During con-
stant temperature experiments, particles were fed into the
calcinator, which was in thermal equilibrium at the experi-
mental temperature. At a predetermined time interval, sam-
ples were withdrawn using a vacuum sampling tube. Titri-
metric method was used to determine the Na2 B4 O7 content
of sample [10]. The Ca2+ content of samples was determined
by Jenway PFP 7 model flame photometer. Bulk density of
BA was measured by a standard method [11].
The particle size of BPH used in the experiments was
chosen in the range 630 + 450 m and bulk density was
determined as 0.91 g cm3 . Technical grade BPH obtained
from ETIHOLDING (Kirka, Turkey) was used in the ex-
periments. Original Na2 B4 O7 content of samples was deter-
mined as 69.95% by titrimetric method [10].
Each series of experiments carried out with 10 g of BPH
washing 20 ml Ca(OH)2 solution with predetermined con-
centration during 1 min. This washed BPH particles were
then dried at 65 C in an oven for 5 h. In another series of
experiments, a mixture containing BPH and 2, 4, 5, 7, 10,
17 and 34 wt.% CaO were fed into the calcinator at the tem-
perature of 500 C for 10 min.
The other series of experiments, particles covering with
different Ca(OH)2 fed into the calcinator to investigate the
thermal decomposition behavior in the calcinator at 300,
 . Sahin et al. / Chemical Engineering and Processing 44 (2005) 16 3

400, 500 C. At each temperature samples was taken at 3,
5, 10, 20, 45 and 60 min intervals to determine the degree
of calcination.
3. Results and discussion
The main aim of this study is that the calcination of BPH
to BA is performed without agglomeration at higher temper-
ature than 300 C at which dehydration and decomposition
steps of it take place very fast. Normally, particles of BPH
dissolve their water and stick to each other to give agglom-
eration.
In this study, agglomeration was prevented by both cov-
ering the surface of BPH particles in Ca(OH)2 aqueous so-
lution and BPHCaO mixture in solid state.
In the first group of experimental work, mixtures contain-
ing BPH and 2, 4, 5, 7, 10, 17 and 34 wt.% CaO were fed
into the batch calcinator for 10 min at 500 C. At the end
of this kind of calcination, the conversion of BPH to BA is
found as about 99.6%. But, the BPH samples containing 2
and 4 wt.% CaO agglomerated slightly. However, agglomer-
ation was not observed at the concentration of CaO higher
than 5 wt.%. Table 1 shows the values found at the end of
this group. As can be seen from this table, the maximum
content of Ca2+ in BA particle was found in the particle size
smaller than 315 m. The particle size of BA is expanded
about twice higher than used for BPH particles at the begin-
ning of experiment by thermal shock. The result obtained
at the end of this step shows that the agglomeration can be
prevented in presence of CaO. However, the obtained prod-
uct (BA) at the end of this kind of operation contains high

Table 1
The effect of different BPHCaO mixture on the calcination of BPH particles at 500 C at the end of 10 min
Particle size (m) %CaO

5 10 17

% fraction Ca2+ % (g cm3 ) % fraction Ca2+ % (g cm3 ) % fraction Ca2+ % (g cm3 )

+1250 40.517 0.51 0.0583 31.609 0.69 0.0612 39.931 1.40 0.0635
+1000 30.747 0.35 30.172 0.69 16.319 0.38
+800 6.897 0.12 10.632 0.35 6.597 0.12
+630 1.724 0.03 2.010 0.06
+560 0.287
+450
+315
315 20.115 4.05 25.287 5.21 18.942 10.16

Fig. 2. TG curves for BPH both in pure and containing 2.46 wt.% Ca2+ .
4 . Sahin et al. / Chemical Engineering and Processing 44 (2005) 16

100

95

90
%Na2B4O7

85 300 C
400 C
80
500 C
75

70
0 10 20 30 40 50
time(min.)

Fig. 3. Changes of calcination time with temperature for the BPH samples containing about 2.5 wt.% Ca2+ .

Table 2
Properties of dehydrated BPH covering with 2.5 wt.% Ca2+ layer
Particle size (m) 300 C 400 C 500 C
% fraction Ca2+ % (g cm3 ) % fraction Ca2+ % (g cm3 ) % fraction Ca2+ % (g cm3 )
+1250 96.407 1.01 0.0552 90.998 1.00 0.0312 81.344 1.00 0.0589
+1000 8.927
+800 0.638
+630
+560
+450
+315
315 3.593 1.88 9.002 1.361 9.009 1.50

values of CaO which is not suitable for industrial applica-
tions.
In the second group of experiment, the surface of BPH
particle were covered with Ca(OH)2 in aqueous solution.
TG analysis of the BPH both in pure state and covered with
2.46 wt.% Ca2+ were performed under nitrogen atmosphere
at 10 C/min heating rate. Fig. 2 shows the results of TG
analysis. As can be seen from Fig. 2, the covered BPH par-
ticles with Ca(OH)2 dehydrate slowly with respect to pure
state. This phenomena involves the simultaneous transfer of
heat to evaporate the liquid and transfer of vapor within the
solid and vapor from the surface into the hot carrier gas. In
the case of pure BPH at temperature higher than 300 C, all
crystal water and some structure water dehydrated quickly
caused agglomeration. Thus, the mass transfer in this step
looked as the constant rate period where moisture movement
within the solid is sufficiently rapid to maintain a saturated
condition at the surface. In the case of dehydration of BPH
particles covered with Ca(OH)2 layer to increase the mass
transfer resistance between the particles and hot carrier gas,
the calcination is mainly controlled by the Ca(OH)2 layer.
Fig. 3 illustrates the Na2 B4 O7 content of the samples
containing about 2.50 wt.% Ca2+ taken from the batch cal-
cinator at different stages during isothermal decomposition
conducted at various temperatures. The total Na2 B4 O7 con-
tent of samples increased with increasing calcination tem-

Table 3
Properties of dehydrated BPH particles containing about 0.670 wt.% Ca2+
Particle size (m) 300 C 400 C 500 C

% fraction Ca2+ % (g cm3 ) % fraction Ca2+ % (g cm3 ) % fraction Ca2+ % (g cm3 )

+1250 33.463 0.201 0.0727 61.318 0.170 0.044 28.940 0.210 0.0824
+1000 39.779 26.689 38.607
+800 18.739 5.574 22.708
+630 4.013 3.482
+560
+450
+315
315 4.006 0.467 6.419 0.498 6.263 0.471
 . Sahin et al. / Chemical Engineering and Processing 44 (2005) 16 5

peratures. The BA contains some CaO, hence the conver- thick and homogenous cover on the surface of the anhy-
sion percentages of BA never attained 100% as seen from drous borax. The thickness of CaO layer controls the trans-
Fig. 3. The conversion of BPH to BA takes place in two fer rate of liberated water from borax pentahydrate to air.
stages namely dehydration and calcination [7]. But dehy- The sieve analysis, bulk density and Ca2+ content of BA
dration stage cannot be observed in Fig. 3 because of rapid obtained at the end of operation having different tempera-
dehydration of BPH. Fig. 3 illustrates that the calcination
of BPH to BA at 300 C completed in 50 min, whereas the
same degree of calcination realized in 5 min at the temper-
ature of 500 C. The Ca2+ content, bulk density and sieve
analysis of BA obtained at the end of this group of exper-
iments are illustrated in Table 2. As can be seen from this
table, the Ca2+ content of calcination realized at 300, 400
and 500 C is about 2.5 Ca2+ wt.%. The chosen particle size
range of BPH before calcination was 630 + 450 m and
the particle size of almost nine to ten part of calcines was
increased to +1210 m shown in Table 2. Thus, it can be
said that puffing is more effective in cause size increment
with respect to shrinking and fragmentation of puffed parti-
cles. The operation of covering BPH particles with Ca(OH)2
can not be attained at the temperature higher than 500 C
because of agglomeration. It is interesting to see that the
bulk density increased from 0.0317 to 0.0599 g cm3 with
increasing the temperature from 400 to 500 C. This behav-
ior of BPH can be easily explained by increasing sintering
effect which depends on the increasing temperature. In the
light of above results, it can also be concluded that better
operation could be possible from the second group of ex-
periment by decreasing the Ca2+ layer on the surface of
BPH particles. Thus, a third group of experiments were car-
ried out using very low Ca(OH)2 aqueous solution to cover
the BPH particles surface. The changes of Na2 B4 O7 con-
tent of samples containing about 0.60 wt.% Ca2+ with time
in this group are given in Fig. 4 at various constant temper-
atures. The Na2 B4 O7 content of the samples increases with
increasing calcination temperature. The calcination rate of
this group is higher than the previous group given in Fig. 3.
This behavior may be attributed to the fact that increas-
ing the percentages of additive caused the formation of a

100

95

90
%Na2B4O7

85
300 C
400 C
80 500 C

75

70
0 10 20 30 40 50
time (min.)

Fig. 4. Effect of temperature on the calcination of BPH samples containing Fig. 5. Microscopic photographs of borax pentahydrate: (a) pure, (b)
about 0.67 wt.% Ca2+ . containing 0.67 wt.% Ca2+ and (c) containing 2.5 wt.% Ca2+ .
6 . Sahin et al. / Chemical Engineering and Processing 44 (2005) 16
tures are given in Table 3. As can be seen in this table, parti-
cle size shows a distribution which is attributed to the Ca2+
content and temperature. In addition, the particles smaller
than 315 m contain twice higher Ca2+ than all other parti-
cle sizes contained. Microscopic photographs of borax pen-
tahydrate in pure, containing 0.67 wt. and 2.5 wt.% Ca2+
are given in Fig. 5, respectively. In pure state, the surface
of BPH particles are smooth whereas the quality of par-
ticles are decreased with increasing Ca2+ quantity on the
surface.
4. Conclusion
In order to obtain anhydrous borax with low bulk den-
sity by calcination in batch calcinator, particular attention is
paid to cover BPH surface with Ca(OH)2 layer. The follow-
ing conclusion can be withdrawn from the results obtained
during the calcination of covered BPH particles in a batch
calcinator:
(a) The calcination of BPH with single stage cannot be
carried out at the temperature higher than 300 C since
agglomeration is started. The calcination of BPH to BA
also takes longer at around 300 C.
(b) BPH can be transformed to BA without agglomeration
by mixing CaO with BPH in solid state at temperature
range 300500 C. In this case, the content of CaO in
BA must be higher than 7%.
(c) Covering the BPH particles with Ca(OH)2 aqueous
solution also prevented the agglomeration in tempera-
ture range 300500 C. In this kind of operation, the
BPH particle surface was covered with a thin layer of
Ca(OH)2 to control the transport rate of liberated water
from inside BPH particle to air. By this method, the
content of BPH can be decreased up to about 0.68 wt.%
Ca2+ .
(d) Calcination of BPH covered with Ca(OH)2 to BA re-
alized in 50, 20 and 5 min intervals at 300, 400 and
500 C, respectively. This result shows that at temper-
atures higher than 400 C, the puffed anhydrous borax
can be obtained by continuous operation. Also, the high
bulk density may be obtained by crashing puffed borax.
(e) The obtained low bulk density puffed borax is preferred
priority in some chemical process such as solid state
reaction, adsorption and high temperature application
of BA.
References
[1] R.P. Douglas, F.G. Donald, Refinement of the structure of tincal-
conite, Acta Cryst. C47 (1991) 22792282.
[2] G. Nencetti, A. Pennacchi, Chim. Ind. Milan 46 (5) (1968) 518525.
[3] Kirk-Othmer, Enc. Chem. Thech., 4 ed., vol. 4, 1992, p. 388.
[4] O. Sahin, A.N. Bulutcu, Turk. J. Chem. 26 (2002) 8996.
[5] S. Kocakusak, K. Akcay, T. Ayok, H.J. Koroglu, O.T. Savasc, R.
Tolun, Ind. Eng. Chem. Res. 35 (1996) 14241428.
[6] O. Sahin, U.G. Beker, A.N. Bulutucu, Int. J. Storing Handl. Process.
Powder 7 (1995) 165167.
[7] O. Sahin, A.N. Bulutucu, Chem. Eng. Process. 41 (2002) 135141.
[8] T.E. Raymond, US Patent 4 412 978 (1983).
[9] Z.P. Li, N. Morigazaki, B.H. Liu, S. Suda, J. Alloys Compd. 349
(2003) 232236.
[10] D.F. Snell, C.L. Hilton, Encyclopedia of Industrial Chemical Anal-
ysis, vol. 7, Wiley, New York, 1968, pp. 373384.
[11] ISO 3424, Sodium perborate for industrial use-determination of bulk
density.
 Chemical Engineering and Processing 44 (2005) 712

The influence of temperature and inlet velocity on cyclone pressure drop:

a CFD study
Jolius Gimbun , T.G. Chuah, A. Fakhrul-Razi, Thomas S.Y. Choong
Department of Chemical and Environmental Engineering, Faculty of Engineering, Universiti Putra Malaysia 43400 UPM Serdang, Selangor D. E., Malaysia

Received 16 February 2004; received in revised form 22 March 2004; accepted 22 March 2004
Available online 18 May 2004

Abstract

This work presents a computational fluid dynamics (CFD) calculation to predict and to evaluate the effects of temperature and inlet velocity
on the pressure drop of gas cyclones. The numerical solutions were carried out using spreadsheet and commercial CFD code Fluent 6.1. This
paper also reviews four empirical models for the prediction of cyclone pressure drop, namely [Air pollution control: a design approach, in: C.
David Cooper, F.C. Alley (Eds.), Cyclones, second ed., Woveland Press Inc., Illinois, 1939, p. 127139] [Chem. Eng. (1983) 99] [Doctoral
Thesis, Havarad University, USA, 1988], and [Chem. Eng. Progress (1993) 51]. All the predictions proved to be satisfactory when compared
with the presented experimental data. The CFD simulations predict excellently the cyclone pressure drop under different temperature and inlet
velocity with a maximum deviation of 3% from the experimental data. Specifically, results obtained from the computer modelling exercise
have demonstrated that CFD is a best method of modelling the cyclones operating pressure drop.
2004 Elsevier B.V. All rights reserved.

Keywords: Cyclone; CFD; Pressure drop; Temperature; Inlet velocity

1. Introduction efficiency of particle and pressure drop through the cy-

Cyclones are devices that employ a centrifugal force
generated by a spinning gas stream to separate particles
from the carrier gas. Their simple design, low capital cost
and nearly maintenance-free operation make them ideal for
use as pre-cleaners for more expensive final control devices
such as baghouses or electrostatic precipitators. Cyclones
are particularly well suited for high temperature and pres-
sure conditions because of their rugged design and flexible
components materials. Cyclone collection efficiencies can
reach 99% for particles bigger than 5 m [12], and can be
operated at very high dust loading. Cyclones are used for
the removal of large particles for both air pollution control
and process use. Application in extreme condition includes
the removing of coal dust in power plant, and the use as a
spray dryer or gasification reactor.
Engineers are generally interested in two parameters in
order to carry out an assessment of the design and perfor-
mance of a cyclone. These parameters are the collection
Corresponding author. Tel.: +60-19-248-9101; fax: +60-38946-7120.
E-mail address: jolius21@yahoo.co.uk (J. Gimbun).
clone. An accurate prediction of cyclone pressure drop is
very important because it relates directly to operating costs.
Higher inlet velocities give higher collection efficiencies for
a given cyclone, but this also increases the pressure drop
across the cyclone. Therefore, a trade off must be made
between higher collection efficiency and low pressure drop
across the cyclone. Computational fluid dynamics (CFD)
has a great potential to predict the flow field characteristics
and particle trajectories inside the cyclone as well as the
pressure drop [8]. The complicated swirling turbulent flow
in a cyclone places great demands on the numerical tech-
niques and the turbulence models employed in the CFD
codes when modelling the cyclone pressure drop.
In this study, pressure drop calculations are performed us-
ing CFD and compared with four empirical model of Shep-
herd and Lapple [11], Casal and Martinez [3], Dirgo [5],
and Coker [4]. These four empirical models and CFD pre-
diction are compared with the experimental data presented
in the literature. In this study, the CFD calculations are
carried out using commercial finite volume code Fluent 6.1
and the empirical models are performed in Microsoft Excel
spreadsheet.

0255-2701/$ see front matter 2004 Elsevier B.V. All rights reserved.
doi:10.1016/j.cep.2004.03.005
8 J. Gimbun et al. / Chemical Engineering and Processing 44 (2005) 712
D flow in a cyclone separator, there are a number of turbulence
b De
a S
h
H for strongly swirling flows found for example in cyclones
B ELEVATION
Fig. 1. Tangential cyclone configuration.
2. Cyclone design
There are a number of different forms of cyclone but
the reverse flow cyclone represented in Fig. 1 is the most
common design used in the industry. The cyclone consists of
four main parts: the inlet, the separation chamber, the dust
chamber and the vortex finder. Tangential inlets are preferred
for the separation of solid particles from gases [1]. In this
study, the numerical simulation deals with the standard case
of reverse flow cyclone with a tangential rectangular inlet.
Cyclone dimension used in this simulation are as shown in
Table 1.
3. Computational fluid dynamics approach
Fluent is a commercially available CFD code which
utilises the finite volume formulation to carry out coupled
or segregated calculations (with reference to the conserva-
tion of mass, momentum and energy equations). It is ideally
suited for incompressible to mildly compressible flows.
The conservation of mass, momentum and energy in fluid
flows are expressed in terms of non-linear partial differen-
tial equations which defy solution by analytical means. The
solution of these equations has been made possible by the
advent of powerful workstations, opening avenues towards
the calculation of complicated flow fields with relative ease.
For the turbulent flow in cyclones, the key to the success
of CFD lies with the accurate description of the turbulent
behaviour of the flow [8]. To model the swirling turbulent
Table 1
Cyclone geometry used in this simulations
Geometry a/D b/D
Stairmand high efficiency 0.5 0.2
Bohnet [2] 0.533 0.133
a Unit in meters.
models available in Fluent. These range from the standard
k
model to the more complicated Reynolds stress model
(RSM). The k
model involves the solution of transport
equations for the kinetic energy of turbulence and its dis-
sipation rate and the calculation of a turbulent contribution
to the viscosity at each computational cell. The standard
k
, RNG k
and realizable k
model was not optimized
[10,6]. Turbulence may be stabilised or destabilised in the
parts of flow domain where strong streamline curvature is
presence. However to reduce the computational effort the
RNG k
model can be used with about 12% deviation on
experimental data [8]. The numerical studies carried out by
PLAN Fredriksson [7] reveal that the RNG k
model under pre-
dicts the variation of the axial velocity profile across the
radial direction and also over predicts the magnitude of the
tangential velocity and the cyclone pressure drop.
The Reynolds stress model requires the solution of trans-
port equations for each of the Reynolds stress components
as well as for dissipation transport without the necessity to
calculate an isotropic turbulent viscosity field. The Reynolds
stress turbulence model yield an accurate prediction on swirl
flow pattern, axial velocity, tangential velocity and pressure
drop on cyclone simulation [7,6,13,10].
The finite volume methods have been used to discretised
the partial differential equations of the model using the Sim-
ple method for pressurevelocity coupling and the second
order upwind scheme to interpolate the variables on the sur-
face of the control volume. The segregated solution algo-
rithm was selected. The Reynolds stress turbulence model
was used in this model due to the anisotropic nature of
the turbulence in cyclones. Standard fluent wall functions
were applied and high order discretisation schemes were
also used.
Under the RSM second order upwind for discretisation
there is a difficulty to reach the convergence in simulation.
The residuals may exhibit cyclic tendencies which mean
that the transient pattern occurs. In this instance, the solver
must be changed to a transient solver and makes the time
step something in the region of 0.025 s or a tiny fraction of
the residence time of the cyclone. The simulation is then
solved with a coupling of unsteady and steady state solver
in Fluent. For the simulation using RNG k
model the
steady state solver is sufficient to reach the convergence. The
CFD simulation was performed with a Pentium IV 2.8 GHz
HP workstation XW8000 with 512 cache-memory, 1 GB
RAM-memory, and 110 GB hard disc memory.
De /D S/D h/D H/D B/D Da
0.5 0.5 1.5 4 0.375 0.305
0.333 0.733 0.693 2.58 0.333 0.15
 J. Gimbun et al. / Chemical Engineering and Processing 44 (2005) 712 9

4. Pressure drop empirical models

The pressure drop across the cyclone is an important pa-

rameter in the evaluation of cyclone performance. It is a
measure of the amount of work that is required to operate the
cyclone at given conditions, which is important for opera-
tional and economical reasons. The total pressure drop over
a cyclone consists of losses at the inlet, outlet and within
the cyclone body. The main part of the pressure drop, i.e.
about 80%, is considered to be pressure losses inside the cy-
clone due to the energy dissipation by the viscous stress of
the turbulent rotational flow [9]. The remaining 20% of the
pressure drop are caused by the contraction of the fluid flow
at the outlet, expansion at the inlet and by fluid friction on
the cyclone wall surface. Fig. 2. CFD surface mesh for (A) Stairmand high efficiency, and (B)
In this study, four empirical models in the literature have Bohnet [2] cyclone.
been chosen to predict the pressure drop over a cyclone,
namely Shepherd and Lapple [11], Casal and Martinez [3], grid as shown in Fig. 2. Several empirical correlation from
Dirgo [5], and Coker [4]. In these four models, the total literature, Shepherd and Lapple [11], Casal and Martinez [3],
pressure drop in cyclone is either assumed equal to the static Dirgo [5] and Coker [4], were also considered to compared
pressure drop or as a function of cyclone dimension and experimental data and numerical solution from Fluent code.
pressure drop coefficient. Generally cyclone pressure drop Figs. 3 and 5 present the comparison. The three-dimension
is proportional to the velocity head and can be written in the map of static pressure of Bohnet and Stairmand cyclones is
form of shown in Figs. 4 and 6, respectively.
g v2i
P = (1) 5.2. Pressure drop prediction under different
2
operating temperature
In the Shepherd and Lapple [11] model, is obtained by
assuming static pressure drop given as Measurement of the cyclone pressure drop of different op-
ab erating temperature was carried out for temperature ranging
= 16 (2) from 293 to 1123 K by Bohnet [2]. The comparison between
De2
the Bohnet experiment, empirical model and CFD prediction
In Casal and Martinez [3], is derived from the statistical is shown in Figs. 7 and 8. Fig. 9 shows the three-dimension
analysis on experimental data given as map of static pressure for operating temperature of 950 K.

2 The calculated static pressure drop of cyclone between
ab
= 11.3 + 3.33 (3) inlet and outlet for the different numerical model is shown
De2 in Figs. 3, 5, 7 and 8. It is shown that good agreement of
In Dirgo [5] model, is a function of cyclone dimension
given as 2500

 1/3
ab S/D
= 20 (4)
De2 (H/D)(h/D)(B/D) 2000

In Coker [4], is given as

Pressure Drop (Pa)

Dirgo CFD RNGk-

1500
ab CFD RSM
= 9.47 (5) Shepherd & Lapple
De2 1000
Coker

5. Result and discussion 500

Casal & Martinez

5.1. Pressure drop prediction under different inlet velocity 0
4 6 8 10 12 14 16
Measurement of the cyclone pressure drop was carried out Inlet gasvelocity (m/s)

for inlet velocity ranging from 4.62 to 14.16 m/s by Bohnet Fig. 3. Evolution of pressure drop with inlet velocity. Comparison between
[2], and from 5.1 to 25 m/s by Griffiths and Boysan [8]. data presented by Bohnet [2], the predictions of CFD and four empirical
The numerical calculation was made with a fine numerical models (P = 1 bar, T = 293 K, D = 150 mm, geometry Bohnet [2]).
10 J. Gimbun et al. / Chemical Engineering and Processing 44 (2005) 712

2500

Dirgo
2000 CFD RSM
Pressure Drop (Pa)

1500 Shepherd & Lapple

CFD RNGk-
1000
Coker
500

Casal & Martinez

0
0 200 400 600 800 1000 1200
Temperature (K)

Fig. 7. Evolution of pressure drop with operating temperature. Compar-

Fig. 4. Evolution of pressure drop with inlet velocity. Comparison between
data presented by Graffiths and Boysan [8], the predictions of CFD and
four empirical models (P = 1 bar, T = 293 K, D = 0.305 m, geometry
Stairmand high efficiency.
ison between data presented by Bohnet [2], the predictions of CFD and
four empirical models (Q = 100 m3 /h, T = 2931123 K, D = 150 mm,
geometry Bohnet [2].

1600
2500 CFD RSM
1400
Dirgo
CFD RNG k- CFD RSM
2000 1200
Pressure Drop (Pa)

1000
Pressure Drop (Pa)

Shepherd & Lapple

1500 800 CFD RNGk-
Shepherd & Lapple
600
1000 Coker
400

200
Casal & Martinez
500 Casal & Martinez
0
Dirgo
0 200 400 600 800 1000 1200
Coker
0 Temperature (K)
5 10 15 20 25
Velocity (m/s) Fig. 8. Evolution of pressure drop with operating temperature. Compar-
ison between data presented by Bohnet [2], the predictions of CFD and
Fig. 5. 2D and 3D map of static pressure of Bohnet [2] cyclone for inlet four empirical models (Q = 80 m3 /h, T = 2931123 K, D = 150 mm,
velocity of 4.62 m/s and temperature 293 K. geometry Bohnet [2].

Fig. 6. 3D map of static pressure of Stairmand cyclone for inlet velocity Fig. 9. 3D Map of static pressure of Bohnet [2] cyclone for inlet velocity
of 20 m/s and temperature 293 K. of 11.48 m/s and temperature 850 K.
 J. Gimbun et al. / Chemical Engineering and Processing 44 (2005) 712
the CFD numerical calculation when compared with experi-
mental data, and predictions from empirical correlation. The
results show that the CFD prediction by using the Fluent
code can be used for pressure drop evaluation in cyclone
design. This low-pressure centre can be responsible for the
flow reversion and deviation of the axial velocity peak to
the wall of the vortex finder pipe as showed in Figs. 4, 6
and 9.
The Fluent code with the RSM turbulence model, predict
very well the pressure drop in cyclones and can be used in
cyclone design for any operational conditions (Figs. 3, 5,
7 and 8). In the CFD numerical calculations a very small
pressure drop deviation were observed, with less than 3%
of deviation at different inlet velocity which probably in
the same magnitude of the experimental error. The CFD
simulations with RNG k
turbulence model still yield a
reasonably good prediction (Figs. 3, 5, 7 and 8) with the
deviation about 1420% of an experimental data. It consid-
erably tolerable since the RNG k
model is much less on
computational time required compared to the complicated
RSM turbulence model. In all cases of the simulation the
RNG k
model considerably underestimates the cyclone
pressure drop as revealed by Griffiths and Boysan [8].
However under extreme temperature (>850 K) there is no
significant difference between RNG k
and RSM model
prediction.
The cyclone pressure drop can be rewritten as a func-
tion of inlet velocity head. The empirical model used for
the prediction of pressure drop is much depends on the cy-
clone operating condition. Shepheard and Lapple [11] and
Dirgo (1990) model show a good prediction on cyclone pres-
sure drop under different operational inlet velocity (Figs. 3
and 4), the prediction within 620% of the measured value.
However, Dirgos model does not take into account temper-
ature in its model: its predictions are, therefore, not reliable
under different operating temperature (Figs. 7 and 8). Under
high temperature Dirgos model considerably overestimates
the cyclone pressure drop with relative error of more than
90%.
The pressure drop decreases significantly with rising
temperature. This effect is mainly due to the decrease of the
density and the increase of the viscosity of the gas. Accord-
ing to Figs. 7 and 8, the models of Shepheard and Lapple
give quite a good approximation of the pressure drop with
an error in the prediction of about 37%. The model of Casal
and Martinez, and Coker were under predicts the cyclone
pressure drop under different operating temperature with
relative error of 72 and 52%, respectively. Since Casal and
Martinez, and Coker models consistently underestimate the
cyclone pressure drop in all the conditions studied, they are
therefore not particularly useful for design purposes. It is
always more practical to design for a larger pressure drop
than for a smaller one. In overall, the cyclone pressure drop
is somewhat closer to Coker model for the operating tem-
perature more than 500 K and close to Shepherd and Lapple
model for the operating temperature lower than 500 K.
11
6. Conclusions
The CFD code FLUENT with the RSM turbulence model,
predict very well the pressure drop in cyclones and can be
used in cyclone design for any operating conditions. In the
CFD numerical calculations a very small pressure drop de-
viation were observed, with about 3% of deviation, probably
in the same magnitude of the experimental error. However
behind the accuracy of the complicated RSM model it does
require much expensive computational effort compared to
the RNG k
model. CFD with RNG k
turbulence model
still yield a reasonably good prediction on cyclone pressure
drop with deviation of 1418% on measured value.
The cyclone pressure drop can be rewritten as a function
of inlet velocity head. The model used for the prediction of
pressure drop depends on the cyclone operating condition.
Both Shepherd and Lapple, and Dirgo models show a good
prediction on cyclone pressure drop under different opera-
tional inlet velocity. However, Dirgos model is unable to
predict accurately the pressure drop under different oper-
ating temperature. For the various temperature conditions,
Shepherd and Lapples pressure drop model prediction is
the best. We therefore, conclude that the Shepherd and Lap-
ple model should be used for estimation of pressure drop
in cyclone design.
Acknowledgements
The authors would like to thank Dr. Tom Fraser, Fluent
India and Fluent Europe UK for their guidance and support.
The authors are grateful to the referees for their useful com-
ments.
Appendix A. Nomenclature
a cyclone inlet height (m)
b cyclone inlet width (m)
B cyclone dust outlet diameter (m)
D cyclone body diameter (m)
De cyclone gas outlet diameter (m)
h cyclone cylinder height (m)
H cyclone height (m)
P cyclone pressure drop (Pa)
S cyclone gas outlet duct length (m)
vi inlet velocity (m/s)
Greek letters
velocity head, pressure drop coefficient (m)
g gas density (kg/m3 )
References
[1] S. Altmeyer, V. Mathieu, S. Jullemier, P. Contal, N. Midoux, S. Rode,
J.-P. Leclerc, Comparison of different models of cyclone prediction
12 J. Gimbun et al. / Chemical Engineering and Processing 44 (2005) 712
performance for various operating conditions using a general soft-
ware, Chem. Eng. Process. 43 (2004) 511522.
[2] M. Bohnet, Influence of the gas temperature on the separation effi-
ciency of aerocyclones, Chem. Eng. Process. 34 (1995) 151156.
[3] J. Casal, J.M. Martinez, A better way to calculate cyclone pressure
drop, Chem. Eng. 90 (1983) 99.
[4] A.K. Coker, Understand cyclone design, Chem. Eng. Progr. 28 (1993)
5155.
[5] J. Dirgo, Relationships between cyclone dimensions and performance.
Doctoral Thesis, Havarad University, USA, 1988.
[6] T. Fraser, personal communication, 2003. fraserta1@cf.ac.uk, http://
www.cfd-online.com.
[7] C. Fredriksson, Exploratory experimental and theoretical studies of
cyclone gasification of wood powder. Doctoral Thesis, Lulea Uni-
versity of Technology, Sweden, 2003.
[8] W.D. Griffiths, F. Boysan, Computational fluid dynamics (CFD) and
empirical modelling of the performance of a number of cyclone
samplers, J. Aerosol Sci. 27 (1996) 281304.
[9] A. Ogawa, Separation of particles from air and gasses, vols. 1 and
2, CRC Press, Boca Raton, Florida, USA, 1984.
[10] M. Reddy, Fluent India, personal communication, 2003, mxr@
fluent.co.in.
[11] C.B. Shepherd, C.E. Lapple, Air pollution control: a design approach.
In: C.D Cooper, F.C. Alley (Eds.), Cyclones, second ed., Woveland
Press Inc., Illinois, 1939, pp. 127139.
[12] P.D. Silva, C. Briens, A. Bernis, Development of a new rapid method
to measure erosion rates in laboratory and pilot plant cyclones,
Powder Technol. 131 (2003) 111119.
[13] M. Slack, Cyclonic separator, QNETCFD application Challenge,
http://www.qnet-cfd.net, Accessed on 8th August 2003.
 Chemical Engineering and Processing 44 (2005) 1321

A pollution reduction methodology in reactor design

Qishi Chen, Xiao Feng
Department of Chemical Engineering, State Key Laboratory of Multi-Phase Flow in Power Engineering, Xian Jiaotong University, Xian 710049, China

Received 20 January 2003; received in revised form 18 March 2004; accepted 26 March 2004
Available online 18 May 2004

Abstract

An algorithm for waste and pollutants reduction in reactor design is presented in this paper. This algorithm use potential environmental
impact balance (PEI) and PEI rate-law expression to track the generated PEI throughout reaction process within a reactor; and study how
reaction conditions (temperature, pressure, concentration, etc.) and various engineering factors (heat and mass transfer, and back-mixing and
so on) affect process environmental performance. The form of PEI rate-law expression should be consistent with the method for calculating
the overall PEI of mixtures of chemicals inside reactors. The algorithm can be used as a tool to aid in designing chemical reactors with
environmentally friendliness processes. The use of the methodology is illustrated with the reaction system of allyl chloride production.
2004 Elsevier B.V. All rights reserved.

Keywords: Reactor design; Potential environmental impact; Chemical reaction engineering; Impact analysis

1. Introduction To address the ideal including environmental impact con-

There is currently a great deal of interest in the devel-
opment of methods that can be used to prevent or at least
reduce the generation of pollution, because it may result in
lower operating costs due to better utilization of raw mate-
rials and energy and reduced waste treatment and disposal
costs. Researches have shown that the most opportunities
of environmental impact minimization exist in the process
research and development and design stage of chemical pro-
cesses [1,2]. Waste and pollutant generation in many chemi-
cal manufacturing process can be traced back to the reaction
system, where reaction pathways that produce high yields
may also generate toxic byproducts or use environmentally
unfriendly raw materials, solvents, and catalystswhich
must be dealt with downstream at a significant cost [3];
and the size, configuration, and composition of the reactor
are very important factors controlling the greenness of a
process [4]. Therefore, the design of new reactor is a very
important element in preventing pollution at the source.
siderations into process design, Cabezas et al. [5] introduced
a potential environmental impact (PEI) balance as an amend-
ment of the Waste Reduction Algorithm [6]. However, this
algorithm is simply a tool to be used to aid in evaluating the
environmental friendliness of a process [7].
Making clear the effects of reaction conditions such
as temperature, pressure and concentration as well as en-
gineering factors involving back-mixing, heat and mass
transfer and so forth, on the performances of reaction pro-
cess is one of critical bases for benign reactor design. In
this paper, potential environmental impact balance and PEI
rate-law expression are proposed as a tool to be used to
effectively discover these effects on environmental perfor-
mance of a reaction process. With this methodology, the
PEI of waste and pollutants could be tracked throughout
reaction process within reactors, and appropriate reactor
type and reaction conditions, which could lead to envi-
ronmentally benign reaction processes, could be selected
expediently.

Abbreviations: CSTR, continuous stirred-tank reactor; EIM, environ- 2. The potential environmental impact of a
mental impact minimization; NP, non-products; PEI, potential environ-
mental impact; PFR, plug-flow reactor
chemical process
Corresponding author. Tel.: +86-29-82668980;

fax: +86-29-83237910. In the algorithm [59], the potential environmental im-

E-mail address: xfeng@mail.xjtu.edu.cn (X. Feng). pact of a certain quantity of a certain material and energy

0255-2701/$ see front matter 2004 Elsevier B.V. All rights reserved.
doi:10.1016/j.cep.2004.03.006
14 Q. Chen, X. Feng / Chemical Engineering and Processing 44 (2005) 1321

is defined as the effect that this material and energy would conditions and engineering factors and the like on the gen-
have on the environment if they were to be emitted into the eration of waste or pollutant is very important. However,
environment. For a steady-state process, the balance equa- studying each of wastes and pollutants within reactor is
tion is as follows: time-consuming and cost-consuming and the results may be
conflicting or competing, that is to say that reducing one pol-
Iin Iout + Igen = 0 (1)
lutant or waste may increasing the others. In addition, this
where Iin and Iout are the input and output rates of PEI, research method does not address the impact of the pollu-
respectively, and Igen denotes the PEI generation rate inside tion generated within a reactor. For example, reactor A may
the processes. produce 50 kg/h of pollutants while reactor B may produce

(in)
100 kg/h. However, the pollutants generated within reactor
Iin = mj xkj k + (2) A may be much more environmentally unfriendly than those
j k generated within reactor B. Thus, the strategies for reduc-

(out)

ing pollutants and wastes generated from reaction process
Iout = mj xkj k + (3) should be studied through their PEI rather than their mass
j k or concentrate, the results would be, then, scientific and be-
(in) (out) yond conflicting.
where mj is the mass flow rate of input stream j, mj Traditionally, the foundation of reactor design is the equa-
the mass flow rate of output stream j, xkj the mass fraction tions such as reaction rate equation, mass, energy and mo-
of chemical k in stream j, and k the overall potential envi- mentum balances as well as the model describing the fluid
ronmental impact (PEI) of chemical k. flow within reactors. With these equations, reactor designers


k = l kl
s
(4) can track compositions of chemicals, pressure and tempera-
l
ture throughout reaction process and discover the effects of
reaction conditions as well as engineering factors on these
where kls is the specific potential environmental impact of
parameters, and then, an appropriate reactor with profitable
chemical k for environmental impact category l, in units of processes can be designed. Nevertheless, this kind of design
PEI/kg of chemical k. Details of the method for calculat- method does not consider environmental perspective of pro-
ing kl
s appeared in reference [7]. The impact categories in-
cesses effectively.
volved in this theory fall into two general area concerning If the PEI generated within reactors could be tracked
with four categories in each area: throughout the process, the influence of reaction conditions
(1) The global atmospheric categories: and engineering factors on the generated PEI could also
Global warming potential (GWP). be clarified, and then reactors that generate the minimum
Ozone depletion potential (ODP). amount of PEI would be easily obtained. The above thinking
Acidification and acid-rain potential (AP). can be implemented by using the PEI balance and the PEI
Photochemical oxidation or smog formation potential rate-law expression in reactor design. In this way, the design
(PCOP). method is similar to the traditional reactor design method
based on balance equation and reaction rate equation.
(2) The local toxicological impact categories:
Human toxicity potential by ingestion (HTPI).
Human toxicity potential by either inhalation or der- 3.1. PEI rate-law expression
mal exposure (HTPE).
Aquatic toxicity potential (ATP). The PEI transformation rate of chemical k can be defined
Terrestrial toxicity potential (TTP). as:

In Eq. (4), l is a relative weighting factor for impact type dIk

rPEI,k = (5)
l independent of chemical k, and represent the value that V dt
society places on particular types of environmental impact
where Ik represents PEI of chemical k, Ik = mk k , mk
[8]. This factor allows Eq. (4) to be customized to specific
is the mass of substance k, k the overall PEI per unit
or local conditions. The suggested procedure is to initially
mass of chemical k in units of PEI/kg, and V the reaction
set all l to a value of one, and allow users to vary indi-
volume. According to Eq. (2), the PEI transformation rate
vidual l from 1 to 10 according to local needs and polices
of a reaction system rPEI , in units of PEI/(L s), is:
[5,8].



rPEI = rPEI,k = rk k Mk 103 (6)
k k
3. The algorithm
where Mk denotes molecular weight of chemical k in units
In order to perform waste reduction and pollution pre- of g/mol, and rk the rate of appearance of chemical k in units
vention in reactor design, studying the effects of reaction of mol/(L s).
 Q. Chen, X. Feng / Chemical Engineering and Processing 44 (2005) 1321
Such an expression, which shows how the rate depends on
the concentrations of reactants, is called PEI rate-law expres-
sion. Note, PEI rate-law expression should be consistent with
the method for calculating the overall PEI in Eqs. (2) and
(3), which ignore the combinational impacts that could be
associate with mixtures of chemicals which accounts for the
additional terms not included into those equations. It should
be realized that if the method for calculating the overall PEI
of mixtures in reactors improve or adopt other method, the
form of the PEI rate-law expression should be adjusted ac-
cordingly.
Obviously, rPEI can be expressed in terms of reaction tem-
perature and concentration or partial pressure of chemicals
and so on. Therefore, the PEI rate-law expression can be
helpful to analyze the effect of concentrations and temper-
ature on the PEI transformation rate, rPEI , and to study the
influence of back-mixing on process environmental perfor-
mance. At least, the PEI rate-law expression can reveal the
factors that control the transformation rate of PEI, so as to
provide guidance for the selection of reactor type and opera-
tion conditions, and the inner structure of the reactor, which
produce desired products while creating minimum undesired
potential environmental impact.
3.2. PEI balance
Since desired products are the substance we desire and the
objective of this paper is to present a methodology of waste
reduction, that is the primary concern is reducing the impact
and the amount of the non-products, the k of the desired
products is not considered in PEI balance. This insures that
the user or producer is not directly penalized for producing
a chemical that has a high PEI value. Auxiliary materials
(catalyst, solvent and so on) used in reaction processes must
leave the production process completely as waste or emis-
sion [10], so the PEI of the auxiliary materials should be
considered. Raw materials that if possible shall be entirely
processed into the desired product are not as a rule com-
pletely converted into the product. The losses that arise are
the cause of generated waste and emission [10], thus their
potential environmental impact should be taken into account.
To sum up, the PEI balance of reaction system under steady
state is:
Igen
NP
= Iout
NP
Iin
NP
(7)
where the superscript NP represents materials except de-
sired products, that is, non-products. Iout
NP and I NP are input
in
and output rates of PEI caused by non-products, and Igen NP
the rate at which impact is generated in a reactor.
For a differential flow reactor, dV, we write a PEI balance
equation over the reactor:
dIgen = rPEI dV (8)
With PEI balance Eq. (7), combining with other equations,
designers can reveal how PEI change with reaction time or
axial and radial direction inside reactors, i.e. can track the
15
changing of PEI throughout reactors; using Eq. (8) combin-
ing with other rate equations, the influence of operation pa-
rameters such as temperature, concentration, pressure and
reaction time, etc. on environmental metrics of a chemical
process could be analyzed facilely, and then, reactors gen-
erating a minimum of PEI could be designed easily.
4. Case study
In this case study, we use the environmental impact indices
below, which can be derived from Eq. (1), to characterize
environmental performance of reaction processes [7,8]:
(1) Potential environmental impact generation rate by non-
products Igen
NP .
(2) Specific potential environmental impact due to non-
products PEI .
Igen
NP
PEI =  (9)
p pp
where pp is the mass flow rate of product p, and PEI has
units of potential environmental impact per mass of prod-
ucts. Obviously, the smaller the values of Igen NP and
PEI ,
the more environmental friendly the process, and all others
factors such as economics being equal, the more desirable.
Igen
NP would be useful in comparing different designs on an
absolute basis, while PEI would be useful in comparing
different designs independently of plant size [8].
These indices characterize some aspects of the genera-
tion of PEI within a reaction process. They are very useful
in addressing questions related to the ability of the reactor
to produce desired products while creating a minimum of
undesired potential environmental impact.
To illustrate the use of the algorithm, a case of allyl chlo-
ride manufacturing was used. Allyl chloride is manufactured
by means of non-catalytic chlorination of propylene. Con-
sider the following reaction system.
The principle reaction is:
Cl2 + CH2 =CHCH3 CH2 =CHCH2 Cl + HCl (10)
The main secondary reaction is further chlorination of allyl
chloride to 1,3-dichloropropene:
Cl2 + CH2 =CHCH2 Cl ClCH=CHCH2 Cl + HCl (11)
Another important side-reaction is the additive chlorination
of propene to 1,2-dichloropropane:
Cl2 + CH2 =CHCH3 CH2 ClCHClCH3 (12)
This reaction system is a kind of complex van de vusse
reaction, a typical reaction process involving consecutive
and parallel reactions. It is, therefore, sufficiently complex
to illustrate the algorithm. Due to the limit to the space of
the article, we only select temperature, concentration and
back-mixing to be the object of research to illustrate the use
of the algorithm.
16 Q. Chen, X. Feng / Chemical Engineering and Processing 44 (2005) 1321

The following calculations and curves drawing can be ac- respectively.The differential PEI balance of PFR reactors
complished by mathematical software such as Mathcad2000 is:
or Matlab.
dIgen = rPEI dV (19)
4.1. Preconditions Dividing Eq. (18) by Eq. (17), then integrate, we obtain:

The kinetics of the reaction system [11] are: b 1a a

(CC )PFR = (CB CB,0 CB ),
a1
r 1 = k1 C A C B (13)
k2 k1
a= , b= (20)
r2 = k2 CA CC (14) k1 + k 3 k1 + k 3
r3 = k3 CA CB (15) Dividing Eq. (19) by Eq. (17), and then integrate this, we
obtain:
where r1 , r2 , and r3 represent the rates of the reactions  CB
rPEI
in reaction Eqs. (10)(12), respectively, expressed in (Igen )PFR = v0 dCB (21)
terms of mol/(L s), and k1 = 1.5 106 e66271/RT , k2 = CB,0 rB
4.4 108 e99410/RT and k3 = 100 e33140/RT . Letters AF
represent Cl2 , propylene, allyl chloride, hydrogen chloride, 4.2.2. Continuous stirred-tank reactors (CSTR)
1,3-dichloropropene and 1,2-dichloropropane, respectively. For CSTR reactors, the mass and PEI balances are:
Since the purpose of this case study is only to illustrate v0 CC = rC V (22)
the use of PEI balance and PEI rate-law expression, not
a practical reactor design, we might as well suppose the v0 (CB,0 CB ) = rB V (23)
reaction process is isovolumetric.
For all cases in this case study, the inlet concentrations Igen = rPEI V (24)
of all chemicals are CA,0 = CB,0 = 0.024 mol/L, CC,0 = From Eqs. (23) and (24), we obtain:
CD,0 = CE,0 = CF,0 = 0 mol/L. The volumetric flow rate rPEI
of the reaction feed at the reactor inlet is v0 = 325 L/s. Igen = (CB CB,0 )v0 (25)
rB
The specific potential environmental impact values for
each impact category for the chemicals involved in the pro- From Eqs. (22) and (23), we obtain:
cess are shown in Table 1. Details of the method for calcu-    
rC CC CC
lating these data appear in reference [7]. Using these data, = =a b
the overall PEI of each chemical can be obtained according rB CSTR CB CB,0 CB
to Eq. (4). Solving this equation, we can get:

4.2. Mathematical models of ideal reactors bCB (CB,0 CB )

(CC )CSTR = (26)
a(CB,0 CB ) + CB
4.2.1. Plug-flow reactors (PFR)
The differential mass balances of PFR reactors are: 4.3. The effect of back-mixing on environmental
v0 dCA = rA dV (16) performance

v0 dCB = rB dV (17) It is well known that PFR represents reactors without

v0 dCC = rC dV (18) back-mixing, CSTR represents reactors with the maximum
back-mixing, we can therefore study effect of back-mixing
where rA = r1 + r2 + r3 , rB = r1 + r3 , and rC = r1 r2 . on process environmental aspect by studying these two kinds
rA , rB and rC are the rate of appearance of A, B and C, of reactor.

Table 1
The specific potential environmental impact values for each category for the chemicals used in this case study [9]
Compound HTPI HTPE ATP TTP GWP PCOP AP

A 0 5.4 104 22 0 0 0 0
B 0 0 3.1 102 0 0 2.1 0
C 0.51 5.4 104 0.10 0.51 0 0 0
D 0.78 2.3 104 4.6 104 0.78 0 0 0.86
E 1.1 0 0.014 1.1 0 0 0
F 0.61 0 0.0059 0.61 0 0 0
Values are in PEI/kg.
 Q. Chen, X. Feng / Chemical Engineering and Processing 44 (2005) 1321 17

4.3.1. The effect on Igen

4.3.1.1. l = 1. When l = 1, namely, the importance of

the above eight environmental impact categories is deemed
the same, and the PEI of each non-product is considered,
the PEI rate-law expression due to non-products is:

rPEI = 1.6k1 CA CB 1.5k3 CA CB 1.2k2 CA CC (27)

rPEI has units of PEI/(L s). The value of rPEI is negative be-
cause the overall PEI of the reactant A is far greater than that
of each reaction products, in addition, the PEI of product C Fig. 1. Curves of rPEI /rB vs. CB , rPEI /rB in units of PEI/mol. The solid
curve is for PFR, the dashed for CSTR.
is not considered. Suppose reaction temperature T = 530 K,
and at the outlet of reactor, CB,out = 0.007 mol/L. Substi-
4.3.1.2. Not considering PEI of A. To illustrate the use of
tute Eqs. (20) and (26) into rPEI /rB , respectively, and then
the algorithm effectively, suppose A = 0 PEI/kg, and the
graph them in Fig. 1, respectively.
weighting factor for each impact type is likewise 1, and the
rPEI 1.6k1 + 1.5k3 + 1.2k2 (CC /CB ) PEI of each non-product except Cl2 is also considered. Note
= (28)
rB k1 + k 3 that this case is probably impractical in real manufacturing
processes. The PEI rate-law expression, rPEI , in units of
According to Eq. (25), Igen of CSTR is: PEI/(L s), is:

Igen = [AABCD + 1.5(0.024 0.007)]v0 rPEI = 1.1 103 k1 CA CB + 0.049k3 CA CB

+ 0.33k2 CA CC (29)
where AABCD is the area of rectangle ABCD in Fig. 1. Ac-
cording to Eq. (21), Igen of PFR is: rPEI 1.1 103 k1 0.049k3 0.33k2 (CC /CB )
= (30)
  rB k1 + k 3
Igen = AA
BCD
+ 1.5(0.024 0.007) v0
Using the same method as mentioned previously, we can
where AA
BCD
is the area of curved trapezoid A
BCD
(the obtain the curves describing the changing of rPEI /rB of
shaded region) in Fig. 1. Obviously, when the outlet concen- PFR and CSTR with CB at various reaction temperature, as
tration CB,out is identical for both reactors, AABCD is greater shown in Figs. 2 and 3, respectively.
than AA
BCD
, thus, lower Igen will be achieved by using According to Eq. (21), if the inlet and outlet concentration
CSTR. The calculation results show that, when CB,out = of B are 0.024 and 0.007 mol/L, respectively, Igen of PFR
0.007 mol/L, and v0 = 325 L/s, Igen of CSTR is 3.7 equals the area of the region below the curves in the range
104 PEI/h, and Igen of PFR is 3.3 104 PEI/h. Apparently, of CB = 0.0070.024 mol/L in Fig. 2 multiplied by v0 . For
Igen of CSTR is about 12% lower than that of PFR. example, when reaction temperature T = 450 K, Igen of

Fig. 2. Curves of rPEI /rB vs. CB of PFR ignoring the overall PEI of A. rPEI /rB in units of PEI/mol. The numbers 17 denote the reaction temperatures
of 450, 470, 490, 500, 510, 520 and 530 K, respectively.
18 Q. Chen, X. Feng / Chemical Engineering and Processing 44 (2005) 1321
Fig. 3. Curves of rPEI /rB vs. CB of CSTR ignoring the overall PEI of A. rPEI /rB in units of PEI/mol. The numbers 17 denote the reaction temperatures
of 450, 470, 490, 500, 510, 520 and 530 K, respectively.
PFR equals the area of the shaded region multiplied by v0 .
From Eq. (25), Igen of CSTR is the product of v0 and the
area of the corresponding rectangle in Fig. 3. For example,
when reaction temperature is 450 K, Igen of CSTR equals
the area of the rectangle in Fig. 3 multiplied by v0 . In such
case, the value of Igen is positive number. Comparing Fig. 2
with Fig. 3, it can be found that lower Igen will be obtained if
we adopt PFR and lower reaction temperature. For example,
when reaction temperature is 500 K, and v0 is 325 L/s, Igen
of PFR is 4.8 101 PEI/h, while for CSTR, Igen = 1.1
103 PEI/h, which is 2.3 times that of PFR.
4.3.2. The effect on PEI
From Eqs. (9), (21) and (25), we obtain the formulas for
calculating PEI . Suppose v0 is 325 L/s, then for PFR:

Igen 13.0 CB rPEI
PEI = = dCB (31)
v0 CC,out MC 103 CC,out CB,0 rB
for CSTR:
Igen 13.0 rPEI
PEI = = (CB CB,0 )
v0 CC,out MC 103 CC,out rB
(32)
where CC,out is the outlet concentration of C.
According to Eqs. (31) and (32), using the curves about
rPEI /rB versus CB and corresponding area, the superior-
ity in environmental index PEI about PFR and CSTR
can be discovered. When l = 1, the outlet concentration
CC,out = 0.011 mol/L and reaction temperature T = 500 K,
using CSTR will obtain lower PEI than using PFR. In
this case, for CSTR, PEI = 31 PEI/kg; for PFR, PEI =
26 PEI/kg. Nevertheless, if not considering A , for the
same CC,out and T, using PFR will obtain lower PEI
than using CSTR, for PFR, PEI = 0.30 PEI/kg, while for
CSTR, PEI = 0.80 PEI/kg, which is 2.6 times that of PFR.
4.4. The effect of temperature on environmental
performance
4.4.1. The effect on Igen
When l = 1, from Eq. (27), it can be seen easily that the
value of rPEI decreases (its absolute value increase) as the
reaction temperature rises, that is, Igen can be lowered by
increasing reaction temperature. For example, when reaction
temperature increases from 500 to 600 K, and the outlet
concentration CB,out = 0.007 mol/K, and v0 is also 325 L/s,
Igen of PFR and CSTR would decrease from 3.2104 and
3.5104 to 3.7104 and 4.2104 PEI/h, respectively.
The effect of reaction temperature on Igen can also be found
from Figs. 2 and 3.
When not considering the PEI of Cl2 , from the PEI
rate-law expression, expressed in Eq. (29), we can find that
in the initial stage of the reaction, the determining part
in the right side of Eq. (29) is 1.1 103 k1 CA CB +
0.049k3 CA CB , due to the concentration of C is very low
while the concentrations of A and B are relatively high.
Comparing the magnitudes of the two activation energies
in the two terms, we conclude that the value of this part
decreases as the reaction temperature increases. Thus, in
the early stage of this reaction process, to lower the value
of rPEI , the reactor should operate at relatively high tem-
perature. As the reaction proceeds, the concentration of C
is increasing while CA and CB is decreasing, so that the
influences of the third term, 0.33k2 CA CC , is becoming sig-
nificant. Since the magnitude of activation energy in this
term is the maximum, operating at relatively low temper-
ature would result in a relatively low value of rPEI in this
stage. Accordingly, for the reaction system in this case, a
PFR operating at higher temperature first and then lower
temperature must generate lower value of Igen than that sim-
ply operating at a corresponding uniform temperature. The
 Q. Chen, X. Feng / Chemical Engineering and Processing 44 (2005) 1321
curves in different temperature in Fig. 2 intersect each other,
this fact also indicate this result. For example, when the inlet
and outlet concentrations of B are 0.024 and 0.007 mol/L,
respectively, and v0 is 325 L/s, for PFR operating at reaction
temperature of 450K, Igen = 3.9 101 PEI/h, and for PFR
at 530 K, Igen = 6.7 101 PEI/h. While for a two-stage
PFR, in which the first stage operates at 530 K, and the
second stage operates at 450 K, the inlet concentration of B
of the second stage is 0.018 mol/L, Igen = 3.4 101 PEI/h.
Obviously, it is 13% lower than that of the PFR simply
operating at reaction temperature of 450 K, and 49% lower
than that of the PFR simply operating at 530 K.
4.4.2. The effect on PEI
Known from Eqs. (31) and (32), rPEI /rB can be used to
study the effect of the reaction temperature. When l = 1,
it is not difficult to find that increasing reaction temperature
would decrease the values of rPEI /rB , accordingly the value
of PEI also decreases. For example, for CSTR operating at
500 K, if its outlet concentration CC,out = 0.011 mol/L, and
v0 is 325 L/s, the value of PEI is 31 PEI/kg, while for
CSTR operating at 530 K and with the same CC,out , the value
of PEI is 36 PEI/kg, that is, this CSTR reduced the value
of PEI by 16%. Similar to this example, for the case of not
considering the potential environmental impact of Cl2 , one
can also conclude the effect of temperature on PEI .
4.5. The effect of concentration on environmental
performance
We here only study the effect of concentration on PEI .
Similar to previous method in Section 4.4.2, one can also use
rPEI /rB to study the effect of concentration. From Eqs. (31)
and (32), it can be found that for a particular outlet concen-
tration of C the value of PEI would be decreased by in-
creasing the value of rPEI /rB . When l = 1, from Eq. (28),
the value of rPEI /rB can be increased by augmenting CC and
decreasing CB within a reactor. Since C is a product whereas
B is a reactant, using CSTR will lead to higher CC and lower
CB within a reactor than using PFR. Therefore, in this case,
using CSTR would result in lower value of PEI than using
PFR operating at the same reaction conditions. When not
considering the potential environmental impact of Cl2 , from
Eq. (30), decreasing CC and increasing CB within a reactor
will increase the value of rPEI /rB , accordingly, using PFR
would result in lower PEI than using CSTR. These results
are also concluded in Section 4.3.2.
5. Discussion
In the case study, the indexes involved are Igen and PEI .
However, these two indexes only express the situation of
PEI generation within processes. The environmental perfor-
mance of a process also relates to its emission of PEI. There-
fore, in the WAR algorithm, the output rate of PEI, Iout , and
19
the total PEI per mass of product streams leaving the system,
Iout , are another category of indices to characterize some
aspects of emission of PEI from a manufacturing process.
According to Eq. (1):
Iout = Iin + Igen (33)
Iout is obtained by dividing Iout by the output of products to
give:
Iout Iin + Igen
Iout =  =  (34)
p p p p
Substituting Eq. (9) into Eq. (34), we obtain:
Iin
Iout =  + PEI (35)
p p
For the case study, it is not difficult to obtain that Iin =
4.6 104 PEI/h, i.e. Iin is a constant. Therefore, when the
feed condition of a reaction process is fixed, back-mixing
and reactiontemperature affect Igen and Iout in the same
way. Since p p = v0 CC,out MC 103 , it is not difficult
to find that, when the feed condition and CC,out are spec-
ified, back-mixing and reaction temperature influence Iout
and PEI in the same way, too.
Thus, in Section 4.3, when l = 1 and the reaction pro-
cess is operated in the specified conditions, the results that
Igen and PEI of CSTR are less than those of PFR indicate
that using CSTR can obtain lower Iout and Iout than us-
ing PFR. That means at the given conditions, back-mixing
make the emission of potential environmental impact de-
crease. By contrast, when not considering the PEI of A, the
results are converse, i.e. using PFR would gain lower PEI
generation rate than using CSTR; and the value of PEI of
CSTR that operate at reaction temperature 500 K and out-
let concentration CC,out = 0.011 mol/L is 2.6 times that of
PFR. The results indicate that at the given reaction condi-
tions, back-mixing will increase the emission of potential
environmental impact of the reaction processes, accordingly,
reactor designers should try to avoid it in such case.
In Section 4.4, when l = 1, the results show that a higher
reaction temperature is favorable for the decrease of Igen
and PEI in the giving conditions. Therefore, increasing the
reaction temperature will lead to the decrease of Iout and
Iout . When not considering the PEI of Cl2 , PFR operating
at a high temperature in the first section and low tempera-
ture in the remainder section can result in lower generation
and emission of PEI than that operating at a corresponding
uniform temperature.
To study the effects of concentration on environmental
performance of reaction processes, two analyses are intro-
duced in Section 4.5. The results show that when l = 1,
operating at relatively low concentration of B and high
concentration of C is favorable for decreasing the poten-
tial environmental impact created in manufacturing a unit
mass of the product C. This also is true for total PEI leav-
ing the system per mass of product streams according to
20 Q. Chen, X. Feng / Chemical Engineering and Processing 44 (2005) 1321

the relations about these two categories of indexes. By
contrast, when neglecting the PEI of Cl2 , the results are
converse.
In the cases study, we have only studied the effects of
back-mixing on the environmental performance. Whereas
the phenomena of back-mixing is only one of engineering
factors, which still involve pre-mixing, heat and mass trans-
fer, thermostability and so on. However, the effects of those
engineering factors not involved in the case study on envi-
ronmental performance can also be simulated and therefore
well studied through a combination of reactor mathematical
model, which relates to corresponding engineering factors,
with PEI balances. As for the effects of other reaction con-
ditions not involved in the case study, such as space time,
pressure, compositions of feeds and so on, which may af-
fect the environmental performance of reaction processes
significantly in some cases, can also be studied with the
help of corresponding PEI balances and the PEI rate-law
expressions, because we can ultimately simulate the change
of the environmental indexes with these reaction conditions
through PEI balance and PEI rate-law expression.
The reaction system involved in the case studied is a kind
of relatively complex van de vusse reaction, nevertheless the
reaction system in real manufacturing process may involve
more reaction types and, therefore, is more complex than
that one. One can, however, simulate the change of envi-
ronmental indexes within a reactor by combining traditional
reactor mathematical model with the PEI balance, and may
also discover the effects of reaction conditions and engineer-
ing factors on environmental performance by PEI rate-law
expression and/or combinations it with other reaction rate
equations as well as other related equations in reactor math-
ematical models.
Note that assuming l = 1 and not considering the PEI of
Cl2 are just for illustrating the algorithm, these assumptions
may be inappropriate for a real reactor design. Other users of
this methodology may choose to weight effects differently
to reflect local needs and polices where the plants locate [8].
In the case study, the quantitative environmental impact
assessment and the PEI indexes are cited from the WAR al-
gorithm, it is worth noting that, however, quantitative impact
assessment and environmental indexes are likely to improve
with time, and that these improvements can be incorporate
into this methodology as they become available. It is suggest
that either the WAR algorithms assessment method and the
PEI indexes, another similar method and indexes, or even a
more sophisticated and comprehensive method and indexes
be used to this pollution reduction methodology.
In the case study, the calculations of the impact indexes
should be accurate to no more than one significant figure
[12]. But two significant figures are used in order to allow
readers to reproduce the calculations if necessary. It is also
worth noting that, even being accurate to one significant
figure, one can still discern significant differences in the
index between different cases.
6. Conclusions
To perform pollution prevention in the process of reac-
tor design, studying the effects of reaction conditions such
as temperature, pressure and concentration as well as vari-
ous engineering factors on environmental performance of a
reaction process is very important. In this paper, potential
environmental impact balance and PEI rate-law expression
are proposed as a tool to be used to carry out this work. The
form of PEI rate-law expression should be consistent with
the method for calculating the overall PEI of mixtures of
chemicals inside reactors. The effectiveness of the method
has been demonstrated through the reaction of allyl chlo-
ride production. From the processes of the analyses in the
case study, analyzing with PEI balance and PEI rate-law ex-
pression combining with other corresponding equations and
some mathematic knowledge can give important insights
into how reaction conditions as well as engineering factors
affect the environmental performance of reaction processes.
This kind of study is one of the bases for developing chem-
ical reactors with environmental benign processes.
Acknowledgements
Financial support provided by the National Natural Sci-
ence Foundation of China under Grant No. 20176405 and
the Major State Basic Research Development Program un-
der Grant No. G2000026307 is gratefully acknowledged.

Appendix A. Nomenclature

AF chlorine, propylene, allyl chloride, hydrogen chloride, 1,3-dichloropropene and

1,2-dichloropropane, respectively
CA , CB , etc. concentration of chemicals A, B, etc. (mol/L)
CA,0 , CB,0 , etc. initial concentration of chemicals A, B, etc. (mol/L)
CB,out , CC,out , etc. outlet concentration of chemicals B and C, etc. (mol/L)
dV, dC, dIgen differential V, C and Igen
Igen potential environmental impact generation rate (PEI/h)
Igen
NP potential environmental impact generation rate of non-products (PEI/h)
 Q. Chen, X. Feng / Chemical Engineering and Processing 44 (2005) 1321 21

Iin potential environmental impact input rate (PEI/h)

Iin
NP potential environmental impact input rate of non-products (PEI/h)
Ik potential environmental impact of chemical k (PEI)
Iout potential environmental impact output rate (PEI/h)
Iout
NP potential environmental impact output rate of non-products (PEI/h)
(in)
mj mass flow rate of stream j into a process (kg/h)
(out)
mj mass flow rate of stream j out of a process (kg/h)
mk mass of chemical k
mk,0 mass of chemical k at initial state of reaction
mk, out mass of chemical k at reactor exit or at the moment of reaction termination
mp mass flow rate of desired product p (kg/h)
Mk molecular weight of chemical k (g/mol)
rA , rB , etc. the rate of appearance of chemicals A, B, etc. (mol/(L s))
rPEI total PEI transformation rate (PEI/(L s))
rPEI,k PEI transformation rate of chemical k (PEI/(L s))
S area of a certain figure
T reaction temperature (K)
v0 volume flow rate entering a reactor (L/s)
xkj mass fraction of chemical k in stream j

Greek letters
l weighting factor for impact category l
PEI generated PEI per unit mass of all desired products (PEI/kg)
kl
s the specific PEI of chemical k for environmental impact category l (PEI/kg of chemical k)
k overall PEI of chemical k (PEI/kg of chemical k)

References
[1] H.T. Kohlbrand, From waste treatment to pollution prevention and
beyond: opportunities for the next 20 years, AIChE Symp. Ser.
94 (320) (1998) 117121.
[2] A.P. Rossister, Waste Minimization Through Design, McGraw-Hill,
New York, 1995.
[3] J.A. Dyer, K.L. Mulholland, Prevent pollution via better reactor
design and operation, Chem. Eng. Prog. 94 (2) (1998) 6166.
[4] A.D. Curzons, D.J.C. Constable, D.N. Mortimer, So you think your
process is green, how do you known? Using principles of sustain-
ability to determine what is greena corporate perspective, Green
Chem. 3 (2001) 16.
[5] H. Cabezas, J.C. Bare, S.K. Mallick, Pollution prevention with chem-
ical process simulator: the generalized waste reduction (WAR) algo-
rithm, Comput. Chem. Eng. 21 (1997) S305S310.
[6] A.K. Hilaly, S.K. Sikdar, Pollution balance: a new methodology for
minimizing waste production in manufacturing process, J. Air Waste
Manage. Assoc. 44 (1994) 13031308.
[7] D.M. Young, H. Cabezas, Designing sustainable process with simu-
lation: the waste reduction (WAR) algorithm, Comput. Chem. Eng.
23 (1999) 14771491.
[8] H. Cabezas, J.C. Bare, S.K. Mallick, Pollution prevention with
chemical process simulator: the generalized waste reduction (WAR)
algorithmfull version, Comput. Chem. Eng. 23 (1999) 623
634.
[9] D.M. Young, R. Scharp, H. Cabezas, The waste reduction (WAR)
algorithm: environmental impacts, energy consumption, and engi-
neering economics, Waste Manage. 20 (2000) 605615.
[10] W. Schramm, New findings on the generation of waste and emissions,
and a modified cleaner production assessment approachillustrated
by leather production, J. Cleaner Prod. 5 (4) (1997) 291300.
[11] B. Pahor, Z. Kravanja, N.I. Bedenik, Synthesis of reactor networks
in overall process flowsheets within the multilevel MINLP approach,
Comput. Chem. Eng. 25 (2001) 765774.
[12] S. Mallick, H. Cabezas, J.C. Bare, et al., A pollution reduction
methodology for chemical process simulator, Ind. Eng. Chem. Res.
35 (11) (1996) 41284138.
 Chemical Engineering and Processing 44 (2005) 2332

Effects of hydraulic residence time on metal uptake by activated sludge

Tlay A. zbelge , H. nder zbelge, Murat Tursun
Chemical Engineering Department, Middle East Technical University, 06531 Ankara, Turkey

Received 28 April 2003; received in revised form 12 April 2004; accepted 13 April 2004
Available online 17 June 2004

Abstract

The combined uptake of Cu2+ and Zn2+ by activated sludge (biomass) was investigated at steady state in an activated sludge process (ASP)
without recycle (namely, once-thro ASP), for different values of influent metal concentrations and hydraulic residence time in the range of
2.540 h. The experiments were performed at a constant pH of 7 and temperature of 25 C. The results showed that the percentage removal of
both copper and zinc by activated sludge increased with the increasing residence time; moreover, the percent metal uptakes were higher for
both metals at the low level of influent metal concentrations [(Cu2+ ) = 1.5 mg/L and (Zn2+ ) = 9 mg/L], than those at the higher level of
concentrations [(Cu2+ ) = 4.5 mg/L and (Zn2+ ) = 27 mg/L]. The least represented forms of both Cu2+ and Zn2+ in activated sludge were
their sulfides, and the dominant metal forms were found to be organically bound for both of the metals, Zn2+ and Cu2+ .
2004 Elsevier B.V. All rights reserved.

Keywords: Biosorption; Activated sludge process; Removal of heavy metals; Sequential extraction of metals; Metal uptake

1. Introduction It was observed that the toxic effects of HMs decreased as

Industrial discharges containing heavy metals (HMs) from
mining, metal refining and other industries are hazardous for
public health unless the HMs are removed at the site of the
waste production by appropriate methods [1]. One of these
techniques for removing the HMs from wastewaters is to
use the well-known activated sludge process (ASP) [2]. The
studies on the treatment of these wastes biologically are not
complete yet, since some metals are toxic to microorganisms
(MOs) even at very low concentrations and their inhibition
effects are not exactly known.
In giving a literature survey on this subject, the following
are the important points to be noted: (i) the form of HM in
ASP affects the degree of its toxicity, the soluble forms be-
ing the most toxic [2]; (ii) different oxidation states of the
same metal can have different binding mechanisms to the
sludge [3]; and (iii) it is known that the dissolved oxygen
concentration in the medium has an important effect on the
oxidation states of the metals present in the system [4]. The
other factors affecting the forms of HMs are the solution pH
and the presence of complexing agencies in the medium [5].
Corresponding author. Tel.: +90 312 210 2621;
fax: +90 312 210 1264.
E-mail address: tozbelge@metu.edu.tr (T.A. zbelge).
the concentrations of complexing agencies increased [5,6],
which was explained with the decrease in free metal ions
due to their reactions with complexing agents. Toxicity char-
acteristics of a certain metal can be greatly modified by
the other HMs because they compete for available organic
ligands in sludge and wastewater [7]. On the other hand,
non-interactive effects among some metals may also occur
in ASP [8]. The diversity of MOs in ASPs may vary from
plant to plant. Individual species of MOs present may even
change within the same system from time to time. There-
fore, responses of different systems, or even the same sys-
tem at different times and operating conditions, to HMs are
unlikely to be identical, and the response mainly depends
on the composition of activated sludge [4].
Complexation, adsorption and precipitation reactions in-
volving HMs are all strongly influenced by hydrogen ion ac-
tivity. The effects of pH value on the degree of dissociation
of potential ligands, competition with metal ions for uptake
and metal precipitation have been investigated [3,9]. Low
pH values tend to reduce the degree of dissociation, which
resulted in decreased adsorption and complexation. At pH
= 7, ligands such as carboxyl and phosphate were favored
for cation uptake, because pKa values were considerably less
than 7. When pH was increased to a level at which other com-
plexing anions could no longer successfully compete with

0255-2701/$ see front matter 2004 Elsevier B.V. All rights reserved.
doi:10.1016/j.cep.2004.04.004
24 T.A. zbelge et al. / Chemical Engineering and Processing 44 (2005) 2332
the hydroxide ions present, metals precipitated as hydrox-
ides. Furthermore, increase in metal inhibition with decrease
in pH was attributed to the increase in free metal ions in the
medium [10,11]. Toxicities of HMs are also affected by cell
residence time in ASP. Nelson et al. [12] reported that toxic
effects of cadmium and zinc decreased considerably by in-
creasing cell residence time from 1 to 15 days in ASP with
recycle. Uloth and Mavinic [13] explained this decrease with
the increasing production of bacterial extracellular polymer
substances (EPS) at high mean cell residence times. Some
investigators [9,14,15] reported that an increase in substrate
and mixed liquor suspended solids (MLSS) concentrations
led to increase in metal uptake and decrease in adverse ef-
fects of HMs to ASP; they attributed this decrease to the
decreasing proportion of active cellular components, which
were tied up as metalligand complexes [15]. Combined ef-
fects of copper, zinc and nickel on the treatment efficiency
of ASP and the combined uptakes of these metals by acti-
vated sludge have been reported [16], as have the effects of
copper and zinc on activated sludge bacteria/growth kinet-
ics [17,18]. However, combined uptakes of only copper and
zinc by sludge, and the effect of hydraulic residence time in
this process have not been reported, although Vian et al. [16]
discussed the effects of long sludge age (kept almost con-
stant at 9 days) on the combined uptakes of copper, nickel
and zinc in an ASP with recycle. In another study [19],
transfer of organic matter from wastewater (WW) to acti-
vated sludge was investigated and observed that the transfer
of soluble fraction always occurred from wastewater to ac-
tivated sludge, whereas bidirectional transfer occurred for
the colloidal fraction. The biosorption capacity of activated
sludge was found to be at a ratio of chemical oxygen de-
mand (COD)/TSS of about 10400 mg/L, and biosorbable
fraction of wastewater was almost 45% of non-settleable
(colloidal + soluble) fraction [19]. A single metal (Zn2+ )
study was reported [20], where zinc biosorption by activated
sludge was more effective than by digested sludge in remov-
ing zinc from metal plating effluent. The objective of this
work is to investigate the effect of hydraulic residence time
on combined uptake of Cu2+ and Zn2+ by activated sludge
in once-thro ASP (OTASP) without recycle, which is not
encountered in the available literature.
2. Experimental
2.1. Experimental set-up
A modular continuously stirred glass tank reactor (from
Cole-Parmer, Inc., USA) with an effective volume of 2 L was
used in the OTASP. Peristaltic pumps were used at the inlet
and at the exit of the reactor for feed/harvest, respectively.
Aeration and efficient mixing was provided in the reactor;
temperature and pH were controlled at 251 C and 70.01,
respectively, by electronic controllers. A synthetic wastew-
ater with known inlet substrate concentration (in terms of
Table 1
Composition of synthetic wastewater
Constituents (mg/L)
Proteose-peptone 1221.7
NaCl 407.4
Na2 SO4 44.6
K2 HPO4 44.6
MgCl2 .6H2 O 3.7
FeCl2 .2H2 O 3.7
CaCl2 .2H2 O 3.7
MnSO4 0.057
H2 MoO4 0.031
NaOH 0.008
ZnSO4 0.046
CoSO4 0.049
CuSO4 0.076
chemical oxygen demand) of a COD of So = 1300 mg/L
was fed, in which Proteose-peptone (1221.7 mg/L) was the
sole carbon and nitrogen source. Other minor nutrients were
also added in sufficient quantities to make carbon the limit-
ing nutrient (Table 1). Solutions of CuSO4 and ZnSO4 were
added to the wastewater to provide Cu2+ and Zn2+ ions at
the desired concentrations. Further details are given else-
where [17,2123].
2.2. Experimental method
Biomass obtained from the wastewater treatment plant
of Middle East Technical University (METU) was used as
seed in the experiments. In the acclimation of MOs to HMs,
maximum concentrations of Cu2+ and Zn2+ reached dur-
ing acclimation were decided from previous work [17,21].
The literature showed that when these metals were present
individually in the system, ASP failed to operate at steady
state values of 15 mg/L for Cu2+ and 90 mg/L for Zn2+ , re-
spectively [5]. Therefore, one-third of threshold concentra-
tions of these metals (that is, 5 mg/L for Cu2+ , and 30 mg/L
for Zn2+ ) were used to constitute the combined threshold
concentration for mixtures of copper and zinc ions. Accli-
mation experiments were performed by gradually increasing
the concentrations of Cu2+ and Zn2+ to two, three and four
times their combined threshold concentration. The optimum
condition for a successful acclimation providing the growth
of healthy MOs, as observed from the increase in turbidity
values, was obtained at the concentrations corresponding to
twice their combined threshold concentration [21].
Experiments without HMs were conducted using syn-
thetic WW inoculated with 10% (v/v) acclimated seed MOs
to twice their combined threshold concentration (that is,
10 mg/L Cu2+ , and 60 mg/L Zn2+ ). The reactor was op-
erated as a batch system for 48 h with the growth of new
biomass. Then, continuous feeding of synthetic WW to the
reactor at a desired volumetric flow rate (or desired resi-
dence time) was started. The system reaches steady state
in about 1244 days depending on the influent metal con-
centrations and the hydraulic residence time (HRT); effluent
 T.A. zbelge et al. / Chemical Engineering and Processing 44 (2005) 2332
samples from the reactor were analyzed for substrate, S, and
biomass, X, concentrations. MLSS concentration was used
as the primary indicator for steady state conditions. At each
residence time, steady state was reached when the effluent
S (COD) and X (MLSS) concentrations remained constant
with respect to time.
Experiments with HMs were performed at two concentra-
tion levels of each metal (low levels: (Cu2+ ) = 1.5 mg/L
and (Zn2+ ) = 9 mg/L; high levels: (Cu2+ ) = 4.5 mg/L
and (Zn2+ ) = 27 mg/L, for different values of HRT in the
range of 240 h using the same synthetic WW having the
composition given in Table 1. In these experiments, 1.8 L
of synthetic WW was inoculated with 0.2 L of acclimated
seed MOs in the activated sludge reactor, and the system
was operated in batch mode for almost 72 h to ensure the
growth of biomass by observing the increase in measured
turbidities of the samples from the reactor. Then, the con-
tinuous operation was started by feeding the reactor with
synthetic WW containing copper and zinc ions at the de-
sired inlet concentrations. Influent flow rate was adjusted
in order to provide a required HRT. Samples were taken
from the effluent stream at different times in order to deter-
mine the effluent S and X concentrations until steady state
was reached. Then, a sample of 1.5 L from effluent stream
was collected for the metal analyses, and the operation was
stopped.
2.3. Analytical methods
Substrate concentrations were determined in centrifuged
sample aliquots by standard COD analysis, and gravimet-
ric method was used to measure the biomass concentrations
[24]. Analyses were repeated at least twice with the sam-
ples under the same experimental conditions and the average
values were obtained.
2.3.1. Distribution of metals among the constituents of
activated sludge mixed liquor
Since metals were distributed between the two phases of
mixed liquor in the reactor, namely solution and biomass, the
amounts of copper and zinc associated with these phases and
the forms of these metals in the biomass were determined
at steady state condition for each HRT.
Firstly, the concentrations of HMs in effluent were de-
termined. At least three samples of 40 mL were acidified
with 10 mL of concentrated HNO3 to decompose MOs.
These samples were diluted to the sensitivity range of the
atomic absorption spectrophotometer (AAS) (Model: Philips
9200X-AAS) (diluted 5 times for Cu2+ analysis, and 25
times for Zn2+ analysis) and then analyzed. The average of
these measurements yielded the total HM concentration in
the effluent, CT (mg/L), for that specific experimental run.
To convert (mg metal/L) to (mg metal/kg sludge), the mea-
sured metal concentration (as, mg metal/L) at each HRT
was divided by the steady state MLSS concentration (as, kg
TSS/L).
25
Secondly, concentrations of the HMs associated with the
solution, and solid phases of the effluent were determined.
Two of 50 mL aliquots were collected from the effluent of
the reactor, centrifuged at 1300 rpm for 15 min, by using
a RC5C type Dupont centrifuge, to separate the solution
and solid phases, and the supernatants transferred to sam-
ple tubes. Then, supernatants and the residual solid phase
were acidified by adding 10 mL concentrated HNO3 to each
sample tube. Samples were then analyzed by AAS after per-
forming the necessary dilutions to the sensitivity range of
the instrument as explained above. The average of the results
from the two samples were recorded as CS (metal concen-
tration in the solution phase), and CA (metal concentration
in the solid phase) values for each HM.
The distribution of different metal forms in the sludge
was determined by using the sequential extraction pro-
cedure [25]. Activated sludge mixed liquor of 1.5 L,
corresponding to 0.2500.100 g of sludge, was collected
from the outlet of the reactor at the steady state. In order
to separate the sludge and supernatant, centrifugation at
1300 rpm for 15 min was carried out using a RC5C type
Dupont centrifuge. Then the extraction process was started
by adding the reagents of KNO3 , KF, Na4 P2 O7 , EDTA
(ethylene-diamine-tetraacetic-acid) and HNO3 on the con-
centrated sludge (order of addition given in Table 2), which
are capable of extracting exchangeable, adsorbed, organi-
cally bound, carbonates, and sulfides forms of the metals
in the sludge sample, respectively. To provide complete
extraction, samples were shaken on a platform shaker, then
centrifuged at 1300 rpm for 15 min, and supernatants were
decanted into separate tubes. Between each extraction, the
sludge residue was washed with 50 mL of de-ionized water
by shaking for 30 min, and the samples centrifuged again
under the same conditions. Both supernatants from the ex-
traction and washing were combined and acidified with two
drops of 5 M HNO3 before metal analysis, in order to pre-
serve the solutions. All metal concentrations were measured
as before and conversions were made to obtain the concen-
trations of each form of each metal as mg metal/ kg sludge,
for which the metal concentration as mg/L was divided by
the dried MLSS concentration (kg TSS/L), determined at
the steady state. The same procedure was followed for the
other reagents, the only exception being that double extrac-
tion was performed with EDTA and supernatants pooled
before the metal analyses. The remaining solid residue in
Table 2
Details about sequential extraction process parameters [25]
Reagents Molarity pH Extraction (Solution/solids)
(M) time (h) ratio
KNO3 1 24 50/1
KF 0.5 6.5 24 80/1
Na4 P2 O7 0.1 24 80/1
EDTA 0.1 6.5 224 80/1
HNO3 1 24 50/1
26 T.A. zbelge et al. / Chemical Engineering and Processing 44 (2005) 2332

the sample tube, at the end of the extraction procedure, was Table 6
decomposed with 5 mL of concentrated HNO3 in order to Material balance on zinc at the influent concentration of (Zn2+ )
= 27 mg/L
determine the levels of metals not extracted by the protocol.
h (h) CT CS CA [(CS + CA ) CT ]
100/CT
3. Results and discussion 40 29.20 29.14 0.72 +2.26
20 27.13 26.23 0.65 0.92
10 29.69 28.75 0.63 +1.04
Two phases exist in the activated sludge reactor; the liq- 5 28.38 27.68 0.52 0.63
uid phase consists of water, soluble ions and soluble sub- 2.5 28.88 28.52 0.44 +0.28
strate while the solid phase (biomass or sludge) has settleable
waste materials, dead and alive MOs in it. Metals associated
with these phases were investigated in this study. The ma- being (Cu2+ ) = 4.5 mg/L, (Zn2+ ) = 27 mg/L. It is seen
terial balance for the metals at the steady state is defined as in these tables that the total metal concentration in the ef-
follows: fluent is greater than that in the influent to the reactor at the
CT = CA + CS (1) steady state due to the experimental error occurred in the
determination of these concentrations in very small amounts
where CT is the total HM concentration in effluent (mg/L), of sludges produced. In spite of this, the obtained results are
CA is the HM concentration taken up by the sludge (mg/L) consistent as indicated by the small differences between the
and CS is the heavy metal concentration in the solution two sides of the material balance at each run, the maximum
(mg/L). This balance was applied to each run using HMs. In percent experimental error being within 5%.
Tables 36, material balance results for copper and zinc are Fig. 1 shows the percent metal removals. Small percentage
presented for two sets of experiments (in the first set, influ- removals (up to 10%) are because only small amounts of
ent HM concentrations being (Cu2+ ) = 1.5 mg/L, (Zn2+ ) biomass were produced in the system under the experimental
= 9 mg/L, and in the second set, influent HM concentrations conditions used. This is expected in a system where toxic

Table 3
Material balance on copper at the influent concentration of (Cu2+ ) 12
= 1.5 mg/L
h (h) CT CS CA [(CS + CA ) CT ]
100/CT
10
17.18 1.47 1.38 0.15 +1.04 Cu: 1.5 mg/L
14.68 1.61 1.42 0.12 4.35 Cu: 4.5 mg/L
9.22 1.63 1.60 0.09 +3.68 Zn: 9 mg/L
5.12 1.63 1.58 0.06 +0.61 Zn: 27 mg/L
3.75 1.43 1.43 0.02 +1.40 8
percent metal removal

Table 4
Material balance on zinc at the influent concentration of (Zn2+ ) = 9 mg/L
6
h (h) CT CS CA [(CS + CA ) CT ]
100/CT
17.18 11.13 10.32 0.54 2.43
14.68 10.57 10.70 0.55 +6.43 4
9.22 10.68 10.70 0.44 +4.31
5.12 9.88 9.09 0.39 4.05
3.75 9.79 9.83 0.29 +3.37

2
Table 5
Material balance on copper at the influent concentration of (Cu2+ )
= 4.5 mg/L
h (h) CT CS CA [(CS + CA ) CT ] 0
100/CT 0 10 20 30 40
40 5.24 4.88 0.22 2.67 residence time (h)
20 4.68 4.53 0.18 +0.64
Fig. 1. Percent removal of each metal [(CA /CT ) 100] in the ex-
10 4.27 4.23 0.13 +1.02
periments of Set-1 (influent concentrations: (Cu2+ ) = 1.5 mg/L,
5 4.20 4.19 0.08 +1.67
(Zn2+ ) = 9 mg/L) and Set-2 (influent concentrations:
2.5 4.58 4.57 0.04 +0.66
(Cu2+ ) = 4.5 mg/L, (Zn2+ ) = 27 mg/L).
 T.A. zbelge et al. / Chemical Engineering and Processing 44 (2005) 2332 27

metals inhibit the microbial community to yield low MLSS 7000

values at the steady state and thus low removals of HMs
[9,1215]. Since the highest value of HRT is 40 h (less than
2 days) in this work, which is equal to the sludge age in 6000
the present OTASP, therefore, the biomass concentrations
amount of metal removed (mg metal / kg sludge)

Cu: 1.5 mg/L

in the effluent stream were low as expected. However, in Cu: 4.5 mg/L
an ASP with recycle, it may be possible to obtain higher 5000 Zn: 9 mg/L
percent removals of copper and zinc by increasing the sludge Zn: 27 mg/L

age up to 812 days [1216] due to the higher biomass

concentrations in the reactor. As seen in Fig. 1, the percent 4000
copper and zinc removals by biomass increased with the
increasing HRT or the decreasing dilution rate (D = 1/h ),
3000
keeping the inlet HM concentrations constant. At the low
level of influent HM concentrations ((Cu2+ ) = 1.5 mg/L,
(Zn2+ ) = 9 mg/L), the percent metal uptakes were higher
2000
than those at the higher level, ((Cu2+ ) = 4.5 mg/L, (Zn2+ )
= 27 mg/L), for both of the metals at each residence time,
due to the lower HM toxicities to biomass at the low level
1000
of the HM concentrations. The higher percent removals of
zinc compared to those of copper at the short residence
times of h <5 h can be attributed to the lower toxicity of 0
zinc in spite of the lower biomass concentrations at these 0 10 20 30 40 50
residence times. Although the toxicity of copper is greater residence time (h)
than that of zinc at the same level of concentration, the
percent copper removals exceeded those of zinc at the longer Fig. 2. Amount of metal bound to the sludge as (mg metal bound/kg
sludge produced) for the experiments of Set-1 and Set-2.
residence times of h >5 h; the reasons for this finding is
because of the lower influent copper concentrations studied
in this experimental work compared to those of zinc, and the
sludge decreases, since the available sites on the enzymes
occurrence of higher biomass concentrations at the longer
to be occupied by the metals are limited. It was reported
residence times which has an effect in decreasing the HM
earlier that both nickel and zinc uptakes decreased when the
toxicities. This may also be due to the greater contribution
inlet concentration of copper in the mixture was increased, at
of the precipitated form of copper to the sludge produced,
constant inlet concentrations of nickel and zinc [16]. It was
as reported by Vian et al. [16].
also reported by Vian et al. [16] that the percentage of zinc
When the amount of HM removed per kg of activated
removal was higher than that of copper due to the increasing
sludge produced (mg metal/kg sludge) for each metal is ob-
percentage of zinc removed by precipitation although the
served in Fig. 2 that these values are higher for Zn2+ be-
percent of zinc uptaken by sludge was lower than that of
cause of the higher influent concentrations of Zn2+ in the
copper for the higher toxicity effect of zinc at its higher
experiments; (mg metal/kg sludge) values for zinc decrease
concentrations. Although these data [16] seem to support
with the increasing residence time due to the increase in the
the above hypothesis, they did not discuss the effect of HRT
amount of sludge produced at the higher HRTs. This de-
on metal uptake by activated sludge in an ASP with recycle
creasing effect on the ratio of masses will be more signifi-
where the sludge age was kept almost constant. Therefore,
cant for zinc than that of copper, since the percent removals
it is necessary to investigate this hypothesis further.
are lower for zinc than those of copper as seen in Fig. 1; but,
still these values are above those of copper due to the higher
influent concentrations of zinc ions studied at both the low 3.1. Distributions of different metal forms in the sludge
and high concentration levels of the HMs in the experiments
compared to those of copper. According to the sequential extraction procedure [25] out-
The increase in the amount of copper and decrease in the lined earlier, the forms of HMs in the sludge are defined as
amount of zinc removed by activated sludge, as the residence exchangeable (extracted by KNO3 ), adsorbed (by KF), or-
time is increased, can be explained by transport processes, ganically bound (by Na4 P2 O7 ), carbonates (by EDTA), and
adsorption onto surfaces or the possible competition between sulfides (by HNO3 ). The results of these experiments are
copper and zinc ions in binding to available sites on the given in Figs. 36. Rankings of the different forms of copper
enzymes. It may be that the interaction between copper and and zinc in the sludge for the two sets of experiments (first
biomass increases as the residence time increases eventually set: influent (Cu2+ ) = 1.5 mg/L, (Zn2+ ) = 9 mg/L; sec-
leading to an increase in the amount of copper taken up ond set: influent (Cu2+ ) = 4.5 mg/L, (Zn2+ ) = 27 mg/L
by activated sludge. As a result, the affinity of zinc for the suggest that,
28 T.A. zbelge et al. / Chemical Engineering and Processing 44 (2005) 2332

Exchangeable 180
800 Adsorbed
Exchangeable
Organically bound Adsorbed
Carbonates 160 Organically bound
700 Sulfides Carbonates
140 Sulfides
mg metal removed / kg sludge

600
120
mg metal removed / kg sludge

500
100

400 80

300 60

40
200

20
100
0
0 2 4 6 8 10 12 14 16 18
2 4 6 8 10 12 14 16 18 residence time (h)
residence time (h)
Fig. 5. Distribution of Cu2+ among its forms in the sludge using the data
Fig. 3. Distribution of Zn2+ among its forms in the sludge using the data of Set-1 experiments.
of Set-1 experiments.

in Fig. 3, for (Zn2+ ) = 9 mg/L: organically bound >

carbonates > exchangeable > adsorbed > sulfides,
in Fig. 4, for (Zn2+ ) = 27 mg/L: organically bound
4000 = exchangeable = adsorbed > carbonates > sulfides,
Exchangeable in Fig. 5, for (Cu2+ ) = 1.5 mg/L: carbonates > organi-
Adsorbed
cally bound > exchangeable > adsorbed > sulfides,
Organically bound
Carbonates in Fig. 6, for (Cu2+ ) = 4.5 mg/L: organically bound >
Sulfides exchangeable > carbonates > adsorbed > sulfides.

3000 As shown in Figs. 36, at low concentration levels of both

metals, namely for (Cu2+ ) = 1.5 mg/L and for (Zn2+ )
mg metal removed / kg sludge

= 9 mg/L, the rankings of metal forms remained unchanged

as the HRT increased. On the other hand, at the high
concentration levels of these metals, namely for (Cu2+ )
2000 = 4.5 mg/L and for (Zn2+ ) = 27 mg/L, the rankings of
the different forms of these metals changed with the HRT;
the least represented forms for both are the sulfides, and the
dominant metal forms are the organically bound forms for
both zinc and copper.
The rankings of the different forms are more marked at
1000
low influent metal concentrations than those at the high level.
However, the dominant and the least represented forms are
the same. It has been reported that the dominant forms of
HMs could vary for different sludges [25]. Therefore, the
rankings of the metal forms for copper and zinc, should not
0 be assumed to be the same for all kinds of sludges.
0 10 20 30 40 To determine the efficiency of the metal extraction pro-
residence time (h) cedure, the sample remaining at the end of the extraction
Fig. 4. Distribution of Zn2+ among its forms in the sludge using the data was acidified with HNO3 to determine the amounts of resid-
of Set-2 experiments. ual metals not extracted by the reagents. Ideally, the total
 T.A. zbelge et al. / Chemical Engineering and Processing 44 (2005) 2332
400
Exchangeable
Adsorbed
Organically bound
Carbonates
Sulfides
300
mg metal removed / kg sludge
200
100
0 tion equilibria. The simultaneous sorption of metal cation
0 10 20
residence time (h)
Fig. 6. Distribution of Cu2+ among its forms in the sludge using the data
of Set-2 experiments.
amount of the metals determined by the extraction proce-
dure should be equal to the amount of metal bound to the
sludge, the discrepancy between the two being denoted by
CH . Thus, the material balance for extraction can be shown
as:
CA = CE + CH
where CA is the concentration of metal bound to the sludge
(mg/L, or mg metal/kg sludge), CE is the concentration
of metal extracted in the analysis (mg/L or mg metal/kg
sludge) and CH is the concentration of metal not extracted
by the used reagents in the extraction (mg/L or mg metal/kg
sludge). Thus, the extraction procedure used is capable of
extracting most of the metals bound to the biomass, varying
between 93.85 and 98.98% of the total amount of the HMs
in the sludge. Also, the total concentrations of the metals
extracted are mostly consistent with the CA values. The ex-
perimental error in the sequential extraction analysis falls
within 10%.
It should be noted that these data were obtained by ex-
tracting the relatively small amounts of biomass, which can
exagerate the experimental error. The reasons for working
with small amounts of biomass are two-fold; firstly, the sam-
ples were collected here at the steady state directly from the
effluent of the system at one day intervals. Samples can de-
generate due to the continuing interaction between metals
and biomass before the analysis, which can cause changes
in the relative amounts of the metal forms in sludge over
the time. Secondly, biomass concentrations in the effluent of
the reactor are quite low anyhow due to metal inhibition ef-
29
fects, a problem which can be overcome if a recycle is used
in the system. Therefore, further studies are planned to be
performed in an ASP with recycle.
In the literature, equilibrium biosorption of Zn2+ and
Cu2+ by nonviable activated sludge was investigated in a
packed column adsorber [26], where metabolically induced
interactions between HMs and MOs were eliminated; thus,
the problems, such as toxicity of metals to MOs, mainte-
nance requirements of MOs in controlling the pH within a
range suitable for the MOs were avoided. It was observed
by Utgikar et al. [26] that equilibrium metal uptakes from
solutions containing single metal ion were 2.5 and 3.4 mg/g
dried biomass for Zn2+ , and 1.9 and 5.9 mg/g dried biomass
for Cu2+ at pH of 3.0 and 3.8, respectively. Equilibrium up-
take of metals was modeled using the Langmuir adsorption
isotherm. Equilibrium uptakes from binary mixtures were
30% lower than single component solution uptakes for both
metals, indicating some competition between the two metals.
No hysteresis was detected between adsorption and desorp-
30 40 and sulfate anion was probably occurring at pH of 3.0, while
proton exchange predominated at pH of 3.8. They pointed
out the usefulness of non-viable activated sludge as a biosor-
bent for removing metal ions from WWs.
On the other hand, the removal of metal ions in an ASP
was affected adversely due to the toxicity of HMs to MOs.
It was reported that the presence of Cu2+ ions in wastewater
inhibited the net specific growth rate of autotrophic biomass
to a larger extent than Zn2+ ions [27]. Therefore, in an ASP
treating such a WW, it is expected that the concentration
and the composition of EPS (consisting of protein, polysac-
(2) charide and DNA) produced by microbial cells will vary
[28], since it is known that the EPS formation rate is pro-
portional to the rate of substrate utilization, thus to the rate
of biomass production [29]. When EPS are not involved in
the process, MOs may not easily display high flocculation
and metal uptake capacity, as the floc forming bacteria can
interact with each other [30]. However, the precise role of
EPS in biosolidliquid separation is not well understood al-
though EPS are thought to be of considerable importance in
bioflocculation and settling of activated sludge [28]. For ex-
ample, the EPS content has been shown to be directly [31],
or conversely [32] related to sludge volume index (SVI).
Besides, the low protein levels were obtained at high car-
bohydrate concentrations depending on microbial consortia
due to the intracellular storage of excess carbon substrate and
lower sludge retention time, particularly, in the pulppaper
and textile industries lab-scale activated sludge systems.
The low protein content was probably due to the inhibition
of EPS production in the aforementioned industies as re-
ported by Dignac et al. [33]. It was noted that the wastewater
composition rather than operating conditions had a strong
influence on EPS composition and the physicochemical
properties of flocs in an ASP. For example, if the ASP was
operated at high F/M ratios, both carbohydrate and DNA
content of EPS increased as the protein content decreased;
30 T.A. zbelge et al. / Chemical Engineering and Processing 44 (2005) 2332
as a result of this, a more hydrophobic (larger contact angle)
and less negatively charged floc surface was observed [28].
The present OTASP was operated at high F/M ratios [23];
although no data were obtained about the concentration and
the composition of EPS produced in this work, it might be
predicted from the available literature that the protein con-
tent of the produced EPS in the flocs was low; thus the set-
tling ability of the resulting sludge, most probably with a
high value of SVI was not good. Actually, low settlability of
the produced sludge was observed in the experiments. The
effect of operational conditions on EPS composition will be
the subject of a future study in which it will be investigated
in an ASP with recycle to compare the effects of sludge
residence time (SRT), HRT, EPS and F/M ratio on the set-
tlability of produced sludge and metal uptake. In this study,
the change in bacterial profile of the MLSS with metal con-
centration was not followed, but it was reported in the lit-
erature [34] that many different species of bacteria isolated
from activated sludge were shown to produce EPS. These
polymers might be in the form of loose slime or capsules
adhered to the cell wall. In activated sludge, slime polymers
remained in the dissolved and colloidal phases of an efflu-
ent, while capsular polymers remained attached to the flocs
and hence settled with the sludges. Some activated sludge
bacteria did not produce capsules or extracellular material.
In these bacteria, metals might be accumulated either in the
cytoplasm or by adsorption on to the cell wall. Some re-
searchers [35] concluded that only the half of the observed
metal uptake could be attributed to EPS and that cell itself
played a role in accumulating metals. In another work [36],
EPS were extracted, cells were broken up using ultrasound
and cell envelopes were sedimented by centrifugation. Only
small quantities of copper were found in the cytoplasm, but
substantial amounts of copper were bound to cell envelopes
and EPS. The metabolism of elements occurred in many
species of bacteria. It was also reported that four major types
of metal transformation in organisms occurred [37], these
being chelate formation by the binding of metals to organic
ligands, shifts in metal valencies, substitution of one metal
for another and biomethylation of metals to microorganisms.
When metals were chelated to substrate molecules, the ad-
sorption of these molecules might represent a means of accu-
mulating metals within a cell. These metabolic transforma-
tions of metal ions by bacteria were different for each metal
and for different combinations of metals in the medium.
Therefore, in such a dynamic system as is ASP, it is not
plausible to use metal uptake as a measure of EPS content.
As it is seen in Fig. 2, the copper uptake increases with the
HRT up to 20 h then it shows a slight decrease between 20
and 40 h. This can be explained with the change in the given
ranking in Fig. 6. The chemical speciation and distribution
of copper in an activated sludge slurry can be quantified
using an equlibrium constant approach including complex-
ation with inorganic and organic ligands and adsorption by
activated sludge solids [38]. It can be observed in Fig. 6 that
the amount of copper carbonate per kg of sludge is higher
than the amount of organically removed copper at a HRT of
40 h, while it was lower than that of organically bound cop-
per at a HRT of 20 h. Therefore, it is shown in Fig. 2 that
the copper uptake at a HRT of 40 h is less than that at 20 h.
The copper uptake is probably less at a HRT of 40 h due to
the effect of changing metabolic activity of microbial flocs
and the solubility constant of copper carbonate complex in
the medium at the prevailing conditions. The desorption of
copper into the solution might have occurred due to the sol-
ubility constant of this complex.
4. Conclusions
The effect of hydraulic residence time (in the range of
2.540 h) on the removal of Zn2+ and Cu2+ ions by acti-
vated sludge was investigated in a once-thro activated sludge
process (OTASP) at a constant pH of 7 and temperature of
25 C. It was found that the percentage removals of both cop-
per and zinc by activated sludge increased with the increas-
ing residence time due to the increasing amounts of sludge
produced with time. Higher percent removal of zinc at the
low level of influent concentrations of both the copper and
zinc, (Cu2+ ) = 1.5 mg/L and for (Zn2+ ) = 9 mg/L, due
to the higher inlet concentration of zinc over that of copper
was replaced by the higher percent removal of copper at the
high level of influent concentrations of each metal, (Cu2+ )
= 4.5 mg/L and for (Zn2+ ) = 27 mg/L, although copper is
known to be more toxic to the microorganisms than zinc.
This suggested that the lower toxicity of zinc at its low con-
centration level is compensated by another factor such as
the removal of both metals by precipitation also contribut-
ing to the removal process. However, it was not possible
to differentiate between the percent removals of each metal
by biosorption or by precipitation in OTASP. The previous
study [16] cannot render this information due to the presence
of a third metal ion such as Ni2+ in the system. Therefore, a
further study is required in an activated sludge system with
recycle which will differentiate the percent removals of each
metal by precipitation and biosorption clearly. In this study,
at short residence times (a range of 2.55 h), the percent-
age of zinc bound to the biomass is higher than for copper
due to the dominating effect of lower toxicity of zinc; this
is also shown by the sequential extraction results yielding
the dominant form of both Zn2+ and Cu2+ as organically
bound indicating that both types of ions are removed by
biosorption rather than by precipitation due to the lower de-
gree of inhibition of MOs at small sludge ages or HRTs in
OTASP. This point can be clarified by investigation in an
activated sludge process with recycle. It is also found that
the least represented forms of both the copper and zinc ions
in the sludge are their sulfides. A further study is under-
way to investigate the effect of operational conditions on the
composition of extracellular polymer substances in an ASP
with recycle to compare the effects of sludge residence time,
hydraulic residence time, EPS and food to microorganism
 T.A. zbelge et al. / Chemical Engineering and Processing 44 (2005) 2332 31

(F/M) ratio on the settlability of produced sludge and the [2] S. Sujarittanonta, J.H. Sherrard, Activated sludge nickel toxicity
mechanism of metal uptake. studies, J. WPCF 53 (1981) 13141322.
[3] W. Stumm, H. Bilinski, Trace metals in natural waters: difficulties
of interpretation arising from our ignorance on their speciation,
Proceedings of the Sixth International Conference, Advances in Water
Acknowledgements Pollution Research, Jerusalem, 1972, pp. 3949.
[4] G. Tchobanoglous, F.L. Burton, Wastewater Engineering: Treatment,
The financial support by TBITAK (Scientific and Tech- Disposal, Reuse, third ed., Metcalf and Eddy, McGraw-Hill, New
nical Research Council of Turkey) and DPT (State Plan- York, 1991.
ning Organization) for the Project (No. KTCAG-DPT-8) is [5] C.P.C. Poon, K.H. Bhayani, Metal toxicity to sewage organisms, J.
greatly appreciated. Sanitary Eng. Div., ASCE 97 (1971) 161169.
[6] J. Gardiner, Complexation of trace metals by EDTA in natural waters,
The invaluable comments of the referees are highly ac- Water Res. 10 (1976) 507517.
knowledged in improving the quality of this contribution [7] Y.P. Ting, F. Lawson, I.G. Prince, Uptake of cadmium and zinc by
and for encouraging the authors to continue a more detailed the Alga Chlorella vulgaris: multi-ion situation, Biotechnol. Bioeng.
experimental study, which will clarify the role of extracel- 39 (1991) 445455.
lular polymer substances in the mechanism of metal uptake [8] E.F. Barth, M.B. Ettinger, B.V. Salotto, G.N. McDermott, Summary
report on the effects of heavy metals on the biological treatment
by activated sludge process. processes, J. WPCF 37 (1965) 8696.
[9] M.H. Cheng, J.W. Patterson, R.A. Minear, Heavy metal uptake by
activated sludge, J. WPCF 47 (1975) 362368.
Appendix A. Nomenclature
[10] J.E. Loveless, H.A. Painter, The influence of metal concentrations
and pH value on the growth of Nitrosomonas strain isolated from
CA concentration of metal bound to the sludge activated sludge, J. Gen. Microbiol. 114 (1968) 144.
(mg/L) [11] F. Braam, A. Klapwijk, Effects of copper on nitrification in activated
CE concentration of metal extracted in the sludge, Water Res. 15 (1981) 10931098.
[12] P.O. Nelson, A.K. Chung, M.C. Hudson, Factors affecting the fate
analysis (mg/L)
of heavy metals in the activated sludge process, J. WPCF 53 (1981)
CH concentration of metal unaccounted for by 13231333.
extract (mg/L) [13] V.C. Uloth, D.S. Mavinic, Aerobic biotreatment of a high strength
CS metal concentration in the solution phase (mg/L) leachate, J. Environ. Eng. Div., ASCE 105 (1979) 647661.
CT total metal concentration in the effluent (mg/L) [14] M.M. Bagby, J.H. Sherrard, Combined effects of cadmium and nickel
COD chemical oxygen demand (mg O2 /L) on the activated sludge process, J. WPCF 53 (1981) 16091619.
[15] R.D. Tyagi, D. Couillard, F.T. Tran, Analysis of final settling tank in
D dilution rate (h1 ) relation to control of metal inhibition in the activated sludge process,
EPS extracellular polymer substances Can. J. Chem. Eng. 69 (1991) 534543.
HM heavy metal (plural: HMs) [16] A. Vian, F. Mirada, J.J. Rodriguez, Copper (2+), zinc (2+) and
HRT hydraulic residence time (h) nickel (2+) uptake by activated sludge, Sep. Sci. Technol. 20 (1985)
587597.
Ka solubility product constant
[17] N.Y. Beyenal, T.A. zbelge (Baser), H.. zbelge, Combined effects
MLSS mixed liquor suspended solids (mg/L) of Cu2+ and Zn2+ on activated sludge process, Water Res. 31 (1997)
MO microorganism (plural: MOs) 699704.
OTASP once-thro activated sludge process [18] A. Cabrero, S. Fernandez, F. Mirada, J. Garcia, Effects of copper
(ASP without recycle) and zinc on the activated sludge bacteria growth kinetics, Water Res.
32 (1998) 13551362.
So substrate concentration in the influent
[19] A. Guellil, F. Thomas, J.C. Block, Transfer of organic matter between
(mg COD/L) wastewater and activated sludge flocs, Water Res. 35 (2001) 143150.
S substrate concentration in the effluent [20] F. Bux, B. Atkinson, H.C. Kasan, Zinc biosorption by waste activated
(mg COD/L) and digested sludge, Water Sci. Technol. 32 (1999) 127130.
SVI sludge volume index [21] N. Yurt, Combined effects of heavy metals on activated sludge
TSS total suspended solid concentration process, M.S. Thesis, Department of Chemical Engineering, Middle
East Technical University, Ankara, 1994.
(mg/L or kg/L) [22] P. Altinten, Combined effects of copper and zinc on the activated
WW wastewater (plural: WWs) sludge process, M.S. Thesis, Department of Chemical Engineering,
X biomass concentration in the effluent (mg/L) Middle East Technical University, Ankara, 1997.
[23] M. Tursun, Effects of heavy metals on activated sludge process, M.S.
Greek letters Thesis, Department of Chemical Engineering, Middle East Technical
University, Ankara, 1997.
h hydraulic residence time (h or days) [24] L.S. Clesceri, A.E. Greenberg, A.D. Eaton (Eds.) Standard Meth-
c mean cell residence time or sludge age (days) ods for Examination of Water and Wastewater, APHA, WPCF, and
AWWA, 20th ed., New York, 1998.
[25] R.C. Stover, L.E. Sommers, D.J. Silviera, Evaluation of metals in
References wastewater sludge, J. WPCF 48 (1976) 21652175.
[26] V. Utgikar, B.Y. Chen, H.H. Tabak, D.F. Bishop, R. Govind, Treat-
[1] R.D. Neufeld, J. Gutierrrez, R.A. Novak, A kinetic model and equi- ment of acid mine drainage: I. Equilibrium biosorption of zinc and
librium relationship for heavy metal accumulation, J. WPCF 49 copper on non-viable activated sludge, Int. Biodeterior. Biodegrad.
(1977) 489498. 46 (2000) 1928.
32 T.A. zbelge et al. / Chemical Engineering and Processing 44 (2005) 2332
[27] S.R. Juliastuti, J. Baeyens, C. Creemers, D. Bixio, E. Lodewyckx,
The inhibitory effects of heavy metals and organic compounds on
the net maximum specific growth rate of the autotrophic biomass in
activated sludge, J. Hazard. Mater. B100 (2003) 271283.
[28] D.T. Sponza, Investigation of extracellular polymer substances (EPS)
and physicochemical properties of different activated sludge flocs
under steady-state conditions, Enzyme Microb. Technol. 32 (2003)
375385.
[29] C.S. Laspidou, E.R. Bruce, A unified theory for extracellular
polymeric substances, soluble microbial products, active and inert
biomass, Water Res. 11 (2002) 27112720.
[30] J.A. Scatt, J. Karanjkar, Immobilized biofilms on granular activated
carbon for removal and accumulation of heavy metals from contam-
inated streams, Water Sci. Technol. 38 (1998) 97204.
[31] V. Urbain, J.C. Block, J. Manem, Bioflocculation in activated sludge:
an analytical approach, Water Res. 27 (1993) 829838.
[32] J.A.S. Goodwin, C.F. Forster, A further examination into the com-
position of activated sludge surfaces in relation to their settlement
characteristics, Water Res. 19 (1985) 527533.
[33] M.F. Dignac, V. Urbain, D. Rybacki, A. Bruchet, D. Snidaro, P.
Scribe, Chemical description of extracellular polymers: implications
on activated sludge floc structure, Water Sci. Technol. 38 (1998)
4553.
[34] M.J. Brown, J.N. Lester, Role of bacterial extracellular polymers in
metal uptake in pure bacterial culture and activated sludge, Water
Res. 16 (1982) 15491560.
[35] B.A. Friedman, P.R. Dugan, Concentration and accumulation of
metallic ions by the bacterium Zoogloea, Dev. Ind. Microbiol. 9
(1968a) 381388.
[36] W.A. Corpe, Metal-binding properties of surface materials from ma-
rine bacteria, Dev. Ind. Microbiol. 16 (1975) 249255.
[37] A. Jernelv, A.L. Martin, Ecological Implications Of Metal
Metabolism By Micro-Organisms, Swedish Water and Air Pollution
Research Laboratory, Stockholm, 1975, p. 32.
[38] B.R. Fristoe, P.O. Nelson, Equilibrium chemical modelling of
heavy metals in activated sludge, Water Res. 17 (1983) 771
778.
 Chemical Engineering and Processing 44 (2005) 3339

Porous catalyst intraparticle status of parallel,

equilibrium-restrained reactions
Guangsuo Yu , Fuchen Wang, Zhenghua Dai, Zunhong Yu
Institute of Clean Coal Technology, East China University of Science and Technology, P.O. Box 272, 130 Meilong Road,
Shanghai 200237, Peoples Republic of China

Received 17 February 2004; received in revised form 11 March 2004; accepted 7 April 2004
Available online 1 June 2004

Abstract

The catalyst intraparticle reaction-diffusion process of parallel, equilibrium-restrained reactions for the methanation system was studied. The
non-isothermal one-dimensional and two-dimensional reaction-diffusion models for the key components have been established, and solved
using an orthogonal collocation method. The simulation values of the effectiveness factors for methanation reaction CH4 and shift reaction CO2
are fairly in agreement with the experimental values. CH4 is large, while CO2 is very small. The shift reaction takes place as direct and reverse
reaction inside the catalyst pellet because of the interaction of methanation and shift reaction. For parallel, equilibrium-restrained reactions,
effectiveness factors are not able to predict the catalyst internal-surface utilization accurately. Therefore, the intraparticle distributions of the
temperature, the concentrations of species and so on should be taken into account.
2004 Elsevier B.V. All rights reserved.

Keywords: Catalyst; Parallel reactions; Equilibrium-restrained; Reaction-diffusion model

1. Introduction Several quantitative analyses of the effect of intraparticle

Gas-phase mass transport in porous catalysts is an im-
portant factor in predicting observed reaction rates for
heterogeneous catalytic systems. Mathematical models of
diffusion and reaction in porous catalysts play a fundamen-
tal role in the design of catalysts and catalytic reactors,
and in the interpretation of experimental rate data. The
phenomenon of simultaneous transport and reaction in-
side a catalyst particle has been widely studied over the
years, but mainly for single reactions and simple kinetics.
The classical diffusion-reaction model for a porous cata-
lyst describes the influence of intraparticle concentration
gradients on the observed rate through the effectiveness
factor, which is defined as the ratio of the volume-averaged
reaction rate over the reaction rate at the external sur-
face. For single reactions, the effectiveness factor can
describe the utilization degree of catalyst internal surface
accurately.
Corresponding author.
E-mail address: gsyu@ecust.edu.cn (G. Yu).
heat and mass transport have been carried out for parallel, ir-
reversible reactions [1]. Roberts and Lamb [2] have worked
on the effect of reversibility on the selectivity of parallel
reactions in a porous catalyst. The reaction selectivity of a
kinetic model of two parallel, first order, irreversible reac-
tions with a second order inhibition kinetic term in one of
them has also been investigated [3].
The reaction-diffusion model inside a catalyst pellet is
a differential equations set of boundary-value problem, the
standard solving methods for which are finite difference,
numerical integration shooting, weighting residual, singular
perturbation [4], Green function [5], false transient [6], or-
thogonal collocation [7], moving grid finite element [8], etc.
Based on the investigation of reaction, heat transfer and
mass transfer of the KD306-type sulfur-resisting methana-
tion catalyst [911], the non-isothermal one-dimensional
and two-dimensional reaction-diffusion models for the key
components have been established, and solved using an or-
thogonal collocation method in this paper. The scope is to
study the catalyst intraparticle reaction-diffusion processes
that involve parallel, non-first order, equilibrium-restrained
reactions.

0255-2701/$ see front matter 2004 Elsevier B.V. All rights reserved.
doi:10.1016/j.cep.2004.04.003
34 G. Yu et al. / Chemical Engineering and Processing 44 (2005) 3339

2. Reaction system and fundamental data under the reaction conditions at 400600 C. The equilib-
rium constant of the WGSR is small and varies insignif-
With the reducing supply of low sulfur petroleum, the icantly with temperature. Due to the water formation and
demand for residual oil and coal with high-sulfur con- the concentration variations in carbon monoxide and hydro-
tent is increasing. The development and application of gen, which are consequences of the methanation reaction,
sulfur-resisting catalysts have received great attention. The the overall WGSR occurs both as the direct and reverse re-
KD306-type sulfur-resisting methanation catalyst, a sulfide action at internal surface of the catalyst. So, the fundamental
catalyst such as Mo and W of group VIB as the primary features of this reaction system are equilibrium-restrained,
species and Fe, Co, Ni of group VIII as the indispens- namely reversible, parallel reactions.
able promoters, can be used in methanation reaction with
sulfur-containing syngas as the reactants at high reactivity 2.1. Reaction kinetics [9]
and selectivity. The cylindrical catalyst pellets (diameter
= 5.422 mm, length = 4.327 mm) were prepared by com- The intrinsic kinetics was measured in an isothermal in-
pressing particles. tegrated reactor and the reaction rate equations in terms of
Methanation reaction and watergas shift reaction power function have been established
(WGSR), which are chosen as the examples for presenta-

46 944
tion, are expressed as follows: (rCH4 )w = 3.8333 106 exp
RT
0.58 P 0.52 mol s1 kg1
CO + 3H2 = CH4 + H2 O, H 298 = 206 kJ/mol PCO H2


CO + H2 O = CO2 + H2 , H 298 = 41 kJ/mol (rCO2 )w = 4.1089 105 exp
58 349
RT
The former is a volume-decreasing reaction, while the PCO0.58 P 0.98 (1 ) mol s1 kg1
H2 O
latter is not. Both reactions are exothermic. Methanation
is a deep hydrogenation reaction for carbon monoxide and where = JP /KP .
WGSR is a complete oxidation reaction in which carbon The macrokinetics were measured in a gradientless reac-
monoxide is oxidized into carbon dioxide and water is re- tor.
duced with the formation of hydrogen. As in the preparation
of methane, other hydrocarbons, low alcohols and particu- 2.2. Effective diffusivity DeffA [10]
larly, carbon dioxide and water are formed. Because of the
presence of water, WGSR always occurs in the methana- Mass transport inside the catalyst has been usually de-
tion process, which reduces the selectivity and yield of the scribed by applying the Fick equation, by means of an effec-
desired product. tive diffusivity DeffA . Based on properties of the interface
Fig. 1 shows reaction equilibrium constants of the metha- and neglecting the composition effect, composite diffusivity
nation reaction and the WGSR [12]. The mole composition of the multi-component gas mixture is calculated through
of town gas is: CO 20%, H2 40%, CO2 10%, H2 O the simplified Wilke model [13]. The effect of pore-radius
0%, N2 the rest. Methanation reaction is almost irreversible distribution on Knudsen diffusivity is taken into account.
The effective diffusivity DeffA is given by
16 1
 N
10000 rmax 1  yj 1
DeffA = + f(r) dr
rmin 1 yA DAj DAk (r)
j=A
12 where is the tortuosity factor of the KD306 catalyst, which
1000
was measured using single pellet string reactor technique.
Kp,methanantion,atm -2

Kp,shift

2.3. Thermal conductivity eff [11,14]

8
100
Based on the Woodside stochastic model, thermal con-
ductivity eff can be calculated from gas composition and
4 temperature of the catalyst exterior surface by
10


eff fm
=
s s
1
673 723 773 823 873
0 where s is given by
T, K
s = 19.90005 + 0.12333T 2.38169( 100
1
T)2
Fig. 1. Reaction equilibrium constants of methanation reaction and shift
+ 0.151983( 100
1
T)3
reaction.
 G. Yu et al. / Chemical Engineering and Processing 44 (2005) 3339 35

3. Theory [15,16] According to the definitions

The assumptions made for the reaction-diffusion models 4R2p Deff,CO2 (dCCO2 /dR)s
are as follows: CO2 =
(4/3)R3p p (rCO2 )w,s
(1) parallel, equilibrium-restrained reactions occur (namely,
methanation reaction is irreversible, while WGSR is re- the effectiveness factors for the catalyst pellet can be calcu-
versible); lated from
(2) non-first order reactions; N+1
(3) non-isothermal; 3 j=1 AN+1,j CCO2 ,j
CO2 = ,
(4) quasi-homogeneous; (Rp /Deff,CO2 )p (rCO2 )w,s
2
(5) only key components are considered (methane and car-
3 N+1
j=1 AN+1,j CCH4 ,j
bon dioxide). CH4 =
(R2p /Deff,CH4 )p (rCH4 )w,s
3.1. One-dimensional model
3.2. Two-dimensional model
Commonly [17], when the length-to-diameter ratio of a
cylindrical catalyst is close to 1, the cylindrical catalyst can RP and HP are the radius and semi-length of cylindri-
be simplified as a sphere, the radius of which, Rp , is calcu- cal catalyst, respectively. Assume a cylindrical ring element
lated by 3Vp /Sp . The one-dimensional, key-component based with the length dH and the radial thickness dR at the axial
reaction-diffusion models of methanation system are as fol- position H and the radial position R in the 2D cylindrical
lows: system. Introducing the dimensionless variables
d2 CCH4 2 dCCH4 R2p
+ = (rCH4 )w p R H
dx2 x dx Deff,CH4 x= , z=
Rp Hp
d2 CCO2 2 dCCO2 R2p
+ = (rCO2 )w p The species (methane and carbon dioxide) mass balance
dx2 x dx Deff,CO2
and energy conservation in the cylindrical ring element at
d2 T 2 dT R2p
+ = p [(rCH4 )w (HCH4 ) steady state lead to the following 2D partial differential equa-
dx2 x dx eff tions:
+ (rCO2 )w (HCO2 )]

2 2
Boundary conditions 2 CCH4 1 CCH4 Rp CCH4
+ +
x2 x x Hp z2
r
x= = 1, CCH4 = (CCH4 )s , CCO2 = (CCO2 )s , Rp2
Rp = p (rCH4 )w
T = Ts Deff,CH4

2 2
dCCH4 dCCO2 dT 2 CCO2 1 CCO2 Rp CCO2
x = 0, = 0, = 0, =0 + +
dx dx dx x2 x x Hp z2
R2p
Applying orthogonal collocation method, the above equa- = p (rCO2 )w
tions become Deff,CO2

2 2
2 T 1 T Rp T


N+1
2

R2p + +
Bjk + Ajk CCH4 ,k = (rCH4 )w,j p x 2 x x Hp z2
x Deff,CH4 Rp2
k=1
= p [(HCH4 )(rCH4 )w + (HCO2 )(rCO2 )w ]


N+1
2

R2p eff
Bjk + Ajk CCO2 ,k = (rCO2 )w,j p
x Deff,CO2
k=1 with the associated boundary conditions


N+1
2

R2p
Bjk + Ajk Tk = p [(rCH4 )w,j (HCH4 ) CCO2 CCH4 T
x eff x = 0, = = =0
k=1 x x x
+ (rCO2 )w,j (HCO2 )] CCO2 CCH4 T
z = 0, = = =0
The solutions of the above non-linear equations can be ob- z z z
tained using the Broyden quasi-Newton method [18]. Based x = 1 or z = 1, CCO2 = CCO2 ,s , CCH4 = CCH4 ,s , T = Ts
on the mass balance, the concentrations of other component,
such as CO, H2 , H2 O and so on, can be calculated from the Each item of the model at collocation point (I, J) may be
key components [9]. expressed as
36 G. Yu et al. / Chemical Engineering and Processing 44 (2005) 3339

F = CCO2 F = CCH4 F =T
 N+1
N+1  N+1
 N+1
 N+1
 N+1

2j2 2j2 2j2
(F)I,J aij xI2i2 zJ bij xI2i2 zJ cij xI2i2 zJ
j=1 i=1 j=1 i=1 j=1 i=1

 N+1
 N+1
 N+1
 N+1
 N+1
 N+1

F 2j2 2j2 2j2
(2i 2)aij xI2i3 zJ (2i 2)bij xI2i3 zJ (2i 2)cij xI2i3 zJ
x I,J j=1 i=1 j=1 i=1 j=1 i=1

 N+1
 N+1
 N+1
 N+1
 N+1
 N+1

2 F 2j2 2j2 2j2
(2i 2)(2i 3)aij xI2i4 zJ (2i 2)(2i 3)bij xI2i4 zJ (2i 2)(2i 3)cij xI2i4 zJ
x2 I,J j=1 i=1 j=1 i=1 j=1 i=1

 N+1
 N+1
 N+1
 N+1
 N+1
 N+1

2 F 2j4 2j4 2j4
(2j 2)(2j 3)aij xI2i2 zJ (2j 2)(2j 3)bij xI2i2 zJ (2j 2)(2j 3)cij xI2i2 zJ
z2 I,J j=1 i=1 j=1 i=1 j=1 i=1
where, I = 1, 2, . . . , NR + 1, J = 1, 2, . . . , NA + 1.
We obtain
NX +1 NX +1
2 N Z +1
1  Rp R2p p
BXIk CCH4 ,kJ + AXIk CCH4 ,kJ + BZJk CCH4 ,kI = (rCH4 )w,IJ
xI Hp Deff,CH4
k=1 k=1 k=1
NX +1 NX +1
2 N Z +1
1  Rp R2p p
BXIk CCO2 ,kJ + AXIk CCO2 ,kJ + BZJk CCO2 ,kI = (rCO2 )w,IJ
xI Hp Deff,CO2
k=1 k=1 k=1
NX +1 NX +1
2 N Z +1
1  Rp R2p p
BXIk TkJ + AXIk TkJ + BZJk TkI = [(HCH4 )IJ (rCH4 )w,IJ + (HCO2 )IJ (rCO2 )w,IJ ]
xI Hp eff
k=1 k=1 k=1

The concentration and temperature profiles are calculated

taking into account the fact that there are errors in measuring
from the above non-linear equations using the Broyden
intrinsic kinetics, macrokinetics, effective diffusivity, and
quasi-Newton method. The effectiveness factors for the
thermal conductivity, the calculated effectiveness factors
catalyst pellet may be expressed as
are in reasonable agreement with the experimental values.
NR +1 NA +1 From Table 1, it can be concluded that the simulation results
I=1 J=1 (rCO2 ,w )I,J WR,I WA,J
CO2 = NR +1 NA +1 , of the one-dimensional model are close to the results of the
(rCO2 ,w )s I=1 WR,I J=1 WA,J two-dimensional model; also, the methanation reaction ef-
NR +1 NA +1
J=1 (rCH4 ,w )I,J WR,I WA,J
fectiveness factors of the two-dimensional model are closer
CH4 = I=1 NR +1 NA +1 to the experimental data than those of the one-dimensional
(rCH4 ,w )s I=1 WR,I J=1 WA,J model. With these models, the concentration profiles and
where WR , WA are weight coefficients of the Gauss integral temperature profile in the catalyst pellet can be predicted.
formulae, defined as
4.2. Results
 1 N+1
 1
x2i2 x1 dx = Wj xj2i2 = , Predicted parameter distributions within catalyst pellet for
0 2i 2 +
j=1 some experimental conditions are presented in Tables 25.
i = 1, 2, . . . , N As shown in Tables 2 and 3, the deeper inside the pellet,
the higher the concentration of carbon dioxide and methane
are, and the lower the concentration of carbon monoxide is.
4. Results and discussion The effectiveness factor CH4 of the methanation reaction
is large, while the effectiveness factor CO2 of the WGSR
4.1. Model testing is very low and nearly equal to zero. Because the equilib-
rium constant of WGSR is small and varies only slightly
The experimentally-determined effectiveness factor is with temperature, and the methanation reaction changes the
determined as the ratio of the experimental macro reaction gas composition, there are zones where the direct reaction
rate to the intrinsic reaction rate under the same interface and the reverse reaction of WGSR dominate at internal sur-
(bulk) composition and temperature. Based on the ex- face of catalyst, which is shown in Tables 2 and 4. The
perimental conditions of the macrokinetics, the predicted low mole-fraction of water results in the reverse reaction of
effectiveness factors of the methanation reaction and the WGSR within the region beneath the exterior surface. Due to
WGSR are obtained by solving the above non-isothermal reaction consumption of carbon monoxide, the mole-fraction
one-dimensional and two-dimensional reaction-diffusion of carbon monoxide within the core region is low, which
models for the key components. Table 1 shows the calcu- leads to the reverse WGSR. The reaction rate of reverse
lated effectiveness factors and the experimental values. By WGSR is large, especially at the exterior surface and the
 G. Yu et al. / Chemical Engineering and Processing 44 (2005) 3339 37

Table 1
Simulation results of one-dimensional and two-dimensional models by orthogonal collocation method
Number T (K) P (atm) Interface composition CH4 ,e CO2 ,e CH4 ,cone CO2 ,cone CH4 ,ctwo CO2 ,ctwo
yCH4 yCO

1 703.45 7.91 0.1774 0.1773 0.9277 0.0440 0.6699 0.0090 0.7192 0.0699
2 702.65 14.22 0.2508 0.1672 0.6992 0.0084 0.6128 0.0098 0.6696 0.0926
3 704.35 10.28 0.2739 0.1499 0.6679 0.0255 0.6490 0.0183 0.7022 0.0824
4 703.15 11.86 0.2816 0.1446 0.8273 0.0198 0.6351 0.0163 0.6901 0.0816
5 738.35 12.05 0.2836 0.1560 0.5774 0.0151 0.5558 0.0085 0.5832 0.0428
6 738.55 8.30 0.2330 0.1730 0.6665 0.0082 0.6081 0.0134 0.6513 0.0727

Table 2
One-dimensional parameter distribution within catalyst pellet for Number 1 (T = 703.45 K, P = 7.91 atm)
x yCO2 yCH4 yCO T (K) rCH4 103 mol s1 kg1 rCO2 103 mol s1 kg1
0.0000 0.1944 0.3443 0.0427 703.79 0.2620 5.4989
0.2154 0.1902 0.3392 0.0486 703.78 0.2862 0.1375
0.4206 0.1846 0.3236 0.0590 703.74 0.3391 0.1907
0.6063 0.1736 0.2977 0.0781 703.69 0.4319 0.3417
0.7635 0.1558 0.2634 0.1066 703.62 0.5625 0.4908
0.8851 0.1330 0.2256 0.1411 703.55 0.7124 0.7889
0.9653 0.1137 0.1935 0.1704 703.48 0.8398 2.3307
1.0000 0.1118 0.1774 0.1773 703.45 0.8905 18.9060

core center. The direct WGSR dominates between the men- not imply that WGSR hardly occurs in the catalyst pellet due
tioned above regions but its rate is low. The region of direct to internal diffusion. WGSR occurs remarkably at all sites of
reaction is big, while the region of reverse reaction is small. catalyst pellet, and the reaction rate of each site is not equal
Internal diffusion is not the major factor leading to low CO2 . to zero. As a result of the existence of direct and reverse
The existence of regions of the direct and reverse WGSR reaction of WGSR, and also the volume-decreasing metha-
contributes to a greater extent to low CO2 , which in turn nation reaction, there is small variation of carbon dioxide
causes the low net reaction rate. This low reaction rate does concentration due to the site change in the pellet.
It is shown in Tables 2 and 5 that the methanation reac-
tion takes place as a direct reaction throughout the whole
Table 3
Two-dimensional parameter distribution within catalyst pellet for No.
internal-surface of the pellet. Carbon monoxide, which is
3-yCH4 (T = 704.35 K, P = 10.28 atm) the product of the reverse WGSR, promotes the methanation
reaction. The interaction between the methanation reaction
z x
and the WGSR enhances the methanation reaction and also
0.0000 0.2390 0.5262 0.7639 0.9275 1.0000 improves CH4 . Internal diffusion also affects the methana-
0.0000 0.4220 0.4183 0.4002 0.3615 0.3085 0.2739 tion reaction.
0.2390 0.4175 0.4141 0.3969 0.3597 0.3081 0.2739 To put it briefly, the effectiveness factors for parallel
0.5262 0.3967 0.3942 0.3809 0.3504 0.3053 0.2739 equilibrium-restrained reactions cannot describe the utiliza-
0.7639 0.3561 0.3548 0.3474 0.3289 0.2981 0.2739 tion degree of catalyst internal-surface accurately. Therefore,
0.9275 0.3052 0.3051 0.3027 0.2966 0.2852 0.2739
the intraparticle distributions of the temperature, the concen-
1.0000 0.2739 0.2739 0.2739 0.2739 0.2739 0.2739
trations of species and so on should be taken into account.

Table 4 Table 5
Two-dimensional parameter distribution within catalyst pellet for No. Two-dimensional parameter distribution within catalyst pellet for No.
3-rCO2 103 (T = 704.35 K, P = 10.28 atm) 3-rCH4 103 (T = 704.35 K, P = 10.28 atm)

z x z x
0.0000 0.2390 0.5262 0.7639 0.9275 1.0000 0.0000 0.2390 0.5262 0.7639 0.9275 1.0000

0.0000 18.909 9.619 8.904 8.136 9.467 52.009 0.0000 0.1672 0.1891 0.2690 0.4486 0.7055 0.8575
0.2390 10.049 0.155 0.206 0.557 0.812 52.009 0.2390 0.1922 0.2129 0.2889 0.4617 0.7112 0.8575
0.5262 9.140 0.306 0.333 0.641 0.767 52.009 0.5262 0.2844 0.3015 0.3621 0.5058 0.7256 0.8575
0.7639 8.375 0.687 0.678 0.905 0.604 52.009 0.7639 0.4739 0.4849 0.5206 0.6098 0.7616 0.8575
0.9275 10.533 16.161 1.648 1.513 3.006 52.009 0.9275 0.7212 0.7260 0.7378 0.7682 0.8251 0.8575
1.0000 52.009 52.009 52.009 52.009 52.009 52.009 1.0000 0.8575 0.8575 0.8575 0.8575 0.8575 0.8575
38 G. Yu et al. / Chemical Engineering and Processing 44 (2005) 3339

5. Conclusions z dimensionless coordinate along

longitudinal distance
Based on the studies on the KD306-type sulfur-resisting
methanation catalyst, the non-isothermal one-dimensional Greek letters
and two-dimensional reaction-diffusion models for the tortuosity
key-components have been established, which were solved CO2 ,cone predicted effectiveness factor of WGSR of
using an orthogonal collocation method. The simulation val- one-dimensional model
ues of the effectiveness factors for the methanation reaction CO2 ,ctwo predicted effectiveness factor of WGSR of
CH4 and the shift reaction CO2 are in fair agreement with two-dimensional model
the experimental values, which indicates that both mod- CO2 ,e experimental effectiveness factor of WGSR
els are able to predict intraparticle transport and reaction CH4 ,cone predicted effectiveness factor of methanation
processes within catalyst pellets. of one-dimensional model
The effectiveness factor CH4 of the methanation reac- CH4 ,ctwo predicted effectiveness factor of methanation
tion is large, while the effectiveness factor CO2 of the of two-dimensional model
WGSR is very small and nearly equal to zero. The inter- CH4 ,e experimental effectiveness factor of
action between the methanation reaction and the WGSR methanation
leads to the existence of the direct and reverse WGSR porosity
in catalyst pellet, which is called equilibrium-restrained fm gas heat conductivity in the pore (W/(m K))
WGSR. s heat conductivity of solid (W/(m K))
The effectiveness factors calculated for the parallel, eff thermal conductivity (W/(m K))
equilibrium-restrained reaction systems are not able to p grain density (kg/m3 )
predict the catalyst internal-surface utilization accurately.
Therefore, the intraparticle distributions of the temperature,
the concentrations of species and so on should be taken into
account. References

[1] G.W. Roberts, The selectivity of porous catalysts: parallel reactions,

Chem. Eng. Sci. 27 (1972) 13481420.
[2] G.W. Roberts, H.H. Lamb, The effect of reversibility on the selectivity
Appendix A. Notations of parallel reactions in a porous catalyst, Chem. Eng. Sci. 51 (1996)
441448.
A, B, AX, transition matrices [3] G.D. Papakonstantopoulos, G.P. Androutsopoulos, C.J. Philippopou-
BX, BZ los, Reaction selectivity in a porous catalyst pellet: analysis of a
Ci concentration of component i (mol/m3 ) kinetic model of two parallel, first order, irreversible reactions with
a second order inhibition kinetic term in one of them, Chem. Eng.
Cis interface concentration of component Sci. 56 (2001) 54135417.
i (mol/m3 ) [4] I. Copelowitz, R. Aris, Communications on the theory of diffusion
DAj binary diffusion coefficient of a mixture A and reaction-VI. The effectiveness of spherical catalyst particles in
and j (m2 /s) steep external gradients, Chem. Eng. Sci. 25 (1970) 885.
DAk Knudsen diffusion coefficient of [5] R.S. Dinit, L.L. Tavlaridis, Integral method of analysis of
FischerTropsch synthesis reactions in a catalyst pellet, Chem. Eng.
component A (m2 /s) Sci. 37 (1982) 539.
DeffA effective diffusivity of component A (m2 /s) [6] M.-S. Xu, Z.-M. Du, X.-F. Gao, False transient method for determin-
f(r) aperture distribution function ing effectiveness factors of high-temperature shift reaction catalyst
HCH4 heat of methanation reaction (J/mol) B109, J. Chem. Ind. Eng. (China) 44 (1993) 465.
[7] J.P. Sorensen, W.E. Stewart, Collocation analysis of multicomponent
HCO2 heat of WGSR (J/mol)
diffusion and reactions in porous catalysts, Chem. Eng. Sci. 37 (1982)
Hp semi-length of pellet (m) 1103.
JP , KP partial pressure ratio and equilibrium [8] A. Madzvamuse, A.J. Wathen, P. Maini, A moving grid finite element
constant of WGSR, respectively method applied to a model biological pattern generator, J. Comput.
NR , NA numbers of radial and longitudinal Phys. 190 (2003) 478500.
[9] G.-S. Yu, Study on Town Gas Methanation of KD306-type
collocation point, respectively
Sulfur-tolerant Catalyst [D] (China), East China University of Sci-
Rp radius of pellet (m) ence and Technology, Shanghai, 1998.
(rCH4 )w , methanation reaction rate and WGSR rate, [10] G.-S. Yu, J.-G. Yu, Z.-H. Yu, The measurement of effective diffusivity
(rCO2 )w respectively (mol s1 kg1 ) for sulfur-tolerant methanation catalyst, Chem. Eng. J. 78 (2000)
Ts interface temperature (K) 141146.
[11] G.-S. Yu, J.-G. Yu, Q.-Z. Tan, Z.-H. Yu, Study on thermal conduc-
T temperature (K)
tivity for sulfur-tolerant methanation catalyst, Chem. Eng. (China)
x dimensionless coordinate along radial 28 (1) (2000) 2224, 32.
distance [12] Z.-H. Yu, Hydrocarbon-steam Conversion Engineering, Hydrocarbon
yi mole-fraction of component i Processing Press, Beijing, 1989 (in Chinese).
 G. Yu et al. / Chemical Engineering and Processing 44 (2005) 3339 39

[13] C.R. Wilke, Diffusional properties of multicomponent gases, Chem.
Eng. Progr. 46 (1950) 95.
[14] W. Woodside, J.H. Messmer, Thermal conductivity of porous
media I. Unconsolidated sands, J. Appl. Phys. 32 (1961)
1688.
[15] B.A. Finlayson, Orthogonal collection in chemical reaction engineer-
ing, Cat. Rev.-Sci. Eng. 10 (1974) 69138.
[16] Z.-X. Xu, Proximate Solution of Differential Equation, East China
Institute of Science and Technology Press, Shanghai, 1990 (in Chi-
nese).
[17] R. Aris, On shape factors for irregular particles-I. The steady state
problem. Diffusion and reaction, Chem. Eng. Sci. 6 (1957) 262.
[18] D.-G. Liu, J.-G. Fei, Y.-J. Yu, FORTRAN Arithmetic Compilation,
National Defence Industry Press, Beijing, 1980 (in Chinese).
 Chemical Engineering and Processing 44 (2005) 4149

PID controller tuning using mathematical programming

George Syrcos a , Ioannis K. Kookos b,
a Controls Department, TEI Pireaus, 250 Petrou Ralli & Thivon Ave, 12244 Athens, Greece
bDepartment of Chemical Engineering, UMIST, PO Box 88, Manchester M60 1QD, UK

Received 19 December 2003; received in revised form 18 February 2004; accepted 1 April 2004
Available online 18 May 2004

Abstract

Proportional integral derivative (PID) controllers are extensively used in the process industry. As a result a large number of general purpose
tuning methodologies are available. These tuning methodologies can offer initial estimates of the parameters of the PID controllers. However,
the design objectives used for the development of these tuning methods can be quite different from the performance objectives specific to a
process under investigation. As a result, the control engineer often needs customized tuning methods in order to speed up or even eliminate the
retuning procedure, and thus, minimize the time and effort needed to design a satisfactory closed loop system. This paper presents a general
mathematical programming formulation for the development of customized PID controller tunings. A reformulation of the mathematical
formulation is proposed that transforms the initially nonlinear formulation to a linear one that can be solved to global optimality. A number
of case studies are presented to clarify the proposed methodology.
2004 Elsevier B.V. All rights reserved.

Keywords: PID-controller tuning; Mathematical programming

1. Introduction a specific application. Thus, control engineers often face the

Proportional integral derivative (PID) controllers are ex-
tensively used in the process industry. It was recently esti-
mated that more than 90% of all control loops involve PID
controllers [1]. The simplicity and transparency of PID con-
trol mechanism, the availability of a large number of highly
efficient, reliable, and cost-effective commercial PID con-
trol modules and their acceptance from the operators are
among the reasons for their success [2]. Tuning of the PID
controllers is achieved in two steps. In the first step a simpli-
fied model of the process is obtained and initial tunings are
calculated using well-known tuning rules (such as the mod-
ified Ziegler-Nichols [2] or IMC-based [3]). In the second
step, the performance of the loop is assessed and the initial
tunings are modified in order to take into account the oper-
ating specifications of the process. The success or failure of
this procedure strongly depends on the judicious choice of
initial tunings. However, all available tuning methods have
been developed to satisfy specific design objectives that, in
most cases, cannot be translated into the design objectives of
need to develop customized tuning rules that speed up the
controller tuning procedure by minimizing the time needed
for retuning or even eliminating retuning.
The aim of this paper is to develop mathematical pro-
gramming formulations for the development of tuning rules
for PID controllers with arbitrary design objectives posed in
the time domain. The resulting mathematical formulation is
a nonlinear programming problem (NLP). A reformulation
of the problem is proposed that transforms the formulation
to a mixed-integer linear problem. If the objective function
chosen is linear (convex in general) then the resulting for-
mulation can be solved to global optimality using currently
available software. The main advantage of the proposed
methodology stems from its generality since it can be ap-
plied to obtain PID controllers tunings for any linear system.
2. PID controller tuning
2.1. Tuning of proportional only (P-only) controllers

Corresponding author. Tel.: +44-1612004346; fax: +44-1612004399. In this work we consider single-input, single output
E-mail address: i.kookos@umist.ac.uk (I.K. Kookos). (SISO) linear time invariant (LTI) systems described by the

0255-2701/$ see front matter 2004 Elsevier B.V. All rights reserved.
doi:10.1016/j.cep.2004.04.001
42 G. Syrcos, I.K. Kookos / Chemical Engineering and Processing 44 (2005) 4149

 q
following discrete-time state space model

uk = ek , k (8)
xk+1 = Axk + buk + Evk
(1) =1
yk = Cxk + duk + Fvk
Finally, we define the variable E ,k to be equal to the product
where x is the n vector of state variables, u the input variable, of the binary variable and the continuous variable ek , i.e.
v the m vector of disturbances, y the output variable, and
E ,k = ek , , k (9)
k is the sampling time. First, we consider the general class
of proportional-only feedback controllers. In this case, the and Eq. (8) can be written in the following linear form
control moves are related to the deviation from the set point q


through the following: uk = E ,k , k (10)
uk = kC ek , k (2) =1

Then, the nonconvex Eq. (9) can be replaced by the fol-

ek = rk yk , k (3)
lowing set of equivalent linear inequalities [4,5]
where r is the reference signal and kC the controller gain. eL E ,k eU (11a)
The design objective is to select the controller gain so as to
minimize a scalar performance index (which is a function ek eU (1 ) E ,k ek eL (1 ) (11b)
of the output and control variables) subject to constraints on It should be noted that when = 0 then E ,k = 0, k,
the input and output variables, i.e. while when = 1 E ,k = ek , k. As a result, the equiva-
lence of Eqs. (9), (11a), and (11b) can be established easily.
xk+1 = Axk + buk + Evk

y = Cx + du + Fv In summary, the nonconvex formulation given by Eq. (4)

k k k k

can be replaced by the following linearly constrained for-

ek = rk yk

mulation

uk = uk1 + uk



N N


min = J[xk , uk ], s.t. uk = kC ek min = J[xk , uk ],
kC
L
k=1
k=0

yk yk ykU

xk+1 = Axk + buk + Evk




uLk uk uU


k
yk = Cxk + duk + Fvk



u uk uU
L


ek = rk yk


(4)




uk = uk1 + uk
Problem (4) is a nonconvex, nonlinear programming prob-



q
lem. The nonconvexities are due to the bilinear form of the
uk = =1 E ,k


control law. This term can be linearized at the expense of

eL E eU ,
optimizing over a discrete (rather than continuous) set of ,k
s.t.
controller gains.


ek e (1 ) E ,k ek e (1 ),
U L

First, we assume that an initial controller gain kC,0 is

q


available and calculated using classical control techniques.
=1 = 1
Then we consider a discrete set of q tuning factors that



lie in the interval [ L , U ] and defined as follows:
ykL yk ykU


 

1

uk uk uk
L U
= +L
( U L ), = 1, 2, . . . , q
q1
(5)


uL u uU

k
Furthermore, we define q binary variables {0, 1} to
{0, 1} ,
denote the selection of tuning factor ( = 1) or not ( =
(12)
0). Then, the control gain can be expressed as follows:
q Depending on the form of the objective function, formu-


kC = (6) lation (12) corresponds to a mixed integer, nonlinear pro-
=1 gramming (MINLP) problem if the objective is nonlinear,
to a mixed integer quadratic programming (MIQP) problem
q

if the objective is quadratic or to a mixed integer linear pro-
= 1 (7)
gramming (MILP) problem if the objective is linear. In this
=1
paper consideration is restricted to the case of convex objec-
where = kC,0 . Substituting Eq. (6) into Eq. (2) we tive functions for which the global optimum solutions can
obtain be found using available algorithms [4].
 G. Syrcos, I.K. Kookos / Chemical Engineering and Processing 44 (2005) 4149 43

2.2. Tuning of proportional-integral (PID) controllers This is a nonlinear formulation due to the bilinear terms
involved in Eq. (17) and the (possibly) nonlinear objective.
The PID control law considered here contains the P, PI, Following the same procedure as in the previous subsection
PD, and PID control laws as special cases. The velocity form we can replace Eq. (17) by a set of linear inequalities.
of the discrete approximation of an ideal PID controller is We first assume that initial values of the controller param-
given by [2] eters have been obtained using any classical control tech-
    nique. We then introduce tuning factors P , I and D as
S D 2D well as the binary variables P , I , and D
uk = kC 1 + + ek 1 + ek1 to denote the se-
I S S lection -th term of the proportional, integral and derivative

D term of the PID controller, i.e.
+ ek2 (13)
S q


m
m
K = m m
, m (20)
where S is the sampling time, kC the controller gain, I the
=1
integral time, and D is the derivative time. We define the
following variables:
qm
m
= 1, m (21)
Pk = ek ek1 (14)
=1

Ik = S ek (15) m = kC,0
m m
, , m (22)
1 ,k = P Pk , , k (23)
Dk = (ek 2ek1 + ek2 ) (16)
S
O ,k = I Ik , , k (24)
and Eq. (13) can be written as
,k = D
Dk , , k (25)
uk = KP Pk + KI Ik + KD Dk (17)
where m {P, I, D}. Eq. (17) can then be written in the
kC
K = kC ,
P
K = I
, K = kC D
D
(18) following linear form:
I
q


P q


I q


D

The complete formulation of the closed loop system op- uk = P ,k + I O ,k + D ,k , k (26)

timization using a PID controller is as follows =1 =1 =1

N


min = J[xk , uk ], P L P ,k P U P
KP ,KI ,KD , , k
k=0 Pk P U (1 P ) E ,k Pk P L (1 P )

xk+1 = Axk + buk + Evk


(27)



yk = Cxk + duk + Fvk



I L I O ,k I U I

ek = rk yk , , k



Ik I U (1 I ) O ,k Ik I L (1 I )



uk = uk1 + uk (28)





uk = KP Pk + KI Ik + KD Dk



D L D U D
,k D

Pk = ek ek1 , , k



Dk DU (1 D
) ,k Dk D (1 )
L D


Ik = S ek (29)
s.t. (19)

S Dk = ek 2ek1 + ek2

Depending on the form of the objective function, the final

L

formulation obtained by replacing the nonlinear Eq. (17) by
yk yk yk
U


the set of linear inequalities corresponds to a MINLP (non-

uL u uU

k
linear objective), to a MIQP (quadratic objective) or to a

k k

MILP (linear objective). For the cases where the objective

u uk uU
L

function is linear, solution to global optimal solution is guar-



anteed using currently available software. The same holds

KP,L KP KP,U

true for the more general case where the objective function



KI,L KI KI,U is a convex function.



D,L The proposed formulation addresses the problem of PID
K KD KD,U controller design solely in the time domain. It is important to
44 G. Syrcos, I.K. Kookos / Chemical Engineering and Processing 44 (2005) 4149

emphasize that no approximations have been introduced in 3

the proposed formulation. This is not the case, for instance,
2
when the design is based on the frequency domain where
y(t) 1
conservative (in most cases) approximations are introduced
in order to translate constraints from the time domain into the 0
frequency domain [6,7]. A further advantage of the proposed
formulation is the fact that the problem can be solved to -1
0 50 100 150
global optimality using direct optimization. Classical control Time (min)
design techniques do not share these desirable characteristics 0

[6]. -0.5

u(t)
-1

3. Case studies -1.5

In order to apply the proposed methodology the control
engineer needs to supply the model of the process under
investigation in discrete form (see Eq. (1)), an initial con-
troller tuning (that can be obtained using any classical tech-
nique) and a performance criterion. Then, the optimization
problem can be solved using any commercial optimization
software. In this work GAMS [8] interface to CPLEX and
MINOS solvers was used to solve the proposed formulation.
All results were obtained on an Intel P4 2.66 GHz/512 Mb
computer and the computational time was of the order of
few seconds.
3.1. Control of integrating processes with time delay
(IPTD)
Integrating processes with time delay are among the most
commonly encountered processes. Liquid levels, for in-
stance, exhibit integrating behavior. A typical plant involves
a large number of liquid level loops whose performance
affects strongly the overall performance of the regulatory
control systems. The transfer function of an integrating
process with time delay is the following:
kP ds
g(s) = e (30)
s
PID controller tunings for this model have been given
by a number of researchers [913]. Chen and Fruehauf
[9] have given an industrial example of the level control
in a distillation column where the open loop dynamics
follows the IPTD model with parameters kP = 0.2 and
d = 7.4 min.
The proposed methodology has been applied using the
following objective function:
N

for a hypothetical system with z = d = 1.6, kP = 1 and
min = S (|ek | + |uk |) (31)
KP ,KI
k=0
where NS = f = 300 min. A unit step disturbance at t =
0 min was assumed followed be a unit step change in the set
point at t = 150 min. The optimization problem was solved
with = 1/ S and S = 0.2 min giving
KP = 0.553, KI = 0.0178 (32)
-2
0 50 100 150
Time (min)
Fig. 1. Load response of the closed loop IPDT system (proposed: con-
tinuous line, Luyben [10]: dotted line).
Figs. 1 and 2 compare the closed loop behavior obtained
using the tuning proposed in this work and the tunings
proposed by Tyreus and Luyben [10]. Fig. 1 shows that a
significant improvement in the closed loop load response is
obtained when the proposed methodology is used. The value
of the objective function (Eq. (31)) is = 59.4 for the pro-
posed and = 197.7 for the Luybens tunings. Fig. 2 com-
pares the two methodologies for a unit step set point change.
The responses are comparable and the proposed method
results in = 28.6 while for Luybens tunings = 29.7.
3.2. Systems with inverse response and time delay
Luyben [14] has recently investigated a class of mod-
els that involve both inverse response and time delay. More
specifically he has considered the following type of pro-
cesses:
1 z s ds
g(s) = kP e (33)
(1 + P s)2
that have a positive zero at s = +1/ z as well as time a
delay d. Using the objective function given by Eq. (31) with
= 1/ S and S = 0.1 we obtained
KP = 0.356, KI = 0.154 (34)
P = 1. A unit step disturbance at t = 0 min was assumed
followed be a unit step change in the set point at t = 50 min.
Fig. 3 compares the closed loop response obtained using the
proposed tunings with the tunings given by Luyben [14].
Luybens tunings correspond to = 19.5 while the ones
proposed in this work to = 16.7. It is interesting to note
that the proposed tunings result in improved load as well as
set point response.
 G. Syrcos, I.K. Kookos / Chemical Engineering and Processing 44 (2005) 4149 45

1.5

y(t) 1

0.5

0
0 50 100 150
Time (min)
1

0.5
u(t)
0

-0.5
0 50 100 150
Time (min)

Fig. 2. Unit step set point response for IPDT system (proposed: continuous line, Luyben [10]: dotted line).

3.3. Control of integrating processes with time delay and
inverse response (IPTD&IV)
The open loop transfer function of an IPDT&IV process
is the following:
1 z s ds
g(s) = kP e (35)
s(1 + P s)
and combines the adverse effects associated with having a
right half plane zero at 1/ z , a pole at zero and a time delay
d. Recently, Luyben [15] proposed a successful and simple
enough tuning techniques for PID controllers and IPDT&IV
systems.
In Figs. 4 and 5 the closed loop response obtained using
the tunings proposed by Luyben (PI controller with kC =
0.854 and I = 23) is compared with the one obtained using
the criterion given by Eq. (31) (kC = 1.117 and I = 6.32)
with = 1/ S and S = 0.025 for the system:
1 0.418s 0.1s
g(s) = 0.457 e (36)
s(1 + 1.06s)
It is clear that for the set point change the two method-
ologies offer comparable performance ( = 7.4 for the
proposed and = 6.6 for Luybens method). However, as it
becomes clear from inspecting Fig. 4, the proposed method-
ology offers massive improvement for the load response

0.5
y(t)

0 10 20 30 40 50 60 70 80 90 100
Time
0.5

u(t) 0

-0.5

-1

-1.5
0 10 20 30 40 50 60 70 80 90 100
Time

Fig. 3. Closed loop responsecase study 2 (proposed: continuous line, Luyben [14]: dotted line).
46 G. Syrcos, I.K. Kookos / Chemical Engineering and Processing 44 (2005) 4149

1.5 1.5

1
y(t) y(t) 1
0.5
0.5
0
0
-0.5 0 5 10 15 20 25 30 35 40 45 50
0 10 20 30 40 50 60 70 80 90 100
Time (min) Time (min)
0.5
1
0
-0.5 0.5
u(t) u(t)
-1
0
-1.5
-2 -0.5
0 10 20 30 40 50 60 70 80 90 100 0 5 10 15 20 25 30 35 40 45 50
Time (min) Time (min)

Fig. 4. Load response of the closed loop IPDT&IV system (proposed: Fig. 5. Unit step set point response for IPDT&IV system (proposed:
continuous line, Luyben [15]: dotted line). continuous line, Luyben [15]: dotted line).

case. The value of the objective function for the Luybens namics. They have, among other case studies, considered
method ( = 28.2) is 3.5 times the value of the objective the following system:
function obtained using the proposed method ( = 8.1). e0.5s
g(s) = (37)
(5s 1)(2s + 1)(0.5s + 1)
3.4. Control of an unstable process
and have demonstrated clearly the superiority of their
Most chemical processes exhibit stable open-loop behav- method over most of the available tuning techniques. The
ior. However, there are some important processes, such as same process was considered here with the objective func-
chemical and biological reactors, that might be operated tion given by Eq. (31) ( = 0.1 and S = 0.01) and the
around an unstable steady state. The tuning of PID controller proposed method gave the following tunings:
for unstable systems has recently attracted attention [1619]. KP = 6.905, K = 1.707, KD = 11.824 (38)
Lee et al. [18] have presented a powerful IMC-based tun-
ing methodology that appears to surpass most of available which are very close to the result reported by Lee et al.
tuning techniques for unstable systems with low order dy- [18]. In Fig. 6, the closed loop response obtained using the

0.2

0.15

0.1
y(t)
0.05

-0.05
0 2 4 6 8 10 12 14 16 18 20
Time
0

-0.5

u(t) -1

-1.5

-2
0 2 4 6 8 10 12 14 16 18 20
Time

Fig. 6. Load response of the open-loop unstable system (proposed: continuous line, Lee et al. [18]: dotted line).
 G. Syrcos, I.K. Kookos / Chemical Engineering and Processing 44 (2005) 4149 47

0.8

0.6
y(t)
0.4

0.2

0
0 10 20 30 40 50
Time

0
-0.2
-0.4
u(t)
-0.6
-0.8
-1
0 10 20 30 40 50
Time

Fig. 7. Load response of the FOPTD/PI system (proposed: continuous line, Tang et al. [20]: dotted line).

proposed methodology is compared to that obtained using kP

g(s) = eds (39)
the PID controller parameters reported in [18]. The improve- P s + 1
ment obtained using the proposed methodology is again
clear. By defining the dimensionless parameter = d/ P
Eq. (39) can be written in the following form:
kP s
3.5. Control of first order processes with large time delay g(s) = e (40)
(FOPTD) s + 1
where s corresponds to time scaled with respect to P .
The transfer function of a first order plus time delay pro- has been used as a measure of the difficulty in controlling
cess is given by FOPTD systems [2]. Systems with close to zero are easy

0.8

0.6
y(t)
0.4

0.2

0
0 5 10 15 20 25 30
Time

0
-0.2
-0.4
u(t)
-0.6
-0.8
-1
0 5 10 15 20 25 30
Time

Fig. 8. Load response of the FOPTD/PID system (proposed: continuous line, Tang et al. [20]: dotted line).
48 G. Syrcos, I.K. Kookos / Chemical Engineering and Processing 44 (2005) 4149

1
0.8
0.6
y(t)
0.4
q =2 q =3 q =4 q =5
0.2
0
0 5 10 15 20 25 30
Time
0
-0.2
-0.4
u(t)
-0.6
-0.8 q =2 q =3 q =4 q =5

-1
0 5 10 15 20 25 30
Time

Fig. 9. Load response of the FOPTD/PID system for different values of .

to control while when approaches or exceeds 1 then the
corresponding system becomes progressively more difficult
to control. A process with = 4 and unit gain is selected
in this case study. The optimization problem (using the ob-
jective function given by Eq. (31)) was solved for a PI con-
trol structure and a unit step load disturbance introduced at
t = 0. The solution obtained is
KP = 0.42, KI = 0.163
Tang et al. [20] have also investigated the same case
study. The closed loop performance obtained using the two
methodologies is compared in Fig. 7. A clear improvement
in the performance is observed from Fig. 7 when the pro-
posed technique is used. In Fig. 8 the same comparison is
shown for the case of PID control.
FOPTD with large time delay systems is an important
class of systems for the process industries. A number of
(41) techniques are indeed available for tuning PID controllers

0.5
y(t)
0

-0.5
0 10 20 30 40 50 60
Time
0

-0.5
u(t)
-1

-1.5
0 10 20 30 40 50 60
Time

Fig. 10. Load response of the FOPTD/PID system using the proposed (continuous line), CM (dash dotted line), and the IMC method (dotted line) for
= 10.
 G. Syrcos, I.K. Kookos / Chemical Engineering and Processing 44 (2005) 4149
that feature some form of predictive capability. However, a
limited number of tuning techniques are available for sim-
ple PID controllers. The proposed mathematical formulation
was solved for a number of process models with [2, 5]
for unit load disturbances and the following tuning rules
were derived:
 
1 [1] K.J. strm, T. Hgglund, The future of PID control, Control Eng.
kC kP = 0.31 + 0.6 (42)
I Society of America, New York, 1995.
= 0.777 + 0.45 (43)
P controller tuning, J. Process Control 13 (2003) 291309.
 2.2
D 1
= 0.44 0.56 (44)
P [5] H.P. Williams, Model Building in Mathematical Programming, third
In Fig. 9 a number of responses to a unit step load dis-
turbances are shown. The performance of the closed loop
system is acceptable since the controller eliminates quickly
the deviation with a reasonable amount of oscillation.
Based on the success of the proposed tuning formulas
(Eqs. (42)(44)) within the limits of for which they were
derived an extreme test was used to investigate its applicabil-
ity by selecting a process with = 10. Luyben [21] has also
applied different PID controller forms and tuning techniques
such as the IMC method [7] and a method reported by Mar-
lin [22] (CianconeMarlin or CM method) to this process
and the reader is referred to [21] for an interesting discus-
sion on the subject. The different closed loop responses to a
unit step load disturbance are compared in Fig. 10. For the
CM method the PID2 from [21] was used to derive the sim-
ulation results (kC = 0.3, I = 5, D = 2.7, a = 0.1). For
the IMC method the PID3 from [21] was used to derive the
simulation results (kC = 0.48, I = 6, D = 0.83, F = 1).
The proposed tuning formulas offer, to our surprise, an im-
proved closed loop response despite the fact that they were
applied to a case which is well outside the limits for which
they were derived.
4. Conclusions
This paper presents a general mathematical programming
formulation the can be used to obtain customized tuning for
PID controllers. A reformulation of the initial NLP prob-
lem is presented that transforms the nonlinear formulation
to a linear one. In the cases where the objective function is
convex then the resulting formulation can be solved easily
to global optimality. The usefulness of the proposed formu-
lation is demonstrated in five case studies where some of
the most commonly used models in the process industry are
employed. It was shown that the proposed methodology of-
fers closed loop performance that is comparable to the one
49
achieved by using the best available tunings for a number
of simplified but yet challenging process models.
References
Practice 9 (2001) 11631175.
[2] K.J. strm, T. Hgglund, PID Controllers, second ed., Instrument
[3] S. Skogestad, Simple analytic rules for model reduction and PID
[4] C.A. Floudas, Nonlinear and Mixed-Integer Optimisation, Oxford
University Press, New York, 1995.
ed., Wiley, Chichester, 1997.
[6] S. Skogestad, I. Postlethwaite, Multivariable Feedback Control, Wi-
ley, Chichester, UK, 1996.
[7] M. Morari, E. Zafiriou, Robust Process Control, Prentice Hall, New
York, 1989.
[8] A. Brooke, D. Kendrick, A. Meeraus, R. Raman, GAMS, a
Users Guide, GAMS Development Corporation, Washington, DC,
1998.
[9] I. Chen, P.S. Fruehauf, Consider IMC tuning to improve controller
performance, Chem. Eng. Prog. 10 (1990) 3341.
[10] B. Tyreus, W.L. Luyben, Tuning PI controllers for integrator/dead
time processes, Ind. Eng. Chem. Res. 31 (1992) 26252628.
[11] I.K. Kookos, A.I. Lygeros, K.G. Arvanitis, On-line pi controller
tuning for integrator/dead time processes, Eur. J. Control 5 (1999)
1931.
[12] W. Zhang, X. Xu, Y. Sun, Quantitative performance design for
integrating processes with time delay, Automatica 35 (1999) 719
723.
[13] M. Chidambaram, R. Padma Sree, A simple method of tuning PID
controllers for integrator/dead-time processes, Comput. Chem. Eng.
27 (2003) 211215.
[14] W.L. Luyben, Tuning of proportional-integral controllers for pro-
cesses with both inverse response and deadtime, Ind. Eng. Chem.
Res. 39 (2000) 973976.
[15] W.L. Luyben, Identification and tuning of integrating processes with
deadtime and inverse response, Ind. Eng. Chem. Res. 42 (2003)
30303035.
[16] G.E. Rotstein, D.R. Lewin, Simple PI and PID tuning for open-loop
unstable systems, Ind. Eng. Chem. Res. 30 (1991) 18641875.
[17] H.P. Huang, C.C. Chen, Control system synthesis for open-loop
unstable processes with time delay, IEE Control Theory Appl. 144
(1997) 334338.
[18] Y. Lee, J. Lee, S. Park, PID controller tuning for integrating and
unstable processes with time delay, Chem. Eng. Sci. 55 (2000) 3481
3493.
[19] V. Saraf, F. Zhao, B.W. Bequette, Relay autotuning of
cascaded-controlled open-loop unstable reactors, Ind. Eng. Chem.
Res. 42 (2003) 44884494.
[20] W. Tang, S. Shi, M. Wang, Simple autotuning PID control for large
time-delay processes and its application to paper basis weight control,
Ind. Eng. Chem. Res. 41 (2002) 43184327.
[21] W.L. Luyben, Effect of derivative algorithm and tuning selection on
the PID control of dead-time processes, Ind. Eng. Chem. Res. 40
(2001) 36053611.
[22] T.E. Marlin, Process Control, McGraw Hill, New York, 1995.
 Chemical Engineering and Processing 44 (2005) 5158

Separation of acetic acidwater mixtures through acrylonitrile

grafted poly(vinyl alcohol) membranes by pervaporation
N. Alghezawi a , O. Sanl a, , L. Aras b , G. Asman a
a Gazi Universitesi Fen-Edebiyat Fakltesi Kimya Blm 06500 Teknik Okullar Ankara, Turkey
b Orta Dogu Teknik Universitesi Fen-Edebiyat Fakltesi Kimya Blm, Ankara, Turkey

Received 19 May 2003; received in revised form 29 March 2004; accepted 30 March 2004
Available online 15 June 2004

Abstract

The pervaporation separation of acetic acidwater mixtures was performed over a range of 1090 wt.% acetic acid in the feed at temperatures
ranging 2550 C using acrylonitrile (AN) grafted poly(vinyl alcohol) (PVA) membranes. The permeation and separation characteristics of
PVA-g-AN membranes were studied as a function of membrane thickness, feed composition, operating temperature and pressure. When the
downstream pressure increased permeation rate increased with decreasing separation factor and at high acetic acid concentrations PVA-g-AN
membranes shows grater tendency for the separation of acetic acidwater mixtures. Depending on the membrane thickness, feed composition
and temperature PVA-g-AN membranes gave separation factors 2.314 and permeation rates 0.181.17 kg/m2 h. It was also determined that
PVA-g-AN membranes were found to have lower permeation rate and grater separation factors than PVA membranes.
2004 Elsevier B.V. All rights reserved.

Keywords: Membranes; Selectivity; Separations; Transport processes; Pervaporation; Acetic acidwater mixtures; Poly(vinyl alcohol) membranes; Graft
copolymerization

1. Introduction PVA is a possible candidate for the separation of aque-

Pervaporation is a membrane separation process where
the liquid feed mixture is in contact with the membrane in
the upstream under atmospheric pressure and permeate is re-
moved from the downstream as vapor by vacuum or a swept
inert gas. Most of the research efforts of the pervaporation
have concentrated on the separation of alcoholwater sys-
tem [120] but the separation of acetic acidwater mixtures
has received relatively little attention [2134]. Acetic acid is
an important basic chemical in the industry ranking among
the top 20 organic intermediates. Because of the small dif-
ferences in the volatilitys of water and acetic acid in dilute
aqueous solutions, azeotropic distillation is used instead of
normal binary distillation so that the process is an energy
intensive process. From this point of view, the pervapora-
tion separation of acetic acidwater mixtures can be one of
the alternate processes for saving energy.
Corresponding author. Tel.: +90-312-2122900;
fax: +90-312-2122279.
E-mail address: osanli@gazi.edu.tr (O. Sanl).
ous mixtures [5,7,1214,2125,35] because of its good
chemical stability, film forming ability and high hy-
drophilicity, except for its poor stability in aqueous solu-
tion. Therefore PVA must be insolubilized by crosslink-
ing, grafting or other modification reactions to create
a stable membrane with good mechanical properties
and selective permeability to water. Huang and Yeom
[23,24] studied the separation of ethanolwater and acetic
acidwater mixtures using amic acid as the crosslink-
ing agent. Yeom and Lee [36] prepared PVA membranes
by crosslinking with gluteraldehyde for the pervapo-
ration of acetic acidwater mixtures. Aminabhavi and
Naik [37] prepared poly(vinyl alcohol)-g-poly(acrylamide)
copolymeric membranes to separate acetic acidwater
mixtures.
Acrylonitrile based membranes were also used in acetic
acidwater separation. Lee and Oh [11] copolymerized
4-vinylpyridine with acrylonitrile in order to prepare a
membrane for the dehydration of wateracetic acid mixture
by pervaporation. Yoshikava et al. [25] reported that mem-
branes prepared from poly (acrylic acid-co-acrylonitrile)

0255-2701/$ see front matter 2004 Elsevier B.V. All rights reserved.
doi:10.1016/j.cep.2004.03.007
52 N. Alghezawi et al. / Chemical Engineering and Processing 44 (2005) 5158

were effective in the selective separation of water from

acetic acidwater mixtures.
In the present study we have aimed to separate acetic
acidwater mixtures using AN grafted PVA membranes and
studied the permeation characteristics of PVA-g-AN mem-
branes as a function of membrane thickness, temperature,
feed composition and pressure.

2. Experimental

2.1. Materials

PVA (Merck) with molecular weight of 72,000 and degree

of saponification of 98.599.2%, was used as supplied, AN
(Merck) was purified by reduced pressure distillation. Ceric
ammonium nitrate (CAN) (Merck) was used after vacuum
drying. Acetic acid (Merck), dimethylformamide (DMF) and
dimethylsulfoxide (DMSO) (Merck) were used without fur-
ther purification. Other reagents used in the study were also
all Merck products.
2.2. Synthesis of PVA-g-AN
The reaction was carried out in a three-necked flask
equipped with stirrer, condenser and nitrogen gas inlet. Five
grams of PVA was dissolved in 95 mL of distilled water at
80 C. Fifteen grams of AN was added to the solution. Stir-
ring was continued for 1 h at room temperature under N2
atmosphere. Ten mililiter portion of 1.0 102 M HNO3
containing 10 mL, 1.0 102 M of CAN was added to the
reaction vessel and polymerization was carried out for a pe-
riod of 3 h at 30 C. The reaction was quenched with 10 mL
of aqueous ferrous salt solution (FeSO4 ) (5.0 102 M)
and pH of graft copolymer latexes were adjusted to 1.5 with
1.0 M HCl solution, then dried in air at room temperature.
The product was extracted with water at 80 C then with
DMF at 55 C for 2 days to remove the unreacted PVA and
polyacrylonitrile (PAN) homopolymer. After the extraction,
graft copolymer was dried in vacuum oven at 60 C [5,38].
The proposed grafting mechanism between PVA and AN is
given in Fig. 1.
Fig. 1. Proposed reaction mechanism between PVA and AN for the
synthesis of PVA-g-AN copolymer.
downstream side was kept at 23 Pa by a vacuum pump
(Vacuu Brand RD 15, GMBH Co., Germany). Pervapora-
tion experiments were conducted at constant temperatures
ranging 2550 C using different feed compositions. Upon
reaching steady state flow conditions, the permeate vapor
was collected in liquid nitrogen traps.
The composition of permeates was deduced refractomet-
rically by digital differential refractometer (Atago DD-5,
Atago Co. Ltd., Japan).
2.5. Representation of the results
The permeation and separation characteristics of
PVA-g-AN membranes toward acetic acidwater mixtures
were expressed as permeation rate (flux) (Q), separation
factor () and pervaporation separation index (PSI).
The permeation rate, Q was determined by using the equa-
tion
m
Q= (1)
At
where m, A and t represent the weight of the permeate (kg),

2.3. Preparation of membranes

1.2% (w/v) solution of PVA-g-AN in DMSO were

prepared. The solutions of 2030 mL were poured on to
petri-dishes (9.2 cm in diameter) and the solvent was evap-
orated at 80 C to form the membrane (15 m40 m).

2.4. Pervaporation experiment

The apparatus used in this study were illustrated in Fig. 2.

The pervaporation cell was a two compartment cell with a Fig. 2. Schematic diagram of the pervaporation apparatus: (1) vacuum
150 mL upper compartment, 75 mL of lower compartment. pump; (24, 6) permeation traps; (5) Mc Leod manometer; (7) vent; (8)
Effective membrane area was 12.5 cm2 . The pressure at the permeation cell; (9) constant temperature water bath.
 N. Alghezawi et al. / Chemical Engineering and Processing 44 (2005) 5158 53

effective membrane area (m2 ) and operating time (h) respec- Table 2
Effect of membrane thickness on permeation rate and separation factor
tively.
Separation factor was calculated using: Thickness (m) Q (kg/m2 h)

15 Low mechanical strength

YH2 O /YCH3 COOH
= (2) 25 1.09 2.30
XH2 O /XCH3 COOH 35 0.66 3.00
40 0.11 5.50
where XH2 O , XCH3 COOH , YH2 O and YCH3 COOH denote the 20 wt.% CH3 COOH; P = 23 Pa, T = 30 C.
weight fractions of water and acetic acid in the feed and in
the permeate, respectively.
Pervaporation separation index (PSI), which is a mea-
sure of the separation ability of a membrane was defined by 3.2. Effect of membrane thickness
Huang and Yeom [7] and expressed as the product of sepa-
ration factor and permeation rate. Membranes with varying thickness (1540 m) were pre-
pared from the copolymer by casting method. The perme-
PSI = Q (3) ation rate and separation factor as a function of membrane
thickness was studied for 20 wt.% acetic acid solutions at
30 C and the results were presented in Table 2. As reflected
from the table as the membrane thickness increases perme-
3. Results and discussion
ation rate decreases whereas separation factor increases as
expected from the Ficks first law [39].
3.1. PVA-g-AN copolymer
Different results concerning the effect of membrane thick-
ness on pervaporation performance was reported in the lit-
PVA-g-AN copolymer was synthesized by using PVA and
erature [39,40,42].
AN in a weight ratio of (PVA/AN) (1/2.5), (1/3.0) and (1/5.0)
Koops et al. [39] investigated the pervaporation selectiv-
in the polymerization solution; compositions of the copoly-
ity as a function of membrane thickness for the polysulfone,
mer were determined by elemental analysis (Table 1). FTIR
poly(vinyl chloride) and poly(acrylonitrile) membranes in
spectrum of the copolymer was taken (Fig. 3). The stretch-
the dehydration of acetic acid and reported that selectiv-
ing vibrations of OH and CH appears at 3340 cm1 and
ity decreases with decreasing membrane thickness, below a
29472910 cm1 , respectively, and stretching vibration of
limiting value of about 15 m.
CN appears at 22402225 cm1 were taken as the evidence
Aptel et al. [40] observed a reduction of separation fac-
of grafting.
tor as a function of decreasing membrane thickness for
grafted polytetrafluoroethylene films using a water/dioxane
Table 1 mixture. They have shown in a different study [41] that
Elemental analysis of PVA-g-AN transport rate is inversely proportional to thicknes in
the pervaporation separation of water through poly(tetra
Membrane PVA/AN (w/w) C (%) H (%) N (%)
fluoroethylene)poly(4-vinylpyridine) membranes.
PVA-1 1/2.5 59.71 6.56 13.62 Brun et al. [42] studied the influence of membrane thick-
PVA-2 1/3.0 60.47 7.13 13.80
ness on the separation factor using nitrile rubber membranes
PVA-3 1/5.0 60.77 7.34 16.29
for 60/40 mixture of butadiene and isobutene. They have
concluded that the separation factor was constant above
100 m membrane thicknesses.
In this present study membranes of 35 m thicknesses
were preferred in the rest of the study due to their acceptable
flux and separation factor.

Table 3
Comparison of PVA and PVA-g-AN membranes
T ( C) Q (kg/m2 h)

PVA PVA-g-AN PVA PVA-g-AN

25 0.95 0.18 1.76 2.58

45 4.47 1.10 1.23 3.03
50 5.70 1.17 1.05 3.03
Fig. 3. FTIR spectrum of PVA-g-AN. 20 wt.% CH3 COOH; P = 23 Pa, t = 1.5 h.
54 N. Alghezawi et al. / Chemical Engineering and Processing 44 (2005) 5158

0.5 16. 0 hydrogen bonding between acetic acid and CN groups of

AN; so it becomes difficult for acetic acid molecules to dif-
14. 0
0.4 fuse through the membrane and separation factor toward wa-
12. 0 ter increases. Decrease in permeation rate can be explained
10. 0 by the dense structure of the grafted membrane. In Fig. 4
Q (kg/m 2 h)

0.
8.0
permeation rate of grafted membranes at different grafting

0.2
ratio with respect to the feed composition was given. As it
6.0
seen from the graph although permeation rate decreases with
4.0 the amount of AN separation factors were almost the same.
0.1
2.0
3.4. Effect of temperature on pervaporation
0 0.0
0 10 30 50 70 90 100
The effect of temperature on the separation performance
CH3 COOH (wt %) of PVA-g-AN (PVA-2) membranes was studied in the tem-
perature range of 2550 C and the results were given in
Fig. 4. Effect of PVA:AN ratio on the permeation rate and separation
factor (T = 30 C): () (PVA-2); () Q (PVA-2); () (PVA-1);
Table 4. As it is reflected from the table, as the tempera-
() Q (PVA-1). ture increases permeation rate increases, however, separa-
tion factor stays constant above 30 C.
Table 4 Aminabhavi and Naik [37] obtained similar results in the
Effect of temperature on permeation rate and separation factor
pervaporation of acetic acidwater mixtures by using PVA
T ( C) Q (kg/m2 h) membranes. They reported that for 20 wt.% acetic acid so-
25 0.18 2.60 lutions separation factor stayed constant at 35 and 45 C al-
30 0.66 3.00 though permeation rate increased with temperature.
40 1.07 3.00 The permeation at various operating temperatures is
50 1.17 3.00
thought to be governed by three factors [7,23].
20 wt.% CH3 COOH; P = 23 Pa, t = 1.5 h.
1. Change in the free volume of the polymer membrane
swollen by pure components, which affects mainly per-
3.3. Effect of AN grafting on the performance of the PVA meation rates.
membranes 2. Change in the free volume due to the plasticizing effect.
3. Change in the interaction between permeants in the mem-
In order to understand the effect of grafting on the perfor- brane.
mance of PVA membranes, PVA-g-AN (PVA-2) membranes
were compared with the heat treated PVA membranes that According to the free volume theory as the temperature
were prepared by casting method from 7% (w/v) PVA solu- increase the frequency and amplitude of chain jumping (i.e.
tion (Table 3). thermal agitation) increase and the resulting free volumes
As it is reflected from the table that the presence of AN de- become larger. Increase in temperature also decreases the
creases the permeation rate while increasing the separation interaction between acetic acid and water molecules so it
factor. Increase in the separation factor is caused from the will be easy for both acetic acid and water molecules to

0. 5 16 .0

14 .0
0. 4
12 .0

0. 3 10 .0
Q (kg/m 2 h)

8. 0

0. 2 6. 0

4. 0
0. 1
2. 0

0. 0 0. 0
0 10 30 50 70 90 100
CH 3 COOH (wt %)

Fig. 5. Change of total permeation rate and separation factor with feed composition: () Q (30 C); () Q (40 C); () (30 C); () (40 C).
 N. Alghezawi et al. / Chemical Engineering and Processing 44 (2005) 5158 55

0.4 5.0

4.0
0.3
2
Q(CH3COOH) (kg/m h)x10
Q (H2O) (kg/m h)

3.0
2
2

0.2

2.0

0.1
1.0

0 0.0
10 30 50 70 90
CH3COOH (wt %)

Fig. 6. Effect of feed composition on permeation rate of water and acetic acid: () Q (H2 O) (30 C); () Q (H2 O) (40 C); () Q (CH3 COOH) (30 C);
() Q (CH3 COOH) (40 C).

diffuse through the free volumes resulted by the increase
in temperature. This situation leads to constant separation
factors.
3.5. Effect of feed composition
At 30 and 40 C the effect of feed concentration, in the
concentration range of 1090 wt.% of acetic acid solutions,
on the permeation rate and the separation factor was inves-
tigated and the results were given in Fig. 5. From the figure
it is clear that when acetic acid concentration in the feed so-
lution increases the permeation rate decreases, whereas the
separation factor increases. It can also be seen that the sep-
aration factors obtained at low temperatures are higher than
those obtained at high operating temperatures and they in-
crease steadily with the increase in acetic acid content of
feed.
Similar results were obtained in the pervaporation separa-
tion of acetic acidwater mixtures using blended polyacrylic
acidnylon 6 membranes [22], poly(4-vinylpyridine-co-
acrylonitrile) membranes [11] and in the permeation and
separation of aqueous alcohol solutions through PVA-AN-
HEMA latex membranes [5].
However, in the studies of Huang and Yeom [23,24] using
chemically crosslinked PVA membranes, maximum perme-
ation rates were reported at low acetic acid concentrations
(10 wt.%).
Looking at the permeation rate of individual components
(Fig. 6), the permeation rate of water component decreases
with the decrease in the water content of the feed solution

1.0

0.8

0.6
PSI

0.4

0.2

0.0
0 10 20 30 40 50 60 70 80 90
CH3COOH (wt%)

Fig. 7. Change of pervaporation separation index with feed composition (T = 30 C).

56 N. Alghezawi et al. / Chemical Engineering and Processing 44 (2005) 5158

Table 5
Comparison of the performance of the membranes based on pva for the pervaporation separation of acetic acidwater mixtures
Polymer Temperature ( C) Mass % of acetic Permeation rate Separation factor () Ref. no.
acid in the feed (kg/m2 h)
PVA-g-PAAm 35 80 0.056 3.90 [45]
PVA/PAA (Blend membranes) 25 90 0.300 6.60 [46]
PVA/PHC 25 90 0.140 7.90 [46]
PVA/PVP 25 90 0.800 2.40 [46]
PVA 35 80 0.056 6.71 [37]
PVA 45 80 0.124 6.39 [37]
PVA-g-AAm (48%) 35 80 0.086 5.38 [37]
PVA-g-AAm (93%) 35 80 0.094 3.81 [37]
PVA-g-AN (52%) (PVA-2) 30 90 0.090 14.60 This study
PAAm, poly(acryl amide); PAA, poly(acrylic acid); PHC, poly(hydroxycarboxylic acid); PVP, poly(N-vinyl-2-pyrolidone); PVA, poly(vinyl alcohol); Aam,
acryl amide; AN, acrylonitrile.

and the permeation rate of water is higher than that of acetic
acid regardless of feed composition. These phenomena can
be explained in terms of plasticizing effect of water. As the
water concentration in the feed is increased, the amorphous
regions of the membrane becomes more swollen; hence the
flexibility of polymer chains increases the energy required
for diffusive transport through the membrane decreases. This
could be the reason why low separation factors were ob-
tained at low acetic acid concentrations.
Additionally in pervaporation separation follows the
solution-diffusion mechanism. Therefore the molecular size
of the permeating molecules becomes very important to
characterize the permeation behavior [43]. It is known that
acetic acid has larger molecular size (0.40 nm) than water
molecules (0.28 nm). As the amount of acetic acid increases
in the feed mixture it becomes difficult for acetic acid
molecules to diffuse through the less swollen membrane, so
separation factor increases at high acid concentrations.
Pervaporation separation index values were calculated us-
ing Eq. (3) and presented in Fig. 7. As it is reflected from
the figure that PSI values of the membranes increases with
the acetic acid concentration which shows that the perfor-
mance of PVA-g-AN (PVA-2) membranes are good at high
acetic acid concentrations
For comparison purposes several results regarding
the separation of acetic acidwater mixtures obtained
by various authors with different type of membranes
listed in Table 5. One can see that PVA-g-AN mem-
branes have high selectivity although they have almost the
same permeation rate depending on the feed composition
and temperature than the membranes studied up in the
table.
3.6. Effect of pressure on pervaporation
Effect of the down stream pressure on the permeation rate
and the separation factor was shown in Fig. 8 for 20 wt.%
acetic acid solution at 30 C. Permeation rate increases with
the pressure which complies well with theoretical prediction.
However separation factor decreases as the down stream

0.6 4.0

3.0
0.4
Q (kg/m h)
2

2.0

0.2
1.0

0.0 0.0
2 4 6 8
-3
P (Pa x 10 )

Fig. 8. Effect of down stream pressure on the total permeation rate and separation factor (20 wt.% CH3 COOH; T = 30 C): () Q () .
 N. Alghezawi et al. / Chemical Engineering and Processing 44 (2005) 5158
pressure increases. Similar results were also reported in the
literature [44].
4. Conclusions
Acrylonitrile grafted poly(vinyl alcohol) membranes were
found to be capable of permeating water in preference to
acetic acid from aqueous acetic acid mixtures. The perme-
ation rate of water increases whereas permeation rate of
acetic acid decreases as the water content of the feed in-
creases. It was also determined that increase in temperature
increased the permeation rate without affecting the separa-
tion factor much. As the downstream pressure increased per-
meation rate increased whereas separation factor decreased
and from the PSI values of the membranes it could be said
that especially at high acetic acid concentrations membranes
behaved more separable.
Acknowledgements
Three of grateful to Gazi University Research Fund for
the support of this study.
References
[1] M. Yoshikawa, T. Yukoshi, K. Sanui, N. Ogata, Separation of
water and ethanol by pervaporation through poly(acrylic acid-co-
acrylonitrile) membrane, J. Polym. Sci. Polym. Sci. Ed. 22 (1984)
473475.
[2] M. Yoshikawa, T. Yukoshi, K. Sanui, N. Ogata, Selective separation
of water from waterethanol solution through quarternized poly(4-
vinyl pyridine-co-acrylonitrile) membranes by pervaporation tech-
nique, J. Appl. Polym. Sci. 33 (1987) 23692392.
[3] M. Yoshikawa, T. Yukoshi, K. Sanui, N. Ogata, Selective separa-
tion of waterethanol mixture through synthetic polymer membranes
having carboxylic acid as a functional group, J. Polym. Sci. Part A:
Polym. Chem. 24 (1986) 15851597.
[4] C.E. Reineke, J.A. Jagodzinski, K.R. Denslow, Highly water selec-
tive cellulosic polyelectrolyte membranes for the pervaporation of
alcoholwater mixtures, J. Membr. Sci. 32 (1987) 207221.
[5] G.H. Hsiue, Y.S. Yang, J.F. Kuo, Permeation and separation of
aqueous alcohol solutions through grafted poly(vinyl alcohol) latex
membranes, J. Appl. Polym. Sci. 34 (1987) 21872196.
[6] Y.F. Xu, R.Y.M. Huang, Pervaporation separation of ethanolwater
mixtures using ionically crosslinked blended poly(acrylic acid)
(PAA)-nylon 6 membranes, J. Appl. Polym. Sci. 36 (1988) 1121
1128.
[7] R.Y. Huang, C.K. Yeom, Pervaporation separation of aqueous mix-
tures using crosslinked poly(vinyl alcohol) (PVA). Part II. Permeation
of ethanolwater mixtures, J. Membr. Sci. 51 (1990) 273292.
[8] T. Uragami, H. Shinomiya, Concentration of aqueous alcoholic solu-
tions through a modified silicone rubber membrane by pervaporation
and evaporation, Makromol. Chem. 192 (1991) 22932305.
[9] T. Uragami, T. Morikawa, Permeation and separation characteristics
of alcoholwater mixtures through poly(dimethyl siloxane) mem-
brane by pervaporation and evaporation, J. Appl. Polym. Sci. 44
(1992) 20092018.
57
[10] H. Ohya, K. Matsumoto, Y. Negishi, T. Hino, H.S. Choi, The sepa-
ration of water and ethanol by pervaporation with PVAPAN com-
posite membranes, J. Membr. Sci. 68 (1992) 141148.
[11] Y.M. Lee, B.K. Oh, Pervaporation of wateracetic acid mixture
through poly(4-vinylpyridine-co-acrylonitrile) membrane, J. Membr.
Sci. 85 (1993) 1320.
[12] B. Will, R.N. Lichtenthaler, Comparison of the separation of mix-
tures by vapor permeation and pervaporation using PVA composite
membranes. Part I. Binary alcoholwater systems, J. Membr. Sci. 68
(1992) 119125.
[13] C.K. Yeom, R.Y.M. Huang, Modelling of the pervaporation separa-
tion of ethanolwater mixtures through crosslinked poly(vinyl alco-
hol) membrane, J. Membr. Sci. 67 (1992) 3955.
[14] W.Y. Chiang, C.C. Huang, Separation of liquid mixtures by using
polymer membranes. Part IV. Wateralcohol separation by pervapo-
ration through modified acrylonitrile poly(vinyl alcohol) copolymer,
J. Appl. Polym. Sci. 48 (1993) 199203.
[15] W.H. Jo, H.J. Kim, Y.S. Kang, Separation of waterethanol mix-
ture through poly(acrylonitrile-co-acrylic acid)/poly(ethylene oxide)
membranes by pervaporation, J. Appl. Polym. Sci. 51 (1994) 529
535.
[16] S.Y. Nam, H.J. Chun, Y.M. Lee, Pervaporation separation of
waterisopropanol mixtures using carboxymethylated poly(vinyl al-
cohol) composite membranes, J. Appl. Polym. Sci. 72 (1999) 241
249.
[17] H.C. Park, R.M. Meertens, M.H.V. Mulder, C.A. Smolders, Perva-
poration of alcoholtoluene mixtures through polymer blend mem-
branes of poly(acrylic acid) and poly(vinyl alcohol), J. Membr. Sci.
90 (1994) 265274.
[18] Y.M. Sun, T.L. Huang, Pervaporation of ethanolwater mix-
tures through temperature sensitive poly(vinyl alcohol-g-n-
isopropyacrylamide) membranes, J. Membr. Sci. 110 (1996) 211
218.
[19] E. Ruckenstein, L. Liang, Pervaporation of ethanolwater mixtures
through poly(vinyl alcohol)poly(acrylamide) interpenetrating poly-
mer network membranes unsupported and supported on polyether-
sulfone ultrafiltration membranes: a comparision, J. Membr. Sci. 110
(1996) 99107.
[20] J.W. Rhim, C.K. Yeom, S.W. Kim, Modification of poly (vinyl
alcohol) membranes using sulfursuccinic acid and its application to
pervaporation separation of water/alcohol mixtures, J. Appl. Polym.
Sci. 68 (1998) 17171723.
[21] T.Q. Nguyen, A. Essamri, R. Clement, J. Nell, Synthesis of mem-
branes for the dehydration of wateracetic acid mixture by perva-
poration. Part 1. Polymer material selection, Makromol. Chem. 188
(1987) 19731984.
[22] R.Y.M. Huang, A. Moreira, R. Notarfonzo, X.F. Xu, Pervaporation
separation of acetic acidwater mixtures using modified membranes.
Part I. Blended poly(acrylic acid) (PAA)nylon 6 membranes, J.
Appl. Polym. Sci. 35 (1988) 11911200.
[23] R.Y.M. Huang, C.K. Yeom, Pervaporation separaion of aqueous mix-
tures using crosslinked poly(vinyl alcohol) membranes. Part III. Per-
meation of acetic acidwater mixtures, J. Membr. Sci. 58 (1991)
3347.
[24] R.Y.M. Huang, C.K. Yeom, Development of crosslinked poly(vinyl
alcohol) (type II) and permeation of acetic acidwater mixtures, J.
Membr. Sci. 62 (1991) 5973.
[25] M. Yoshikawa, T. Shimidzu, Y. Maeda, K. Magara, H. Tsugaya,
Pervaporation of aqueous acetic acid mixture through poly(acrylic
acid-co-acrylonitrile) membranes, J. Membr. Sci. 82 (1993) 157162.
[26] I.J. Ball, S.C. Huang, R.A. Wolf, J.Y. Shimano, R.B. Kaner, Perva-
poration studies with polyaniline membranes and blends, J. Membr.
Sci. 174 (2000) 161176.
[27] U.S. Toti, M.Y. Kariduraganavar, K.S. Soppimath, T.M. Aminabhavi,
Sorption, diffusion and pervaporation separation of water/acetic acid
mixtures through the blend membranes of sodium alginate and guar
gum-g-poly(acrylamide), J. Appl. Polym. Sci. 83 (2002) 259272.
58 N. Alghezawi et al. / Chemical Engineering and Processing 44 (2005) 5158
[28] S. Li, V.A. Tuan, R.D. Noble, J.L. Falconer, A Ge-substituted ZSM-
5 zeolite membrane for the separation of acetic acid from water, Ind.
Eng. Chem. Res. 40 (2001) 61656171.
[29] T. Sano, S. Ejiri, K. Yamada, Y. Kawakami, H. Yanagishita, Sepa-
ration of acetic acidwater by pervaporation through silicate mem-
brane, J. Membr. Sci. 123 (1997) 225233.
[30] S.K. Ray, S.B. Sawant, J.B. Joshi, V.G. Pangarkar, Dehydration of
acetic acid by pervaporation, J. Membr. Sci. 138 (1998) 117.
[31] S.P. Kusumocahyo, M. Sudoh, Dehydration of acetic acid by perva-
poration with charged membranes, J. Membr. Sci. 161 (1999) 77
83.
[32] T.M. Aminabhavi, H.G. Naik, Pervaporation separation of wa-
ter/dimethyl formamide mixtures using poly(vinyl alcohol)-g-
poly(acrylamide) copolymeric membranes, J. Appl. Polym. Sci. 83
(2002) 273282.
[33] S.Y. Lu, C.P. Chiu, H.Y. Huang, Pervaporation of acetic acid/water
mixtures through silicate filled polydimethylsiloxane membranes, J.
Membr. Sci. 176 (2002) 159167.
[34] X.P. Wang, Modified alginate composite membranes for the dehy-
dration of acetic acid, J. Membr. Sci. 170 (2000) 7179.
[35] G. Asman, O. Sanl, Ultrafiltration of Fe (III) solutions in the presence
of poly(vinyl alcohol) using modified poly(methyl methacrylate-co-
methacrylic acid) membranes, J. Appl. Polym. Sci. 64 (1997) 1115
1121.
[36] C.K. Yeom, K.H. Lee, Pervaporation separation of wateracetic acid
mixtures through poly(vinyl alcohol) membranes crosslinked with
gluteraldehyde, J. Membr. Sci. 109 (1996) 257265.
[37] T.M. Aminabhavi, H.G. Naik, Synthesis of graft copolymeric mem-
branes of poly(vinyl alcohol) and poly(acrylamide) for the pervapo-
ration separation of water/acetic acid mixtures, J. Appl. Polym. Sci.
83 (2002) 244258.
[38] M.D. Kurkuri, T.M. Aminabhavi, Polyacrylonitrile-g-poly(vinyl al-
cohol) membranes for the pervaporation separation of dimethyl for-
mamide and water mixtures, J. Appl. Polym. Sci. 91 (2004) 4091
4097.
[39] G.H. Koops, J.A.M. Nolten, M.H.V. Mulder, G.A. Smolders, Se-
lectivity as a function of membrane thickness: gas separation and
permeation, J. Appl. Polym. Sci. 53 (1994) 16391651.
[40] P. Aptel, J. Cuny, J. Jozefonvicz, G. Morel, J. Neel, Liquid transport
through membranes prepared by grafting of polar monomers onto
PTFE films. Part II. Some factors determing pervaporation rate and
selectivity, J. Appl. Polym. Sci. 18 (1974) 351364.
[41] P. Aptel, J. Cuny, J. Jozefenvicz, G. Morel, J. Neel, Liquid transport
through membranes prepared by grafting of polar monomers onto
PTFE films. Part III. Steady state distribution in membrane during
pervaporation, J. Appl. Polym. Sci. 18 (1974) 365378.
[42] J.P. Brun, G. Bulvestre, A. Kergreis, M. Guillou, Hydrocarbon sep-
arations with poly membranes. Part I. Butadiene-isobutene separa-
tion with nitrile rubber membranes, J. Appl. Polym. Sci. 18 (1974)
16631683.
[43] G. Asman, O. Sanl, Characteristics of permeation and separation for
acetic acidwater mixtures through poly(vinyl alcohol) membranes
modified with poly(acrylic acid), Sep. Sci. Technol. 38 (2003) 1963
1980.
[44] I. Blume, J.G. Wijmans, R.W. Baker, The separation of dissolved
organics from water by pervaporation, J. Membr. Sci. 49 (1990)
253286.
[45] G.H. Koops, J.A.M. Nolten, M.H.V. Mulder, G.A. Smolders,
Poly(vinyl chloride) poly(acrylonitrile) composite membranes for the
dehydration of acetic acid, J. Membr. Sci. 81 (1993) 5770.
[46] A. Rudin, The Elements of Polymer Science and Engineering, Aca-
demic Press, New York, 1982, Chapter 3, 102 pp.
 Chemical Engineering and Processing 44 (2004) 5969

Countercurrent flow distribution in structured packing

via computed tomography
Shaibal Roy a , A. Kemoun a , M.H. Al-Dahhan a, , M.P. Dudukovic a ,
Thomas B. Skourlis b , Frits M. Dautzenberg b
a Chemical Reaction Engineering Laboratory, One Brookings Drive, Campus Box 1198, Washington University, St. Louis, MO 63130-4899, USA
b ABB Lummus Global, Inc, 1515 Broad Street, Bloomfield, NJ 07003-3096, USA

Received 21 July 2003; received in revised form 9 March 2004; accepted 25 March 2004
Available online 26 June 2004

Abstract

A newly developed -ray computed tomography scanner was implemented to study liquid flow distribution in a 30.48-cm-diameter column
packed with corrugated structure using countercurrent airwater flow. Validation experiments confirm that spatial resolution of 2.5 mm can
be achieved by the new CT unit.
Within the gas (010 cm/s superficial velocity) and liquid (0.62.5 cm/s superficial velocity) flow ranges investigated, a good liquid distri-
bution was observed at all conditions, as manifested by uniformity factor in excess of 70%. The liquid saturation increases with increasing
superficial liquid velocity as well as down the column height. Within the conditions studied, the effect of gas velocity was, in general, found
to be minimal.
2004 Elsevier B.V. All rights reserved.

Keywords: Hydrodynamics; Structured packing; Holdup; Computed tomography

1. Introduction (CT) scanner with -ray source at the Chemical Reaction

Fixed-bed catalytic reactors and reactive distillation
columns are widely used in many industrial processes. Re-
cently, structured packing (e.g., monoliths, katapak, mella-
pak etc.) has been suggested for various chemical processes
[14,14].One of the major challenges in the design and op-
eration of reactors with structured packing is the prevention
of liquid flow maldistribution, which could cause portions
of the bed to be incompletely wetted. Such maldistribution,
when it occurs, causes severe under-performance of reac-
tors or catalytic distillation columns. It also can lead to hot
spot formation, reactor runaway in exothermic reactions,
decreased selectivity to desired products, in addition to the
general underutilization of the catalyst bed.
To overcome these problems, flow distribution in reactors
with structured packing of a desired type needs to be quan-
titatively studied and understood over a desired range of op-
erating conditions. The non-invasive computed tomography
Corresponding author. Tel.: +1 314 935 7187; fax: +1 314 935 7211.
E-mail address: muthanna@che.wustl.edu (M.H. Al-Dahhan).
Engineering Laboratory (CREL), Washington University [5]
has been used successfully in the past to characterize flow
distribution of liquid and gas over non-structured packing
[6]. Hence, it is the tool of choice for studies on structured
packing.
Many studies on the flow distribution in random packed
beds have been reported in the literature. Mercandelli et al.
[8] published a short review of the flow distribution work in
random packed trickle bed, which includes the list of various
techniques used to determine and quantify the flow distri-
bution. Conventional methods include, for example, collect-
ing liquid at the bottom of the column from different zones
while advanced methods include tomographic techniques.
Mercandelli et al. [8] used several techniques to quantify
liquid distribution in columns of diameters up to 30 cm with
three different distributor designs. They used global pres-
sure drop measurements, global residence time distribution
(RTD) of the liquid, local heat transfer probes, capacitance
tomography and a collector at the bottom of the column.
Boyer and Fanget [7] used computed tomography to
quantify liquid distribution in a 60-cm-diameter trickle bed

0255-2701/$ see front matter 2004 Elsevier B.V. All rights reserved.
doi:10.1016/j.cep.2004.03.010
60 S. Roy et al. / Chemical Engineering and Processing 44 (2004) 5969

reactor (downflow of gas and liquid over randomly packed
particles). The apparatus was very similar to that being used
in the Chemical Reaction Engineering Laboratory (CREL)
at Washington University in St Louis, albeit, with a higher
number of detectors (32 of 1.27-cm-diameter BGO detectors
as against 9 of 5-cm-diameter Nal detectors) and increased
radioactive source strength (300 mCi as against 85 mCi).
A comprehensive study was performed to quantify the error
associated with the measurements and the authors were able
to detect solids holdup within 3% accuracy. The authors
also proposed a correlation to predict liquid holdup based
on a force balance on liquid film and showed that tomog-
raphy can reproduce the liquid holdup as predicted by the
model for liquid flow rates up to 1.7 cm/s (no gas flowing).
Yin et al. [15] investigated liquid distribution in a col-
umn of 60 cm in diameter containing pall rings. The authors
found that the distributor type had a significant effect on the
distribution, which was verified using computational fluid
dynamics (CFD).
Examples of the use of tomographic techniques to deter-
mine gas/liquid distribution in structured packing is limited
in the open literature. Mewes et al. [9] studied flow distribu-
tion in monoliths using capacitance tomography. A spatial
resolution of about 510% of the column diameter in the
measurement plane was claimed [912]. Mewes et al. [9]
used water which was introduced through a large number of
injection nozzles into a 120 mm diameter reactor containing
monolith. With increasing liquid velocity, the liquid distri-
bution improved. The authors did not demonstrate the effect
of gas velocity on the liquid distribution.
Toya et al. [13] used X-ray tomography to scan gas/liquid
flow in a corrugated structured packing known as Sulzer
Mellapak 250 Y. The solids profile of the dry structured
packing was reproduced with fair accuracy. The scan was
performed in a 60-cm-diameter column. Very thin liquid film
flowing past the structure at 0.006 m/s was also detected by
the tomography unit.
Iluita and Larachi [4] developed a hydraulic mechanistic
model for prediction of the irrigated pressure drop, liquid
holdup and the effective interfacial area in the preloading
zone of structured packing operated in countercurrent flow
of gas and liquid under partial wetting condition. The model
mimics the two phase flow situation by a network of double
slits consisting of dry and wet slits. The model was success-
ful in predicting the irrigated pressure drop, liquid holdup
and effective interfacial area under various conditions.

Fig. 1. Schematic diagram of the experimental setup: (a) Process diagram (all dimensions in centimeter); (b) packing element; (c) liquid distributor. DP,
differential pressure transmitter; FI, flow indicator; TI, temperature indicator.
 S. Roy et al. / Chemical Engineering and Processing 44 (2004) 5969 61

This work focuses on investigating the flow distribution in

a structured packed bed using gamma ray computed tomog-
raphy. Gas and liquid flow in a countercurrent mode over
a bed packed with corrugated structured packing is usually
used in reactive and catalytic distillation columns. This in-
vestigation includes determining the liquid holdup and its
distribution in the cross-section of the reactor at a range of
gas and liquid velocities.

2. Experimental work

2.1. Experimental setup and operating conditions

The experimental setup, schematically shown in Fig. 1a,

was developed to measure gasliquid flow distribution in
countercurrent flow structured packed bed. It consists of a
packing section, distributor and liquid and gas delivery sys-
tems. The Plexiglas column is 152.4 cm high and 30.48 cm
in diameter and holds four packing elements each 27.3 cm
in diameter and height. The packing is of corrugated type
made of stainless steel and frequently used in modern re-
active distillation columns, commercially known as Norton
packing (Fig. 1b). The corrugations are at a 45 angle to the
vertical axis. The packing elements are held in position us-
ing a donut ring at the bottom of the Plexiglas column. The
liquid is recirculated and introduced at the top through a dis-
tributor having seven holes of 2.5-cm-diameter; the middle
one is covered with a mesh (Fig. 1c).
Water was used as the liquid phase. The liquid delivery
system consists of a feed tank, pump, and a flow indicator.
Water is recycled, as well as added if necessary, to maintain
a constant suction head at the pump. The in-house air system
was used as the gas delivery system through a rotameter.
Air enters the setup at a point below the packing. Two taps
are mounted at the inlet and outlet of the packed column for Fig. 2. Schematic diagram of the newly developed CREL computed
tomography.
pressure drop measurement.
Flow scans were performed at three axial levels (1.5, 2.5
and 3.5D) from the top packing surface (D being the diam- oped by Kumar et al. [5] at CREL. The key system elements
eter of the packing), for various flow conditions. Superficial are illustrated in Fig. 2. The CT scanner consists of an ar-
liquid and gas velocities used, denoted as UL and Ug , re- ray of Nal (TI) detectors with a diameter of 5 cm as well
spectively, are listed in Table 1. as an encapsulated 70 mCi Cs137 source located opposite
to the center of the array of detectors. Nine detectors are
2.2. The newly developed gamma ray computed used in the present study to cover the cross-section of the
tomography (CT) scanner design 30.48-cm-diameter column. The detectors and the source
are mounted on a plate, which can be rotated around the
A newly installed computed tomography (CT) scan- axis of the column by a stepping motor that is controlled
ner was used in this study. This is a modified unit of the through a microprocessor. Moreover, the whole assembly
third-generation fan-beam configuration originally devel- can be moved in the axial direction along the column to
perform a scan at different axial levels of the column. The
source collimator provides a collimated fan beam of 40
Table 1 in the horizontal plane and 5 mm in the vertical plane.
Range of operating conditions available for the experimental setup used.
The detector side collimator provides 2 mm width colli-
UG 0 to 10 cm/s mated -rays to the detector. The movement of the colli-
UL 0.63 to 2.17 cm/s mator rack and detector assembly is controlled by another
Scan Levels: 1.5D, 2.5D, 3.5D from the top, D 27.3 cm. stepping motor. In each movement, this assembly rotates
62 S. Roy et al. / Chemical Engineering and Processing 44 (2004) 5969

by 0.2 , thus producing large number of projections per
detector.
The new scanner introduced the following modifications
(Fig. 3).
2.2.1. Collimator and detectors rotate simultaneously
In the old CT unit, the detectors are stationary in a partic-
ular view and the collimators move by a desired angle to ob-
tain projections for various chords. The intensity of radiation
as detected by the detectors is not uniform; it is maximum
when the collimator passes the center of the detector and
minimum at the two edges. To overcome this non-uniformity
in the new setup, the collimator position has been fixed at
the center of the detector and the whole detectorcollimator
rack is rotated (Fig. 3a).
2.2.2. Eliminating the need to interpolate between detectors
In the old CT unit, the angular distance between two de-
tectors is 0.5 whereas a typical angle between two pro-
jections is approximately 0.2 . Therefore, the source in-
tensity in the space between the detectors has to be in-
terpolated. Hence, two artificial projections are incorpo-
rated into the reconstruction process, which may lead to
some errors. In the new setup, the angular distance between
two consecutive detectors is 0.2 , which is the same as
the angle between two consecutive projections. Therefore,
there is no need to interpolate between the detector spaces
(Fig. 3b).
2.2.3. Collimator slit width and radioactive source strength
In order to achieve the best possible resolution, the thick-
ness of the ray being attenuated should have infinitely small
width. However, this is practically impossible since in that
case, we need to have an infinitely small slit width of the
detector collimator also. If the slit width is too small, the
counts registered by the detectors will be very small, which
in turn will affect the resolution. Generally, the maximum
resolution that can be attained is equivalent to the slit width.
In the new setup, a collimator with a slit width of 2 mm is
used instead of 5 mm, which was the case in old CT unit.
This is expected to result in a higher resolution of the order
of 2 mm (Fig. 3c).
Moreover, the new CT along with all the associated elec-
tronics is mounted on wheels. Being completely made of
aluminum instead of stainless steel, the whole assembly is
lightweight and can be moved around in different locations

Fig. 3. (a) New CT: the whole assembly moves; old CT: only the collimator rack moves; (b) new CT: no interpolation needed; old CT: interpolation
needed between detectors; (c) new CT: 2 mm collimator slit thickness; old CT: 5 mm collimator slit thickness.
 S. Roy et al. / Chemical Engineering and Processing 44 (2004) 5969
Fig. 4. CT validation scan using: (a) jar filled with water; (b) cross-sectional attenuation profile of the liquid in the jar; (c) azimuthally averaged attenuation
in the radial direction (- actual attenuation, average attenuation).
for conducting holdup measurements at various experimen-
tal setups.
The beam attenuation is measured along a number of
beam paths through the column, which originate from dif-
ferent angles. Once a set of attenuation measurements is
completed, the density distribution image is reconstructed
by using the EstimationMaximization algorithm [5]. This
reconstruction algorithm is used, since it has the following
advantages: (1) it accounts for statistical variations associ-
ated with radiation measurements; (2) it readily incorporates
non-uniform beam effects; and (3) it ensures that the final
reconstruction will contain positive values. To obtain statis-
tically significant results, and to reduce the effect of position,
the CT scans were obtained by scanning 360 around the
column using a total scanning time of about five hours and
collecting numerous beam path attenuations (approximately
2500 projections). The final holdup distribution images are
obtained from the attenuation data using the relationships
and the procedure developed by Chen et al. (2000) [6] for
packed bed.
2.3. CT scan evaluation
Since this was a new CT scanner developed to be used
for the first time, it was imperative to evaluate the perfor-
mance of the unit in terms of spatial and density resolution.
Two CT scans were performed to evaluate the system. In the
first case, a 5.08 cm jar was filled with water and scanned
(Fig. 4). In the second case, a hollow ball dipped in water
was scanned and gas holdup was evaluated (Fig. 5). Further,
a third validation, which will be discussed in the following
section of the Results and discussion, was also performed
63
by determining the solids fraction of the packing using com-
puted tomography, which was compared to the one measured
by water displacement method.
For the first case of a jar filled with water, the results are
as follows.
The collimator used in the present CT has an aperture of
2 mm and therefore, theoretically, spatial resolution should
be around 2 mm. In this test case, a 70 mm domain con-
taining 5.1 cm outer diameter glass jar filled with water was
reproduced (as shown in Fig. 4b). The diameter of the ob-
ject obtained from CT was 5.3 cm, which indicates that we
have a spatial resolution of 2 mm in this case. The attenua-
tion in the water region should be uniform. In this case, the
scan gives a slightly non-uniform attenuation as shown in
Fig. 4c. The calculated error in the attenuation coefficient
was found to be within 6.5%.
For the second case, a hollow ball was dipped inside a
jar filled with water and scanned. The diameters of the ball
and jar were 6.98 and 18.95 cm, respectively, as shown in
Fig. 5a. A scan area of 27 cm in diameter was reproduced
using the reconstruction algorithm. The dimensions of the
objects as reproduced by the scan were 6.97 and 19.20 cm,
respectively, as shown in Fig. 5b, which give a maximum
spatial error of about 2.5 mm. This is good enough to re-
solve relatively small maldistribution, if it exists, inside the
30.48-cm-diameter column used in this study. The figure
shows that the overall error in the estimated total holdup is
within 12.8%.
It must be noted that the resolution changes with the size
of the object being scanned due to different attenuation of
the gamma rays. However, the obtained resolutions should
be adequate for a column of 30.5-cm-diameter.
64 S. Roy et al. / Chemical Engineering and Processing 44 (2004) 5969

Fig. 5. CT validation using: (a) hollow ball in water Jar; (b) density distribution of the object; (c) gas holdup profile of the system.

3. Results and discussion
3.1. Solids fraction of the packing
Solids fraction of the packing was first evaluated using CT
at the same axial levels used for the actual flow scans, i.e.,
at 1.5, 2.5 and 3.5D, from the top packing surface, where
D denote the diameter of the packing (30.48 cm). Fig. 6a
shows the solids distribution at those three axial positions.
The azimuthally averaged solids fraction is shown in Fig. 6b
and is approximately equal to 2.5%. The solid fraction was
also measured experimentally by immersing the packing el-
ement in water-filled bath and measuring the volume of wa-
ter displaced, which was found to be approximately 3%.
This overall average value is close to the values obtained
by CT scanning. Considering the error associated with the
measurement of small amount of the displaced water, this
indicates that CT provides adequate measurements of the
solids distribution in the bed despite its small values.
3.2. Liquid distribution in the bed of corrugated structured
packing
Fig. 7ac shows the time-averaged cross-sectional liquid
saturation distribution in the column at superficial gas ve-
locities of 0, 5 and 10 cm/s, respectively. For each gas veloc-
ity, the liquid saturation distribution is shown at three axial
positions and three liquid velocities. Also provided are the
 S. Roy et al. / Chemical Engineering and Processing 44 (2004) 5969 65

Fig. 6. Determination of solid distribution at three axial positions: (a) cross-sectional distribution; (b) azimuthally averaged solids holdup and (c) overall
solids holdup. The horizontal line indicates experimentally determined solids fraction.

time and cross-sectionally averaged liquid saturation values,
L , as well as the uniformity factor expressed as percentage
f (discussed later). Liquid saturation is the ratio of liquid
volume to the bed void volume. It can be seen that, for the
flow conditions used, as superficial liquid velocity increases,
the cross-sectional averaged liquid saturation increases. For
all the superficial gas and liquid velocities used, the liquid
saturation varies between 6 to 15%.
Fig. 7ac also illustrates the liquid saturation distribution
along the length of the column. In general, it is observed
that as the liquid moves downward, liquid saturation in-
creases. At all conditions studied, the uniformity factor
shows that the liquid distribution improves at the bottom
section of the bed (i.e. 3.5D). This could be due to the
liquid distributor design, which produces seven liquid jets
at the top of the bed. These jets disperse while moving
downward. At high liquid superficial velocity (2.17 cm/s),
the uniformity factors are smaller than those obtained at
lower liquid superficial velocities, since, at this high liquid
velocity, the jets require longer bed length to disperse and
distribute.
Fig. 8 shows the time and azimuthally averaged radial
liquid saturation profiles at varying superficial gas and liq-
uid velocities at the middle axial position (2.5D). The figure
shows that liquid saturation is nearly flat, which suggests
a fair uniformity of liquid distribution. Moreover, with in-
creasing liquid velocities, liquid saturation increases. Simi-
lar trends were obtained at all scan heights.
66 S. Roy et al. / Chemical Engineering and Processing 44 (2004) 5969

Fig. 7. Time averaged cross-sectional liquid saturation distribution in structured bed at (a) Ug: 0 cm/s; (b) Ug: 5 cm/s; (c) Ug: 10 cm/s. In the table, L
shows the liquid saturation and f % show uniformity factor percentage.
 S. Roy et al. / Chemical Engineering and Processing 44 (2004) 5969 67

Fig. 7. (Continued ).

Fig. 9 illustrates the effect of superficial gas and liquid ve-

locities on the cross-sectionally averaged liquid saturation at
the middle of the column (2.5D axial position). It is obvious
that the effect of gas velocity on the liquid saturation is not
significant within the range of flows studied. This could be
due to the fact that solid and liquid holdups are very small,
leaving enough space for the gas to flow upwards without
significant interactions with the liquid phase flowing down-
ward.
The measured cross-sectionally averaged liquid holdup
(i.e. measured liquid saturation times the bed voidage) are
consistent with previously reported results [4]. For a sim-
ilar packing structure, Illiuta and Larachi [4] found liquid
holdups, as calculated by their mechanistic model and com-
pared with the reported experimental results, slightly lower
than the values obtained in this study. However, there are
some differences between the two studies: this work was
performed in a smaller diameter column (30 cm versus

Fig. 8. Effect of superficial gas and liquid velocities on the liquid saturation Fig. 9. Effects of gas and liquid superficial velocities on the cross-
radial profile at axial position of 2.5D. sectionally averaged liquid saturation.
68 S. Roy et al. / Chemical Engineering and Processing 44 (2004) 5969
100 cm), and higher superficial gas velocity was used. Il-
liuta and Larachi [4] mentioned that gas flow rate had a
marginal effect on liquid holdup, a fact reflected in this work
as well.
Experiments were also performed to compare the holdup
and flow distribution in a bed, randomly packed with 3 mm
spherical alumina particle, under the same flow conditions
as was done for structured packing. However, it was evident
that the successful operating conditions for structured pack-
ing were too severe for random packed bed, due to very high
pressure drop. For very low liquid velocity (1 mm/s) and
no gas flow, when the experiment was possible, the liquid
distribution was poor as indicated by a low uniformity factor
(40%). However, this information is insufficient to com-
pare the distribution characteristics of structured and random
packings.
3.3. Uniformity factor (f )
The uniformity of the liquid flow distribution across the
column cross-section was quantified in this work. This is
presented in Fig. 7ac in a tabular form along with time and
cross-sectionally averaged liquid saturation values. Various
methods have been used in the literature to quantify the flow
uniformity across a column cross-section. Mercandelli et al.
[8] used a maldistribution factor, which was a variation of
the standard deviation among nine sectors that the flow do-
main was divided into. The factor varied between 0 and 1.
This method is suitable for flow domains, which are divided
into small number of sectors. For cases where large num-
ber of sectors are used (e.g. holdup obtained by tomography
where large number of pixels are used), the maldistribution
values invariably give small values, thus making it impos-
sible to ascertain the quality of distribution. Jiang proposed
the relative standard deviation to quantify the uniformity of
flow [1]. However, since standard deviation can take values
up to infinity, this quantity will also not provide a proper
feel for the uniformity.
In the present work, the degree of uniformity is cal-
culated based on the t-test method. The circular domain
representing the reactor is divided into several sub-domains,
each having a fixed number of pixels. In this work, 10 pix-
els were arbitrarily taken. This constitutes several groups
of data points for liquid saturation; t-test was then carried
out between the group representing the overall averaged
cross-sectional distribution of liquid saturation and the av-
erage of one sub-domain group. If the group passes the
t-test, this indicates that the null hypothesis is satisfied and
the two groups are statistically the same. Then 1 is assigned
to that pair. If the two groups are statistically not simi-
lar, then 0 is assigned. Uniformity factor is then defined
as the percentage number of groups that are statistically
similar (i.e. passes the t-test) to the group representing the
whole cross-sectional distribution of liquid saturation and
its average. In this method, a uniform distribution yields
a uniformity factor of 100%, and a completely maldis-
tributed domain would give a uniformity factor very close
to 0%.
The domain consisted of 60 by 60 pixels (3600) out of
which 2936 pixels represent the actual cross-section of the
column, and these are the pixels which represent the liquid
saturation distribution. Fig. 7ac show the uniformity factor
in percentage (f %) for different liquid and gas superficial
velocities for the scans performed in this work. In most of the
cases, the uniformity factor is large (above 70%), suggesting
a fairly uniform distribution of liquid achieved in the studied
setup.
3.4. Pressure drop in corrugated structured packing
The pressure drop across the length of the packing was
extremely small and was beyond the detectable limit of the
available pressure gauges. The full scale of the available
transducer (validyne differential transducer) was 200 units,
which correspond to a pressure drop of 1 cm of water. In
our study, it was observed that the maximum pressure drop
within the gas and liquid flow rates studied was about 1 mm
of water (20 calibration units). A trend was seen wherein
the pressure drop increases with increasing gas and liquid
velocities.
4. Final remarks
It was established that the newly developed CT unit has
a spatial resolution of about 2 2.5 mm. This is quite ac-
ceptable given the size of the column used in this study. The
solid fraction as determined by CT scan was about 2.5%,
which was close to the value of 3% as determined by water
displacement method.
The liquid saturation increases with increasing superficial
liquid velocity. Moreover, the liquid saturation increases as
the liquid phase moves downward. The liquid distribution
was found to be fairly uniform in general as expressed by the
uniformity factor which was relatively large, between 70 to
95%. Liquid distribution was better at the bottom of the bed,
compared to the upper section. The effect of gas velocity,
was in general, found to be very small at the conditions used
in this study.
Acknowledgements
The financial support provided by ABB Lummus Global,
Inc. is acknowledged which made this work possible.
References
[1] Y. Jiang, Flow distribution and its impact on performance of
packed-bed reactors, Ph.D thesis, Washington University in St Louis,
Missouri, 2000.
 S. Roy et al. / Chemical Engineering and Processing 44 (2004) 5969
[2] R. Krishna, Reactive Separation: a New Paradigm in an Old Bottle,
Report from Chemical Engineering department, University of Ams-
terdam, 1999.
[3] G.G. Podrebarac, F.T.T. Ng, G.L. Rempel, The production of diace-
ton alcohol with catalytic distillation:part II. A rate-based catalytic
distillation model for the reaction zone, Chem. Eng. Sci. 53 (3)
(1998) 1077.
[4] I. Iliuta, F. Larachi, Mechanistic model for structured-packing con-
taining columns: Irrigated pressure drop, liquid holdup and pack-
ing fractional wetted area, Ind. Eng. Chem. Res. 40 (2000) 5140
5146.
[5] S.B. Kumar, M.P. Dudukovic, Jamal Chaouki, Faical Larachi, Milorad
P. Dudukovic, Computer assisted gamma and X-ray tomography:
applications to multiphase flow systems, in: Non-Invasive Monit.
Multiphase Flows, Elsevier, 1997, pp. 47103.
[6] J. Chen, R. Novica, M.H. Al-Dahhan, M.P. Dudukovic, Study of
particle motion in packed/ebullated beds by computed tomography
(CT) and computer automated radioactive particle tracking (CARPT),
AIChE J. 47 (5) (2000) 9941004.
[7] C. Boyer, B. Fanget, Measurement of liquid flow distribu-
tion in trickle bed reactor of large diameter with a new
gamma-ray tomographic system, Chem. Eng. Sci. 57 (2002) 1079
1089.
69
[8] C. Marcandelli, A.S. Lamine, J.R. Bernard, G. Wild, Liquid distri-
bution in trickle-bed reactor, Oil Gas Sci. Technol.-Rev IFP 55 (4)
(2000) 407415.
[9] D. Mewes, T. Loser, M. Millies, Modeling of two-phase flow in
packings and monoliths, Chem. Eng. Sci. 54 (21) (1999) 47294747.
[10] N. Reinecke, D. Mewes, Flow regimes of two phase flow in mono-
lith catalyst, in: Proceedings of 5th World Congress on Chemical
Engineering, San Diego, CA, 1996.
[11] N. Reinecke, G. Petritsch, D. Schmitz, D. Mewes, Tomographic
measurement techniquesvisualization of multiphase flows, Chem.
Eng. Technol 21 (1998) 718.
[12] N. Reinecke, D. Mewes, Oscillatory transient two-phase flows in
single channels with reference to monolithic catalyst supports, Int.
J. Multiphase Flow 25 (67) (1999) 13731393.
[13] D. Toye, P. Marchot, M. Crine, M. Pelsser, G. LHomme, Local
measurements of void fraction and liquid holdup in packed columns
using X-ray computed tomography, Chem. Eng. Process. 37 (6)
(1998) 511520.
[14] L. Spiegel, W. Meier, Distillation columns with structured packings
in the next decade, Chem. Eng. Res. Des. 81 (A1) (2003) 3947.
[15] F. Yin, A. Afacan, K. Nandakumar, K.T. Chuang, Liquid holdup
distribution in packed columns: gamma ray tomography and CFD
simulation, Chem. Eng. Process. 41 (5) (2002) 473483.
 Chemical Engineering and Processing 44 (2005) 7179

Effect of cycling operations on an adsorbed natural gas storage

O. Pupier, V. Goetz , R. Fiscal
IMP-CNRS UPR 8521, Institut de Science et Gnie des Matriaux et Procds, Rambla de la Thermodynamique, 66100 Perpignan, France

Received 18 February 2004; received in revised form 7 May 2004; accepted 7 May 2004
Available online 1 July 2004

Abstract

Adsorbed natural gas (ANG) is an interesting opportunity for developing natural gas vehicles technology. In this case, adsorbents such as
activated carbons are used to store natural gas at moderate pressure, 3.5 MPa, compared to the high-pressure (20 MPa) required for current
compressed natural gas technology. Many studies are devoted to the elaboration of suitable adsorbent materials to optimise the methane storage
capacity. Nevertheless, since natural gas (NG) is composed of about 95% of methane mixed with other components, an important deterioration
of the storage performance is observed after successive cycles of an ANG system (filling and delivery). It is the result of adsorption of the
other components present in NG that are mainly higher molecular-weight hydrocarbons, carbon dioxide and nitrogen. In the present study,
the evolution of the storage capacity of a 2 l vessel cycled with NG is experimentally measured. As a function of the cycle number, the gas
composition at the outlet of the ANG system is determined using gas chromatography analysis.
2004 Elsevier B.V. All rights reserved.

Keywords: Activated carbon; Adsorption; Storage; Natural gas; Cycling operation

1. Introduction nature of the microporous solid. The microporous structure

Natural gas is a possible alternative as a transportation
fuel. Moreover, natural gas vehicles (NVGs) provide bene-
fits as cleanliness and safety [1,2]. However, the disadvan-
tage of NG is its low volumetric energy density compared to
that of conventional liquid fuels. Current gas vehicles em-
ploy storage vessels at high pressure (20 MPa). This implies
high manufacturing and filling costs. Adsorbed natural gas
(ANG) on a suitable microporous adsorbent offers an inter-
esting opportunity of developing NGVs technology. ANG
technology can provide adequate energy density at a low
pressure, that is, 3.5 MPa and at room temperature. Natural
gas storage capacity of an adsorbent is currently expressed
in terms of the delivered volume of natural gas per unit of
volume containing the adsorbent, measured at standard con-
ditions (P = 0.101 MPa, T = 273 K). To be a commercially
attractive system, the final objective is to reach a delivered
quantity equal to 150 (v/v) under real operating conditions
[3]. Numerous aspects occur in ANG storage that affect its
performance, and hence, its viability. The first one is the
Corresponding author. Tel.: +33 4 68 68 22 36;
fax: +33 4 68 68 22 13.
E-mail address: goetz@univ-perp.fr (V. Goetz).
of the activated carbon defines the adsorption capacity per
unit of mass of adsorbent. Many theoretical [4,5] and ex-
perimental [6,7] studies and researches have been devoted
to the definition and the practical elaboration of the microp-
orous texture best adapted for methane storage. The second
one is the packing adsorbent scale that defines the storage
capacity per unit of volume. Densification is necessary to
try to reach the requirement of 150 (v/v) delivered by the
system. Numerous methods of densification are provided
in literature [810]. To be really effective, the densification
procedure should lead to a compromise between a high vol-
umetric capacity and high heat and mass transfer properties.
The third aspect corresponds to the scale of the ANG ves-
sel. Good adequacy is necessary between the design of the
ANG vessel, the transfer properties of the packing adsorbent
and the dynamic working conditions of the storage process
[11,12]. Finally, the last aspect concerns the influence of the
composition of NG during successive cycles of charge and
discharge. This aspect is taken into consideration in the gen-
eral design of a NGVs process through, for example, the
implementation of a guard bed [13,14]. A theoretical study
[15] has shown the determining influence of the gas compo-
sition. Nevertheless, in the literature, very few experimental
results quantify and analyse the evolution of performance

0255-2701/$ see front matter 2004 Elsevier B.V. All rights reserved.
doi:10.1016/j.cep.2004.05.005
72 O. Pupier et al. / Chemical Engineering and Processing 44 (2005) 7179
of a storage vessel during real dynamic cycling operation
with NG [1618]. The purpose of the present work is to in-
vestigate this last point. An automatic cycling experimental
device has been developed. The global storage performance
of the vessel is measured as a function of the cycle number.
For each cycle, the gas composition at the outlet of the ves-
sel during the discharge is determined by gas chromatogra-
phy. The influence on the performance of the system of the
different gas species that compose NG is described.
2. Experimental
2.1. Experimental device and procedure
Fig. 1 shows the experimental laboratory device. A cylin-
drical stainless steel vessel of about 2 l (diameter: 110.3 mm;
length: 215 mm) equipped with a central gas diffuser was
filled with the adsorbent packing. Four thermocouples were
distributed radially throughout the vessel with a spacing of
11 mm and one on the reactor wall to evaluate the thermal
gradient during charge and discharge phases. Heat exchange
through the vessel wall took place with water at 298 K inlet
temperature flowing in a double casing around the reservoir.
Two mass flow controllers (MFC) recorded and controlled
the gas flow at the inlet and the outlet of the vessel. A pres-
sure transducer recorded the pressure in the vessel. At the
outlet of the vessel a micro-gas chromatograph allowed the
analysis of the composition of the delivered gas during dis-
charge. For safety reasons, a safety valve (SV) (calibrated at
4.0 MPa) was on the top of the storage vessel, and the appa-
ratus was placed in a ventilated safety closed room equipped
with gas detectors.
The experimental procedure is described as follows. Be-
fore the start of the first run, the storage vessel was degassed
during several hours at a temperature higher than 423 K at a
pressure lower than 10 Pa. Before the cycling operation, the
data acquisition system was activated. During each charge,
P
MFC
T
NG
VESSEL
Fig. 1. Experimental laboratory device for cycling operation.
0.12
0.10
Q(CH4) adsorbed (g/g)
0.08
0.06
0.04
0.02
0.00
0 1 2 3 4
Pressure (MPa)
Fig. 2. Adsorption isotherm of methane in the activated carbon at 298 K.
a constant volumetric flow rate equal to 25 sl min1 was in-
troduced into the vessel. Charge was considered over when
the pressure in the vessel was equal to 3.5 MPa and when the
difference between the mean temperature of the adsorbent
minus the heat transfer fluid temperature was lower than
0.5 K. During each discharge, an almost constant volumet-
ric flow rate of 1 sl min1 was delivered at the outlet of the
vessel. Every 4 min, the gas composition was measured with
a micro-gas chromatograph. Discharge was considered over
when the pressure in the vessel was lower than 0.15 MPa
and when the difference between the mean temperature of
the adsorbent minus the heat transfer fluid temperature was
lower than 0.5 K. As soon as these conditions were reached
a charge procedure happened.
2.2. Preparation and characterisation of the adsorbent
A commercial activated carbon, manufactured by NORIT
was selected for the cycling operation. It is available in a
grain form with particle sizes in the range of few millimetres.
Fig. 2 presents the adsorption isotherm of pure methane at
micro-GC
MFC
SV
T Micro-Computer
Water
 O. Pupier et al. / Chemical Engineering and Processing 44 (2005) 7179 73

Table 1 Cycle 1 Cycle 5 Cycle 10

Characteristics of the adsorbent composite block with and k, respectively, 4
the heat transfer conductivity and the permeability 90

AC (kg m3 ) ENG (kg m3 ) (W m1 K1 ) k 1012 (m2 ) 75

3
720 90 2.4 0.023
Pressure (MPa)

60

298 K measured with a high-pressure volumetric device [19]
under pressure up to 5 MPa.
From this raw material, a consolidated adsorbent com-
posite block was prepared following a procedure previously
detailed [10] in the case of Maxsorb like activated car-
bon (produced and sold by the Kansai Coke and Chemical
Company Ltd). First, the activated carbon crushed in powder
form with particle sizes around 100 m was coated with a
thermoplastic binder according to Boses patent [20]. Then,
it was mixed with expanded natural graphite (ENG) and
the obtained mixture was compressed directly in the storage
vessel and then heated up to the melting point of the me-
chanical binder. Thereafter, the vessel was air-cooled under
applied stress. The mechanical binder solidification ensures
all mechanical cohesion of the composite. ENG, which is
an excellent thermal binder ensures high values of the heat
transfer conductivity of the composite and of the heat ex-
change coefficient between the composite block and the wall
of the vessel. As discussed in previous papers [10,12], this
procedure of preparation of composite block leads to two
main advantages. The density of the activated carbon in the
composite is increased. The thermal properties are improved
avoiding too much severe thermal limitation during dynamic
charge and discharge.
The two binders, the ENG and the thermoplastic polymer,
are inert concerning the methane adsorption. The methane
quantity delivered depends only on the activated carbon ad-
sorption isotherm and its apparent density in the adsorbent
composite block (Table 1). For pure methane, the quantity
delivered at 298 K between 3.5 and 0.1 MPa is equal to 89
(v/v).
3. Results and discussion
3.1. Cycling operation with pure methane
A preliminary cycling test was performed with pure
methane. It is illustrated in Fig. 3. According to the de-
scribed experimental procedure, in the first cycle, the vessel
is charged from vacuum to 3.5 MPa at 298 K. During the
first discharge, when the pressure decreases from 3.5 to
0.1 MPa, around 20% of the total amount of gas adsorbed
Qdd (V/V)
2
45
30
1
15
0 0
0 10 20 30
Time (hours)
Fig. 3. Cycling operation with pure methane: pressure profiles (continuous
line and left hand scale), adsorbed volume of methane per volume of
composite block (symbol () and right-hand scale).
into the composite at 3.5 MPa is retained. This corresponds
to the difference between the delivered and the stored quan-
tity of methane. As expected, during cycling operation, the
delivered quantity of methane is constant and is not influ-
enced by the number of cycles. This is the result of the
reversible adsorption/desorption phenomena of a pure gas
on an activated carbon. The experimental delivered quantity
(Fig. 3), directly obtained by the integration of the instanta-
neous flow rate of methane at the outlet of the vessel is in
agreement with the calculated one.
3.2. Cycling operation with natural gas
Gaz De France provided natural gas. It is composed of
92.2% of methane. The other components analysed with
the micro-gas chromatograph are: CO2 , N2 , C2 H6 , C3 H8 ,
i-C4 H10 , n-C4 H10 , i-C5 H12 , n-C5 H12 and C6+ (Table 2).
3.2.1. Comparison of three typical cycles
Fig. 4 shows experimental profiles of the mean tempera-
ture, the pressure and the flow rate during charges of the cy-
cles 5, 100 and 700. Whatever the cycle, during the charge
performed at high flow rate, the exothermic adsorption of
the different components of natural gas entails an increase
of the mean temperature of the vessel (Fig. 4a). The range
in variation of the mean temperatures is the consequence
of the coupling between the power delivered by the gas ad-
sorption and the heat transfer inside the composite block.
The decrease of this range during the charge as a function of
the cycle number is a first indication of the evolution of the

Table 2
Natural gas composition (mol%)
CH4 N2 CO2 C2 H6 C3 H8 i-C4 H10 n-C4 H10 i-C5 H12 n-C5 H12 C6+

92.18 1.82 0.75 4.23 0.78 0.13 0.09 0.012 0.006 0.002
74 O. Pupier et al. / Chemical Engineering and Processing 44 (2005) 7179

310 duration necessary to return to 298 K is very short and less

than 10 min. During discharge, the thermal properties of the
composite associated with the low regulated flow rate, al-
low a quite isothermal working mode. The three presented
Mean temperature (K)

305 charges were realised for the same regulated flow rates at the
inlet of the vessel (Fig. 4c). The evolution of the duration
necessary for complete charge as well as the pressure pro-
file (Fig. 4b) demonstrate the decrease of the cycled mass
300 of natural gas at the vessel scale.
Fig. 5 presents in the case of the selected cycles, the
evolution of the molar ratio for three different components.
It is defined as follows,
295
Mi outlet
0 2 4 6 8 10 12 14
ri = ,
(a) Charge time (min) Mi inlet
where Mi outlet and Mi inlet are, respectively, the outlet and
4
the inlet molar percentage of the component i. While the
inlet molar percentage is constant (Table 2), the outlet mo-
lar percentage of each species was measured during the dis-
3 charges with the micro-gas chromatograph. Every 4 min, an
analysis was performed. In order to be able to compare these
Pressure (MPa)

profiles, they are presented as a function of the pressure at

2 the outlet of the vessel.
Whatever the gas and the cycle are, at the beginning of the
discharge, the ratios of all components are all equal to one.
1 This is the consequence of the dynamic cycling operation
with a discharge happening immediately at the end of the
charge. The first volume of gas flowing out of the vessel is
0 not desorbed by the activated carbon, but comes from the
0 2 4 6 8 10 12 14
dead gas volume of the vessel. Its composition is equal to
(b) Charge time (min)
the inlet composition of the gas.
During a discharge, the composition of the gas at the
30
outlet is a consequence of the multi-component adsorption
equilibria between the previously adsorbed phase and the
25 gas phase. The ratio profiles are necessarily linked to the
dynamic working mode of the vessel (like the flow rates
20 during charge and discharge) and the heat and mass transfer
Flow rate (sl/min)

properties of the composite adsorbent. Nevertheless, general

15 tendencies can be observed.
The outlet gas composition is not constant and depends
10 on the level of pressure. Desorption of the high-molecular
weight hydrocarbons happens mainly for pressure lower than
5 1 MPa. Between 3.5 MPa and 1 MPa, the equilibrium condi-
tions between the gas and the adsorbed phase entail values
0 of the different ratios systematically lower than 1. This is
0 2 4 6 8 10 12 14
especially pronounced for the i-butane and the i-pentane.
(c) Charge time (min)
Ratio profiles depend highly on the cycle number. They
Fig. 4. Experimental profiles of the mean temperature (a), the pressure clearly indicate the evolution of the composition of the ad-
(b) and the flow rate (c) during the cycle numbers 5 (), 100 () and sorbed phase during cycling operation. As an example, at
700 (). cycle number 5, whatever the level of the pressure, the
i-pentane (Fig. 5c) is never detected in gas mixture at the
behaviour of the storage vessel. At the end of the charge, as outlet of the vessel. This means that the whole i-pentane in-
soon as the constant flow rate at the inlet of the vessel can- troduced in the vessel during the previous charges remains
not be ensured any more, the mean temperature decreases in the adsorbed phase and is stored inside the activated car-
(Figs. 4a and c). Due to the high values of the heat con- bon. This is not any longer the case for cycles 100 and 700.
ductivity and the heat exchange coefficient at the wall, the As the cycling operation goes on, the composition of the ad-
 O. Pupier et al. / Chemical Engineering and Processing 44 (2005) 7179 75

3.0 Table 3
Evolution of the mean ratio (rm) during all the study
2.5 rmC2 H6 rmi-C4 H10 rmi-C5 H12

Cycle 5 0.80 0.35 0.03

2.0 Cycle 100 1.04 0.78 0.48
Cycle 700 1.04 0.99 0.77
rC2H6

1.5

1.0 or is no longer stored at the scale of the vessel during a suc-

cessive phase of charge and discharge. In the case of ethane,
0.5 at cycle number five, the value of 0.80 for rmC2 H6 points
out that around 20% of the ethane introduced in the vessel
0.0 remains stored during the discharge. At cycle number 100,
3.5 3.0 2.5 2.0 1.5 1.0 0.5 0.0 the value close to 1 indicates that a steady state is reached
(a) Pressure during desorption (MPa) for this gas. It is confirmed by the same value of rmC2 H6 at
the cycle number 700. Moreover, the ratio profile at cycle
3.0 number 700 is strictly identical if compared to that of cycle
100 (Fig. 5a). The i-butane corresponds to the intermediate
2.5 case (Fig. 5b, Table 3). This gas continues to be stored in
the adsorbed phase after cycle number 100. After 700 cy-
2.0 cles, the steady state is reached. Finally, as demonstrated by
its ratio profile (Fig. 5c), even after 700 cycles, the activated
ri-C4H10

1.5 carbon is not saturated by i-pentane with around 20% of this

gas remained in the adsorbent during the discharge phase
1.0 (Table 3).

0.5 3.2.2. Efficiency of the ANG process

The performance of ANG systems is usually defined as
0.0 the volume of methane or natural gas delivered, measured
3.5 3.0 2.5 2.0 1.5 1.0 0.5 0.0 at standard conditions per adsorbent volume. The impact of
(b) Pressure during desorption (MPa) the cycling operation on performances is evaluated by the
efficiency (Fig. 6) defined as follows,
3.0
Qn
= ,
2.5 Q1
where Q1 and Qn are the delivered gas volume per adsor-
2.0
bent volume at cycle number 1 and n, respectively. Q1 is
ri-C5H12

1.5

1.0

0.5

0.0
3.5 3.0 2.5 2.0 1.5 1.0 0.5 0.0
(c) Pressure during desorption (MPa)

Fig. 5. Evolution of the ratio C2 H6 (a), i-C4 H10 (b), i-C5 H12 (c) at the
cycle numbers 5 (), 100 () and 700 ().

sorbed phase changes. It is progressively saturated by the hy-

drocarbons with molecular weight higher than methane and,
as a consequence, the ratio of these components increase.
During the discharges, the mean values (rm) of the ratios
corresponding to the pressures ranging from 3.5 to 0.1 MPa Fig. 6. Experimental profile of the efficiency as a function of the cycle
were calculated (Table 3). These values indicate if the gas is number (H.H. is for heavy hydrocarbons).
76 O. Pupier et al. / Chemical Engineering and Processing 44 (2005) 7179

1.2
1.0
1.0
0.9
0.8
0.8
rmCO2

0.6
0.7
Carbon dioxide ( CO2 )
0.4
0.6
0.2
0.5
0.0
0 10 20 30 40 50 0 10 20 30 40 50
(a) Cycle number (b) Cycle number

Fig. 7. Relation between the mean ratio of CO2 in the gas at the outlet of the vessel (a) and the rate of variation of the efficiency (b).

equal to 83 (v/v), close to the delivered quantity of pure
methane.
First it is interesting to note the continuous decrease of
the efficiency. After 700 cycles, the steady state conditions
are not yet completely reached and the efficiency goes down
very slowly. At the end of the cycling operation, less than a
half of the initial delivered quantity of gas is still cycled by
the storage vessel. This is the result of the adsorption of the
gases other than methane present in the mixture at low per-
centage. Secondly, the rate of decrease is not constant and
depends on the number of cycles already performed. The
straight lines in Fig. 6 show that several rates of variation
can be drawn. These regimes are linked to the storage in the
vessel of the different species that compose natural gas. It is
confirmed by the evolution of the mean ratios of the gas as a
function of the number of cycles. The first rate of variation
corresponds to the progressive saturation of the adsorbent by
the carbon dioxide. During this step CO2 , like all the gases
with a molecular weight higher than methane is stored in the
vessel. As demonstrated by the Fig. 7a, at the scale of the
vessel a steady state is reached for CO2 at a cycle number
around 10. As soon as the mean ratio of CO2 is equal to one,
the rate of variation of the efficiency changes (Fig. 7b). The
second step corresponds to the number of cycle necessary
to reach the steady state for ethane (Fig. 8a), which is after
methane, the lightest hydrocarbon in the gas mixture. Be-
tween the cycles 10 and 30, a second rate of variation char-
acterised by a second straight line can be drawn (Fig. 8b). As
for CO2 , as soon as the quantity of C2 H6 stored in the vessel
at the end of the discharge phase does not change any more,
another regime begins. Saturation of the vessel with propane
that ends at cycle number 150 defines the third rate of vari-
ation (Fig. 9). Then, the evolution of the global efficiency is
closely linked to the successive saturations of the activated
carbon with the different gases. At the end of the cycling op-
eration, the slow decrease of the efficiency is the result of the
slow accumulation of n-butane and gases with higher molec-
ular weight (cf. Appendix A). If the cycling operation had
been continued, the last part of the efficiency profile (Fig. 6)
should probably be decomposed with the different gases still
stored.
The global efficiency of the ANG process is influenced
greatly by the natural gas composition. Nevertheless, the
molar percentage of methane at the outlet of the vessel dur-
ing discharge changes little during the cycling operation
(Fig. 10). It is a little bit higher than 0.92, the gas supply
composition, during the first 50 cycles. Since nitrogen is not
adsorbed at the working levels of pressure and temperature
(cf. Appendix A), this is the consequence of the storage of
CO2 and C2 H6 , the gas having the highest molar percent-
ages in the gas feed. When this process is finished, because
of the very low percentage of the others components of

1.2
1.0
1.0
0.9
0.8
rmC2H6

0.8
0.6 CO2
Ethane ( C2H6 ) 0.7
0.4
0.6
0.2
0.5
0.0
0 10 20 30 40 50 0 10 20 30 40 50
(a) Cycle number (b) Cycle number
Fig. 8. Relation between the mean ratio of C2 H6 in the gas at the outlet of the vessel (a) and the rate of variation of the efficiency (b).
 O. Pupier et al. / Chemical Engineering and Processing 44 (2005) 7179 77

1.2
1.0
1.0
0.9
0.8
rmC3H8

0.8
0.6
Propane (C3H8) 0.7
0.4
0.6
0.2
CO2
0.0
0.5 C2H6
0 50 100 150 200 0 50 100 150 200
(a) Cycle number (b) Cycle number
Fig. 9. Relation between the mean ratio of C3 H8 in the gas at the outlet of the vessel (a) and the rate of variation of the efficiency (b).

95 the system of analysis retained, it was not possible to detect

the admixed sulphur-containing odorants which are present
in natural gas at the level of some parts per million. Like the
CH4 concentration (molar %)
94 alkanes with molecular weight higher than that of methane,
at the outlet of the vessel
they can be suspected to participate in the slow decrease of
CH4
CH4 concentration (molar %) the cycled mass of gas. Adsorption of these components by
concentration 93
(molar %) at the inlet of the vessel the activated carbon is not irreversible. Modifications in the
working conditions make it possible to restore the initial per-
92
formance. As an example, at the end of the discharge when
the pressure is equal to 0.1 MPa, heating the vessel to a tem-
perature equal to 473 K entails a partial desorption of the
91 different species. Following this treatment, a cycle leads to
0 200 400 600 800 an experimental value of the efficiency equal to 0.65. Nev-
Cycle number
ertheless, a complete regeneration of the activated carbon
Fig. 10. Mean molar percentage of methane during discharge as a function necessitates heating the vessel at 473 K under vacuum. Stor-
of the cycles number. age capacity is currently the main limitation to the practical
development of ANG technology. It is therefore crucial to
natural gas, the molar percentage of methane at the outlet is design an efficient process, adapted to the practical use of
almost constant and equal to the inlet one. this storage system. At the present time, a guard bed at the
inlet of the storage vessel is the proposed solution.

4. Conclusion
The obtained experimental results display the great im-
pact of the gas composition on the performance of an ANG
storage system. After 700 cycles, corresponding approxi-
mately to 250,000 km for a vehicle with a fuel tank allowing
400 km of autonomy, the efficiency of the storage system
falls down by 50%. The number of cycles necessary to reach
a steady state working mode, as well as the accurate value of
the final performance, probably depends on the composition
of natural gas and the dynamic working mode selected for
the cycling operation. Moreover, the microporous structure
of the activated carbon should influence the result. Never-
theless, basically, the final performance is the result of the
multi-component adsorption equilibriums at a temperature
of 298 K for the two characteristic working pressures, 3.5
and 0.1 MPa. Gas chromatography analysis of the stream
composition at the outlet of the vessel shows that, as ex-
pected, the different components contained in the natural
gas, slowly obstruct the adsorption sites for methane. With
Acknowledgements
This work has been done with the financial support of
Gaz De France and ADEME.
Appendix A. Experimental profiles of the mean ratios
for nitrogen, i-butane, n-butane, i-pentane, n-pentane,
and the hydrocarbons with molecular weight higher
than n-pentane (C6+ )
For safety reasons, the cycling operation is interrupted
during the weekend. After an interruption, always happening
at the end of a discharge phase, several cycles are necessary
to join the continuous curve representative of the evolution of
the behaviour of the tank. This phenomenon tends to demon-
strate the importance of the kinetics of the multi-component
adsorption equilibrium in the case of a complex mixture
of gas.
78 O. Pupier et al. / Chemical Engineering and Processing 44 (2005) 7179
 O. Pupier et al. / Chemical Engineering and Processing 44 (2005) 7179 79

References [11] K.J. Chang, O. Talu, Behaviour and performances of adsorptive

natural gas storage cylinders during discharge, Appl. Therm. Eng.
16 (1996) 359374.
[1] J. Wegrzyn, M. Gurevich, Adsorbent storage of natural gas, Appl.
[12] S. Bilo S, V. Goetz, S. Mauran, Dynamic discharge and performance
Energy 55 (1996) 7183.
of a new adsorbant for natural gas storage, AIChE J. 47 (2001)
[2] O. Talu, An overview of adsorptive storage of natural gas, in: M.
28192830.
Suzuki (Ed.), Proceedings of the fourth International Conference on
[13] K. Dvorak, R.C. Hodrien, Development of adsorbed natural gas
Fundamentals of Adsorption, Kyoto, 1992, pp. 655622.
technology for large scale diurnal storage application, in: Proceedings
[3] N.D. Parkins, D.F. Quinn, Natural gas adsorbed on carbon, in: J.W.
of IGRC, Amsterdam, 2001.
Patrick (Ed.), Porosity in Carbons, Great Britain, Edward Arnold,
[14] K. Seki, Y. Sumie, Development of adsorptive natural gas storage
1995, pp. 291325.
systemApplication to gas holder and natural gas vehicle, in: Pro-
[4] K.R. Matranga, A.L. Myers, E.D. Glandt, Storage of natural gas by
ceedings of IGRC, Amsterdam, 2001.
adsorption on activated carbon, Chem. Eng. Sci. 47 (1992) 1569
[15] J.P.B. Mota, Impact of gas composition on natural gas storage by
1579.
adsorption, AIChE J. 45 (1999) 986996.
[5] S. Bilo, V. Goetz, A. Guillot, Optimal design of an activated carbon
[16] A. Golovoy, E.J. Blais, Natural gas storage on activated carbon, in:
for an adsorbed natural gas storage system, Carbon 40 (2002) 1295
Proceedings of SAE Conference, Warrendale, 1983, pp. 4753.
1308.
[17] A.S. Pedersen, B. Larsen, Adsorption of methane and natural gas on
[6] D. Lozano-Castello, D. Cazorla-Amoros, A. Linares-Solano, D.F.
six carbons, Rap. Riso-M-2781, Riso National Laboratory, Denmark,
Quinn, Influence of pore size distribution on methane storage at
1989.
relatively low pressure: preparation of activated carbon with optimum
[18] L. Czepirski, Some aspects of a sorbent-containing storage system
pore size, Carbon 40 (2002) 9891002.
for natural gas, Indian J. Tech. 29 (1991) 266268.
[7] D. Lozano-Castello, D. Cazorla-Amoros, A. Linares-Solano, Can
[19] A. Guillot, S. Follin, L. Poujardieu, Adsorption studies of carbon
highly activated carbons be prepared with homogeneous micropore
dioxide on KF1500, X2MH6/8 and PX21 by means of a high-pressure
size distribution, Fuel Process. Technol. 7778 (2002) 325330.
volumetric device, in: B. Mc Enaney, T.J. Mays, J. Rouquerol, F.
[8] J.A.F. Mac Donald, D.F. Quinn, Carbon adsorbent for natural gas
Rodriguez-Reinoso, K.S.W. Sing, K.K. Unger (Eds.), Proceedings
storage, Fuels 77 (1998) 6164.
on Characterisation of Porous Solid IV, London, 1997, pp. 573
[9] T.D. Burchell, M.R. Rogers, Carbon fibers based natural gas stor-
580.
age monoliths, Extended Abstract, Gas storage workshop, Kingston,
[20] T.K. Bose, R. Chahine, J.M. St-Arnaud, High density Adsorbent and
2001.
Method Producing the Same, US Patent 4,999,330, 1991.
[10] S. Bilo, V. Goetz, S. Mauran, Characterisation of adsorbent com-
posite blocks for methane storage, Carbon 39 (2001) 16531662.
 Chemical Engineering and Processing 44 (2005) 8187

Effect of internal on the hydrodynamics in external-loop airlift reactors

Tongwang Zhang, Jinfu Wang , Tiefeng Wang, Jing Lin, Yong Jin
Department of Chemical Engineering, Tsinghua University, Beijing 100084, PR China

Received 12 March 2004; received in revised form 4 May 2004; accepted 5 May 2004
Available online 19 June 2004

Abstract

Small bubbles and flow uniformity are important for gasliquid and gasliquidsolid multiphase reactors. A reactor internal was designed
and installed in an external-loop airlift reactor (EL-ALR) to enhance bubble breakup and flow redistribution and improve reactor performance.
Hydrodynamic parameters, including local gas holdup, bubble rise velocity, bubble Sauter diameter and liquid velocity were measured. A
radial maldistribution index was introduced to describe radial non-uniformity in the hydrodynamic parameters. The influence of the internal
on this index was studied. Experimental results show that The effect of the internal is to make the radial profiles of the gas holdup, bubble
rise velocity and liquid velocity radially uniform. The bubble Sauter diameter decreases and the bubble size distribution is narrower. With
increasing distance away from the internal, the radial profiles change back to be similar to those before contact with it. The internal improves
the flow behavior up to a distance of 1.4 m.
2004 Elsevier B.V. All rights reserved.

Keywords: Airlift reactor; Internal; Radial profile; Gas holdup; Bubble size; Bubble breakup

1. Introduction by the mass transfer coefficient and the interfacial area.

External-loop airlift reactors (EL-ALRs) have drawn
much attention due to their simple construction, good heat
and mass transfer, and good mixing characteristics as the
gas phase in the reactor serves the dual functions of aer-
ation and agitation [1]. EL-ALRs have been used in a
variety of industrial applications where intimate contact
of gasliquidsolid phases is necessary in the chemical,
petrochemical, minerals processing and biochemical pro-
cesses [2]. The special feature distinguishing EL-ALRs
from bubble columns is the recirculation of the liquid
through a downcomer that connects the gasliquid separa-
tor and the bottom of the riser. The density difference due
to the gas passing through the riser induces circulation of
the liquid and the solid in the EL-ALR. The circulation
creates good mixing in all phases and provides good mass
transfer.
Mass transfer is an important issue in the reaction pro-
cesses in EL-ALRs, especially in the heterogeneous flow
regime. The volumetric mass transfer rate is determined
Corresponding author. Tel.: +86 10 62785464;
fax: +86 10 62772051.
E-mail address: wangjf@flotu.org (J. Wang).
Since smaller bubbles and larger gas holdups have higher
specific interfacial area, a decrease in bubble size or an
increase in gas holdup can improve mass transfer and in-
crease liquid production at the same superficial gas velocity
[3]. A uniform radial profile of the gas holdup decreases
bubble-bubble collisions, which in turn decreases bubble
coalescence. Therefore, it is valuable to provide an approach
to reduce bubble size and make the gas holdup radial profile
more uniform.
Previous workers have studied the influence of the ratio
of the cross-section area of the downcomer to the riser [4,5],
the reactor height [6,7], the gasliquid separator configura-
tion [8], and the distributor type and location [9]. All these
affect the flow characteristics and mass transfer. Most pre-
vious works focus on global parameters, such as the liquid
circulation velocity [1013] and the average gas holdup in
the riser [1416]. Although much work has been carried out
on EL-ALRs, the proper design and scale-up of an EL-ALR
is still difficult because any variation in the physical proper-
ties of the gas or the liquid and the reactor structural feath-
ers can have a considerable effect on the hydrodynamics
[7].
Smaller bubbles and higher gasliquid interface re-
newal frequency are beneficial to improving mass transfer.

0255-2701/$ see front matter 2004 Elsevier B.V. All rights reserved.
doi:10.1016/j.cep.2004.05.003
82 T. Zhang et al. / Chemical Engineering and Processing 44 (2005) 8187

Intensified turbulence can destroy the stability of large

bubbles, which leads to smaller bubbles, and increase the
frequency of bubble coalescence and breakup, which im-
proves the surface renewal frequency of bubbles [17,18].
Intensified turbulence can be achieved by changing the gas
sparger to decrease the initial bubble size and improve its
radial distribution, but the effective region of the gas sparger
is only limited to a certain height above the distributor [19].
Liquid mixing time decreases sharply for an initial in-
crease in the gas sparging rate and approach an asymptotic
value that is determined by the height and diameter of the
downcomer and the liquid properties [5]. A higher liquid
velocity shortens the gas residence time and results in a de-
crease of gas holdup and interfacial area. The radial pro-
file of the liquid is parabolic. These are disadvantageous for
mass transfer. The mounting of internals in a fixed bed is
often used to improve the radial profile of the liquid veloc-
ity. This motivates us to mount internals in an EL-ALRs to
improve the radial profile of the gas holdup and the liquid
velocity and to intensify turbulence.
As noted above, small bubbles, a uniform gas holdup ra-
dial distribution and an appropriate liquid circulating veloc-
ity can intensify mass transfer between the gas phase and
the continuous phase and improve the production efficiency
in EL-ALRs. In order to reduce the bubble size and obtain a
more uniform radial distribution of the local gas holdup and
the liquid and bubble rise velocities, and regulate the liq-
uid circulating velocity appropriately as well, in this work, a
novel internal is used and mounted in the riser column to im- Fig. 1. Schematic of the experimental equipment.
prove hydrodynamics and mass transfer. The hydrodynamic
behavior and mass transfer characteristics of an EL-ALR
with the new designed internal are investigated. flow of the downcomer, the flow near the sparger is not axial
symmetry. But the unsymmetry will disappear about 130 cm
above the sparger. So the internal is installed 165 cm above
2. Experimental setup the sparger.

An EL-ALR was set up for the experimental study and

fabricated using plexiglas. The schematic diagram of the
experimental setup is shown in Fig. 1. The riser and the
downcomer are 0.230 and 0.190 m in diameter, respectively,
and 4.8 m in height. The height and the diameter of the
airliquid separator are 0.960 and 0.480 m, respectively.
The downcomer is connected to the riser 0.2 m above the
distributor.
Air was used as the gas phase and was introduced into the
system through a distributor with holes of diameter 1 mm
and a perforation of 0.25%. The superficial gas velocity,
based on the riser cross-section area, varied from 0.0067 and
0.0535 m/s.
The internal has been described in patent CN 1,403,190A
[20], and is shown in Fig. 2. It is 230 mm in diameter and
100 mm in height. The angle between the baffle of the in-
ternal and the vertical axis is 45 . Each baffle is 30 mm in
width and 1 mm in thickness. There are some semicircular
holes on the plate and each hole has a tongue-like plate fac-
ing the upstream to break bubbles. Because of the returning Fig. 2. Photo of the internal used in this work.
 T. Zhang et al. / Chemical Engineering and Processing 44 (2005) 8187
The local gas holdup and bubble behavior were measured
by a reflective optic fiber probe developed by Wang and
co-workers [21,22]. It can be known whether the probe is im-
merging in the gas. The rate of the time that probe immerg-
ing in the gas and the total sample time is gas holdup. Gas
velocity can be got by the time difference that one bub-
ble touch two probes and the distance between two probes.
Chord length can be obtained from one bubble velocity and
the time that the probe stays in the bubble. Bubble size dis-
tribution is got from the probability density of the chord
length based on some numerical method. The local liquid
velocity in the riser was measured by a backward scattering
LDA system (system 9100-8, model TSI). Details have been
given by Lin et al. [23].
3. Results and discussion
3.1. Gas holdup
The upper surface of the internal is defined as the zero
of the axial position. Axial positions are positive above this
and negative below this. The radial profiles of the gas holdup
at five axial positions were measured, as shown in Fig. 3.
The radial profile of the gas holdup becomes much flatter
after flowing through the internal, with an increase in the
gas holdup near the wall and a decrease in the center region
as compared with the gas holdup below the internal. As
the distance above and away from the internal increases,
the profile becomes more and more similar to that before
contact with the internal. The local radial profiles of the gas
holdup at axial positions 144 and 209 cm show almost no
difference. This shows that the influence of the internal on
the radial profile of the gas holdup becomes weaker and
weaker with increasing distance away from the internal and
is no longer felt beyond a certain distance. The analysis of
the local gas holdup profiles shows that the influence of
the internal persists to a distance of about 1.4 m. In order
to show the effect of the internal on the radial profile of
Fig. 3. Radial profile of the gas holdup at different axial positions.
83
Fig. 4. Radial maldistribution index of the local gas holdup at different
axial positions.
the gas holdup clearly, a radial maldistribution index, , is
introduced, defined as:


 n
 1 
= (yi y)2 (1)
n1
i=1
where y can be gas holdup, bubble velocity, bubble diameter
or liquid velocity. Fig. 4 shows the change in maldistribu-
tion index with axial position at different superficial gas
velocities. It can be seen that the radial maldistribution has
a minimum a short distance after passing the internal and
then it increases gradually. The difference between 144
and 209 cm is negligible. Therefore, it can be concluded that
the region of the influence of the internal on the gas holdup
is within 140 cm.
Fig. 5 shows axial changes in the cross-sectional aver-
aged gas holdup at different superficial gas velocities with
an increase in the distance above the internal. The average
gas holdup first increases to a maximum at about 40 mm
above the internal and then decreases gradually. Note that
at the superficial gas velocity of 0.0067 m/s, the maximum
gas holdup appears somewhat earlier. At a low superficial
Fig. 5. Cross-sectional gas holdup at different axial positions.
84 T. Zhang et al. / Chemical Engineering and Processing 44 (2005) 8187
velocity, the influence of the internal on liquid turbulence
and its effect on breaking bubbles are weak, which results
in the maximum gas holdup appearing somewhat earlier. As
expected, the average gas holdup is almost unchanged when
the axial height exceeds 140 cm. An enhancement in turbu-
lence intensity was observed after liquid flowed through the
internal. Enhanced turbulence intensity effectively forces the
bubble breakup and results in an increase in the gas holdup
[24]. In the population balance model [17], bubble size is
determined by an equilibrium between bubble breakup and
coalescence. Due to enhanced turbulence by the internal,
bubble breakup will be dominant in the downstream region
after the internal, which leads to a decrease in the bubble
size and an increase in the local gas holdup. With increasing
axial height, the turbulent intensity decreases and the local
gas holdup decreases. When the axial height is more than
144 cm above the internal, the influence of the internal has
vanished and the local gas holdup is the same as in the state
below the internal.
Fig. 6 presents the experimental results on the gas holdup
profiles with and without the internal at the axial position
of 74 cm for different superficial gas velocities. The inter-
nal causes extra flow resistance, which in turn decreases the
liquid circulation velocity and increases gas holdup [5]. In
addition, the experimental results show that the radial pro-
files of the local gas holdup with an internal are flatter than
those without the internal. Therefore, a properly designed
internal can have dual function of increasing the gas holdup
and improving its radial profile.
3.2. Bubble rise velocity
It is important to study the bubble rise velocity and its
radial profile in a gasliquid system as these are closely re-
lated to the hydrodynamics, and mass and heat transfer [25].
Bubble rise velocity and its radial profile have also signif-
icant influences on gas and liquid residence time distribu-
tions. A suitable bubble rise velocity and radial profile can
improve production efficiency. Bubble rise velocities in a
Fig. 6. Influence of the internal on the local gas holdup for different
superficial gas velocity.
Fig. 7. Bubble rise velocity radial profile with axial positions.
gasliquid system are affected by the physical properties of
each phase, bubble size, local gas holdup and liquid veloc-
ity. Larger bubbles have higher rise velocities and the bubble
swarm velocity is higher than that of a single isolated bubble
[26]. Bubble breakup is caused by collisions between bub-
bles and turbulent eddies [24]. Thus, the turbulence inten-
sity has an important effect on the bubble rise velocity and
methods that intensify the turbulence intensity can improve
the bubble rise velocity and its radial profile.
Fig. 7 shows the radial profiles of the bubble rise velocity
at different axial positions. The radial profiles of bubble rise
velocity become much more uniform after the gasliquid
mixture flowed through the internal, with an increase near
the wall and a decrease in the central region compared with
that below the internal. As the distance above the internal
increases, the radial profile of the bubble rise velocity be-
comes more and more similar to that below the internal. At
the axial position of 144 cm, the radial profile of the bubble
rise velocity is about the same as that below the internal.
To show the effect of the internal on the average bubble
rise velocity more clearly, the cross-sectional average bub-
ble rise velocity, ub , and the radial maldistribution index of
bubble rise velocity, ub are introduced. ub is defined as:
R
2 0 g (r)ub (r)rdr
ub = (2)
Ag
where A is the cross-sectional area.
Fig. 8 shows the average bubble velocity as a function
of the axial distance at different superficial gas velocities.
The average bubble velocity decreases after contact with the
internal to a minimum, then it increases gradually to the
original velocity above the position of 144 cm. Fig. 9 shows
ub as a function of the axial distance at different superficial
gas velocities. It decreases notably after contact with the
internal to a minimum at 39 cm above the internal, then
increases gradually. According to the analysis of Figs. 79,
the radial profile of the bubble rise velocity above the internal
are much flatter than that below, and the average bubble rise
velocity increases after the bubble had passed through the
internal. The change caused by the internal gets smaller and
 T. Zhang et al. / Chemical Engineering and Processing 44 (2005) 8187
Fig. 8. Average bubble rise velocity as a function of the axial position.
then vanishes gradually with increasing distance from the
internal.
3.3. Bubble Sauter diameter
Precise knowledge of the bubble size is important for a
better understanding of the hydrodynamics, mass transfer,
and reactor design. Zun [27] concluded that small bubbles
tend to have uniform radial profiles while large bubbles tend
to rise in the central region. Smaller bubbles and a uniform
radial profile lead to a larger interfacial area and a lower rise
velocity, which increases mass transfer.
The Sauter bubble diameter, d32 , is commonly used to
show the statistic size distribution of the bubbles [28].

ni d 3
d32 =  i2 (3)
ni di
Fig. 10 shows the radial profile of the bubble Sauter di-
ameter at five axial positions. The bubble size is very much
decreased after flowing through the internal and then in-
creases with an increase of the distance from the internal.
The difference between the values at the positions of 144
and 209 cm above the internal is negligible and their radial
Fig. 9. Axial change in the radial maldistribution index of the bubble rise
velocities.
85
Fig. 10. Bubble Sauter diameter radial profile at different axial positions.
profile is similar to that below the internal. According to
the analysis of bubble diameter, it can be concluded that the
effective distance of the internal is about 140 cm.
While flowing through the internal, bubbles rise along the
undersurface of the baffles and collide with the tongue-like
bars, and are broken up into smaller bubbles as shown in
Fig. 10. The difference in the direction of adjacent baffles
increases the liquid turbulent intensity, which is beneficial
to the breakup of large bubbles. With increasing distance
from the internal, the turbulent intensity decreases and coa-
lescence becomes dominant until a new equilibrium between
the breakup and coalescence of bubbles is reached.
3.4. Bubble size distribution
The bubble size detected by the probe is subject to a
probability distribution [28,29] and only the mean statistical
value of the bubbles is given by d32 . The analysis of the
bubble size distribution helps us understand the uniformity
of the bubble sizes and its axial evolution. Fig. 11 shows
the axial change of the bubble size distribution at r/R = 0
and r/R = 0.97, respectively. Below the internal, the peak
Fig. 11. Bubble size distribution at different axial positions.
86 T. Zhang et al. / Chemical Engineering and Processing 44 (2005) 8187
of the distribution at r/R = 0.97 is to the left of that at r/R
= 0. Due of bubble breakup by the internal, the bubble size
distributions at the center and near the wall are narrower than
that obtained below the internal and the peaks are shifted to
the left after passing through the internal. This means that the
bubble size decreases and the bubble size distribution is more
uniform above as compared with that below. With increasing
distance from the internal, the effect of the internal vanishes
gradually and the bubble size distributions become wider and
the peaks shift to the right. There is no difference between
the size distributions obtained at the position of 144 and
209 cm above the internal, which means that the region of
the influence of the internal on bubble size distribution is
about 1.4 m.
3.5. Liquid velocity
Fig. 12 shows the radial profile of the liquid velocity at
different axial positions. Due of the baffles, the liquid is re-
distributed in the radial direction and the turbulent intensity
is increased. The radial profile of the liquid velocity is al-
most uniform after passing the internal. Liquid velocity is
lower at the center and higher near the wall as compared
with that below the internal. With increasing distance from
the internal, the turbulence intensity diminishes and the wall
effect becomes more apparent, that is, the liquid velocity in-
creases at the center and decreases near the wall. The radial
profile obtained at the position of 114 cm from the internal
is similar to that obtained below the internal and is the same
as that at the position of 144 cm.
The radial maldistribution index of the liquid velocity,
ul , is defined according to Eq. (1). Fig. 13 shows the ax-
ial changes in ul at different superficial gas velocities. ul
is much decreased after passing through the internal and
reaches a minimum at 5 cm above the internal, then it in-
creases gradually with increasing axial height. According to
the analysis of the radial profile of the liquid velocities, it
can be concluded that the region in which the internal influ-
ences the liquid velocity is only within 110 cm.
Fig. 12. Radial profile of the liquid velocity.
Fig. 13. Axial change of the radial maldistribution index of liquid veloc-
ities.
4. Conclusions
Mass transfer is essential in EL-ALRs. Smaller bubbles
and a uniform gas holdup radial distribution increase the in-
terfacial area and improve mass transfer. Intensified turbu-
lence increases the surface renewal frequency and decreases
bubble size. A novel internal to improve mass transfer and
the hydrodynamic behavior in a gasliquid system is re-
ported. Experiments were carried out to study the effect of
the internal on the bubble behavior and liquid velocity in an
EL-ALR.
The radial profiles of the gas holdup and bubble rise veloc-
ity become more uniform after passing through the inter-
nal. With increasing distance from the internal, the radial
profiles of the gas holdup and bubble rise velocity change
back to be similar to that below the internal. At 144 cm
above the internal, the radial profiles of the gas holdup
and the bubble rise velocity are the same as those below,
while andub reach their minimum 39 cm above the
internal.
The bubble Sauter diameter very much decreases after
passing the internal, then increases with increasing dis-
tance from it. The bubble size distribution becomes nar-
rower after the internal and its peak is shifted to the left.
The bubble size at 144 cm above the internal has increased
back to be the same as that below.
Turbulence is intensified after passing through the inter-
nal. The radial profile of the liquid velocity becomes flat-
ter. ul reaches a minimum 5 cm above the internal. The
liquid velocity 114 cm above the internal is the same as
that below, that is, the internal affects the liquid velocity
up to about 110 cm above the internal.
The internal used in this work has good performance in
intensification of gasliquid mass transfer and in improv-
ing the flow distribution, and has a good perspective in
application in airlift reactors. Meanwhile, further study on
the internal should be carried out.
 T. Zhang et al. / Chemical Engineering and Processing 44 (2005) 8187
Acknowledgements
The authors gratefully acknowledge financial support by
the National Natural Science Foundation of China (No.
20276035) and by the Ministry of Education Ph.D. Fellow-
ship Program (No. 20020003060).
Appendix A. Notations
A cross-section of the reactor (m2 )
d32 Sauter bubble diameter (mm)
H axial distance from the upper surface of
the internal (m)
r radial position in the reactor (m)
R radius of the riser (m)
u local velocity (m/s)
U superficial velocity (m/s)
local gas holdup
radial maldistribution index
Subscriptions
b bubble
g gas phase
l liquid phase
u velocity
gas holdup
References
[1] J.C. Merchuk, S. Yehuda, Local holdup and liquid velocity in air-lift
reactors, AIChE J. 27 (1981) 377388.
[2] A.G. Livingston, S.F. Zhang, Hydrodynamic behavior of three-phase
(gasliquidsolid) airlift reactor, Chem. Eng. Sci. 48 (1993) 1641
1654.
[3] S. Guet, G. Ooms, R.V.A. Oliemans, R.F. Mudde, Bubble injector
effect on the gas-lift efficiency, AIChE J. 49 (2003) 22422252.
[4] K.H. Choi, W.K. Lee, Circulation liquid velocity, gas hold-up, vol-
umetric oxygen transfer coefficient in external-loop airlift reactors,
J. Chem. Technol. Biotechnol. 56 (1993) 5158.
[5] M.A. Young, R.G. Carbonell, D.F. Ollis, Airlift bioreactors: analysis
of local two-phase hydrodynamics, AIChE J. 37 (1991) 403428.
[6] A.B. Russell, C.R. Thomas, M.D. Lilly, The influence of vessel
height and top-section size on the hydrodynamic characteristics of
airlift fermentors, Biotechnol. Bioeng. 43 (1994) 6976.
[7] C. Bentifraouine, C. Xuereb, J.P. Riba, An experimental study of
the hydrodynamic characteristics of external-loop airlift contactors,
J. Chem. Technol. Biotechnol. 69 (1997) 345349.
87
[8] M.H. Siegel, J.C. Merchuk, Hydrodynamics in rectangular airlift
reactors: scale-up and the influence of gasliquid separator design,
Can. J. Chem. Eng. 69 (1991) 465473.
[9] S. Becker, A. Sokolichin, G. Eigenberger, Gasliquid flow in bubble
column and loop reactors. Part II. Comparison of detailed experiments
and flow simulations, Chem. Eng. Sci. 49 (1994) 57475762.
[10] M.Y. Chisti, B. Hallard, M. Moo-Young, Liquid circulation in airlift
reactors, Chem. Eng. Sci. 43 (1988) 451457.
[11] A.G. Jones, Liquid circulation in a draft tube bubble column, Chem.
Eng. Sci. 40 (1985) 449462.
[12] Y.C. Hsu, M.P. Dudukovic, Gas holdup and liquid recirculation in
gas-lift reactors, Chem. Eng. Sci. 35 (1980) 134141.
[13] M. Gavrilescu, R.Z. Tudose, Study of the liquid velocity in
external-loop airlift reactor, Bioprocess Eng. 14 (1995) 183193.
[14] J.H. Hills, The operation of a bubble column at high throughputs. I.
Gas hold-up measurement, Chem. Eng. J. 12 (1976) 8999.
[15] M.H. Siegel, J.C. Merchuk, K. Schugerl, Airlift reactor analysis
interrelationships between riser, downcomer and gasliquid behavior,
including gas recirculation effects, AIChE J. 32 (1986) 15851595.
[16] J.B. Joshi, V.V. Ranade, S.D. Gharat, S.S. Lele, Sparged loop reactors,
Can. J. Chem. Eng. 68 (1990) 705741.
[17] M.J. Prince, H.W. Blanch, Bubble coalescence and break-up in
air-sparged bubble columns, AIChE J. 36 (1990) 14851499.
[18] C. Maretto, R. Krishna, Modeling of a bubble column slurry reactor
for FischerTropsch synthesis, Catal. Today 52 (1999) 279289.
[19] S.M. Shavaraju, T.W.F. Russel, H.W. Blanch, The design of gas
sparged devices for viscous liquid systems, AIChE J. 24 (1978)
454466.
[20] Y. Jin, T. F. Wang, J. F. Wang, T. W. Zhang, CN Patent 1,403,190A
(2003).
[21] T.F. Wang, J.F. Wang, W.G. Yang, Y. Jin, Bubble behavior in
gasliquidsolid three-phase circulating fluidized beds, Chem. Eng.
J. 84 (2001a) 397404.
[22] T.F. Wang, J.F. Wang, W.G. Yang, Y. Jin, Experimental study on
gas-holdup and gasliquid interfacial area in TPCFBs, Chem. Eng.
Commun. 187 (2001b) 251263.
[23] J. Lin, M.H. Han, T.F. Wang, T.W. Zhang, J.F. Wang, Y. Jin, Exper-
imental study on the local hydrodynamic behavior of a three-phase
external-loop airlift reactor, Ind. Eng. Chem. Res., in press.
[24] T.F. Wang, J.F. Wang, Y. Jin, A novel theoretical breakup kernel
function for bubbles/droplets in a turbulent flow, Chem. Eng. Sci.
58 (2003) 46294637.
[25] K. Tsuchiya, A. Furumoto, L.S. Fan, J. Zhang, Suspension viscosity
and bubble rise velocity in liquidsolid fluidized beds, Chem. Eng.
Sci. 52 (1997) 30533066.
[26] R. Krishna, M.I. Urseanu, J.M. van Baten, J. Ellenberger, Rise
velocity of a swarm of large gas bubble in liquids, Chem. Eng. Sci.
54 (1999) 171183.
[27] I. Zun, The transverse migration of bubble influenced by walls in
vertical bubbly flow, Int. J. Multiphase Flow 6 (1980) 583588.
[28] S.A. Patel, J.G. Daly, D.B. Bukur, Holdup and interfacial area mea-
surements using dynamic gas disengagement, AIChE J. 35 (1989)
931942.
[29] D. Sabtana, A. Macias-Machin, Local bubble-size distribution in
fluidized beds, AIChE J. 46 (2000) 13401347.
 Chemical Engineering and Processing 44 (2005) 89100

Multicriteria synthesis of flexible heat-exchanger networks

with uncertain source-stream temperatures
Cheng-Liang Chen , Ping-Sung Hung
Department of Chemical Engineering, National Taiwan University, Taipei 10617, Taiwan, ROC

Received 23 May 2003; received in revised form 28 July 2003; accepted 31 March 2004
Available online 2 July 2004

Abstract

A multi-criteria synthesis strategy for heat-exchanger networks (HENs) simultaneously considering minimum utility consumption, max-
imum source-stream temperature flexibility, and even minimum number of matches is proposed. The flexible HEN synthesis problem is
formulated as a multi-objective mixed-integer linear programming (MO-MILP). For handling the multiple conflict design targets, a two-phase
fuzzy multi-criteria decision-making method is presented to attain a best compromised solution. Two numerical examples with flexibility pref-
erences in source-stream temperatures are supplied, demonstrating that the proposed strategy can provide definite and feasible compensatory
solutions for multi-criteria HEN synthesis problems.
2004 Elsevier B.V. All rights reserved.

Keywords: Heat-exchanger network; Synthesis; Flexibility; Superstructure; MILP; Multi-criteria decision-making; Fuzzy optimization

1. Introduction rameters from their nominal values [10], however, attracts

A heat exchanger network (HEN) synthesis problem can
be described as the one that synthesize a HEN configura-
tion to reach some assigned targets such as minimum util-
ity consumption, minimum total number of heat exchangers,
etc., with given heating/cooling utilities and hot/cold process
streams be cooled/heated from nominal inlet temperatures
to specified target temperatures [1].
Most of the existing HEN synthesis methods rely on ei-
ther heuristic rules (for example, pinch analysis method [2])
or mathematical programming (for example, simultaneous
optimization approach [36]). And further, to some typical
objectives considered in the HEN synthesis such as utility
consumption, total number of matches, and total exchanger
area, the flexibility of the HENs for feasible operation un-
der possible variation of source-stream temperatures and/or
heat-capacity flow rates has been emphasized in some re-
cent articles [610]. For HEN synthesis, the analysis of this
flexibility, defined as the size of the region of feasible oper-
ation in the space of desired or undesired deviations of pa-
Corresponding author. Tel.: +886 2 23636194;
fax: +886 2 23623040.
E-mail address: ccl@ntu.edu.tw (C.-L. Chen).
attention usually indirectly as test examples basing on math-
ematical programming synthesis. Therein, [10] explored the
HEN synthesis problem with simultaneous flexibility target-
ing and minimum-utility objective based on an MILP for-
mulation. The optimal solution is examined on the basis of
the vertices of the polyhedral uncertainty region in the space
of source-stream temperatures. It is found, however, that the
resulting HEN structures with increasing flexibility require-
ments are prone to variation, and the true maximum flexi-
bilities of resulting HEN structures are usually greater than
the assigned targets. Thus, for a given flexibility target, it is
very often to obtain a more conservative HEN design by the
method of simultaneous flexibility targeting and synthesis
of minimum-utility HENs proposed in [10].
In this paper, we extend the work of [10] by simultane-
ously considering minimization of the total utility consump-
tion, maximization of operational flexibility to source-stream
temperatures, and even minimum number of matches as
multiple design objectives. The flexible HEN synthesis
problem is thus formulated as the one of multi-objective
mixed-integer linear programming (MO-MILP). This for-
mulation also assumes that the feasible region in the space
of uncertain input parameters is convex, so that the optimal
solution can thus be explored on the basis of the vertices

0255-2701/$ see front matter 2004 Elsevier B.V. All rights reserved.
doi:10.1016/j.cep.2004.03.017
90 C.-L. Chen, P.-S. Hung / Chemical Engineering and Processing 44 (2005) 89100
of the polyhedral region of uncertainty [10]. Under the as-
sumption of convexity, only the source-stream temperatures
of the HENs are considered to be the uncertain input pa-
rameters. With this formulation, the standard definition of
the HEN synthesis problem with minimal total utility con-
sumption and even minimum number of units is extended
to include a flexibility specification for the potential HEN
structure and can be stated as: given hot/cold streams to be
cooled/heated from nominal supply temperatures to speci-
fied target temperatures and hot/cold utility specifications,
synthesize a HEN such that it has minimal utility consump-
tion (considering nominal case or average of all vertical
operating points), minimal number of matches if desired,
and maximal flexibility for feasible operation.
For handling the multiple and conflict design objectives,
a fuzzy decision-making method is adopted to attain the
compromised solution among all conflict objectives [11].
Therein each design objective is treated as a fuzzy goal,
and a specific membership function is used to characterize
the transition from the numerical objective value to the de-
gree of satisfaction for the fuzzy objective. The final deci-
sion, therefore, is interpreted as a fuzzy aggregation of these
multiple objectives and measured by the overall degree of
satisfaction. And the best compromised solution is finally
reached by maximizing the overall degree of satisfaction for
the decision. In the course of finding the solution, two pop-
ular operators, the minimum and the average, are applied as
the fuzzy intersection operators, the effects of which are ex-
amined as well. We also proposed an interactive two-phase
fuzzy decision-making method by combining these two op-
erators to take advantages of the both [11,12]. The minimum
operator is used in phase I to maximize the degree of satis-
faction for the worst objective, and the average operator is
Fig. 1. The two-stage superstructure.
then applied in phase II to simultaneously promote satisfac-
tion levels of all objectives with guaranteed least satisfaction
value.
Two numerical examples with flexibility preference in
source-stream temperatures is presented here to demonstrate
that the proposed interactive two-phase fuzzy optimization
method can provide a feasible and better compensatory so-
lution for multi-objective HEN synthesis.
2. Model formulation
Consider the standard HEN synthesis problem with NH
hot and NC cold process streams along with hot and cold
utilities. Since it is suitable for formulating the simultaneous
solution which involving the consideration of both utility
consumption and operational flexibility, the HEN superstruc-
ture proposed by [3,4] is applied for modeling the structure.
Therein, the isothermal mixing assumption in the simplified
superstructure eliminates the need for nonlinear/nonconvex
energy balance. The minimum number of superstructure
stages, NT , corresponds to max{NH , NC }, as suggested by
[3]. Fig. 1 illustrates a 2-hot/2-cold/2-stage superstructure.
The mathematical programming formulation for minimizing
utility consumption with specified uncertain source-stream
temperature ranges can be summarized as follows: [3,4,10]:

(0)

(0)
minJ (0) = qcui + qhuj (1)
x
i HP j CP
zijk , zcui , zhuj ; tik , tjk ; dtijk , dtcui , dthuj ;
x qijk , qcui , qhuj ; (2)
i HP, j CP, k ST
 C.-L. Chen, P.-S. Hung / Chemical Engineering and Processing 44 (2005) 89100 91

(T in T out )FCpi =
qijk + qcui
i

i




kST jCP









overall heat balances

(T out T in )FCp =


j

j j qijk + qhuj










kST iHP










(tik ti,k+1 )FCpi =

q





ijk





jCP









stage heat balances

(tjk tj,k+1 )FCpj =

qijk







iHP











in(0) in(0)


ti,1 = Tiin {Ti

, Ti + }








uncertain inlet temperatures

t

in(0) in(0)

j,NT +1 = Tj {Tj , Tj + }
in







t t



ik

i,k+1







t t




jk

j,k+1



feasibility of temperatures

T out t



i

i,NT +1





out




Tj tj,1











(t



i,NT +1 Ti )FCpi = qcui
out
= x utility loads (3)

(Tjout tj,1 )FCpj = qhuj









qijk zijk 0











qcui zcui 0



logical constraints




qhuj zhuj 0









dt t t + (1 z )


ijk ik jk ijk







dt ti,k+1 t j,k+1 + (1 z )


ij,k+1 ijk
approach temperatures

dtcu t
i,NT +1 TCU + (1 zcui )
out

i






dthu j T out t
j,1 + (1 zhu j )


HU


(0)


dt T


ijk min




dtcu
(0)
T nominal approach temp. bounds



i min


dthu(0) Tmin



j











z + +


ijk zcu i zhu j MEU max

maximum exchanger units

i HP,j CP,k ST



i HP j CP







zijk , zcui , zhuj
{0, 1}




tik , tjk , dtijk , dtcui , dthuj , qijk , qcui , qhuj ;
0

i HP, j CP, k ST

where x and denote variables for design and the feasible

peratures, the feasible region defined by the reduced inequal-
searching space, respectively; is the targeted flexibility for
ity constraints is convex [6], so the critical point that limits
source-stream temperatures [10]; the superscript (0) denotes
the solution lies at a vertex of the polyhedral region of un-
the nominal condition; and the upper bound for unit num-
certainty. For the problem of Eq. (7) with N ( NH + NC )
bers is MEUmax . The searching space is comprised of all
uncertain source-stream temperatures, the vertex-based for-
heat balances constraints and relevant logical constraints. As
mulation is given as follows [10]:
pointed out in [610], this problem is difficult to solve di-
rectly since it involves a max-min-max constraint that leads


(0) (0)
to a non-differentiable global optimization problem. For the min J (0) = qcui + qhuj (4)
HEN synthesis problem with uncertain source-stream tem- xV V
iHP jCP
92 C.-L. Chen, P.-S. Hung / Chemical Engineering and Processing 44 (2005) 89100

or (n) (n)
Here, ri and rj are the vertex identifiers which take val-
ues of NV = 2(NH +NC ) combinations of +1 and 1, see
min J (ave)
xV V Table 1 in [10]. Therein, Eq. (4) considers nominal utilities
1

(n)

(n) and Eq. (5) takes into account the average of all vertical op-
= qcui + qhuj (5)
NV + 1 erating points.
n{0}VT iHP jCP
For the benchmark example mentioned in [10], a
(n) (n) (n)
2-hot/2-cold streams problem along with heating steam and
zijk , zcui , zhuj ; dtijk , dtcui , dthuj ;
cooling water, the maximal allowable variation of various
xV = (n) (n) (n) (n) (n)
tik , tjk ; q ijk , qcui , qhuj ;
(6) possible minimum-utility HEN structures is examined by

i HP, j CP, k ST, n {0} VT



(n)

(Ti
in(n)
Tiout )FCpi = qijk + qcui
(n)










k ST jCP





(n)



(n)




in(n)
(Tjout Tj )FCpj = qijk + qhuj







k ST iHP








(n)




(n) (n)
(tik ti,k+1 )FCpi = qijk







jCP








(n)



(n) (n)



(tjk tj,k+1 )FCpj = qijk







iHP







(n) (n)



Ti
in(n)
= (Ti
in(0)
+ ri ) = ti,1







in(n) in(0) (n) (n)



Tj = (Tj + rj ) = tj,NT +1







(n) (n)



tik ti,k+1











(n) (n)
tjk tj,k+1











(n)



Tiout ti,NT +1







(n)



Tjout tj,1












(n)
(ti,NT +1 Tiout )FCpi = qcui
(n)



V = xV (7)
out(n) (n) (n)

(Tj tj,1 )FCpj = qhuj





(n)


qijk zijk 0





(n)


qcui zcui 0





(n)


qhuj zhuj 0





(n) (n) (n)


dtijk tik tjk + (1 zijk )





(n) (n) (n)


dtij,k+1 ti,k+1 tj,k+1 + (1 zijk )





(n) (n)


dtcui ti,NT +1 TCU out + (1 zcu )
i





(n) (n)


dthuj THU tj,1 + (1 zhuj )
out






(0)
dtijk Tmin








(0)
dtcui Tmin





(0)


dthuj Tmin











zijk + zcui + zhuj MEUmax





iHP,jCP,k ST iHP jCP





zijk , zcui , zhuj {0, 1}





(n) (n) (n) (n) (n) (n) (n) (n)


tik , tjk , dtijk , dtcui , dthuj , qijk , qcui , qhuj 0




i HP, j CP, k ST, n {0} VT
 C.-L. Chen, P.-S. Hung / Chemical Engineering and Processing 44 (2005) 89100 93

Table 1
Problem data of example 1
Process streams and utilities Heat-capacity flow rate FCp (kW/K) Input temperature T in (K) Output temperature T out (K)

Hot stream 1 (H1) 10 650 370

Hot stream 2 (H2) 20 590 370
Cold stream 1 (C1) 15 410 650
Cold stream 2 (C) 13 350 500
Hot utility (HU) 680 680
Cold utility (CU) 300 320
Tmin = 10 K.

Table 2 In this paper, the targeted source-stream temperatures

The resulting HEN structure of example 1 with increasing flexibility target are directly treated as individual design objective, and the
( 0, see Table 2 of [10] for labeling HEN structures)
multi-criteria optimization approach is adopted for HEN
Specified value Resulting HEN structure Maximum flexibility synthesis. The minimizing utility and the maximizing oper-
0 36.95 A 36.95 ational flexibility can be simultaneously considered as two
36.95 +  B1a 70.63 conflict objectives for synthesis of the network structure.
40 B7a 90.00 Furthermore, other targets such as minimizing number of
50 B6 90.00 matches can also be considered, such as,
60 B1a 70.63
(0)  (0)  (0)
70 B7a 90.00 min J1 = iHP qcui + jCP qhuj or
80 B7a 90.00 xV V
90 B7a 90.00 (ave) 1

90 + 
min J1 =
C2 120.0 xV V NV + 1
120150 D 150.0 n{0}VT

(n)

(n)
qcui + qhuj and
increasing the value over [0, 150] and solving Eq. (4) iHP jCP (11)
or Eq. (5). There are 18 sets of structure-determining bi- max J2 = and
xV V



nary variables and 11 of them possess unique structures
min J3 = zijk
(see Table 2 of [10]). However, we found that the resulting xV V
iHP jCP kST
HEN structures with increasing flexibility targets are prone


to variation, and the true maximal flexibilities of these HEN + zcui + zhuj
iHP jCP
structures are usually greater than the required targets, as
illustrated in Table 2. Similar results can also be found In such a case, a unique HEN structure with satisfactory
should restricted heat-load constraints on vertices, as shown levels in nominal or average utility consumption and opera-
in Eqs. (8) and (9), be taken into consideration in the HEN tional flexibility as well as unit numbers will be obtained. A
synthesis. two-phase fuzzy optimization method is proposed to find a

(0)

(0) best compromised solution for the multi-criteria HEN syn-
min J (0) = qcui + qhuj (8) thesis problem, as discussed in the next section. The basic
xV V load
iHP jCP
number of constraints and variables for the multi-objective
or MILP formulation are summarized in the following.

1
1. The number of constraints:

min J (ave) = (a) for linear equality constraints: (NV + 1)[NT (NH +
xV V load NV + 1
n{0}VT NC ) + 4(NH + NC )];


(n)

(n)
qcui + qhuj (9)
iHP jCP

q(0) (1 ) q(n) q(0) (1 + )

ijk


ijk ijk









qcu(0) (1 ) qcu(n) qcu(0) (1 + )

(n)
i i i
(n) (n)
load = qijk , qcui , qhuj (10)




qhu(0) (n) (0)


j (1 ) qhuj qhuj (1 + )







i HP, j CP, k ST, n VT
94 C.-L. Chen, P.-S. Hung / Chemical Engineering and Processing 44 (2005) 89100

(b) for linear inequality constraints: (NV + 1)[(NH + Js (Js (xV ))

NC )(NT + 3) + 3NT NH NC ] + NT NH NC + NH + NC . 1; for Js1 Js

J0 J
2. The number of variables: s s
= for Js1 Js Js0 for Js to be minimized

(a) binary variables: NH NC NT + NH + NC ; Js J s
0 1

(b) positive continuous variables: (NV +1)[NT (2NH NC + 0; for Js Js0
NH + NC ) + 2(NH + NC )] + 1. (12)
Js (Js (xV ))


1; for Js Js1
3. Fuzzy multi-criteria optimization
1
Js J s
= for Js0 Js Js1 for Js to be maximized

J 1 J0
Consider the multi-criteria optimization problem de-
s s
fined in Eq. (11). Because of the fact that these objective 0; for Js0 Js
functions usually conflict with each other in practice, the (13)
optimization of one objective implies the sacrifice of other
The original multi-criteria optimization problem is now con-
targets; it is thus impossible to attain their own optima, Js ,
verted to the one that looks for a suitable decision variable
s S = [1, . . . , S], simultaneously. Therefore, the decision
vector that can provide the maximal degree-of-satisfaction
maker (DM) must make some compromise among these
for the multiple fuzzy objectives.
goals. In contrast to the optimality used in single objec-
tive optimization problems, Pareto optimality characterizes max (J1 (xV ), . . . , JS (xV )) (14)
xV V
the solutions in a multi-objective optimization problem
[13]. Under incompatible objective circumstances, a DM must
The weighting-sum method, among methods found in lit- make a compromise decision that provides a maximal
eratures for solving multi-objective optimization problems, degree-of-satisfaction for all these conflict objectives. The
is the one that is used most often. Basing on the subjective new optimization problem, Eq. (14), can be interpreted as
comprehension for each objective, the DM of this method the synthetic notation of a conjunction statement (maximize
can weigh and sum up these objectives into a scalar form, jointly all objectives). The result of this aggregation can be
and then find the solution by any existing single-objective viewed as a fuzzy intersection of all fuzzy goals, Js , s S,
optimization method. However, as the situation of combin- and is still a fuzzy set, D.
ing weighting factors becomes more complex, this method D = J1 . . . JS (15)
becomes more tedious and the solution could be still in-
valid. Moreover, it is difficult for the DM to attribute a set The final degree-of-satisfaction resulting from certain vari-
of incompatible objectives, such as utility consumption, able set, D (xV ) can be determined by aggregating the
operational flexibility, or number of matches in a heat ex- degree-of-satisfaction for all objectives, Js (xV ), s S, via
changer network, without knowledge of the possible level specific fuzzy intersection operator, T.
 
of attainment for these objectives. The physical mean- D (xV ) = T J1 (xV ), . . . , JS (xV ) (16)
ing of the final scalar objective function is thus usually
vague. The fundamental properties for a fuzzy set and the related
In this work, we adopt the fuzzy set theory [14] to deal operators can be found in [15]. As the firing level for each
with the multi-objective optimization problem. By consid- policy is determined by the above procedure, the best so-
ering the uncertain property of human thinking, it is quite lution, xV , with the maximal firing level, D (xV ), can be
natural to assume that the DM has a fuzzy goal, Js , to selected.
describe the objective Js with an interval [Js1 , Js0 ]. For the max D (xV ) = max T(J1 (xV ), . . . , JS (xV )) (17)
sth objective to be minimized, it is quite satisfied as the xV V xV V
objective value Js Js1 , and is unacceptable as Js Js0 . Using the fuzzy intersection operator, the original multi-
The degree of satisfaction decreases as the objective value objective optimization problem, Eq. (11), is converted into a
increases from Js1 to Js0 . A strictly monotonic decreas- single objective problem, Eq. (17). Several operators for im-
ing membership function, Js (Js (xV )) [0, 1], can be plementing fuzzy intersection can be selected for T, therein
used to characterize such a transition from the objective two most popular ones are shown below.
value to the degree of satisfaction, therein value of 1 de-
notes absolutely satisfactory, and 0 means unacceptable. T(J1 , . . . , JS )

Notably, an interval of [Js0 , Js1 ] and a monotonic increas- min(J1 , . . . , JS ) T = minimum
ing membership function should be used for defining a = (J1 + . . . + JS ) (18)
T = average
fuzzy objective to be maximized. Without loss of gener- S
ality, we will adopt linear membership functions in the The average operator simultaneously takes all membership
following. values into account. But, no guarantee can be made for any
 C.-L. Chen, P.-S. Hung / Chemical Engineering and Processing 44 (2005) 89100
single objective. The minimum operator concentrates on im-
proving the worst scenario. However, the minimum operator
may result in multiple solutions since contribution of those
objectives with membership values greater than the mini-
mum one would not be cared. We thus combine these two
fuzzy intersection operators to take advantages of both and
propose a two-phase optimization procedure [11,12]. The
minimum operator is first used in phase I to find the least
degree of satisfaction for the worst objective, then the aver-
age operator is applied in phase II to promote satisfaction
levels of all objectives with guaranteed least membership
value. The merit of this method is that we can not only ob-
tain the unique optimal solution by using average operator
but also guarantee each objective to go after their own max-
imum on the basis of taking the least degree of satisfaction
as the lower-bound constraint. So, now we can summarize
the procedure of the two-phase fuzzy satisfying approach
for the multi-criteria optimization problem.
Step 1. For a function to be minimized, determine its ideal
solution and anti-ideal solutions by directly mini-
mizing and maximizing the objective function.
min Js = J 1s
xV V
(ideal solution ofJs , totally acceptable
value) (19)
max Js = Js0
xV V
(anti ideal solution ofJs ,
unacceptable value) (20)
Notably, direct maximization and minimization
should be taken for a maximizing objective to
obtain the ideal and anti-ideal solutions.
Step 2. Based on the importance of different objective func-
tions and the acceptable ranges for objective val-
ues, subjectively select suitable lower/upper bounds,
J 1s Js1 Js0 Js0 for minimizing objective and
J 0s Js0 Js1 Js1 for maximizing objective. De-
membership functions for multiple fuzzy ob-
fine
jectives as given in Eqs. (12) and (13).
Step 3. (Phase I). Use the minimum operator to find the
maximal degree of satisfaction for the worst objec-
tive, min .
max D = max
xV V xV V (21)
min(J1 , . . . , JS ) = min
Step 4. (Phase II). Use the average operator to simultane-
ously promote satisfaction levels of all objectives
with new constraints of guaranteed minimal degree
of satisfaction.
J1 + + JS
max D = max (22)
+
xV V +
xV V S
where
+ J3 units
V = V {Js min , s S} (23)
95
The new feasible space guarantees the least degree
of satisfaction for each fuzzy objective.
4. Numerical example
Two numerical examples adapted from [4,10] are supplied
to demonstrate the efficacy of proposed HEN synthesis strat-
egy. To solve this MO-MILP for the HEN model, GAMS
[16] and CPLEX are used as the modeling environment and
the MIP solver, respectively. The computing machine is a
personal computer with an Intel Pentium IV 2.26 GHz CPU.
Example 1. The 2-hot/2-cold streams example studied by
[10], with problem data presented in Table 1, is illustrated.
With these parameters, the multi-objective MILP formula-
tion has 408 linear equality constraints, 760 linear inequal-
ity constraints, 12 binary variables, and 545 positive con-
tinuous variables. Notably, the restriction of MEUmax = 6
in Eq. (7) will be removed should the minimum number of
matches be simultaneously taken into account as one of the
design objectives.
According to the model formulation and the problem data,
we solve the multi-criteria MILP synthesis problem by us-
ing the fuzzy optimization procedure discussed in Section
3. The ideal and anti-ideal solutions, obtained by sequen-
tially minimizing and maximizing each objective function,
are shown in Table 3. Within these lower/upper limiting val-
ues, several preference intervals are subjectively selected to
establish linear membership functions for the fuzzy objec-
tives. In implementing the two-phase fuzzy optimization, we
firstly use the minimum operator to maximize the degree of
satisfaction for the worst target, min . The average operator
is then applied to optimize the aggregated objectives with
guaranteed level of satisfaction.
At first, only two conflict objectives are considered: the
minimal utility consumption and the maximal flexibility to
all source-stream temperatures. And then the third objective,
the minimal number of matches, would be appended. Results
of two-phase fuzzy optimization with preference intervals of
[2550, 12750] or [2550, 8850] for utility, [0, 150], [40, 90]
or [40, 70] for flexibility, and [4, 7] for unit numbers, along
with either or not considering restrictions on heat loads at
extreme operating points, are listed in Table 4. The resulting
HEN structures are also depicted in Fig. 2. Notably, the re-
duced range of flexibility, [40, 90], implies that the required
minimum tolerance for temperature deviation is at least 40 K
and a tolerance of maximum temperature deviation for 90 K
Table 3
The ideal/anti-ideal solutions of various objectives of example 1
Objective function Ideal solution J 1i Anti-ideal solution Ji0
(0)
J1 utility (kW) 2550 12750
J2 flexibility (K) 150 0
4 12
96 C.-L. Chen, P.-S. Hung / Chemical Engineering and Processing 44 (2005) 89100

Table 4
Results of HEN synthesis for example 1 using two-phase optimization when simultaneously considering minimal utility and maximal flexibility (cases
IIV), and additional objective of minimal units (cases V and VI) with different preference intervals, and with or without considering restriction on heat
loads at vertices ( = = 0.6)
Case Preference intervals Heat load Phase Minimum utility Maximum flexibility Minimum units

Hot Cold Total Satisfaction Flexibility Satisfaction Units Satisfaction

I [2550, 12750] No I 1300 2950 4250 0.833 120.0 0.800 6

(C2) [0, 150], [, 6] II 1300 2950 4250 0.833 120.0 0.800 6
II [2550, 8850] No I 1050 2700 3750 0.810 80.5 0.810 6
(B7) [40, 90], [, 6] II 1050 2700 3750 0.810 90.0 1.00 6
III [2550, 8850] Yes I 2190 3840 6030 0.448 62.4 0.448 6
[40, 90], [, 6] II 2190 3840 6030 0.448 62.4 0.448 6
IV [2550, 8850] Yes I 1893 3543 5436 0.542 56.3 0.542 6
[40, 70], [, 6] II 1893 3543 5436 0.542 56.3 0.542 6
V [2550, 8850] No I 1300 2950 4250 0.730 73.3 0.667 5 0.667
[40, 90], [4, 7] II 1300 2950 4250 0.730 74.2 0.685 5 0.667
VI [2550, 8850] Yes I 2550 4200 6750 0.333 56.7 0.333 6 0.333
[40, 90], [4, 7] II 2550 4200 6750 0.333 69.8 0.597 6 0.333

Fig. 2. The HEN structures for cases IVI of example 1.

 C.-L. Chen, P.-S. Hung / Chemical Engineering and Processing 44 (2005) 89100 97

Table 5 to the restriction of preference intervals, and the flexibility,

Problem data of example 2 90, is equivalent to the maximum of the B7 structure.
Process streams Heat-capacity Input Output All other conditions are equal to those in case II, case
and utilities flow rate FCp temperature temperature III includes heat-load restrictions ( = = 0.6) on vari-
(kW/K) T in (K) T out (K) ous vertical operating points as additional constraints. The
Hot stream 1 (H1) 6 500 320 prices of such additional restrictions are increased utility
Hot stream 2 (H2) 4 480 380
consumption from 3750 to 6030 and decreased flexibility to
Hot stream 3 (H3) 6 460 360
Hot stream 4 (H4) 20 380 360 temperature deviation from 90 to 62.4, which is still signifi-
Hot stream 5 (H5) 12 380 320 cantly greater than the minimum targeted value, 40. In case
Cold stream 1 (C1) 18 290 660 IV, the preference interval for temperature deviation is fur-
Hot utility (HU) 700 700 ther reduced to [40, 70]. It is found that the resulting HEN
Cold utility (CU) 300 320
has smaller utility consumption, 5438, with the expense of
Tmin = 10 K. further reduction on flexibility, 56.3, since our desideratum
Table 6
for flexibility has been made lower.
The ideal/anti-ideal solutions of various objectives of example 2 Cases V and VI take into account the unit number as
the third design objective, where heat-load restrictions on
Objective function Ideal solution J 1i Anti-ideal solutionJi0
vertices are either included or not. In case V, it is found
(0)
J1 utility (kW) 3780 9860 that the unit number is only five, flexibility to temperature
J2 flexibility (K) 20 0 deviation is 74.2, and the utility consumption is 4250, a little
J3 units 6 31
more than B4 and B5 of [10]. In case VI where heat-load
restrictions on vertices are further included, the total unit
is absolutely satisfied. We choose 90 as the maximum de- number becomes six, the total utility is increased to 6750
viation for most cases since it is the maximal tolerance for with a reduced maximum temperature flexibility of 69.8.
the HEN with heat load restrictions on vertices. From these results, it is found that the proposed
As shown in Table 4 and Fig. 2, the HEN structure in multi-criteria synthesis strategy can attain a definite and
case I is the same as C2 of [10], therein the maximum flex- compromised solution for a problem with assorted conflict
ibility to temperatures is explicitly given as 120. In case II, objectives. The preference intervals of various objectives
the B7 structure of [10] can be obtained when the tolera- have significant effects on final HEN structures. Such ac-
ble upper bound for utility consumption is decreased from ceptable and/or preference intervals can also reflect the
12750 to 8850, and the preference interval for flexibility is importance of different objective functions. Should one
reduced from [0,150] to [40,90]. The resulting utility con- specific objective is emphasized, a tighter restriction or
sumption will be reduced from 4250 (Case I) to 3750 due shrinking span should be placed on its acceptable ranges.

Table 7
Results of HEN synthesis for example 2 using two-phase optimization when simultaneously considering minimal utility, maximal flexibility and minimal
units with different preference intervals, and with or without considering restriction on heat loads ( = = 0.6)
Case Preference intervals Heat load Phase Minimum utility Maximum flexibility Minimum units
Hot Cold Total Satisfaction Flexibility Satisfaction Units Satisfaction

I [3780, 9860] No I 3863 403 4266 0.920 18.4 0.920 8 0.920

[0, 20], [6, 31] II 3793 333 4126 0.943 20.0 1.00 8 0.920
II [3780, 4930] No I 3712 252 3964 0.840 16.8 0.840 10 0.840
[0, 20], [6, 31] II 3660 200 3860 0.930 20.0 1.00 10 0.840
III [3780, 4930] No I 3811 351 4162 0.667 13.3 0.667 8 0.667
[0, 20], [6, 12] II 3793 333 4126 0.699 20.0 1.00 8 0.667
IV [3780, 4930] No I 3811 351 4162 0.667 16.7 0.667 8 0.667
[10, 20], [6, 12] II 3793 333 4126 0.699 20.0 1.00 8 0.667
V [3780, 9860] Yes I 3980 520 4500 0.882 10.6 0.882 8 0.920
[0, 12], [6, 31] II 3980 520 4500 0.882 12.0 1.00 8 0.920
VI [3780, 4930] Yes I 3742 282 4024 0.787 9.4 0.787 11 0.800
[0, 12], [6, 31] II 3742 282 4024 0.787 9.4 0.787 11 0.800
VII [3780, 4930] Yes I 3860 400 4260 0.583 6.99 0.583 8 0.667
[0, 12], [6, 12] II 3860 400 4260 0.583 10.0 0.833 8 0.667
VIII [3780, 4930] Yes I 3860 400 4260 0.583 9.5 0.583 8 0.667
[6, 12], [6, 12] II 3860 400 4260 0.583 10.0 0.667 8 0.667
98 C.-L. Chen, P.-S. Hung / Chemical Engineering and Processing 44 (2005) 89100

Fig. 3. The HEN structures for cases IVIII of example 2.

Observing the results shown in Table 4, we also discover
that the integrated two-phase method can bring the merits of
the minimum and average operators together. The minimum
operator maximize the degree of satisfaction for the worst
objective, and can result in a harmonious solution with
satisfaction levels for objective functions that are equal or
close to each other. The average operator, on the other hand,
can sometimes promote satisfaction levels for other objec-
tives with guaranteed minimum value, such as Cases II, V,
and VI. Owing to these advantages, the two-phase method
can thus provide the best compromised HEN configuration.
Example 2. This problem consists of five hot streams and
one cold stream, (NH = 5, NC = 1), along with steam and
cooling water as utilities [3]. The problem data are listed
in Table 5. The number of superstructure stages is set as
NT = 5. With these parameters, the MO-MILP formulation
has 3510 linear equality constraints, 8026 linear inequality
constraints, 31 binary variables, and 5981 positive continu-
ous variables. The ideal and anti-ideal solutions are shown
in Table 6. Various preference intervals are sequentially se-
lected for defining the membership functions, as shown in
Table 7.
We directly use the ideal/anti-ideal solutions as the pref-
erence intervals in Case I. The acceptable utility range is
reduced from [3780,9860] to [3780,4930] in Case II. With
such a smaller allowance interval, the resulting utility con-
sumption will be reduced slightly from 3793 to 3660 at the
expense of unit numbers increased from 8 to 10. In Case III,
 C.-L. Chen, P.-S. Hung / Chemical Engineering and Processing 44 (2005) 89100 99

the maximum allowable unit number is changed from 31 to NC number of cold streams
12. The resulting HEN structure is slightly different to that NH number of hot streams
of Case I, but the required unit numbers, utilities, and flexi- NT number of superstructure stages
bility are the same as Case I. In Case IV, when the minimum NV number of vertices, = 2N
flexibility requirement is upgraded from 0 to 10, the same qijk heat exchanged between stream i and j in stage k
HEN structure of Case I results with the same levels of util- qcui heat exchanged between stream i and cold utility
ity, flexibility and unit numbers. Cases VVIII give similar qhuj heat exchanged between stream i and hot utility
results with additional heat-load restrictions ( = = 0.6) r directional identifier for vertices
on various vertical operating points. With such additional ST index set of superstructure stages
constraints, the HEN structures use similar levels of utilities S number of objectives
and unit numbers, but the flexibility levels are dramatically Tmin minimum approach temperature
reduced. The resulting HEN structures are depicted in Fig. 3. tik temperature of stream i at hot end of stage k
tjk temperature of stream j at hot end of stage k
T temperature
5. Conclusion VT index set of vertices
x, xV vector of variables
In this paper, we use the fuzzy multi-criteria optimization zijk binary variable for existence of unit for match i
approach to synthesize the heat-exchanger network where and j in stage k
some conflict design objectives such as the total utility zcui binary variable for existence of unit for match i
consumption, the flexibilities to source-stream temperature and cold utility in stage k
variations, and even the total number of heat exchange units zhui binary variable for existence of unit for match j
can be considered simultaneously. Such a flexible HEN and hot utility in stage k
synthesis problem can be formulated as a multi-objective
mixed-integer linear programming (MO-MILP). For han- Greek letters
dling the multiple conflict design objectives, a two-phase , parameters used for restriction of heat-load
fuzzy optimization method is proposed to attain the best deviations
compromised solution. The attractive features of the pro- flexibility index
posed MO-MILP model are that it not only considers the flexibility index (scalar)
trade-off among the utility consumption, the source-stream Js membership function for Js
temperature flexibility, and even the number of matches, Js a fuzzy goal
but also avoids the determination of structural boundaries, S index set of multiple objectives
as discussed in [10]. Two numerical examples with various D a fuzzy set
cases are studied, demonstrating that the proposed strat- D degree of satisfaction
egy can provide a feasible compensatory solution for the upper bound for temperature difference
multi-criteria HEN synthesis problem. upper bound for heat exchange
the feasible searching region

Acknowledgements Superscripts
in inlet
(n) identifier for vertices
This work is supported by the National Science Council
out outlet
(ROC) under Contract NSC91-ET-7-002-004-ET. Partial fi-
(0) identifier for nominal operating condition
nancial support of Ministry of Economic Affairs under grant
92-EC-17-A-09-S1-019 is also acknowledged.
Subscripts
CU cold utility
Appendix A. Nomenclature HU hot utility
i index for hot process streams
CP index set of cold process stream j index for cold process streams
dtijk temperature approach for match i and j in stage k k index for superstructure stages
dtcui temperature approach for match i and cold utility s index for objectives
dthuj temperature approach for match j and hot utility
FCp heat capacity flowrate References
HP index set of hot process stream
J objective function [1] K.C. Furman, N.V. Sahinidis, A critical review and annotated bib-
MEUmax maximum number of heat-exchange units liography for heat exchanger network synthesis in the 20th century,
N number of uncertain parameters Ind. Eng. Chem. Res. 41 (2002) 23352370.
100 C.-L. Chen, P.-S. Hung / Chemical Engineering and Processing 44 (2005) 89100
[2] B. Linnhoff, E. Hindmarsh, The pinch design method for heat ex-
changer networks, Chem. Eng. Sci. 38 (5) (1983) 745763.
[3] T.F. Yee, I.E. Grossmann, Z. Kravanja, Simultaneous optimization
models for heat integration-I. Area and energy targeting and modeling
of multi-stream exchangers, Comp. Chem. Eng. 14 (10) (1990) 1151
1164.
[4] T.F. Yee, I.E. Grossmann, Simultaneous optimization models for heat
integration-II. Heat exchanger network synthesis, Comp. Chem. Eng.
14 (10) (1990) 11651184.
[5] C.A. Floudas, Nonlinear and Mixed-Integer Optimization: Funda-
mentals and Applications, Oxford University Press, New York, 1995.
[6] L.T. Biegler, I.E. Grossmann, A.W. Westerberg, Systematic Methods
of Chemical Process Design, Prentice Hall, Englewood Cliffs, NJ,
1997.
[7] K.P. Halemane, I.E. Grossmann, Optimal process design under un-
certainty, AIChE J. 29 (3) (1983) 425433.
[8] R.E. Swaney, I.E. Grossmann, An index for operational flexibility
in chemical process design, AIChE J. 31 (4) (1985) 621630.
[9] I.E. Grossmann, C.A. Floudas, Active constraint strategy for flexi-
bility analysis in chemical processes, Computers Chem. Eng. 11 (6)
(1987) 675693.
[10] A.E.S. Konukman, M.C. Camurdan, U. Akman, Simultaneous flex-
ibility targeting and synthesis of minimum-utility heat exchanger
networks with superstructure-based MILP formulation, Chem. Eng.
Processing 41 (2002) 501518.
[11] C.L. Chen, B.W. Wang, W.C. Lee, Multi-objective Optimization for
Multi-enterprise Supply Chain Networks, Ind. Eng. Chem. Res. 42
(2003) 18791889.
[12] R.J. Li, E.S. Lee, Fuzzy multiple objective programming and com-
promise programming with Pareto optimum, Fuzzy Sets and Systems
53 (1993) 275288.
[13] Sakawa, M., Fuzzy Sets and Interactive Multi-Objective Optimiza-
tion, Plenum Press, New York, 1993.
[14] L.A. Zadeh, Fuzzy sets, Information and Control 8 (3) (1965) 338
353.
[15] G.L. Klir, B. Yuan, Fuzzy Sets and Fuzzy Logics-Theory and Ap-
plication, Prentice Hall, New York, 1995.
[16] A. Brooke, D. Kendrick, A. Meeraus, R. Raman, R.E. Rosen-
thal, GMAS: A Users Guide, GAMS Development Corporation,
1988.
 Chemical Engineering and Processing 44 (2005) 101114

On-line dynamic optimization and control strategy for

improving the performance of batch reactors
A. Arpornwichanop a , P. Kittisupakorn a , I.M. Mujtaba b,
a Department of Chemical Engineering, Chulalongkorn University, Bangkok 10330, Thailand
b School of Engineering, Design & Technology (EDT 3), University of Bradford, Bradford, West Yorkshire BD7 1DP, UK

Received 23 December 2002; received in revised form 25 August 2003; accepted 28 April 2004
Available online 2 July 2004

Abstract

Since batch reactors are generally applied to produce a wide variety of specialty products, there is a great deal of interest to enhance batch
operation to achieve high quality and purity product while minimizing the conversion of undesired by-product. The use of process optimization
in the control of batch reactors presents a useful tool for operating batch reactors efficiently and optimally. In this work, we develop an approach,
based on an on-line dynamic optimization strategy, to modify optimal temperature set point profile for batch reactors. Two different optimization
problems concerning batch operation: maximization of product concentration and minimization of batch time, are formulated and solved using
a sequential optimization approach. An Extended Kalman Filter (EKF) is incorporated into the proposed approach in order to update current
states from their delayed measurement and to estimate unmeasurable state variables. A nonlinear model-based controller: generic model
control algorithm (GMC) is applied to drive the temperature of the batch reactor to follow the desired profile. A batch reactor with complex
exothermic reaction scheme is used to demonstrate the effectiveness of the proposed approach. The simulation results indicate that with the
proposed strategy, large improvement in batch reactor performance, in term of the amount of a desired product and batch operation time, can
be achieved compared to the method where the optimal temperature set point is pre-determined.
This paper gathers and integrates efficient well-known methodologies such as EKF, GMC and on-line optimization together, resulting in
an applicable, reliable and robust control technique for batch reactors.
2004 Elsevier B.V. All rights reserved.

1. Introduction also especially suitable to carry out reactions where materi-

In many chemical industries, there is an increasing trend to
place a consideration on the production of high value prod-
ucts (e.g. polymers, pharmaceuticals, and specialty chemi-
cals) in batch processes. As an important main unit in such
processes, a batch reactor is generally involved in manufac-
turing of these products. The use of batch reactors offers
many advantages. Firstly, a batch reactor is quite flexible, it
can adapt to small volume production of various products,
which are greatly submitted to the rapid changes in market
conditions and the advent of new technology. Secondly, a
batch reactor provides the natural way to scale-up processes
from laboratory experiment where complex chemical syn-
thesis is studied, to industrial manufacturing. Finally, it is
Corresponding author. Tel.: +44 1274 233645;
fax: +44 1274 235700.
E-mail address: i.m.mujtaba@bradford.ac.uk (I.M. Mujtaba).
als involved are dangerous and difficult to handle [1,2].
As batch reactors are used to produce a wide variety of ex-
pensive products, it is known that this process involves sev-
eral competing reactions which may cause undesired prod-
uct or waste. As a result, there is a great deal of interest
to enhance batch operation to achieve high quality and pu-
rity products while minimizing the conversion of undesired
by-products. Recently, the use of process optimization in the
control of batch reactors has received much attention in the
literature. This provides a useful tool for operating batch re-
actors efficiently and optimally. For this purpose, it is desir-
able to optimize the process conditions during its operation
in order to meet the desired product and safety specifica-
tions. A control system is an essential part to ensure that the
desired operating conditions can be maintained as close as
possible during the course of a batch run.
However, achieving such a proposed method for an op-
timal operation of batch reactors is quite difficult, and still

0255-2701/$ see front matter 2004 Elsevier B.V. All rights reserved.
doi:10.1016/j.cep.2004.04.010
102 A. Arpornwichanop et al. / Chemical Engineering and Processing 44 (2005) 101114
provides challenging and interesting problems. This is
mainly due to the inherent complexity of the batch reac-
tors which can be characterized by (i) highly nonlinear
behavior resulting from the dependence of reaction rates on
concentrations and temperature, (ii) time-varying system;
the process variables (e.g. concentration, temperature) and
parameters change with time, (iii) no steady state operating
condition, (iv) imperfect model; complex kinetic reactions
occurring within batch reactors are rarely well understood
that leads to an inaccuracy in developing the system model,
and (v) lack of measurement information; the product
qualities or key properties to be controlled (e.g. molecular
weight) cannot be measured until the end of batch run or
even if they can be measured sometime (e.g. concentration),
there is a significant time delay. Only a few of physical
quantities such as temperature and pressure are available
for direct on-line measurement. That makes direct con-
trol of product properties difficult [2]. Although, in recent
years there have been significant advances in developing
new sensors for measuring these product properties, they
have rarely been used in industrial processes due to high
operating cost and expensive investment on the measure-
ment devices. Thus, the usual practice is to control other
variables that can be measured rapidly in order to obtain
desired product properties instead.
In general, optimal batch reactor operation can be car-
ried out by two-step approach; firstly, determining an opti-
mal set point profile of key operating process variables such
as temperature [3]and secondly, tracking the desired profile
by a control system [4]. The optimal profile can usually be
determined off-line by solving an optimal control problem.
However, as mentioned above, because of the complexity
of chemical reaction schemes, modeling errors are always
present and in addition, process disturbances can occur dur-
ing the process operation. Due to the existence of this error
and disturbance, the final product may significantly differ
from the desired value, even though the pre-specified opti-
mal profile is tracked perfectly [5]. To realize this fact, it
is necessary to recalculate the optimal profile as an on-line
optimization strategy whenever new feedback information
is available. This strategy could compensate the modeling
error leading to process operation improvement.
In order to perform the on-line optimization strategy, the
knowledge of current state variables and/or parameters in
the process models is required. Due to the fact that some of
these variables cannot be known exactly or sometime can
be measured with time delay, it is essential to include an
on-line estimator to estimate these process variables using
available process measurements as well. The sequence of
an estimation and optimization procedure is known as an
estimation-optimization task [6]. As in several estimation
techniques, an Extended Kalman Filter (EKF) has become
increasingly popular because it is relatively easy to imple-
ment. It has been found that the EKF can be applied to a
number of chemical process applications with great success.
Once the estimate of unknown process variables is deter-
mined and then the model is updated, the optimization is
performed on-line to generate a new optimal input profile.
With the modified optimal profile, a designed controller is
used to control the system to follow this profile until the
new one is available.
Apart from specifying the optimal set point profile, a con-
trol system used to implement such a profile is another im-
portant issue to be considered. This is because the deviation
from the desired profile may cause an off-spec product. How-
ever, since it is well known that the control of batch reactors
is difficult due to the inherently nonlinear behavior, the use
of a linear control technique may give a poor performance.
For this reason, many advanced control techniques have been
developed and applied to the control of batch reactor. These
include, for example, nonlinear feedforward-feedback con-
trol [7], generic model control [8], adaptive control [1], glob-
ally linearizing control [9], dynamic matrix control [10],
linear model predictive control [11], or inverse model con-
trol [12]. Reviews on the progress in control techniques that
have been applied to the control of batch reactors as well as
their importance and performance are given by [13]. Among
these advanced control methods, a generic model control
(GMC) technique is one of the most studied control algo-
rithm. This is because nonlinear process models can be in-
terpreted straightaway in the GMC control algorithm so that
they do not need to be linearized. Furthermore, its implemen-
tation is relatively easy when compared to other model-based
control methods; consequently, the application of this con-
trol technique appears in many chemical processes.
In this work, an approach based on an on-line dynamic op-
timization strategy to modify optimal temperature set point
profile is developed for improving batch operation perfor-
mance. To demonstrate the effectiveness of the developed
approach, the batch reactor studied by [8], where two paral-
lel exothermic reactions occur, is chosen here as a case study.
For solving the on-line optimization problem, it needs the
knowledge of the current states of the system. Although most
physical quantities (e.g. temperature, flowrate) can be mea-
sured frequently and are available as on line measurement,
some other properties (e.g. concentration) are measured in-
frequently with time delay. To overcome this difficulty, the
Extended Kalman Filter (EKF) is incorporated into the pro-
posed strategy in order to estimate the concentrations from
their delayed measurements. The resulting optimal control
problem is solved by the sequential solution and optimiza-
tion method which is often referred to as the control vector
parameterization (CPV) method. Once the optimal tempera-
ture profile is modified, a controller based on generic model
control algorithm (GMC) is applied to control the batch re-
actor temperature following the desired profile. In the GMC
formulation, the EKF is also used to estimate the heat re-
leased from the chemical reactions. It should be noted that
the proposed strategy would be one of several strategy stud-
ied to promote the applicability of on-line dynamic opti-
mization with set point tracking for improving a batch pro-
cess; the aim of this work is to integrate efficient methods
 A. Arpornwichanop et al. / Chemical Engineering and Processing 44 (2005) 101114 103

including the EKF and GMC and on-line optimization lead- with
ing to an applicable, reliable and successful strategy for an

k12
actual implementation. k1 = exp k11
Tr + 273.15


2. Batch reactor k22
k2 = exp k21
Tr + 273.15
A reactor system considered by [8] which consists of a
batch reactor and jacket cooling system is chosen here as W = MWA MA + MWB MB + MWC MC + MWD MD
a case study. The typical diagram of this system is shown M r = MA + M B + M C + M D
in Fig. 1. It is assumed that two parallel highly exothermic
reactions are carried out in the reactor (CpA MA + CpB MB + CpC MC + CpD MD )
Cpr =
k1 Mr
A + B
C
Qr = H1 (k1 MA MB ) H2 (k2 MA MC )
k2
A + C
D Qj = UA(Tj Tr )

where A and B are raw materials, C and D are desirable 2W

A=
product and undesirable by-product, respectively. The rate r
constant k1 and k2 are temperature dependence according to where Mi is the amount of mole of component i, Tr is the
the Arrhenius relation. reactor temperature, Tj is the jacket temperature, and Tjsp is
a set point value of the jacket temperature control system.
2.1. Reactor model The meaning of other variables and parameters are explained
in the nomenclature.
The batch reactor is modeled by the following equations: The dynamic behavior of the reactor can be simulated
Material balances in the reactor by solving Eqs. (1)(6). The differential-algebraic solver
dMA DASSL [14] is used to give the solution of these equations.
= k1 MA MB k2 MA MC (1)
dt The initial conditions for MA , MB , MC , MD used in all
dMB simulation studies are 12, 12, 0, and 0 kmol, respectively.
= k1 MA MB (2) The initial values of both reactor and jacket temperature are
dt
set to 20 C. Other process parameter values used in the
dMC
= +k1 MA MB k2 MA MC (3) reactor models are listed in Table 1.
dt In this work, it was assumed that all temperature can be
dMD measured frequently without delay. The sampling time of
= +k2 MA MC (4)
dt
Energy balances around the reactor Table 1
Process parameter values
dTr Qr + Q j
= (5) MWA = 30 kg/kmol
dt Mr Cpr MWB = 100 kg/kmol
Fj j Cpj (Tjsp Tj ) Qj MWC = 130 kg/kmol
dTj
= (6) MWD = 160 kg/kmol
dt Vj j Cpj k11 = 20.9057
k12 = 10000
k21 = 38.9057
TRC Tr
k22 = 17000
r = 0.5 m
TJC Tj Fj = 0.348 m3 /min
U = 40.842 kJ/(min m2 C)
CpA = 75.31 kJ/(kmol C)
CpB = 167.36 kJ/(kmol C)
CpC = 217.57 kJ/(kmol C)
Cooling fluid CpD = 334.73 kJ/(kmol C)
H1 = 41840 kJ/kmol
Heating fluid H2 = 25105 kJ/kmol
= 1000 kg/m3
j = 1000 kg/m3
Heat exchanger Batch reactor
Cpj = 1.8828 kJ/(kg C)
Fig. 1. Batch reactor system. Vj = 0.6912 m3
104 A. Arpornwichanop et al. / Chemical Engineering and Processing 44 (2005) 101114
0.2 min is used for temperature measurements. It is noted
that measurement noise is also included in the temperature
sensors having Gaussian noise with zero mean and 1 C stan-
dard deviation. In addition, it was assumed that the concen-
tration (amount of mol) of reactants in the reactor is mea-
sured infrequently and has a sampling time and measure-
ment delay of 10 min.
3. On-line dynamic optimization strategy
The aim of a dynamic optimization problem is to deter-
mine a control profile minimizing (or maximizing) a given
objective function subject to process constraints. With the
optimal control policy, the controlled system is driven from
the initial state to a final desired state in an optimal way.
However, in the presence of modeling error, the pre-specified
control profile may lose its optimal character [2]. For this
reason, an on-line optimization strategy through a receding
horizon scheme is employed in this work to compensate
such an error. The basic concept is to compute the optimal
control profile based on current feedback information. How-
ever, only the initial value of the optimal trajectory is sent
to the system as a set point for the controller. After new
information of states is available from either measurement
or estimation, the optimization is repeated again to generate
updated optimal set point profile at the next time interval.
The method proposed for improving the batch operation
can be divided into two phases: on-line modification of the
reactor temperature trajectory and on-line tracking of the de-
sired temperature trajectory. The first phase involves deter-
mining an optimal temperature set point profile by solving
the on-line dynamic optimization problem and will be de-
scribed in this section. The other phase involves designing a
nonlinear model-based controller to track the obtained tem-
perature set point and will be presented in the next section.
Since both the on-line dynamic optimization and the
model-based control strategy rely on process models, the
knowledge of current states and/or model parameters is
required. However, in most industrial processes, state vari-
ables are not all measurable and some parameters are not
known exactly. As a consequence, there is a need for es-
timating these states and parameters. In this work, two
Extended Kalman Filters (EKF) are implemented. The first
one is applied to predict the reactant concentration, which
will be used for on-line dynamic optimization, from its
delayed measurement. The other one is applied to esti-
mate the unknown heat of reaction, which will be used
for model-based controller, from the frequently available
measurements of temperature.
3.1. Problem statements
As a batch reactor is utilized for the production of a wide
variety of high value products, an optimization of batch op-
erating conditions, e.g. temperature, operating time, etc. is
the more efficient approach to obtain a maximum yield in a
minimum time or minimum cost and also to reach the spe-
cific final conditions of the products in terms of quality and
quantity.
In general, an objective function in the optimization prob-
lem can be chosen, depending on the nature of the problem.
Here, two practical optimization problems related to batch
operation: maximization of product concentration in a fixed
batch time and minimization of batch operation time given
amount of desired product, are considered to determine an
optimal reactor temperature profile. The first problem for-
mulation is applied to a situation where we need to increase
the amount of desired product while batch operation time is
fixed. This is due to the limitation of complete production
line in a sequential processing. However, in some circum-
stances, we need to reduce the duration of batch run to allow
the operation of more runs per day. This requirement leads
to the minimum time optimization problem. These problems
can be described in details as follows.
3.1.1. Maximum product concentration problem (P1)
In this type of problem, the objective is to compute the
optimal temperature policy maximizing the amount of a de-
sired product concentration for a given fixed batch time sub-
ject to bounds on the reactor temperature. The problem can
be written mathematically as
max J = X(tf )
T(t)
subject to
x = f(x(t), T, p, t) process model
x(t0 ) = x(0) initial conditions
TL T TU
tf = tf
where X is the amount of the desired product at a given final
batch time, x is the vector of state variables, x is the derivative
of x with respect to time (t), T is the reactor temperature, p
are process parameters, tf is the fixed batch time, and TL and
TU are lower and upper bounds of the reactor temperature.
3.1.2. Minimum batch time problem (P2)
The purpose of this optimization problem is to determine
the optimal temperature profiles to achieve the desired fi-
nal product concentration in minimum batch time, thus the
performance index is the final time whereas the desired pro-
duction concentration is defined as a terminal constraints.
The formulation of the minimum batch time problem can be
shown as
min J = tf
T(t)
subject to
x = f(x(t), T, p, t) process model
 A. Arpornwichanop et al. / Chemical Engineering and Processing 44 (2005) 101114
x(t0 ) = x(0) initial conditions
TL T TU
X(tf ) = X
where X is the desired product concentration at the end of
batch run and tf is final batch time.
3.2. Solution of dynamic optimization problems
Computational techniques for the solution of a dynamic
optimization problem as formulated above have been an ac-
tive area of research. There are a number of different tech-
niques that have been proposed in the literature to solve this
class of problems. In general, they are mainly classified into
three classes.
The first one is based on a classical variation method.
This approach is also known as an indirect method as it
focuses on obtaining the solution of the necessary conditions
rather than solving the optimization directly. Solution of
these conditions often results in a two-point boundary value
problem (TPBVP), which is accepted that it is difficult to
solve [15]. Although several numerical techniques have been
developed to address the solution of TPBVP, e.g. control
vector iteration (CVI) and single/multiple shooting method,
these methods are generally based on an iterative integration
of the state and adjoint equations and are usually inefficient
[16]. Another difficulty relies on the fact that it requires an
analytical differentiation to derive the necessary conditions.
The second class of solutions is based on dynamic pro-
gramming. Unlike the variation method, this approach
applies the principle of optimality to formulate an op-
timization problem, leading to the development of the
HamiltonJacobiBellman partial equations that determine
the solution of the optimal control problem. However,
this approach is quite limited to a simple control prob-
lem [17] because of a difficulty in obtaining the solution
of the optimality equations. [18] extended the idea of the
optimality principle to develop an alternative technique,
named as iterative dynamic programming (IDP). Although
the implementation of the IDP for solving many optimal
control problems can be found in the literature, it is known
that the IDP algorithm would be slower than most other
gradient-based algorithms [19].
The last one is based on discretization techniques, re-
ceived major attention and considered as an efficient solu-
tion method. The concept of this approach is to transform the
original optimal control problem into a finite dimensional
optimization problem, typically a nonlinear programming
problem (NLP). Then, the optimal control solution is given
by applying a standard NLP solver to directly solve the op-
timization problem. For this reason, the method is known as
a direct method. The transformation of the problem can be
made by using discretization technique on either only con-
trol variables (partial discretization) or both state and con-
trol variables (complete discretization). Based on this con-
105
sideration, this approach can be divided into two categories:
sequential and simultaneous strategy.
In the sequential strategy, a control (manipulated) variable
profile is discretized over a time interval. The discretized
control profile can be represented as a piecewise constant,
a piecewise linear, or a piecewise polynomial function. The
parameters in such functions and the length of time subin-
terval become decision variables in optimization problem.
This strategy is also referred to a control vector parameter-
ization (CVP).
One advantage in the sequential approach is that only the
parameters that are used to discretize the control variable
profile are considered as the decision variables. The opti-
mization formulated by this approach is a small scale NLP
that makes it attractive to apply for solving the optimal con-
trol with large dimensional systems that are modeled by a
large number of differential equations. In addition, this ap-
proach can take the advantage of available IVP solvers. How-
ever, the limitation of the sequential method is a difficulty
to handle a constraint on state variables (path constraint).
This is because the state variables are not directly included
in NLP.
In contrast to the sequential solution method, the simul-
taneous strategy solves the dynamic process model and the
optimization problem at one step. This avoids solving the
model equations at each iteration in the optimization algo-
rithm as in the sequential approach. In this approach, the dy-
namic process model constraints in the optimal control prob-
lem are transformed to a set of algebraic equations which
is treated as equality constraints in NLP problem [20]. To
apply the simultaneous strategy, both state and control vari-
able profiles are discretized by approximating functions and
treated as the decision variables in optimization algorithms.
The main advantage of the simultaneous approach is a ca-
pability in handing constraints on the state variables. This
is because these constraints can be dealt with by including
them directly in the NLP as additional constraints. However,
due to the discretization on both state and control variables,
this leads the simultaneous approach to a large scale opti-
mization problem consisting of a large set of algebraic con-
straints and decision variables and needs a special solution
strategy.
As the ability of the sequential approach to handle large
systems without the need to solve excessively large opti-
mization problem, this approach is utilized in this study to
solve the optimal control problem. The formulation of the
optimal control problem as a nonlinear programming is de-
scribed below.
3.2.1. Sequential approach
Consider a dynamic process model as in the form of an
implicit function
f(t, x(t), x(t), u(t), p) = 0 [t0 , tF ] (7)
where x(t) is the set of all state variables, x denotes the
derivative of x(t) with respect to time, u(t) is a vector of con-
106 A. Arpornwichanop et al. / Chemical Engineering and Processing 44 (2005) 101114
trol variables, and p is a vector of time invariant parameters.
The time interval of interest is [t0 , tF ] and the function f is
assumed to be continuously differentiable with respect to all
its arguments. For given initial condition x(t0 ), the optimal
control of a dynamic process represented by Eq. (7) can be
determined by computing u(t) maximizing (or minimizing)
an objective function J of the form
J = F(x(tF ); x(tF ); u(tF ); tF )
subject to bounds on u(t) and terminal constraints. To trans-
form such a dynamic optimization problem into a nonlin-
ear programming problem via the sequential approach, the
control u(t) is approximated by a finite dimensional repre-
sentation. The time interval [t0 , tF ] is divided into a finite
number of subintervals (P). In each subinterval, the control
u(t) is represented by a set of basis functions involving a
finite number of parameters
u(t) = j (t, zj ), t [tj1 , tj ] with j = 1, 2, . . . , J
where tJ = tF . The control profile is defined by the param-
eters zj and switching times tj . In this study, the piecewise
constant control is assumed and used because the form of
the solution is ideally suited for implementation on a digital
computer. Thus the set of decision variables for the nonlin-
ear program can be written as
y = {z1 , z2 , . . . , zJ , t1 , t2 , . . . , tJ }
The resulting nonlinear programming problem of the form
min(or max ) J(y)
y delay of one sampling time; that is, at time k of the reactor,
subject to equality constraints (process model), inequality
constraints, bounds on control, etc. is solved using an effec-
tive decomposition algorithm named the successive reduced
quadratic programming developed by [21]. Computational
procedure of the sequential approach for the optimization
problem (Eq. (11)) is illustrated in Fig. 2. With the initial
guess of the decision variables (y), an integrator based on
Set initial conditions
Guess initial control parameters
Model Solver
Evaluate
Objective function
Constraints
Gradient
no
Check
NLP
tolerance
yes
Optimal control
parameters
Fig. 2. Optimal control solution via the sequential approach.
Gears type method is used for solving the process model
which provides the value of the objective function and con-
straints. Gradient information of the objective function and
constraints with respect to the decision variables is eval-
uated in an efficient way using adjoint variable approach.
Then, a NLP solver determines a new set of control param-
eters and sends it back to the model solver. This procedure
is repeated until the optimal value is found satisfying a
(8)
specified accuracy.
3.3. State and parameter estimation
The implementation of the on-line optimization strategy
requires the knowledge of current states and/or parameters in
nonlinear process models in order to modify a new optimal
profile defined as a set point for a controller. It is known
that some measurements i.e. concentration are available at
low sampling rate with significant time delay. To overcome
this difficulty, state and parameter estimation is incorporated
(9) into the proposed on-line optimization algorithm.
In this work, an Extended Kalman Filter (EKF), an ex-
tension of the Kalman Filter, is designed to reconstruct the
current state variables from the delayed state measurement.
The advantage of the EKF is that it requires information only
from the previous sampling time and allows prior knowledge
of a system via process models to be used for the estimation.
The algorithm of the EKF can be seen in Appendix A.
(10)
3.3.1. Application to the batch reactor
Since the concentrations of reactants (MA , MB , MC and
(11) MD ) in batch reactor are assumed to be measurable with a
only information at time k 1 is available. Thus, the EKF
is applied to estimate the value of reactant concentration at
current time k from their delayed measurements at sampling
time k 1. However, since it is expected to exhibit uncer-
tainty in reaction rate constant (i.e. k21 and k22 ) in real plant,
the EKF is also used to estimate these uncertain parame-
ters. The following equations, therefore, are appended for
parameter estimation
dk21
=0 (12)
dt
dk22
=0 (13)
dt
Eqs. (1)(4), (12) and (13) correspond to Eq. (A.1) in
the EKF algorithm. Based on the estimate of the current
information, the dynamic optimization problem is resolved
to generate a new optimal temperature trajectory.
4. Generic model control (GMC)
This section presents the design of a controller to control
the reactor temperature following a desired temperature tra-
 A. Arpornwichanop et al. / Chemical Engineering and Processing 44 (2005) 101114 107

jectory. It is accepted that the use of linear control techniques is negligible compared to the heat transferred in the reactor,
in highly nonlinear and time variant chemical processes (e.g. the energy balance equation becomes
batch reactors) is quite limited to their performances and Qr + Ur Ar (Tj Tr )
dTr
may give a poor control response. Therefore, in this work, = (18)
a nonlinear control technique based on generic model con- dt Wr Cpr
trol (GMC) algorithm is utilized. This model-based control where Ur is the heat transfer coefficient, Ar is the heat trans-
methodology has been received much interest during the last fer area, Wr is the mass of the reactor contents, Cpr is the
decade and a number of applications of GMC to the con- mass heat capacity of the reactor content, and Qr is the heat
trol of batch processes have been reported in the literature released by the reactions.
(e.g. [4,8,22,23], etc.). However, most of these works focus Rearranging the Eq. (18) as in the form of GMC algo-
on using the GMC to track the pre-determined optimal pro- rithm, the following functions, f, g, and h can be defined
file (off-line calculation) of a reactor temperature. No effort
Qr Ur Ar Tr
has been made to apply the GMC to implement an on-line f(x, p, t) = (19)
optimal set point profile. Therefore, in this work, the perfor- Wr Cpr
mance of the GMC controller with optimal temperature set Ur Ar
points determined by on-line optimization strategy is also g(x, t) = (20)
Wr Cpr
evaluated and compared to that of the GMC controller with
pre-determined set point. h(x) = Tr (21)

4.1. Control algorithm Replacing these equations in Eq. (17), we have


Let us consider a process based on the following model Wr Cpr
Tj = Tr + K1 (Trsp Tr )
equations: U r Ar
 t 
dx Qr
= f(x, p, t) + g(x, t)u (14) + K2 (Trsp Tr ) dt (22)
dt 0 Ur A r
y = h(x) (15) In order to make the GMC control law available for an
on-line control implementation, the integral term in Eq. (22)
where x is a vector of state variables, y is a vector of output
is approximated by numerical integration. This leads to the
variables, u is a vector of input variables, p is vector of
discrete-time form of the GMC algorithm as given in the
process parameters, and f, g, and h are generally nonlinear
following equation.
function vectors. The general form of GMC algorithm can
be written as

 tf Wr Cpr
dy Tj (k) = Tr (k) + K1 (Trsp Tr (k))
= K1 (ysp y) + K2 (ysp y) dt (16) U r Ar
dt 0 
k
The GMC control response can be designed via the tuning Qr (k)
+ K2 (Trsp Tr (k)) t (23)
parameters K1 and K2 based on the tuning curve given by Ur A r
0
[24]. It should be noted that the GMC approach is a special
case of the global input output linearizing control technique where t is the sampling time chosen to be equal to the
in which a transformed control action is chosen properly frequency of the temperature measurement. It is sufficiently
with the external PI controller. The use of Eq. (16) forces y small to approximate the continuous-time form without re-
toward its set point, ysp , with zero offset. If Eq. (15) is dif- quiring a large number of sampled values.
ferentiated, and the Eq. (16) is substituted into the resulting However, Eq. (23) gives the actual jacket temperature
equation, the GMC control law is (Tj (k)) required at the next sampling time to control the re-
 actor temperature (Tr (k)) at the desired trajectory (Trsp ). As
[K1 (ysp y) + K2 (ysp y) dt (dh/dx)f(x, d, t)] in usual practice, the reactor temperature control is cascaded
u=
(dh/dx)g(x, t) with the jacket temperature control (heating and cooling sys-
(17) tem); the output of reactor temperature controller (master
loop) is the set point of the jacket temperature controller
4.2. Application of GMC controller to the batch reactor (slave loop), as demonstrated in Fig. 3. In addition, as the
model of the heat exchanger system for heating and cooling
To implement the GMC, an energy balance around the is not included in Eq. (18), if Tj (k) is applied directly as the
reactor is required; it gives the relation between the reactor set point for jacket temperature control system without con-
temperature (controlled variable) and the jacket temperature sidering on its dynamic, the resulting control response would
(manipulated variable). Based on the assumption that the be sluggish. To accommodate such an effect, it is reason-
amount of the heat accumulated in the walls of the reactor able to assume that the dynamic of the jacket control system
108 A. Arpornwichanop et al. / Chemical Engineering and Processing 44 (2005) 101114

Initial
setpoint Tj Tjsp Tr(k) Tj(k)
Dynamic
GMC REACTOR
compensation

Mi (K-1)

Qr(k) EKF
with simplified
reactor model

EKF

Updated Mi (K)

Trsp (K)
On-line dynamic
optimization problem
New reactor
temperature setpoint

Fig. 3. The proposed strategy for on-line update and control of reactor temperature profile.

can be approximated by a first order model with time con- states as shown below is used instead.
stant ( j ) [25]. Consequently, the Tjsp (k) can be computed dTr Qr + Ur Ar (Tj Tr )
by = (25)
dt Wr Cpr
 
Tj (k) Tj (k 1) dTj Fj j Cpj (Tjsp Tj ) Ur Ar (Tj Tr )
Tjsp (k) = Tj (k 1) + j (24) =
t dt Vj j Cpj
(26)

With this jacket temperature set point, a jacket temper- dN

ature controller (setting as a PI controller) through a heat = bNTr (27)
dt
exchanger system opens or closes control valve reflecting to
the flowrate of heating and cooling fluid. However, in real- dQr dTr dN
=N + Tr (28)
ity the ability of the heat exchanger in adjusting the jacket dt dt dt
temperature is always limited, thus, in this work the jacket db
temperature is bounded between 0 C and 120 C. The tun- =0 (29)
dt
ing parameters of GMC are given in Table 2.
As shown in Eq. (23), the heat released (Qr (k)), which where N = bMr (H), b is a pseudo reaction rate constant,
cannot be measured, is needed in the GMC algorithm. Here, Mr is the total reactant concentration, and H is heat of reac-
the EKF algorithm, as used in on-line optimization strategy, tion. It should be noted that the variable N representing two
coupled with the simplified reactor model, given by [26], unknown parameters (Mr and H) can be estimated instead
is also applied to estimate the heat released (Qr (k)). The of these parameters so that the number of state equations for
reason of using the simplified model, not the exact model estimation decrease and an error of estimation correspond-
of the plant, is because if the exact model were used, too ing to the uncertainty of each parameters can be reduced.
many uncertain/unknown parameters as well as too many Eqs. (25)(29) correspond to Eq. (A.1) in the EKF algo-
unmeasurable states would be involved. That may lead to rithm. Once the reactor and jacket temperature measurement
poor performance of the EKF. Hence, the simplified model are available, the EKF with simplified model estimates the
with less uncertain/unknown parameters and unmeasurable heat release of reactions (Qr (k)). Table 3 summarizes the
initial conditions and tuning parameters of the EKF used in
this simulation work.
Table 2
Parameters in GMC algorithm
5. Simulation results
Wr = 1560 kg
Cpr = 1.8828 kJ/(kg C)
5.1. Maximum conversion problem (P1)
Ur = 40.842 kJ/(min m2 C)
Ar = 6.24 m2
j = 2 min All simulation results given here are based on the opti-
K1 = 2.4 mization problem P1. The objective in the problem formu-
K2 = 104 lation is to find the optimal reactor temperature profile, such
 A. Arpornwichanop et al. / Chemical Engineering and Processing 44 (2005) 101114 109

Table 3 98
Parameters and initial conditions in EKF
Estimate Mi , k21 and k22 96
MA (0) = 12 kmol P = diag[100 10 100 100 100 2000] 5
MB (0) = 12 kmol Q = diag[100 1 100 1 500 5000]
MC (0) = 0 kmol R = diag[10 10 10 10] 94
Temperature (C)

MD (0) = 0 kmol
k21 (0) = 38.9057 4
k22 (0) = 17000 92

Estimate Qr
Tr (0) = 20 C P = diag[1 1 100 20 10] 90 3

Tj (0) = 20 C Q = diag[10 10 2500 100 100]

N(0) = 1.8462 R = diag[10 10] 2 1
88
Qr (0) = 0
b(0) = 1.8386 106
86
0 20 40 60 80 100 120 140 160 180 200

that the amount of mole of product C is maximized in a fixed Time (min)

batch time with respect to a constraint on the temperature
due to the reason of safe operation. In this case study, the
specified final batch time (tf ) of 200 min is used and the reac-
tor temperature is bounded according to 20 T ( C) 120.
5.1.1. Temperature set point profile determined off-line
with perfect tracking
First, simulation studies have investigated the case where
the optimal temperature profile is determined by off-line
computation and perfect tracking of such a profile is as-
sumed. This results in the maximum amount of product C
(maximum conversion) that can be achieved at the end of
batch run and is served as a reference to be compared with
results obtained from the proposed strategy. The optimal
control problems have been solved using time interval with
equal length varied from one to 40 intervals to discretize the
profile. The switching time is fixed and the length of each
interval is specified by dividing the batch operation time (tf )
by a number of time intervals (P). Thus, the problem is to
seek an optimal temperature value (decision variables) in
each subinterval.
Simulation results with different time interval (P) are re-
ported in Table 4. Optimal control policy in reactor tem-
perature for each case is shown in Fig. 4. As shown in
Table 4, when one time interval (P = 1) is used, the amount
of product C obtained at the final time (tf = 200 min) is
7.0171 kmol and the optimal temperature (isothermal opera-
tion) set point is 88.01 C whereas using P = 20, the amount
of product C achieved is 7.0379 kmol. It was found from
Table 4
Summary of the results: off-line optimization and perfect tracking
Time Product By-product
interval C (kmol) D (kmol)
1 7.0171 1.3464
5 7.0281 1.3605
10 7.0339 1.3594
20 7.0379 1.3585
40 7.0402 1.3579
Run on Pentium III/850 computer.
Fig. 4. Optimal temperature profile: 1 interval (1), 5 intervals (2), 10
intervals (3), 20 intervals (4), 40 intervals (5)
these results that the amount of the desired product C in-
creases as a number of interval increases. This is due to that
as the number of intervals enlarges (more degrees of free-
dom in the optimization), the approximated optimal profile
with piecewise constant policy is closer to the actual optimal
profile.
5.1.2. Temperature set point profile determined on-line
with GMC controller
Next, the proposed strategy using an on-line dynamic op-
timization to update optimal temperature set point profile is
implemented. Rather than assuming the reactor temperature
trajectory can be tracked perfectly as in previous studies, the
GMC controller is applied here to drive the system to follow
the desired trajectory. Considering the time elapsed in the
determination of the optimal control problem, with P = 20,
the temperature set point profile is updated every 10 min.
This means that, to apply this strategy on-line, the computa-
tional time for the updated temperature set point profile must
be less than 10 min. Here, the computational time based on
Pentium III/850 mHz is approximately 3 s. As a result, this
strategy is applicable for on-line implementation. Regard-
ing to the GMC control performance, it was found that the
GMC controller can drive the system from current set point
value to a new one and maintain it at this set point. There-
fore, the GMC controller can be used for tracking the profile
obtained from the proposed strategy.
Simulation studies carried out to compare the amount
of desired product C obtained from on-line dynamic opti-
CPU
time (s)
mization strategy with that from off-line strategy, are cases
where the perfect model (all parameters correctly specified)
0.1591
1.2600 is used (nominal case), and where plant/model mismatch
1.9778 is introduced by changing parameters in actual plant i.e.
2.8066 pre-exponential rate constant (k0 ) decreased by 50% and ac-
5.0519 tivation energy (Ea) increased by 20% from their nominal
values, as shown in Table 5.
110 A. Arpornwichanop et al. / Chemical Engineering and Processing 44 (2005) 101114
Table 5
Comparisons of the results obtained from off-line and on-line optimization strategy with GMC controller (problem P1)
Case studies
1. Nominal case
All parameters specified correctly
2. Plants/model mismatch case
50% k0 of reaction 2 in plant model (k21 = 38.2125)
+20% Ea of reaction 2 in plant model (k22 = 20400)
50% k0 and +20% Ea of reaction 2 in plant model
Note: k0 = exp(k21 ) and Ea = (k22 )(R) where k0 is pre-exponential rate constant, Ea = activation energy, and R is ideal gas constant.
In the nominal case, the product obtained from the off-line
strategy (C = 6.9478) is close to that obtained by the on-line
strategy (C = 6.9584). Fig. 5(a) and (b) shows the response
of GMC controller to track the reactor temperature trajectory
that is pre-specified by off-line calculation, and the compar-
Fig. 5. (a) Control response (off-line, nominal case): Trsp (dash), Tr
(solid), Tj (dot). (b) Heat released (off-line, nominal case): actual (solid),
estimated (dot).
Product C (kmol)
Off-line On-line
6.9478 6.9584
7.6751 7.8851
8.5825 10.2131
8.5827 10.2137
ison of the actual and estimated heat released by reactions,
respectively. It can be seen that the EKF provides an excel-
lent estimation of the heat released. With this heat released,
the GMC controller gives very good temperature control.
Similarly, in the case that optimal temperature is modified
via the on-line optimization strategy based on the current in-
formation of Mi , k21 and k22 obtained from delayed measure-
ment of Mi , the GMC controller is able to successfully track
the reactor temperature (Fig. 6(b)). The performance of the
EKF to predict the amount of MA , MB , MC , and MD at cur-
rent time from their measurement with time delay of 10 min
is illustrated in Fig. 6(a). Also, the EKF gives good estima-
tions of k21 and k22 as shown in Fig. 6(c). It is interesting to
note that since the initial reactor temperature starts at 20 C
which is below the optimal temperature set point, the GMC
controller attempts to drive the reactor temperature from this
condition to the specified set point. However, due to the dy-
namics of the reactor temperature, the reactor temperature
differs from the optimal set point during the first 20 min,
causing that the amount of the product C obtained from the
on-line strategy with the GMC controller (C = 6.9584) is
less than that obtained from the off-line strategy with perfect
tracking (see Table 4 for P = 20, C = 7.0379)
For the mismatch in k21 , the value of the desired prod-
uct C = 7.8851 can be achieved at the end of batch for the
on-line optimization strategy which is higher than that ob-
tained from off-line strategy where the mismatch is not no-
ticed (C = 7.6751). Similar results can be observed under
the case of plant model mismatch in k22 as shown in Table 5.
These results indicate clearly that the performance of batch
reactor operation is improved via the proposed strategy. Due
to similarity in their control responses, only the result for
change in k21 is shown in Fig. 7.
Finally, with a change in both k21 (50% k0 ) and k22
(+20% Ea) in plant model, the results using the on-line opti-
mization strategy show that the GMC controller is able to ac-
commodate this change very well as can be seen in Fig. 8(a).
Fig. 8(b) presents the performance of the EKF for estima-
tion of k21 and k22 . Since the EKF estimates these parameters
close to the true values, the mismatch is eliminated. That
leads to high product C obtained at the final batch time (C
= 10.2137) compared to the value of C = 8.5827 obtained
from the off-line optimization strategy.
 A. Arpornwichanop et al. / Chemical Engineering and Processing 44 (2005) 101114 111

Fig. 6. (a) Product profile (on-line, nominal case): actual (solid), estimated (). (b) Control response (on-line, nominal case): Trsp (dash), Tr (solid), Tj
(dot). (c) Estimate of k21 and k22 (on-line, nominal case): actual (solid), estimated ().

5.2. Minimum time problem (P2)

The results presented here correspond to the case where

Fig. 7. Control response (on-line, mismatch in k21 ): Trsp (dash), Tr (solid),
Tj (dot).
the objective is to minimize the batch time of operation
subject to the terminal constraints on the desired amount
of mole of product C (MC (tf ) = 6.00 kmol). The reactor
temperature constraints are same as in problem P1.
Several simulations have been carried out under process
parameter uncertainties e.g. in pre-exponential rate constant
(k0 ) and activation energy (Ea). In all case studies we consid-
ered 10 time intervals when reactor temperature and switch-
ing time are optimized while minimizing the final batch op-
eration time. Results, reported in the value of minimum batch
time to obtain the desired product C and the amount of the
desired product C at the end of batch operation, from on-line
dynamic optimization strategy are also compared with those
from the off-line strategy.
With an 20% increase of parameter k21 in plant model, it
can be seen from Table 6 that the final batch time needed
to achieve the desired product C from the proposed on-line
112 A. Arpornwichanop et al. / Chemical Engineering and Processing 44 (2005) 101114

Fig. 9. Control response (off-line, mismatch in k21 , problem P2): Trsp

(dash), Tr (solid), Tj (dot).

Fig. 8. (a) Control response (on-line, mismatch in k21 and k22 ): Trsp (dash),
Tr (solid), Tj (dot). (b) Estimate of k21 and k22 (on-line, mismatch in k21
and k22 ): actual (solid), estimated ().

modification of temperature set point profile (tf = 49.4 min)
is shorter compared to the result with the off-line strategy (tf
= 64.8 min). This is because the EKF can acknowledge this
parameter uncertainty, so that the temperature set point pro-
file is updated corresponding to a modified parameter value
close to the actual value. Fig. 9 shows, for this mismatch,
the control response of GMC controller to deliver the reac-
tor temperature from initial condition to the temperature set
point determined off-line. For the on-line temperature set
Fig. 10. Control response (on-line, mismatch in k21 , problem P2): Trsp
(dash), Tr (solid), Tj (dot).
point modification, as expected, GMC controller can also
control the reactor corresponding to the changes in temper-
ature set point as can be seen in Fig. 10
The results for the remaining case studies are summa-
rized in Table 6. An important aspect obtained from these
results is that in all cases, the minimum batch time to obtain
the desired product concentration in the on-line set point

Table 6
Comparisons of the results obtained from off-line and on-line optimization strategy with GMC controller (problem P2)
Case studies Final time (min)/product C (kmol)

Off-line On-line

(1) 50% k0 of reaction 2 in plant model (k21 = 38.2125) 64.8/6.0049 49.4/6.0089

(1) +20% Ea of reaction 2 in plant model (k22 = 20400) 56.4/6.0147 44.6/6.0104
(1) 50% k0 and + 20% Ea of reaction 2 in plant model 56.7/6.0144 45.2/6.0121
 A. Arpornwichanop et al. / Chemical Engineering and Processing 44 (2005) 101114 113

modification strategy decreases. This points out the effec- For nonlinear systems, the process model can be described
tiveness of the proposed method to improve the operation by differential equations
of batch reactor.
x = F(x(t), u(t), t) + (t) (A.1)
y = G(x(t)) + (t) (A.2)
6. Conclusions
where F is a vector of system function, G is a vector of
measurement function, is a zero mean Gaussian process
In this work, the method using an on-line dynamic opti-
noise with covariance Q, and is a zero mean Gaussian
mization and control strategy to enhance batch reactor oper-
measurement noise with covariance R.
ation has been proposed. The on-line dynamic optimization
The equations for the EKF are given by a set of correction
by the idea of receding horizon scheme is performed. The ap-
and prediction equations as shown in the following:
proach is based on the updated current information of states
Correction phase: Correct the prior estimates of state at
of the system which are estimated from the delayed mea-
k 1 and update the weighting matrix
surement of the amount of reactants in the reactor using an
Extended Kalman Filter (EKF) technique, at specified time Kk1 = Pk1/k2 Ck1 T [Ck1 Pk1/k2 Ck1 T + R]1
interval to provide a new updated optimal reactor temper-
(A.3)
ature set point profile. Two types of optimization formula-
tion related to batch operation (maximum concentration and
minimum time problem) were considered in the proposed xk1/k1 = xk1/k2 + Kk1 [yk1 Ck1 xk1/k2 ]
on-line set point modification strategy. The obtained reac- (A.4)
tor temperature set point was implemented using a generic
model control (GMC). The EKF was also incorporated into
the GMC algorithm in order to estimate the heat released by Pk1/k1 = [I Kk1 Ck1 ]Pk1/k2 [I Kk1 Ck1 ]T
reactions using the direct measurement of reactor and jacket + Kk1 RKTk1 (A.5)
temperature. A batch reactor with highly exothermic reac-
tions was used as a simulation case study to demonstrate the Prediction phase: Integrate the nonlinear state and co-
effectiveness of the proposed approach. Simulation studies variance equations from time k 1 to k in order to acquire
have been carried out in both nominal case and plant/model estimate xk/k1 and Pk/k1
x = F(x, u)
mismatch case and the results show that the performance of
(A.6)
the batch reactor in terms of the amount of a desired prod-
uct and batch operation time can be improved significantly P = Ak1 P + PATk1 + Q (A.7)
by the proposed strategy. In addition, they also clearly in-
dicate the capability of the GMC controller to control the where xk/k1 denotes the estimate of state x at t = k from
reactor temperature along the specified trajectory and that information at t = k 1, K is Kalman gain matrix, P is
of the EKF to estimate the states and parameters of the covariance matrix, and matrices
system.
 
 
F  G 
Ak1 = and C =
x xk1|k1 x xk1|k2
Acknowledgements are the Jacobians of function F and G with respect to the
state vector, respectively.
The financial support to A. Arpornwichanop through local
graduate scholarship from the national science and technol-
ogy development agency (NSTDA) is gratefully acknowl- Appendix B. Nomenclature
edged. A. Arpornwichanop also acknowledges the support
of the Department of Chemical Engineering, University of ki rate constant for reaction i (kmol1 /s)
Bradford, UK during his stay as a visiting researcher at the ki1 rate constant 1 for reaction i
University in 2000. ki2 rate constant 2 for reaction i
t time (min)
t sampling time (min)
Appendix A. The EKF algorithm u input variables
x state variables
Since in this work the process models used are in time y output variables
continuous form and measurements are in discrete form, the A heat transfer area (m2 )
EKF in discrete/continuous formulation [27] is used. The Cp mass heat capacity (kJ/(kg C))
basic algorithm of the EKF can be summarized as follows: Cpi molar heat capacity of component i (kJ/(kmol C))
114 A. Arpornwichanop et al. / Chemical Engineering and Processing 44 (2005) 101114

F flowrate (m3 /min) [9] Z.H. Liu, S. Macchietto, Model based control of a multipurpose
batch reactor an experimental study, Comp. Chem. Eng. 19(Suppl)
Hi heat of reaction of reaction i (kJ/kmol) (1995) s477s482.
K1 , K2 GMC controller parameters [10] S. Yuce, A. Hasaltun, S. Erdogan, M. Alpbaz, Temperature control
Mi number of moles of component i (kmol) of a batch polymerization reactor, Trans. IChemE 77 (Part A) (1999)
MWi molecular weight of component i (kg/kmol) 413420.
P covariance matrix [11] P. Kittisupakorn, M.A. Hussain, A. Arpornwichanop, Temperature
control of an exothermic batch reactor by model predictive exother-
Q covariance matrix of process noise mic batch reactor by model predictive control, in: Proceedings of the
Qr heat released from reactions (kJ/min) Third Asian Control Conference, Shanghai, China, 47 July 2000,
R covariance matrix of measurement noise pp. 25062511.
Ri rate of reaction i (kmol/min) [12] N. Aziz, M.A. Hussain, I.M. Mujtaba, Optimal control of batch
T reactor temperature ( C) reactor: comparison of neural network based GMC and inverse model
control approach, in: Proceedings of the Sixth World Congress of
U heat transfer coefficient (kJ/(min m2 C)) Chemical Engineering, Melbourne, Australia, 2327 September 2001.
V reactor volume (m3 ) [13] R. Berber, Control of batch reactors: a review, Trans. IChemE 74 (Part
W Reactor content (kg) A) (1996) 320.
density (kg/m3 ) [14] L.R. Petzold, A description of DASSL: a differential/algebraic system
solver, SAND82-8637, Sandia National Laboratories, 1982.
[15] W.H. Ray, Advanced Process Control, McGraw Hill, New York,
Subscripts 1981.
f filter [16] J.G. Renfro, A.M. Morshedi, O.A. Asbjornsen, Simultaneous opti-
j jacket mization and solution of systems described by differential/algebraic
r reactor equations, Comp. Chem. Eng. 11 (5) (1987) 503517.
[17] V. Nevistic, Ph.D. thesis, Constrained control of nonlinear systems,
sp set point
Swiss Federal Institute of Technology, 1997.
[18] R. Luus, Optimal control by dynamic programming using systematic
reduction in grid size, Int. J. Control 51 (1990) 9951013.
References [19] S.A. Dadebo, K.B. Mcauley, Dynamic optimization of con-
strained chemical engineering problems using dynamic programming,
[1] G.E. Rotstein, D.R. Lewin, Control of an unstable batch chemical Comp. Chem. Eng. 19 (5) (1995) 513525.
reactor, Comp. Chem. Eng. 16 (1) (1992) 2749. [20] J.E. Cuthrell, L.T. Biegler, On the optimization of differential-
[2] D. Bonvin, Optimal operation of batch reactors a personal view, algebraic process systems, AIChE J. 33 (8) (1987) 1257
J. Proc. Cont. 8 (5/6) (1998) 355368. 1270.
[3] N. Aziz, I.M. Mujtaba, Optimal operation policies in batch reactors, [21] C.L. Chen, Ph.D. thesis, A class of successive quadratic programming
Chem. Eng. J. 85 (2002) 313325. methods for flowsheet optimization, University of London, 1988.
[4] N. Aziz, M.A. Hussain, I.M. Mujtaba, Performance of different [22] L.S. Kershenbaum, P. Kittisupakorn, The use of a partially simulated
types of controllers in tracking optimal temperature profiles in batch exothermic (PARSEX) reactor for experimental testing of control
reactors, Comp. Chem. Eng. 24 (2000) 10691075. algorithms, Trans. IChemE. 72 (Part A) (1994) 5563.
[5] C. Loeblein, J.D. Perkins, B. Srinivasan, D. Bonvin, Performance [23] J.X. Shen, M.S. Chiu, Q.G. Wang, A comparative study of model-
analysis of on-line batch optimization systems, Comp. Chem. Eng. based control techniques for batch crystallization process, J. Chem.
21 (Suppl) (1997) s867s872. Eng. Jpn. 32 (4) (1999) 456464.
[6] D. Ruppen, D. Bonvin, D.W.T. Rippin, Implementation of adap- [24] P.L. Lee, G.R. Sullivan, Generic model control (GMC), Comp. Chem.
tive optimal control operation for a semi-batch reaction system, Eng. 12 (6) (1988) 573580.
Comp. Chem. Eng. 22 (1998) 185189. [25] B.G. Liptak, Controlling and optimization chemical reactors, Chem.
[7] C. Kravaris, R.A. Wright, J.F. Carrier, Nonlinear controller for tra- Eng. Magazine (May 1986) 6981.
jectory tracking in batch processes, Comp. Chem. Eng. 13 (1/2) [26] P. Kittisupakorn, Ph.D. thesis, The use of nonlinear model predic-
(1989) 7382. tive control techniques for the control of a reactor with exothermic
[8] B.J. Cott, S. Macchietto, Temperature control of exothermic batch reactions, University of London, 1995.
reactors using generic model control, Ind. Eng. Chem. Res. 28 (1989) [27] A. Gelb, Applied Optimal Estimation, The MIT Press, Cambridge,
11771184. MA, 1974.
 Chemical Engineering and Processing 44 (2005) 115119

Synthesis of nano-sized particles from metal carbonates

by the method of reversed mycelles
Christo Karagiozov, Dafina Momchilova
Departement of Chemical Engineering, Faculty of Technical Sciences, University Prof. Dr. Assen Zlatarov Bourgas,
Bul: Prof. Iakimov 1, 8010 Bourgas, Bulgaria

Received 2 December 2003; received in revised form 30 April 2004; accepted 5 May 2004
Available online 25 August 2004

Abstract

The nano-sized particles of calcium carbonate and barium carbonate have specific characteristics. They are important materials for the
industry. The main object of this investigation is to obtain nanoparticles of calcium carbonate and barium carbonate by chemical reaction
carried out in microemulsion of water in oil. The nanoparticles obtained are spherical. Their sizes vary from 20 to 30 nm. The shape and size
of particles are determinated by electron microscopy.
2004 Elsevier B.V. All rights reserved.

Keywords: Calcium carbonate; Barium carbonate; Nanoparticles; Microemulsion; Surface tension

1. Introduction [5,6], electrolysis [10], synthesis in microemulsion or the so

The concepts of the nano-world suggested recently re-
vealed new fields of scientific research. The number of the
publications, patents, projects, implementations and compa-
nies involved with nano-technology has been exponentially
increasing during the last 10 years. Nano-technology is a
transition field between the concepts of the classic quantum
mechanics and physical chemistry of solid state [1].
The investigations of materials obtained from nano-sized
particles showed that they have unknown properties or
enhanced characteristics compared to common materials.
Many alkaline carbonates have take in especially application
in practice. Calcium carbonate is widely used for production
of toothpaste, binding agents, etc. [12,13]. Barium carbon-
ate is used mainly in glass production industries, as well as
water softening agent, etc. Both substances are crystalline,
with rhombic lattice and practically insoluble in water [12].
Many methods have been used for synthesis of nano-sized
particles: vapour transport [2], precipitation of homogeneous
solution [3], synthesis by sol-gel process [4], freeze-drying
Corresponding author. Tel.: +359 56 858351/424;
fax: +359 56 880249.
E-mail addresses: chcar@abv.bgs, dafi mom@yahoo.com
(D. Momchilova).
called method of reversed micelles [710].
Microemulsion is used as a special microreactor to limit
the nano-sized particles growth. The shape of the microre-
actor depends on reaction conditions [9]. This method in-
creases the homogeneity of the chemical composition at
nano-level and facilitates the preparation of nano-particles
with comparatively equal sizes [11]. The specific properties
of the nano-particles make them suitable for microelectron-
ics, ceramics, catalysis, medicine, cosmetics, as piezoelec-
tric materials, conductors, etc.
The aim of the present investigation is to synthesize
nano-sized particles from CaCO3 and BaCO3 by the mi-
croemulsion method.
2. Experimental
The synthesis of nano-sized particles from CaCO3 and
BaCO3 was carried out in a recurrent reactor equipped with
stirrer and was close to reactor with ideal mixing. Chemical
interaction between CO2 and the corresponding alkali sus-
pension (dispersed in organic phase) takes place in the reac-
tor. The process is endothermic so the device was equipped
with cooling jacket. A diagram of the device is shown in
Fig. 1.

0255-2701/$ see front matter 2004 Elsevier B.V. All rights reserved.
doi:10.1016/j.cep.2004.05.004
116 C. Karagiozov, D. Momchilova / Chemical Engineering and Processing 44 (2005) 115119

action in a microemulsion of type water in oil. The qual-

Fig. 1. Diagram of the facility for preparation of nano-sized particles. 1:
Bottle with carbon dioxide, 2: rotameter, 3: recurrent reactior with jacket,
4: bubbler, 5: magnetic stirrers, 6: drecssel bottle.
The experiments started with preparation of solutions of
calcium (barium) hydroxide in water and storing them for 1
day. Their exact concentration was determined by titration
with hydrochloric acid. The microemulsion used was aque-
ous alkali solution/oil (w/o) where the water drops contained
solved calcium or barium hydroxide and organic phase
n-hexane.
The total volume of the microemulsion was 50 ml for all
the experiments. The volumes of the two phases were varied
from 10 to 40 ml, respectively. The volumes of both organic
and inorganic phases of the system for each experiment are
shown in Table 1.
The liquid mixture was stirred with the magnetic stirrer
for 60 min at constant speed. The emulsion formed con-
tained only one of the reagents. The other was fed into the
reactor as gas (CO2 ) at constant flow rate for 60 min. The
second reagent and the cooling water were counterflowing
(Fig. 1). The experiments were carried out at temperature of
2023 C.
3. Results and discussion
Solid nano-sized particles of CaCO3 and BaCO3 were ob-
tained from the experiments carried out using chemical re-
itative analyses of the organic and inorganic phases for all
the experiments showed the presence of calcium or barium
carbonate. The shape and size of the particles were deter-
mined by electron microscopy. Photographs of nano-sized
particles of BaCO3 are presented in Fig. 2.
The nano-particles synthesized were with spherical shape
and almost equal sizes varying from 20 to 30 nm.
The electron-microscopy analysis (Fig. 2) revealed that
processes of particles agglomeration took place. Since the
opalescence observed with the organic phase from the ex-
periments was not very good, the agglomeration was sup-
posed to occur during the electron microscopy analysis it-
self. It may result from the irradiation of the samples by the
high-energy electron beam used for the analysis. Photograph
A was taken from samples obtained from the experiment
carried out at a ratio of inorganic phase Ba(OH)2 to organic
phase n-C6 H14 15/35 ml. Photograph B was taken at phase
ratio Ba(OH)2 /n-C6 H14 = 40/10 ml.
The results sustained the opinion that the synthesis of
nano-particles in microemulsion (w/o) is to be preferred.
In common emulsions, both phases spontaneously separate
from each other while microemulsions are thermodynami-
cally stable, do not segregate and appear to be transparent.
This can be explained with the size of the water droplets
[14,11]. Microemulsions are also characterized by the so
called dynamic exchange process. The emulsion droplets
in such emulsions constantly integrate and disintegrate, thus
exchanging substance between each other.
The substance dissolved in the emulsion droplets was
quite small amount, i.e. the reagent concentration in the
droplet was low so it could be expected to obtain small par-
ticles with almost equal sizes.
In the case studied, the following physical model was sug-
gested. The microemulsion water solution/organic phase was
formed preliminarily. It contained only one of the reagents
(in this case solution of Ca(OH)2 or Ba(OH)2 ). The sec-
ond reagent was added in the form of a gas (CO2 ). The
gas passed through the organic liquid, diffused through the

Table 1
Ratio of volumes of organic and inorganic phases
Experiment Inorganic Organic Volume of inorganic Volume of organic V1 /V2
no. phase phase phase V1 (ml) phase V2 (ml)
1 2 3 4 5 6
1 Ca(OH)2 n-C6 H14 5 45 0.111
2 Ca(OH)2 n-C6 H14 10 40 0.250
3 Ca(OH)2 n-C6 H14 15 35 0.429
4 Ca(OH)2 n-C6 H14 20 30 0.667
1 Ba(OH)2 n-C6 H14 30 20 1.5
2 Ba(OH)2 n-C6 H14 25 25 1
3 Ba(OH)2 n-C6 H14 20 30 0.667
4 Ba(OH)2 n-C6 H14 15 35 0.429
5 Ba(OH)2 n-C6 H14 35 15 2.333
7 Ba(OH)2 n-C6 H14 40 10 4
Note: 1. All the experiments were carried out under stirring for 60 min at 400 rpm, 2. CO2 bubbling time was 60 min.
 C. Karagiozov, D. Momchilova / Chemical Engineering and Processing 44 (2005) 115119 117

Fig. 2. Electron microscopy photographs of nano-sized particles of BaCO3 .

interphase surface oil/water solution in the drops and then The size of the nanoparticles of BaCO3 obtained was
interacts with the inorganic phase present in them. Thus, about 2030 nm. Their specific area was calculated to be
a chemical reaction was initiated which gave slightly solu- from 6.82 107 to 4.55 107 m2 /kg. The particles with
ble products (CaCO3 and BaCO3 ). This hypothesis can be 20 nm sizes have specific area of 6.82 107 m2 /kg while
schematically represented as follows (Fig. 3). particles with 30 nm size 4.55 107 m2 /kg. Therefore, the
The summarized equation of the chemical reaction pro- specific area decreased with the increase of particle radius
ceeding between the corresponding hydroxide and CO2 is (Fig. 3).
For the microemulsion system inorganic solution/
Me(OH)2 + CO2 MeCO3 + H2 O (1)
n-C6 H14 , it is important to prove that the inorganic so-
The chemistry of the separate stages of the process within a lution was dispersed in the organic phase in the form of
droplet can be described by the following reactions: microdrops. To determine the distribution of the phases in
the microemulsion, the forces of surface tension for each
Me(OH)2 Me2+ + 2OH (2)
microdrop should be estimated, as well as the forces acting
CO2 + H2 O H2 CO3 (3) at the interphase surface.
Forces of surface tension act on the surface of the parti-
H2 CO3 HCO
3 +H
+
(4) cles in the liquid phase. It can be expressed as the work for
the formation of a unit of interphase surface under constant
CO2 + OH HCO
3 (5)
thermodynamic parameters of the state (temperature, pres-
HCO + 2
3 H + CO3 (6) sure, chemical potentials of the components). This process
is reversible and isothermal. The surface tension forces can
Me2+ + CO2
3 MeCO3 (7) be regarded also as free energy per unit area, i.e. specific
free energy (GS ). Then, the free energy per unit weight of
where Me = Ca, Ba. particles would be
The basic parameter characterizing nano-particles is their
specific area, which can be determined by the formula [15] GS = GS A (11)
A = 4r N(m /kg)
2 2
(8)
where A: specific area (m2 /kg), r: particles radius (m), N:
number of particles per unit of weight.
The number of particles can be calculated from
1
N= (9)
(4/3)r 3
Substituting Eq. (11) into Eq. (10) gives the following ex-
pression for calculation of the specific area:
3 2
A= (m /kg) (10)
r Fig. 3. Probable mechanism of contact between the reagents.
118 C. Karagiozov, D. Momchilova / Chemical Engineering and Processing 44 (2005) 115119
Assuming that GS is constant value, then free energy
GS should increase with the decrease of particle radius A
[16,17].
The surface tension of 0.02 N solution of Ba(OH)2 was
determined by the formula [18]
Ba(OH)2 = 0 + 0.049(1 aW ) (12)
where : surface tension of Ba(OH)2 (dyne/cm), 0 : sur-
face tension of water at 25 C. It is considered to be
71.97 dyne/cm [18], aw : activity of water in the solution.
The water activity in the solution was determined by the
well-known equation [18]

n-C6 H14 remained lower than that of the inorganic solution,
tmBa(OH)2 MBa(OH)2
ln aW = B1 + at + bt2
55510RT
 
B2
+
1000 + MBa(OH)2 mBa(OH)2
 
1 4. Conclusions
mBa(OH)2 MBa(OH)2 + (13)
E
where R: universal gas constant = 8.314441 J/mol, T: tem-
perature (K), mi : solution molality (g/l), Mi : mole weight
(g/mol), B1 , a, b, B2 , E: coefficients calculated by mathe-
matical treatment of the experimental data for heat capacity.
Value of the surface tension of Ba(OH)2 was calculated to
be 71.88 dyne/cm. Reference data showed that n-C6 H14 =
32.21 [18]. The value of Ba(OH)2 surface tension was close
to that of water ( 0 ) since the concentration of Ba(OH)2 was
low and the solution was very diluted.
Forces of interphase tension T1 /T2 act at the contact sur-
face between the two liquid phases of Ba(OH)2 (or Ca(OH)2 )
and n-C6 H14 . The thermodynamic condition for a liquid to
spread over the surface of another liquid is [19]
T2 / > T1 / + T1 /T2 (14)
where T1 / , T2 / : surface tensions of the liquids related
to their environment (dyne/cm), T1 /T2 : surface tension at
the interphase boundary between the liquids (dyne/cm).
The interphase tension at the phase boundary between an
inorganic solution and oil was assumed to be 50.5 dyne/cm,
according to literary data [19]. The two phases are quite
different by polarity. The basic process taking place at the
contact surface between them is the change of the total con-
tact surface area due to the decrease of the total energy of
the system, i.e. the conditions became suitable for process
of wetting to take place. The coefficient of spreading S can
be calculated from the values of the surface and interphase
tensions according to [20]
S = T2 / T1 / T1 /T2 , S = 10.853 (15)
At S < 0, the wetting angle can be determined by the equa-
tion [20]
T22 / T22 / T21 /T2
cos 0 = (16)
2T1 / T2 /
i.e. cos 0 = 0.3717
Its value was calculated to be 0 = 68.18 . At 0 < 90
(as is in this case), the process called limited wetting or
wetting with impregnation. Therefore, the change of the free
energy per unit area can be found by the expression
G = TB /T T/ (17)
The interphase tension T1 /T2 corresponds to the so-called
boundary coefficient of spreading Sk , after equilibrium was
established at the interphase surface between the two liq-
uids. The higher the coefficient of spreading is, the faster is
the spreading of the wetting liquid. In the system studied,
T1 /T2 = Sk = 50.5 > 0. It means that the surface tension of
i.e. n-C6 H14 wetted the polar phase of aqueous solution of
Me(OH)2 .
The following conclusions can be formulated on the basis
of the results obtained
1. A technique and process for preparation of nano-sized
particles from alkali carbonates was developed on the
basis of the method of reversed micelles and a probable
mechanism was suggested.
2. Nano-sized particles of barium and calcium carbonate
were obtained by a chemical reaction in a microemulsion.
The particles were studied by electron microscopy and
were found to possess spherical shape and diameters from
20 to 30 nm.
3. The nano-sized particles were obtained at various ratios
of the phases forming the microemulsion. They were
about the same size, which confirmed the advantage of
the method selected.
4. The particles synthesized were observed to agglomer-
ate, probably due to effects induced by the electron mi-
croscopy analysis.
5. The values of the surface tensions of the individual phases
were calculated, as well as the value of the interphase
tension. The results confirmed that the inorganic phase
was dispersed in the organic phase.
Appendix A. Nomenclature
aw activity of water in the solution
mi solution molality (g/l)
r particles radius (m)
o oil
w water
A specific area (m2 /kg)
B1 , a, b, B2 , E coefficients calculated by mathematical
treatment of the experimental data for
heat capacity
G change of the free energy per unit area
 C. Karagiozov, D. Momchilova / Chemical Engineering and Processing 44 (2005) 115119
GS specific free energy
Mi mole weight (g/mol)
N number of particles per unit of
weight
R Universal gas constant (J/mol)
T temperature (K)
S coefficient of spreading
surface tension of Ba(OH)2 (dyne/cm)
0 surface tension of water at 25 C (dyne/cm)
T1 /T2 surface tension at the interphase boundary
between the liquids (dyne/cm)
T1/ , T2 / surface tensions of the liquids related
to their environment (dyne/cm)
References
[1] Xp. Kapaoob, . Momoba, Hahoact ha akah [14] M. Roth, Dissertation, Synthhesen zur Herstellungneuartiger-
kapohat chtepah ocpectbom mkpoemycoheh meto, nanoskaliger Materialien und Dispersionen mittels kontrollierter Fal-
Scientific Conference with International Participation Stara Zagora,
Technical Sciences, Sea Sciences and Ecology, vol. II, 2003,
pp. 179184.
[2] X. Dong, C. Choi, B. Kim, Chemical synthesis of Co nanoparticles
by chemical vapor condensation, Powder Technol. 113 (2002) 75
79.
[3] Y. Du, J. Fang, M. Zhang, J. Hong, Z. Yin, Q. Zhang, Phase
character and structural anomaly of Bi4 Ti3 O12 nanoparticles prepared
by chemical coprecipitation, Mater. Lett. 57 (2002) 802806.
[4] A. Hartridge, A. Bhattacharya, Preparation and analysis of zirconia
doped ceria nanocrystal dispersions, J. Nanoparticle Res. 1 (2001)
7580.
[5] E. Bermejo, T. Becue, C. Lacour, M. Quarton, Synthesis of
nanoscaled iron particles from freeze-dried precursors, Powder Tech-
nol. 94 (Issues 1) (1997) 2934.
119
[6] R. Valencuela, G. Bueno, A. Solbes, Nanostructured ceria-based
catalysts for oxydehydrogenation of ethane with CO2 , Topics Catal.
15 (2-4) (2001) 181187.
[7] Y. Mori, Y. Okastu, Y. Tsujimoto, Titanium dioxide nanoparticles
produced in water-in-oil emulsion, J. Nanoparticle Res. 3 (23)
(2001) 219225.
[8] B. Palla, D. Shah, P. Garcia-Casillas, J. Matutes-Aquino, Preparation
of nanoparticles of barium ferrite from precipitation in microemul-
sions, J. Nanoparticle Res. 1 (2) (1999) 215221.
[9] K. Song, J. Kim, Synthesis of high surface area tin (Sn) oxide
powders via water-in-oil microemulsions, Powder Technol. 103 (3)
(2000) 268272.
[10] W. Wang, X. Fu, J. Tang, L. Jiang, Preparation of submicron spherical
particles of silica by the water-in-oil microemulsion method, Colloids
Surf. A: Physicochem. Eng. Aspects 81 (1993) 177180.
[11] J. Shmidt, Reaktionstechnishe Untersuchungen zur Herstellung von
Nanopartikeln in Mikroemulsionen, Dissertation, TU Berlin, 2000.
[12] Xp. Koapob, Heopahha xm, Texhka, C., 1970.
[13] M. Dagaonkar, Effect of a Microphase on GasLiquid Mass Transfer,
Proefschrift (Rijksuniversiteit Groningen) Printed by: Print Partners
IPSKamp, Enschede, 3 December 2001.
lung, 1998.
[15] K. Tamapy, Kapha xm, Mockba Mp, 1983.
[16] A. Akoh, Xmecka tepmohamka, Mockba: Bica
koa, 1963.
[17] A. pock, 1. Cboctba ha matepta. Xmecka
tepmohamka, 2. Patbop, Eektpoxm, Xmha
khetka, 1951
[18] . a eb, .. Aceeb ko xmecke cboctba haphix
mhookomohehthix pactbopob heopaheckx beectb,
Mockba "Xm", 1988.
[19] H. Backes, J. Ma, E. Bender, G. Maurer, Interfacial tensions in binary
and ternary liquidliquid systems, Chem. Eng. Sci. 45 (1) (1990)
275286.
[20] . Cymm, .B. ophob, ko xmecke ochobi
cmabah pactekah, Mockba Xm, 1976.
 Chemical Engineering and Processing 44 (2005) 121130

Mechanism of mass transfer from bubbles in dispersions

Part II: Mass transfer coefficients in stirred gasliquid reactor
and bubble column
V. Linek , M. Kordac, T. Moucha
Department of Chemical Engineering, Prague Institute of Chemical Technology, Technick 5, 166 28 Prague 6, Czech Republic

Received 15 December 2003; accepted 14 May 2004

Available online 2 July 2004

Abstract

Experimental data on the average mass transfer liquid film coefficient (kL ) in an aerated tank stirred by Rushton turbine and in bubble
column are presented. Liquid media were used as 0.8 M sodium sulphite solution, pure or with the addition of Sokrat 44 (copolymer of
acrylonitrile and acrylic acid) or short-fiber carboxymethylcellulose (CMC) for the Newtonian and long-fiber CMC for the non-Newtonian
viscosity enhancement and ocenol (cis-9-octadecen-1-ol) or polyethylenglycol (PEG) 1000 for surface tension change. Volumetric mass
transfer coefficient (kL a) and specific interfacial area (a) were measured by the Danckwerts plot method. Coefficients kL measured by pure
oxygen absorption in pure sulphite solution and Newtonian viscous liquids are well fitted by the eddy model in the form of kL = 0.448
(ev/)0.25 (D/v)0.5 with a mean deviation of 20%. Surface-active agents (ocenol and PEG) and non-Newtonian additive (long-fiber CMC)
reduced kL value significantly but their effect was not described satisfactorily neither by surface tension nor by surface pressure. It is shown
that the decisive quantity to correlate kL in the stirred tank and bubble column is power dissipated in the liquid phase rather than the bubble
diameter and the slip velocity. Absorption of air did not yield correct kL data, which did not depend on or slightly decreased with increasing
power. This is due to the application of an improper gas phase mixing model for absorption data evaluation.
2004 Elsevier B.V. All rights reserved.

Keywords: Gasliquid contactor; Bubble column; Agitated vessel; Mass transfer coefficient; Viscosity; Surfactants

1. Introduction
Mass transfer from swarm of bubbles into turbulent liq-
uid controls the rate of many chemical and biochemical pro-
cesses. It is assumed that the mechanism of mass transport
in liquid phase is due to a renewal of the liquid at the bub-
ble surface. Models of the process differ in the scale of flow,
which is responsible for the renewal.
The first group of models (eddy models) assumes that
the liquid renewal is due to small-scale eddies of the turbu-
lent field. These models are based on idealized eddy struc-
tures of turbulence in the bubble vicinity. Lamont and Scott
[1] have assumed that the small scales of turbulent motion,
which extend from the smallest viscous motions to the iner-
Corresponding author. Tel. +420 2 2435 3298;
fax: +420 2 3333 7335.
E-mail address: linekv@vscht.cz (V. Linek).
tial ones, affect the mass transfer. In any case, these motions
are much smaller in scale than the gas bubbles. As a result,
the size of the gas bubble is not a very critical parameter for
the estimation of kL . They deduced [1] the following rela-
tion for kL :

0.25
0.5
ev D
kL = c1 (1)
v
Different authors predict different values of the constant c1 :
0.301 [2], 0.4 [1], 0.531 [3], 0.592 [4] and 1.13 [5].
The second group of models (slip velocity models) as-
sumes a gross mean flow of fluid relative to the bubble (slip
velocity) and a bubble surface mobility control of this re-
newal rather than the small-scale eddies of the turbulent
field. The model proposed by Calderbank and Moo-Young
[6] belongs to this group. They [6] have divided bubbles by
size into two categories: the small bubbles (d < 1 mm),
which behave always as a rigid sphere, and the large
ones (d > 2.5 mm), which always have a completely mobile

0255-2701/$ see front matter 2004 Elsevier B.V. All rights reserved.
doi:10.1016/j.cep.2004.05.009
122 V. Linek et al. / Chemical Engineering and Processing 44 (2005) 121130
surface. An expression for the calculation of kL for the
large bubbles has been deduced from dimensionless analy-
sis of mass transfer from rising particle in gravitational field.



g 1/3 D 1/2
kL = 0.42 (2)
v where d is an average bubble diameter, k is an empirical con-
For the small bubbles, an expression for the calculation of
kL has been obtained by the equation proposed by Frssling
[7] for dissolution of a rigid sphere.



g 1/3 D 2/3
kL = 0.31 (3)
The transition between these two bubble categories was
unclear and strongly dependent on the presence of surfac-
tants. An expression for calculation of the mass transfer
coefficients in this transition region was not given.
Recently, Vasconcelos et al. [8] and Alves et al. [9] pro-
posed another variant of the slip velocity model. Their
model is based on experimentally observed phenomena of
an abrupt change of the dissolution rate of free-floating bub-
bles held stationary in a downward water flow. Starting from
these experiments, they have interpreted the mass transport
from bubbles in terms of bubble contamination kinetics, us-
ing a stagnant cap model, according to which bubbles sud-
denly change from a mobile interface to a rigid condition
when surface tension gradients, caused by surfactant accu-
mulation, balance out shear stress. The fresh bubble entering
the dispersion has clean, completely mobile interface. For
the bubbles with completely mobile surface, kL is given by
the equation, which follows from the well-known penetra-
tion model with exposition time equal to d/vsl :

vsl 1/2
kLmobil = 1.13 D (4)
d
After some time ( mobile ), enough contaminant for transition
to rigidity is accumulated on the surface. For the rigid bub-
bles, an expression for the calculation of kL follows from
Frssling [7] equation.
 their results in terms of bubble contamination kinetics.
rigid vsl 2/3 1/6
kL = 0.6 D v (5)
d thors found an increase of kL with increasing energy dissi-
Gasliquid slip velocities vsl are assumed to be close to
single-bubble terminal velocities in still water, vt , on which a
correction for turbulence is introduced, vsl = 0.65vt [9]. As-
suming monosized bubble dispersion and considering only
rigid
two possible values of kL (kLmobil and kL , depending on
surface mobility), they deduced the following relation for
the average mass transfer coefficient:
rigid
kmobil mobile + kL (R mobile )
kL = L (6)
R depend on power dissipated and Hassan and Robinson [23]
where R is the average bubble residence time.
VL
R = (7)
q(1 )
The time mobile depends on bubble diameter and surfactant
concentration.
d 1/2 ln (d/ htrans )
mobile = k (8)
csurf
stant related with surfactant properties, csurf is the concen-
tration of surfactant and htrans is the bubble clean segment
height at the transition point from mobile to rigid.
The main difference between the eddy and the slip
velocity models is in the influence of turbulence inten-
sity on mass transfer coefficient: the eddy models predict
an increase while the slip velocity models a decrease of
kL with increasing turbulence intensity, i.e. with increasing
power dissipated in the liquid. This is due to the fact that
kL from rigid bubbles is cca four times lower than from the
ones with a mobile surface. The small bubbles, which ei-
ther have lower kL (according to Calderbanks conception)
or become rigid more quickly (according to Alves concep-
tion, mobile diminishes with diminishing bubble diameter
d, see Eq. (8)), are generated in large amount just at higher
turbulence intensities. As a result, this leads to the decrease
of average mass transfer coefficient with increasing power
dissipated in the liquid. Literature data on kL in stirred tanks
supporting both these models can be found.
In dispersions with no mechanical agitation (bubble
columns, air-lifts), the only source of energy is the expansion
of the entering gas which gives the relation e = gvs . Reith
[10], Dillon and Harris [11], Kawase et al. [5] and Vzquez
et al. [12] revealed strong dependence of liquid-side mass
transfer coefficient on gas flow rate in bubble columns:
kL vs 0.25/0.5 . Slight dependence or independence of kL
on vs was found by Schumpe and Deckwer [13], Bouaifi
et al. [14], Wongsuchoto et al. [15] and Vasconcelos et
al. [8]. The values of kL corresponded to values predicted
by Calderbanks model for large bubbles. Values which
correspond also to small bubbles were found by Vzquez
et al. [12] and Vasconcelos et al. [8] only, who worked with
solutions of surfactants. Vasconcelos et al. [8] interpreted
In dispersions with mechanical agitation, following au-
pation e: Prasher and Wills [4] (kL e0.25 ); Figueiredo and
Calderbank [16] (kL e0.33 ); Bouaifi and Roustan [17] (kL
e0.22 ); Panja and Rao [18] (kL e0.15 ); and Linek et al.
[19] (kL e0.14 ). Power dissipated in the liquid by an agi-
tator is proportional to the third power of impeller rate (e
f3 ). Thus, we can add the results of authors that presented kL
as a function of impeller rate f: Koetsier and Thoenes [20]
(kL f0.9 e0.3 ) and Yoshida and Miura [21] (kL f0.6
e0.2 ). Robinson and Wilke [22] measured kL that did not
obtained coefficients that decreased with increasing power
dissipated. It was shown [32] that the reported independence
on or the decrease of kL with increasing power dissipated
was a result of misinterpretation of kL a data.
 V. Linek et al. / Chemical Engineering and Processing 44 (2005) 121130
Figueiredo and Calderbank [16] have analysed the suit-
ability of Calderbanks model [6] for small and large
bubbles to the interpretation of mass transfer coefficients in
stirred vessels. They analysed an effect of the gas phase mix-
ing model used for kL a evaluation by the dynamic method
based on the inlet gas exchange N2 /air. In their original
work [6], the kL a evaluation assumed that the interchange
of the gases in gas hold-up is immediate (steady-state model
of the gas phase). Corresponding kL values were indepen-
dent of power dissipated in the liquid by the agitator and
matched the constant value of 4.03.104 m s1 as predicted
by Calderbanks model [6] for large bubbles (Eq. (2)).
When the gas phase dynamics was taken into account (the
flushing out of an old gas by a new one), the correspond-
ing values of kL were higher and well fitted by the eddy
model (Eq. (1)) with c1 = 0.5, see Fig. 1. (It was shown
by Linek et al. [24] that faulty sampling of liquid used by
Figueiredo and Calderbank [16] leads to considerably over-
estimated kL a values for kL a > 0.1 s1 . Therefore, kL values
corresponding to this range are not included in the figure.)
They concluded [16] (in difference with the original work
[6]) that kL is dependent on the turbulent intensity expressed
as a function of the power dissipated per unit volume and
that the coefficient is considerably higher than would be
found for bubbles in free rise under gravitational force as
given by the model of small and large bubbles.
Vasconcelos variant of the slip velocity model based
on bubble contamination kinetics, Eqs. (4)(8), was used
by Alves et al. [9] to interpret kL data in a double Rush-
ton turbine-stirred tank. The application of Vasconcelos
model to the interpretation of unreliable mass transfer co-
Fig. 1. Effect of gas phase mixing model used for kL a evaluation; kL values calculated from kL a and a presented by Figueiredo and Calderbank [16].
123
efficients in the stirred vessel has been analysed by Linek
et al. [3]. Alves et al. [9] measured the overall volumetric
mass transfer coefficients kL a by the peroxide decomposi-
tion steady-state technique with air as a gas phase, show-
ing that kL decreased slightly with power dissipated. Linek
et al. [25] showed, however, that the steady-state methods
with air absorption/desorption give substantially lower kL a
values at high mixing intensities as compared to pure oxy-
gen absorption. If the coefficients kL a were measured by the
non-problematic pressure dynamic method [26] with pure
oxygen absorption in the same apparatus as in Alves et al.
[9], the relevant kL data were well interpreted in terms of the
eddy model (Eq. (1)) with c1 = 0.531 [3]. The corrected
data have shown that the decisive quantity to correlate kL in
stirred tanks is power dissipated in the liquid phase rather
than the bubble diameter and their rigidity and that kL in-
creases with increasing turbulence intensity.
It seems from the literature survey that in mechanically
agitated dispersions, the intensity of energy dissipation plays
a dominant role in the value of the mass transfer coefficient
and that the successful applications of the slip velocity
models in stirred tanks are probably only a result of errors
in evaluation of kL a. On the contrary, in dispersions with no
mechanical agitation, both conceptions (eddy and slip
velocity) are equally successful. This may indicate that the
mechanism of mass transfer without mechanical agitation
differs from that in the presence of agitation, as was sug-
gested by Bouaifi et al. [14]. The same indicate the results by
Lamont and Scott [1]. They compared rates of mass trans-
fer from turbulent liquid into bubbles flowing in vertical
and horizontal tube where the bubbles can and cannot rise
124 V. Linek et al. / Chemical Engineering and Processing 44 (2005) 121130
freely through the turbulent liquid, respectively. If the bub-
bles could rise freely through the liquid, the mass transfer
coefficient did not depend on intensity of turbulence at low
Reynolds number. With increasing Re number, the role of
the intensity increased and at high Re number, the dependen-
cies for both arrangements (with and without the free rise of
bubbles) coincided. It means that both mechanisms operate
simultaneously but they apply at different extent, depending
on intensity of turbulence.
Surface-active substances reduce the mass transfer coef-
ficient markedly in gasliquid dispersions. Although a con-
siderable effort was made to identify the mechanism of mass
transfer in the presence of surfactants, no consistent conclu-
sion had been drawn. Two possibilities were offered [27]:
surfactant molecules affect mass transfer via hydrodynam-
ics effects or via the formation of an interfacial barrier layer.
The barrier effect is due to a monolayer of the surfactants
formed at the interface, which can offer a resistance to be
crossed by the gas molecule. The resistance exerted by the
monolayer is a complex phenomenon, possibly depending on
the effects of absorbing gassurfactantwater interactions,
which are related to the molecular structure of the surfac-
tant: the polarity of the hydrophilic group, the molecular
weight of the hydrophilic group, the hydrophilic chain length
and the orientation of the molecules at interface. Hydrody-
namics effects are induced by the difference in composition
and physical properties of the liquid at the interface and in
the bulk of liquid. Surfactants (but also electrolytes [28,29])
tend to create different spatially oriented structures in close
proximity of the interface and in bulk liquid and thus, their
presence causes an energetic barrier for the surface renewal
due to additional requirements of reordering the liquid en-
tering or leaving the interface. For example, if the liquid
turbulent eddies penetrate to the interface, they would bring
about an increase in the free energy per unit area of inter-
face due to the replacement of the interfacial liquid layer
by inner liquid of higher surface tension. This dampens the
rejuvenation of the interfacial liquid by fresh bulk liquid,
renders mass transfer more difficult and lowers the value
of mass transfer coefficient. Llorens et al. [27] decided the
hydrodynamic effect to be the principal. They offered the
surface pressure (i.e. the difference in surface tension of
the solution without and with s , the surfactant) to rep-
resent a measure of the additional work needed to replace
a surface element: (kL )s /kL = 1/(1.25 + 0.073 ) for 0.4 <
< 16 mN m1 and (kL )s /kL = 0.41 for > 16 mN m1 .
The form of the correlation did not depend on the type of
surfactant. Zieminsky and Whittemore [28] offered the ionic
strength and Machon and Linek [29] the electrical surface
potential to correlate the reduction of kL in electrolyte solu-
tions as measures of the additional work needed to replace a
surface element in these solutions. Vzquez et al. [12] used
a simple correlation between the mass transfer coefficient
and surface tension of solution under the form kL 1.35 .
Vasconcelos et al. [8] interpreted mass transfer coefficients
in air-lift and bubble columns with solution of surfactants by
the model of gradual surface contamination (Eqs. (4)(8))
using three empirical parameters: htrans , k/csurf and vt /vsl .
The aim of this paper is to make measurements with liq-
uids of various physical properties in order to define the
effect of the liquid properties and operating conditions on
the parameter kL and the limits of validity of the literature
models for the interpretation of mass transfer coefficients in
bubble dispersions. The method, which is used for the mea-
surements, was verified in Part I to minimize misinterpreta-
tions.
2. Experimental
The experiments were performed in a flat-bottomed
cylindrical vessel with internal diameter 0.19 m, equipped
with standard Rushton turbine of 0.075 m in diameter lo-
cated 0.08 m above the bottom. Four baffles (1/10 of the
vessel diameter width) were symmetrically mounted. A
membrane-covered polarographic oxygen probe was sit-
uated in the bottom to measure the equilibrium partial
pressure of oxygen in the bulk of liquid. Gas (pure oxygen
or air) was taken from a pressure flask and introduced 2 cm
under the impeller through a tube of i.d. 4 mm. The bubble
column was realized by keeping zero rotational frequency of
the agitator. The experiments were performed with the same
solutions as those used in the kinetics measurement: 0.8 M
sodium sulphite solution, pure or with the addition of vis-
cosity or surface tension-changing additives (see Table 1).
Physical properties of the solutions are given in Part I of
this work. The vessel was filled with liquid to the height of
226 mm, which amounts to 5.9 l. Superficial velocities of
aeration gas were 1.8, 3.6 or 5.4 mm s1 , which correspond
to volumetric flow rates 5.1 105 , 10.2 105 and 15.3
105 m3 s1 , respectively. The impeller rotational speeds
were 0, 5.5, 10, 14.33 and 18.83 l/s. Temperature was kept
at 30 1 C by a cooling coil.
The specific power dissipated by agitators in the liquid,
eagit , was measured by a strain gauge mounted on the im-
peller shaft. The total specific power e dissipated in the liq-
uid by the agitator and the rising bubbles was calculated as
e = eagit + gvs . The volumetric mass transfer coefficient
and interfacial area were measured by the Danckwerts plot
method described in detail in Part I.
3. Results
3.1. Volumetric mass transfer coefficient
The overall volumetric mass transfer coefficients mea-
sured by pure oxygen absorption were expressed as a func-
tion of the total specific power dissipated in the liquid as
follows:
kL a = 1 e2 (9)
 V. Linek et al. / Chemical Engineering and Processing 44 (2005) 121130 125

Table 1
Solutions used in this work and pertinent symbols and lines used in figures and parameters , and in correlations (9), (10) and (11) obtained by
regression of the experimental data
Solution Symbol Lines Parameter i =1 i =2 Mean Mean
deviation (%) deviationa
0.8 kmol m3 Na2 SO3 i 3.35 105 1.18 34 21
i 0.523 0.863 36 36
i 1.06 104 0.238 19 20
0.8 kmol m3 Na2 SO3 + Sokrat 44 (3 vol.%) i 3.26 105 1.16 44 43
i 0.507 0.913 40 43
i 6.70 105 0.243 11 11
0.8 kmol m3 Na2 SO3 + CMC TS.5 (0.2 wt.%)
i 3.65 105 1.08 30 29
i 1.21 0.632 33 49
i 1.04 104 0.276 25 26
0.8 kmol m3 Na2 SO3 + CMC TS.20 (0.6 wt.%)  i 2.21 104 0.771 14 14
i 1.80 0.689 13 14
i 1.15 104 0.090 5 5
0.8 kmol m3 Na2 SO3 + ocenol (3 ppm by volume) + i 1.27 104 0.732 27 14
i 2.35 0.492 12 9
i 7.97 105 0.188 17 13
0.8 kmol m3 Na2 SO3 + PEG 1000 (100 ppm by mass) i 2.65 105 1.15 1
i 1.14 0.903 2
i 0.237 0.246 1

a Mean deviation when the gas flow rate vs is taken into account as an additional parameter (e.g., kL a = 1 e2 vs 3 ).

1, 2 values together with mean deviations of the data are
given for all batches in Table 1. Taking into account gas
flow rate vs as additional independent parameter has not
improved fidelity of the correlation as follows from a com-
parison of the mean deviations of the relation (9) with and
without taking into account vs (see Table 1). Fig. 2 clearly
shows that kL a data for pure sulphite solution agree well
with a recommended correlation for non-coalescent batches
of electrolyte solutions [24]. Marked reductions of kL a
are observed with ocenol (up to 10 times) and long-fiber
carboxymethylcellulose (CMC) TS.20 (up to four times).
The ocenol has a strong antifoaming effect that enhances
the coalescence rate of bubbles in dispersion. The resulting
coefficients are even lower than those predicted by a corre-
lation suggested for coalescent batches like pure water [24].
The reduction of kL a due to CMC TS.20 cannot be ascribed

Fig. 2. kL a/(vs )0.4 measured by pure oxygen absorption as a function of total power dissipated in the liquid, e. Symbols and lines (see Table 1).
126 V. Linek et al. / Chemical Engineering and Processing 44 (2005) 121130
Fig. 3. Interfacial area a measured by pure oxygen absorption as a function of total power dissipated in the liquid, e. Symbols and lines (see Table 1).
unequivocally to higher viscosity of the batch as the addi-
tion of Sokrat 44 (which increased the viscosity to the same
extent as CMC) made no reduction in kL a.
3.2. Specific interfacial area
The specific interfacial areas measured by pure oxygen
absorption were expressed as a function of the total specific
power dissipated in the liquid as follows:
a = 1 e2
1, 2 values together with mean deviations of the data are
given for all batches in Table 1. Taking into account gas
flow rate vs as an additional independent parameter has not
improved fidelity of the correlation as follows from a com-
parison of the mean deviations of the relation (10) with and
without taking into account vs (see Table 1). This agrees
with the finding of Westersterp et al. [30], who stated that
the interfacial area is independent of gas flow rate at im-
peller speeds higher than a critical one. The critical im-
peller speed for Rushton turbine and sulphite solution equals
9.4 s1 [29]. All but one of the impeller speeds used in this
work were higher than this critical value. The results given
in Fig. 3 show that ocenol and short-fiber CMC TS.5 pro-
duce a marked decrease of the interfacial area (up to five and
three times, respectively), and that a marked increase is pro-
duced by polyethylenglycol (PEG) (up to three times) and
a smaller one by the addition of Sokrat (up to two times).
3.3. Mass transfer coefficient
Mass transfer coefficients were calculated by the ratio
kL a/a. The coefficients evaluated from the data measured by
pure oxygen absorption were expressed as a function of the
total specific power dissipated in the liquid as follows:
kL = 1 e2 (11)
1, 2 values together with mean deviations of the data are
given for all batches in Table 1. The results show clearly that
kL increases with increasing power dfissipated in the liq-
uid with exponent 2 equal approximately to 0.25 in agree-
ment with the eddy model (1) (see Fig. 4). Exceptions
(10) are the solutions with the addition of long-fiber CMC TS.20
and ocenol. In these solutions, the coefficients increase with
power but the exponent 2 is lower than 0.25 ( 2 = 0.09
and 0.188, respectively). The correlation (11) also fits well
the data measured in the bubble column, which are repre-
sented by the points for e < 100 W m3 in Fig. 4. kL data
for all Newtonian batches without surface-active agents (i.e.
pure sulphite solution and with the addition of Sokrat and
short-fiber CMC) are well fitted by the equation

0.243

ev D
kL = 0.915 0.623 (12)
v
with a mean deviation of 19%. If the theoretical value 0.25 of
the power exponent and 0.5 of Schmidt number exponent
are used, the equation changes to

0.25
0.5
ev D
kL = 0.448 (13)
v
with a mean deviation of 20%. Since the deviation is prac-
tically the same as the deviations of correlation (11), cor-
relations (12) and (13) describe the effects of viscosity and
diffusivity satisfactorily. Experimental and calculated values
from Eq. (13) of kL are compared in Fig. 5.
 V. Linek et al. / Chemical Engineering and Processing 44 (2005) 121130 127

Fig. 4. Liquid film mass transfer coefficient measured by pure oxygen absorption vs. total power dissipated in the liquid, e. Symbols and lines (see Table 1).

Literature data for Newtonian batches (water [3,4,16,17] Table 2

and for sodium sulphite solutions [3,19]) together with Parameters c1 , c2 and c3 in correlation (14) obtained by regression of
literature and our data for Newtonian batches
our Newtonian data were also evaluated according to
relation kL = c1 (ev/)c2 (D/v)c3
c1 c2 c3 Mean deviation (%)

c2
c3
ev D 0.363 0.213 0.542 19
kL = c1 (14)
v 0.265 0.210 0.500 19
0.463 0.250 0.500 22

yielding the regression constants c1 , c2 and c3 given in

Table 2. Correlation (14) with the theoretical exponents c2
= 0.25 and c3 = 0.5 fits our and literature data on Newto-
nian batches with a mean deviation of 20%. Experimental
and calculated values of kL from Eq. (14) are compared in
Fig. 6.
3.4. Effect of surfactants
Additions of substances for surface tension depression
(ocenol, PEG, CMC TS.20, see Table 1 of Part I) induced
a significant reduction of the mass transfer coefficient (by

Fig. 6. Comparison of our and literature experimental data for Newtonian

Fig. 5. Comparison of experimental and calculated values of mass transfer batches with those calculated from Eq. (14) with c1 = 0.463, c2 = 0.25
coefficient kL from Eq. (13). Symbols (see Table 1). and c3 = 0.5. Hollow symbols (see Table 1) [31].
128 V. Linek et al. / Chemical Engineering and Processing 44 (2005) 121130
Fig. 7. Relative decrease of mass transfer coefficient in the presence of
surfactant (in comparison to its value in pure sulphite solution) as a
function of surface pressure . Comparison with literature data measured
in bubble columns [12,8] and in wetted-wall column [27].
40 up to 80%). The results are given in Fig. 7 together with
literature data measured in bubble columns [12,8] and in
wetted-wall column [27]. In the figure, the relative decrease
of kL (in comparison to its value in pure sulphite solution,
(kL )s /kL ) is plotted as a function of surface pressure . It is
apparent that neither the surface pressure nor surface tension
alone (cf. the values of given in Table 1 of Part I) is
Fig. 8. Comparison of kL a, a and kL data measured by pure oxygen and air absorption in pure sulphite solution and with addition of ocenol. Air
absorption, filled symbols; oxygen absorption, hollow symbols; see Table 1.
the proper parameter to correlate the effect of surfactants
on kL .
3.5. Effect of diluted gas (air) absorption
Experiments with air were performed in pure sulphite so-
lution (non-coalescent batch) and with the addition of ocenol
(coalescent). The results are plotted in Fig. 8 together with
those measured by pure oxygen absorption. While the mass
transfer coefficients measured with oxygen increase with in-
creasing dissipated power, the kL values determined with air
either do not depend (in pure sulphite solution) or slightly
decrease with increasing power (in sulphite with ocenol).
The coefficients roughly coincide at the lowest stirring
intensity (330 rpm). With increasing stirring intensity, the
difference between them increases up to 70% at the highest
stirring intensity (1130 rpm). The differences can be caused
by application of an improper gas phase mixing model of the
dispersion, leading to an incorrect value of the driving force
for absorption, which proliferates to kL values. Application
of pure oxygen absorption avoids the problem as the oxygen
concentration and, therefore, the absorption efficiency are
the same for all bubbles. Thus, kL data used in this work are
measured by pure oxygen. The absorption of air has only
been used to demonstrate the possible reason of discrepancy
in the literature data. Fig. 9 shows that kL data measured
with air can be interpreted in terms of Calderbanks model:
data measured in pure sulphite solution are independent
of bubble diameter and correspond to Eq. (2) for large
 V. Linek et al. / Chemical Engineering and Processing 44 (2005) 121130 129

Fig. 9. Comparison of mass transfer coefficients measured by air absorption in pure sulphite solution and in the solution with addition of ocenol with
Calderbanks model.

bubbles. (The average bubble diameter was calculated from
gas hold-up and interfacial area as d = 6/[(1)a].) The
data measured in solution with ocenol fall between the
values for large and small bubbles. This is usually
interpreted by partial contamination of bubble surface by
surfactant, which shifts the transition between large and
small bubbles to higher bubble diameters. The decrease
of kL values with increasing stirring intensity (see Fig. 8 for
ocenol) can be explained by an increasing amount of small
bubbles. Such interpretation is frequently used in literature
for the data measured by absorption of diluted gas.
be the result of interpretation of physically inconsistent kL
data.
Acknowledgements
This work was supported by the Grant agency of Czech
Republic through the project no. 104/98/1126 and by Czech
Ministry of Education through the scientific project no.
223400007.
Appendix A. Nomenclature

4. Conclusions a specific interfacial area based on the liquid

volume, m1
Mass transfer coefficients measured by absorption of pure c1, 2, 3 constants in correlation (14)
oxygen increased in all batches with increasing power dis- d bubble diameter, m
sipated in the liquid raised approximately to 0.25. An ex- D gas diffusivity in the liquid, m2 s1
ception was the addition of long-fiber CMC for which the e total specific power dissipated in the liquid
exponent is only 0.09. kL data for all Newtonian batches volume, W m3
without surface-active agents (i.e. pure sulphite solution and eagit specific power dissipated by agitators in the
with addition of Sokrat and short-fiber CMC) are well fit- liquid volume, W m3
ted by the eddy model of Lamont and Scott [4], Eq. (13). f impeller rate, s1
Substances that depress the surface tension of the absorp- g gravitational constant, m s2
tion solution (ocenol, PEG, CMC TS.20) induced a signif- kL liquid-side mass transfer coefficient measured by
icant reduction of the mass transfer coefficient (by 40 up oxygen absorption, m s1
to 80%). Absorption of air (diluted gas) did not yield cor- (kL )s liquid-side mass transfer coefficient in the
rect kL data, not depending on or slightly decreasing with presence of surfactant, m s1
increasing power. This is due to the application of an im- (kL )air liquid-side mass transfer coefficient measured by
proper model of gas flow pattern in the dispersion (i.e. per- air absorption, m s1
fect mixing of gas phase), which leads to an incorrect value kL a volumetric mass transfer coefficient referred to
of the driving force for absorption that proliferates to kL the liquid volume, s1
values. Since this data are well interpreted by Calderbanks k empirical constant of the model of bubble
model of small and large bubbles, this model seems to contamination kinetics, mol m7/2 s
130 V. Linek et al. / Chemical Engineering and Processing 44 (2005) 121130
q volumetric gas flow rate, m3 s1
vs superficial velocity, m s1
vt single bubble terminal velocity in still water, m s1
vsl slip velocity, m s1
Greek symbols
, , constants in correlations (9), (10) and (11)
gas hold-up
kinematics viscosity of the liquid phase, m2 s1
=( s ), surface pressure, N m1
density of the liquid phase, kg m3
 =(liquid gas ), kg m3
s, surface tension of solution with or without
surfactant, N m1
mobile time for transition of bubble surface to rigidity, s
R average bubble residence time, s
References
[1] J.C. Lamont, D.S. Scott, An eddy cell model of mass transfer into
surface of a turbulent liquid, AIChE J. 16 (1970) 513519.
[2] Y. Kawase, B. Halard, M. Moo-Young, Liquid-phase mass trans-
fer coefficients in bioreactors, Biotech. Bioeng. 39 (1992) 1133
1140.
[3] V. Linek, M. Fujasov, M. Korda, T. Moucha, Gasliquid mass
transfer coefficient in stirred tank interpreted through models of
idealized eddy structure of turbulence in the bubble vicinity, Chem.
Eng. Proc. (in press).
[4] B.D. Prasher, G.B. Wills, Mass transfer in an agitated vessel, I&EC
Process Des. Develop. 12 (1973) 351354.
[5] Y. Kawase, B. Halard, M. Moo-Young, Theoretical prediction of
volumetric mass transfer coefficients in bubble columns for Newto-
nian and non-newtonian fluids, Chem. Eng. Sci. 42 (1987) 1609
1617.
[6] P.H. Calderbank, M.B. Moo-Young, The continuous phase heat and
mass transfer properties of dispersions, Chem. Eng. Sci. 16 (1961)
3961.
[7] N. Frssling, ber die Verdnstung fallenden Tropfen, Gerlands
Beitage Geophys. 52 (1938) 170216.
[8] J.M.T. Vasconcelos, J.M.I. Rodrigues, S.C.P. Orvalho, S.S. Alves,
R.L. Mendes, A. Reis, Effect of contaminants on mass transfer
coefficients in bubble column and airlift contactors, Chem. Eng. Sci.
58 (2003) 14311440.
[9] S.S. Alves, C.I. Maia, J.M.T. Vasconcelos, Gasliquid mass transfer
coefficient in stirred tank interpreted through bubble contamination
kinetics, Chem. Eng. Proc. 43 (7) (2004) 823830.
[10] T. Reith, Physical aspects of bubble dispersions in liquids, Ph.D.
Thesis, Delft Techn. Univ., 1968.
[11] G.B. Dillon, I.J. Harris, The determination of mass transfer coeffi-
cients and interfacial areas in gasliquid contacting systems, Can. J.
Chem. Eng. 44 (1966) 307312.
[12] G. Vzquez, M.A. Cancela, C. Riverol, E. Alvarez, J.M. Navaza,
Application of the Danckwerts method in a bubble column. Effects
of surfactants on mass transfer coefficient and interfacial area, Chem.
Eng. J. 78 (2000) 1319.
[13] A. Schumpe, W.D. Deckwer, Gas holdups, specific interfacial areas,
and mass transfer coefficients of aerated carboxymethyl cellulose
solutions in a bubble column, I&EC Process Des. Develop. 21 (1982)
706711.
[14] M. Bouaifi, G. Hebrard, D. Bostoul, M. Roustan, A comparative
study of gas hold-up, bubble size, interfacial area and mass transfer
coefficients in gasliquid reactors and bubble columns, Chem. Eng.
Proc. 40 (2001) 97111.
[15] P. Wongsuchoto, T. Chrinpanitkul, P. Pavasant, Bubble size distribu-
tion and gas/liquid mass transfer in airlift contactors, Chem. Eng. J.
92 (2003) 8190.
[16] M.M.L. Figueiredo, P.H. Calderbank, The scale-up of aerated mixing
vessels for specified oxygen dissolution rates, Chem. Eng. Sci. 34
(1979) 13331338.
[17] M. Bouaifi, M. Roustan, Bubble size and mass transfer coefficients
in dual-impeller agitated reactors, Can. J. Chem. Eng. 76 (1998)
390397.
[18] N.C. Panja, D.P. Rao, Measurement of gasliquid parameters in a
mechanically agitated contactor, Chem. Eng. J. 52 (1993) 121129.
[19] V. Linek, J. Mayrhoferov, J. Monerov, The influence of diffusivity
on liquid phase mass transfer in solutions of electrolytes, Chem.
Eng. Sci. 25 (1970) 10331045.
[20] W.T. Koetsier, D. Thoenes, Mass transfer in a closed stirred gas/liquid
contactor part 2: the liquid phase mass transfer coefficient kL , Chem.
Eng. J. 5 (1973) 7175.
[21] F. Yoshida, Y. Miura, Gas absorption in agitated gasliquid contac-
tors, I&EC Process Des. Develop. 2 (1963) 263268.
[22] C.W. Robinson, C.R. Wilke, Simultaneous measurement of interfacial
area and mass transfer coefficients for a well-mixed gas dispersion
in aqueous electrolyte solutions, AIChE J. 20 (1974) 285294.
[23] I.T.M. Hassan, C.W. Robinson, Mass transfer coefficients in mechan-
ically agitated gasaqueous electrolyte dispersions, Can. J. Chem.
Eng. 58 (1980) 198205.
[24] V. Linek, V. Vacek, P. Bene, A critical review and experimental
verification of the correct use of the dynamic method for the de-
termination of oxygen transfer in aerated agitated vessels to water,
electrolyte solutions and viscous liquids, Chem. Eng. J. 34 (1987)
1134.
[25] V. Linek, P. Bene, J. Sinkule, Critical assessment of the steady-
state Na2 SO3 feeding method for kL a measurement in fermentors,
Biotechnol. Bioeng. 35 (1990) 766770.
[26] V. Linek, P. Bene, J. Sinkule, T. Moucha, Non-ideal pressure step
method for kL a measurement, Chem. Eng. Sci. 48 (1993) 15931599.
[27] J. Llorens, C. Mans, J. Costa, Discrimination of the effect of sur-
factants in gas absorption, Chem. Eng. Sci. 43 (1988) 443450.
[28] S.A. Zieminski, R.C. Whittemore, Behavior of gas bubbles in aqueous
electrolyte solutions, Chem. Eng. Sci. 26 (1971) 509520.
[29] V. Machon, V. Linek, Effect of salts on the rate of mass transfer
across a plane interface between a gas and mechanically agitated
aqueous solutions of inorganic electrolytes, Chem. Eng. J. 8 (1974)
5361.
[30] K.R. Westerterp, L.L. Van Dierendonck, J.A. de Kraa, Interfacial
areas in agitated gasliquid contactors, Chem. Eng. Sci. 18 (1963)
157176.
[31] A.A. Kulkarni, J.B. Joshi, Simultaneous measurement of flow pattern
and mass transfer coefficient in bubble columns, Chem. Eng. Sci.
59 (2004) 271281.
[32] V. Linek, P. Bene, V. Vacek, F. Hovorka, Analysis of differences
in kL a values determined by steady-state and dynamic methods in
stirred tanks, Chem. Eng. J. 25 (1982) 7788.
 Chemical Engineering and Processing 44 (2005) 131137

Separation of n-hexaneethyl acetate mixtures by azeotropic batch

distillation with heterogeneous entrainers
I. Rodriguez-Donis a , J. Acosta-Esquijarosa a , V. Gerbaud b, ,
E. Pardillo-Fondevila a , X. Joulia b
a Centro de Quimica Farmaceutica, Ave. 200 y 21 Atabey Apdo. 16042, Playa, C. Habana, Cuba
b Laboratoire de Gnie Chimique UMR CNRS 5503, ENSIACETUPSCNRS, BP1301, 5 rue Paulin Talabot, 31106 Toulouse Cedex 1, France

Received 26 February 2004; received in revised form 4 May 2004; accepted 14 May 2004
Available online 2 July 2004

Abstract

In this article, a systematic study of the separation of the n-hexaneethyl acetate mixture with an entrainer by heterogeneous azeotropic
batch distillation is performed. Based upon the thermodynamic behaviour of the ternary mixtures, potential entrainers partially miscible with
one or two original azeotropic components are chosen. In all cases, the entrainer adds a heterogeneous binary or ternary azeotrope that is the
lowest boiling point in the ternary diagram. Therefore, it leaves the column by the overhead stream which is subcooled to get two liquid phases
in the decanter. The phase with the highest amount of the original component is removed as distillate product whereas the entrainer-rich phase
is continuously refluxed to the column. Considering methanol, acetonitrile, water and nitromethane as heterogeneous entrainers, screening was
performed based on the composition of the unstable heteroazeotropic mixture, the ratio of both liquid phases in the condensed top vapour and the
purity of the distillate product determined by the liquidliquid envelope at the decanter temperature. The process feasibility analysis is validated
by using rigorous simulation with the batch process simulator ProSimBatch. Simulation results are then corroborated in a bench experimental
column for the selected entrainer, showing several advantages of heterogeneous batch distillation (HBD) compared to homogeneous systems.
2004 Elsevier B.V. All rights reserved.

Keywords: Batch distillation; Heterogeneous entrainer; Azeotropic distillation

1. Introduction substance in the original mixture have been developed and

Batch processes are again becoming important because
of recent expansion of the pharmaceutical and speciality
chemical industries in many developing countries. Besides,
recycling of liquid waste streams is becoming a key issue in
all processing plants to meet stricter environmental regula-
tions. Batch distillation is a widely used separation method
in batch processes because of its inherent operational flex-
ibility due to some extent to its time dependent operation.
The separation of azeotropic mixtures or close boiling com-
ponents is a challenging task in many chemical processes
as it is impossible using a single conventional column or
as a pressure swing distillation is an uneconomical process.
Alternative techniques based on the addition of an auxiliary
Corresponding author. Tel.: +33 5 62 88 58 26;
fax: +33 5 62 88 58 29.
E-mail address: Vincent.Gerbaud@ensiacet.fr (V. Gerbaud).
are usually known as azeotropic and extractive distillation.
Process synthesis and design of these non-conventional
distillation processes proceed in two steps. The first
stepprocess synthesisis the selection of one or more
candidate entrainers along with the computation of ther-
modynamic properties like residue curve maps that help
assess many column features such as the adequate column
configuration and the corresponding product cuts sequence.
The second stepprocess designinvolves the search for
optimal values of batch distillation parameters such as the
entrainer amount, reflux ratio, boiler duty and number of
stages. The complexity of the second step depends on the
solutions obtained at the previous level, because efficiency
in azeotropic and extractive distillation is largely deter-
mined by the mixture thermodynamic properties that are
closely linked to the nature of the entrainer. Hence, we
have established a complete set of rules for the selection of
feasible entrainers for the separation of non ideal mixtures

0255-2701/$ see front matter 2004 Elsevier B.V. All rights reserved.
doi:10.1016/j.cep.2004.05.006
132 I. Rodriguez-Donis et al. / Chemical Engineering and Processing 44 (2005) 131137
by homogeneous azeotropic batch distillation [1]. In con-
trast to batch homogeneous distillation processes, studies
on heterogeneous batch distillation (HBD) were scarce
[24] until a new method for the synthesis of heterogeneous
distillation in a batch rectifier was defined and a complete
set of rules for the selection of heterogeneous entrainers for
the separation of non-ideal mixtures was established [5,6].
Simulation and experimental results have shown key dif-
ferences between homogeneous and heterogeneous batch
distillation [6]. Indeed, some advantages of heterogeneous
batch processes compared to homogeneous systems are: (i)
more suitable candidate entrainers and hence, more design
alternatives for the separation of non-ideal mixtures, (ii)
simplified distillation sequences thanks to the liquidliquid
phase split occurring in some parts of the column and in the
decanter, (iii) the addition of a smaller amount of entrainer to
the original mixture, (iv) the use of more flexible reflux poli-
cies through any combination of the entrainer-rich phase and
the distillate-rich phase, (v) the still path can cross unstable
separatrices allowing the separation of components located
in different basic distillation regions. The overall cost re-
duction generated by these advantages makes heterogeneous
batch distillation an attractive option for the separation of
non-ideal mixture generated by specialty industries.
The aim of this paper is to study the performance of het-
erogeneous batch distillation for the separation of the mix-
ture n-hexaneethyl acetate. This azeotropic mixture is often
used in pharmaceutical industries for separating bioactive
substances by liquid chromatographic processes. The sepa-
ration of this mixture by batch distillation was first studied
by Rodriguez-Donis et al. [7], by using acetone as a homo-
geneous entrainer. As the separation by using a single batch
distillation process with acetone was not possible, a hybrid
process was proposed involving a homogeneous azeotropic
distillation column with acetone and a liquidliquid extrac-
tion column with water. To circumvent this complex pro-
cess, we propose in this paper a simpler process by using
heterogeneous azeotropic batch distillation. Based on en-
trainer selection rules established by Rodriguez-Donis et al.
[5], which consider the thermodynamic behaviour of mul-
ticomponent mixtures, an initial list of potential entrainers
partially miscible with one or both original azeotropic com-
Table 1
Potential heterogeneous entrainers for the separation of n-hexaneethyl acetate mixture with thermodynamics data from Gmehling et al. [8]
Entrainer bp Binary azeotrope with
n-hexane (68.7 C)
( C) xmolar a T ( C)
Methanol 64.7 0.5100b 50.6
Acetonitrile 81.5 0.5755b 54.4
Water 100.0 0.7860b 61.8
Nitromethane 101.2 0.7271b 62.0
a The mole fraction indicated is that of the lightest component in the mixture.
b Heteroazeotrope.
ponents are selected. From the analysis of residue curve map
features, in particular the decanter phases composition and
the distillation region separatrices curvature, the sequence of
products to be drawn from the rectifier column can be pre-
dicted. The number of distillation tasks in the whole process
is set by the products targeted purity.
2. Selection of a suitable entrainer for the separation
of n-hexaneethyl acetate mixtures by heterogeneous
azeotropic batch distillation
Liquidvapour equilibrium data for the binary mixture
n-hexaneethyl acetate have been scarcely reported in the
scientific literature [8]. Those authors reported a mini-
mum boiling temperature azeotrope with an average mole
fraction of n-hexane equal to 0.6565 and a temperature
of 65.15 C at 101.3 kPa. A more detailed experimental
study, at atmospheric pressure, of liquidvapour equilib-
rium was performed by Acosta et al. [9]. Their estimation
of the azeotropic mixture corresponds to a temperature of
64.85 C with a molar fraction of n-hexane of 0.657 with
an average experimental error for the temperature and com-
position measurements of 0.02 C and 0.003 mole fraction,
respectively.
The candidate homogeneous and heterogeneous en-
trainers are investigated applying the rules reported by
Rodriguez-Donis et al. [1,5]. The existence of binary and
ternary azeotropes is sought by using literature data [8]
and by simulating liquidvapour and liquidliquidvapour
equilibrium by using the BibPhyAddIn Macro in an Excel
spreadsheet [10]. Confirming the need for a complex hybrid
process in case a homogeneous entrainer like acetone is
used [7], no homogeneous entrainer matching the reported
rules has been found whereas some suitable heteroge-
neous entrainers have been identified; the composition and
temperature of their azeotropic mixtures with the original
components are displayed in Table 1 [8].
As seen in Table 1, the boiling temperature of the pos-
sible entrainers is either lower or higher than that of the
original components. All candidates form a binary hetero-
geneous azeotrope with n-hexane. Methanol and acetonitrile
Binary azeotrope with Ternary azeotrope
ethyl acetate (77.1 C)
xmolar a T ( C) xmolar a T ( C)
0.708 62.05
0.6730 76.1
0.7000b 70.5 0.5764b 60.7
0.2558
Zeotropic
 I. Rodriguez-Donis et al. / Chemical Engineering and Processing 44 (2005) 131137
Table 2
Binary interaction parameters for NRTL and UNIQUAC models
Binary parameters Aij Aji aij
Binary coefficients for NRTL model (cal/mol)
n-Hexaneethyl acetate 720.731 33.844 0.1521
n-Hexaneacetonitrile 924.616 1212.64 0.2000
Ethyl acetateacetonitrile 580.783 253.247 0.1997
Binary coefficients for UNIQUAC model (cal/mol)
n-Hexaneethyl acetate 371.892 133.73
n-Hexanemethanol 1451.225 7.831
Ethyl acetatemethanol 770.688 147.777
n-Hexanewater 2292.190 324.669
Ethyl acetatewater 768.058 214.271
n-Hexanenitromethane 1080.270 7.335
Ethyl acetatenitromethane 181.416 42.973
add a homogeneous minimum boiling azeotrope with ethyl
acetate. Nitromethane forms a zeotropic mixture with ethyl
acetate, whereas water is the only component that forms an
additional heterogeneous azeotrope with ethyl acetate and
also adds a heterogeneous ternary azeotrope to the result-
ing mixture. Experimental liquidvapour equilibrium data
are available for all binary entrainer- ethyl acetate mixtures
[11]. However, for the binary system entrainer-n-hexane that
exhibits a significant miscibility gap, the only experimental
thermodynamic information is given in Table 1, except for
methanol for which liquidvapour equilibrium data were re-
ported by Gmehling et al. [11]. All previous experimental
data are used to regress binary interaction parameters for the
UNIQUAC and NRTL models with ChemCAD IV minimiz-
ing the mean square deviation between the experimental and
calculated mole fraction in the liquid-phase (Table 2). Then,
ternary liquidliquidvapour equilibrium is determined by
using these binary coefficients with BibPhyAddIn [10]. As
a result of the use of these binary interaction parameters
regressed on binary systems to predict the behavior of a
ternary system, we have detected the existence of the ex-
perimental heterogeneous ternary azeotrope in the mixture
n-hexaneethyl acetatewater, which has the lower boiling
point in the system (unstable node).
For the synthesis of heterogeneous batch distillation
the liquidliquid envelope at the decanter temperature is
considered in addition to the residue curve map. There-
fore, the binary interaction parameters used in predicting
liquidliquid equilibrium are estimated from binary hetero-
geneous azeotrope or liquidliquid equilibrium data [8,10].
Table 3 shows the calculated purity of original compo-
nents in each phase split at 25 C for all heterogeneous
azeotropes reported in Table 1. The thermodynamic mod-
els and binary coefficients used in the calculation of the
liquidliquidvapour equilibrium, liquidliquid equilibrium
at 25 C and the separatrices are reported in Table 2.
From the thermodynamic information given in Tables 13,
the residue curve maps are drawn in Fig. 1 for each entrainer
and the batch distillation task sequence required to perform
the separation of the original components is deduced using
published rules [5]. The residue curve maps of the ternary
133
Table 3
Purity of n-hexane-rich phase (xI ) and entrainer-rich phase (xII ) at 25 C
Entrainer Binary azeotrope with Ternary azeotrope
n-hexane
xnhexane
I xnhexane
II xI xII
Methanol 0.9152 0.1198 0.5346
Acetonitrile 0.9176 0.0605 0.3991
Water 0.7028 0.2481 0.1265
0.1754 0.0324
Nitromethane 0.9356 0.0893 0.2464 -
mixture n-hexaneethyl acetate-entrainer displayed in Fig. 1
indicate the temperature and the topological stability of all
singular points. As the heteroazeotropes are unstable nodes,
they are recovered in the decanter where the phase rich in
one of the original components LW is removed as distillate,
whereas the other original component is retained into the
still at the process end. In some cases, the composition of the
phase rich in original component defined by the liquidliquid
equilibrium at the decanter temperature may not meet purity
requirements; then, further distillation tasks are needed.
One of the key thermodynamic parameter in heteroge-
neous batch distillation is the decanter split ratio . It is de-
fined by the position of the heteroazeotrope composition on
its liquidliquid tie line at the decanter temperature or alter-
natively by the mole ratio of the entrainer-rich phase LR to
the overall liquid phase Lo into the decanter as follows:
LR = Lo (1)
Then, the molar quantity of entrainer-weak/original
component-rich phase LW is given by equation (2):
LW = (1 )Lo (2)
As stated by Rodriguez-Donis et al. [6], the reflux policy
to be used is strongly influenced by the split ratio at the
decanter. If the liquid reflux needed at the top of the column
is lower than LR , then the distillation can be performed by
using only the reflux of entrainer-rich phase. Otherwise, the
separation of original components requires the reflux of a
combination of both decanted phases.
Nitromethane shows the simplest residue curve map with
one unstable curved separatrix dividing the triangle in two
basic distillation regions. Methanol and acetonitrile give rise
two binary azeotropic mixtures and three distillation regions
that are bounded by two unstable curved separatrices. Water
shows the most complicated residue curve maps, due to the
presence of a ternary azeotrope and a miscibility gap with
both the n-hexane and the ethyl acetate component. In all
four cases, the heteroazeotrope (binary or ternary) has the
lowest boiling temperature of the system. As it can be seen in
Table 3, all entrainers except water provide the n-hexane-rich
phase LW as distillate product with a purity better than 0.91.
Water is not a desirable entrainer because of the existence
of ternary azeotrope whose n-hexane-rich phase has a water
purity much lower (0.70). Considering in Table 3 the split
134 I. Rodriguez-Donis et al. / Chemical Engineering and Processing 44 (2005) 131137

Fig. 1. Residue curve maps of ternary systems n-hexaneethyl acetate-heterogeneous entrainer.

ratio at the decanter (inside the range 0.30.6) and the
purity of both liquid phases at 25 C, methanol and acetoni-
trile are better entrainers than nitromethane. Methanol is a
light entrainer that would result in a ternary system with a
low boiling temperature that in turn will likely increase the
cost of the condenser cooling system and the make-up of
fresh entrainer to be added in the next batch. Besides, the
mole ratio of the product-to-be n-hexane in the methanol
II
rich phase (xnhexane in Table 3) is higher than the one cal-
culated for the acetonitrile. Therefore, acetonitrile seems the
best entrainer for the separation of n-hexaneethyl acetate
by heterogeneous azeotropic batch distillation.
(MESH) equations which are solved for the whole column,
decanter included and taking into account the liquidliquid
phase split. Numerical treatment of the Differential Alge-
braic Equation (DAE) system and discrete events handling is
performed with DISCo, a numerical package for hybrid sys-
tems with a DAE solver based on Gears method. The col-
umn technical features and operating conditions are shown in
Table 4. A sequence of two operational batch steps, namely
Table 4
Operating conditions of heteroazeotropic batch distillation with acetonitrile
Parameters Value

Initial charge (mol) 12.2

3. Simulation of the separation of the mixture Initial composition (molar) 0.705/0.173/0.122
n-hexaneethyl acetate by heteroazeotropic batch Number of theoretical trays 50
distillation using acetonitrile Operating pressure (atm) 1
Holdup on the trays (mol) 0.0072
We study the separation of n-hexaneethyl acetate mix- Tray efficiency 1
ture by using acetonitrile as a heavy heterogeneous entrainer. Vapour flow rate (mol/h) from 9.7 to 7.5
Extra entrainer-rich phase (mol/h) 1.2
The simulation of the process is performed with the batch Decanter holdup (mol) 3.4
process simulator ProSimBatch [10]. It enables to evaluate Decanter temperature ( C) 25
operational parameters like the entrainer amount that are Distillation time (h) 4.8
not provided by the feasibility and synthesis analysis The 0.3991
column model consists of usual plate by plate Material bal- 1.01
0.7
ance, Equilibrium, Summation of fractions and Heat balance
 I. Rodriguez-Donis et al. / Chemical Engineering and Processing 44 (2005) 131137 135

bination of both decanted liquid phases. In this example, the

Fig. 2. Simulation and experimental results of the separation of n-hexane
ethyl acetate by using batch distillation with acetonitrile as a heavy
heterogeneous entrainer.
an infinite reflux step and a distillate removal step described
in detail below are simulated.
The amount of heterogeneous entrainer to be added to
the original binary mixture is established from material bal-
ances considering that the entrainer composition in the still
should be less than 0.01 at the end of infinite reflux batch
step 1. Simulation shows that this composition is reached
after two hours, which is the end of batch step 1. It should
be noted that the initial ternary composition F is placed in
distillation region I that is not limited by the vertex of ethyl
acetate (see Fig. 2). Therefore, in order to reach the vertex
of ethyl acetate located in region II at the end of the distil-
lation process, the still path must cross the unstable separa-
trix connecting the heteroazeotrope with the homoazeotrope
n-hexaneethyl acetate. As established by Rodriguez-Donis
et al. [6], such a crossing is possible with different reflux
policies, either the reflux of entrainer-rich phase or a com-
distillate removal step (batch step 2) is performed refluxing
a combination of entrainer-rich LR and entrainer-weak LW
phases coming from the decanter, the reflux policy 2.3 of
Rodriguez-Donis et al. [6]. As in Rodriguez-Donis et al. [6]
the acetonitrile-rich phase L R  and n-hexane-rich phase L

W reflux flow rates are a fraction and of the amount
of each phase (LR and LW ) contained in Lo , the condensed
vapour flow rate entering the decanter. Table 4 displays the
and values that enable to withdraw the binary heteroge-
neous azeotrope during the whole distillation process. The
process requires that the reflux of the acetonitrile-rich phase
is larger ( >1) than the acetonitrile quantity in the vapour
flow rate leaving the column top Lo , while the amount of
n-hexane to be recycled at the top of the column is deter-
mined by the calculated value. > 1 means that drying
of the acetonitrile-rich phase in the decanter during opera-
tion can occur and care should be taken to avoid it in the
definition of an optimal reflux policy.
Fig. 2 shows the simulation results, the still path and the
column liquid profile, at the end of batch step 1 (infinite
reflux) and at the beginning and end of batch step 2 (dis-
tillate removal). Table 5 displays significant process values.
As can be seen in Fig. 2 and Table 5, only 12.2% of acetoni-
trile is added to the initial azeotropic binary mixture and the
resulting ternary mixture lies in the distillation region I. it
has been observed [6] that heterogeneous batch distillation
requires less entrainer than homogeneous batch distillation.
The liquid profile in the column at the end of the infinite
reflux operation links the still composition FR to the bi-
nary heteroazeotrope n-hexaneacetonitrile in the first top
tray of the column through the vapour line. Then, applying
the reflux policy detailed in Table 4, distillate removal starts.
Five minutes after the beginning of the distillation step,
the column liquid profile changes suddenly and crosses the

Table 5
Simulation and experimental results (x1 : n-hexane; x2 : ethyl acetate; x3 : acetonitrile)
Simulation results Experimental results

Quantity (mol) Molar fraction Quantity (mol) Molar fraction

Initial mixture 12.2 x1 0.705 12.2 x1 0.705

x2 0.173 x2 0.173
x3 0.122 x3 0.122
n-Hexane-rich phase (distillate + decanter) 8.6 x1 0.951 8.4 x1 0.968
x2 0.002 x2 0.001
x3 0.047 x3 0.031
Acetonitrile-rich phase (decanter) 1.1 x1 0.107 0.99 x1 0.082
x2 0.157 x2 0.190
x3 0.736 x3 0.728
Final still content (ethyl acetate) 1.7 x1 0.001 1.6 x1 0.001
x2 0.992 x2 0.991
x3 0.007 x3 0.008
Ethyl acetate recovery yield 81.04% 75.1%
n-Hexane recovery yield 95.1% 94.5%
Acetonitrile recovery yield 54.5% 48.4%
136 I. Rodriguez-Donis et al. / Chemical Engineering and Processing 44 (2005) 131137
unstable separatrix due to the recycle of both decanted liquid
phases. The reason is that the liquid phase leaving the first
tray of the column top has a composition (xT ) inbetween the
heteroazeotrope and the acetonitrile-rich phase composition
and is located in a distillation region (III) different from re-
gion (I) which contains the still composition at infinite reflux
FR . Consequently, the simulated still path crosses the un-
stable separatrix separating distillation regions I and II (full
black line in Fig. 2). At the end of batch step 2, the still
path reaches the ethyl acetate vertex when all n-hexane is
drawn as distillate product. The operating parameters and
are set to obtain the binary heteroazeotrope at the column
top during the whole step 2. Therefore, the split ratio is
kept constant. The n-hexane-rich phase is drawn as distillate
product whereas the acetonitrile-rich phase can be recov-
ered at the decanter at the end of the step 2. Table 5 reports
the simulated purity and recovery yield of all components.
Simulation computation time is typically less than 4 min on
a Pentium III-based computer.
4. Experimental verification of the separation of the
mixture n-hexaneethyl acetate by heteroazeotropic
batch distillation in a bench scale rectification column
The technical features of the column and the operating
conditions are similar for the experiments and the simula-
tions. The glass SCHOTT distillation column has a total
height of 1.7 m and an internal diameter of 0.026 m. The to-
tal height is packed with stainless steel wire mesh rings of
3 mm 3 mm. The total packed height is considered to be
equivalent to 45 theoretical plates and the total liquid hold
goes up of 52.5 ml. The condenser is subcooled at 25 C and
total liquid reflux is provided at the top of the column by a
solenoid valve. The open/close time of the solenoid valve is
defined considering that 70% of the condensed vapour is re-
turned to the column. The 30% remaining condensed vapour
is collected into the decanter (70 ml) where the light phase
is drawn as distillate (n-hexane-rich phase) whereas a por-
tion of the heavy entrainer-rich phase is recycled to the top
of the column with a peristaltic pump.
Experiments taking into account these operating condi-
tions (Table 4) are carried out in the bench distillation col-
umn and the results are presented in Fig. 2 and Table 5.
Analysis of the still composition is performed every 0.5 h.
The analysis of purity of the light and heavy phases in the
decanter and of the distillate product is made at the end of
the process (4.8 h). The acetonitrile, n-hexane and ethyl ac-
etate compositions are determined using a FISONS HRGC
chromatograph with a FID detector and with a column HP
INNOWax (30 m 0.53 mm, 0.25 m film). The injector
and detector temperature are held at 180 C while the col-
umn temperature is set equal to 65 C. Prior calibration is
done using internal standards; the samples were prepared by
weighing on a SARTORIUS BP 211D balance with a pre-
cision of 10 mg. The standard accuracy and reproducibility
in the composition for all components are found to be 0.1
and 0.3%, respectively.
As shown in Fig. 2, the still path obtained experimentally
(circles) is in excellent agreement with the still trajectory
calculated by simulation. The selected reflux policy permits
the still path to cross the separatrix into another distillation
region than the initial feed region. Hence, the still path is
able to reach the ethyl acetate vertex and this component re-
mains pure into the still at the end of the distillate removal
step. Such a behavior is not possible with a homogeneous
entrainer that gives rise to a similar residue curve map be-
cause the distillation process is restricted to the distillation
region where the initial composition of the mixture is lo-
cated. In this case, ethyl acetate could not be obtained as an
isolated product.
Table 5 displays the quantity and composition of the main
streams of the process. n-Hexane is drawn of as distillate
product with a molar purity of 96.8%. Another binary batch
distillation step is required to obtain n-hexane with a high-
est purity. Ethyl acetate with 99.6% of purity is recovered in
the still at the end of the process, while the acetonitrilerich
phase obtained from decanter can be reused in the next batch
distillation process. The make-up of fresh entrainer for the
next cycle of n-hexaneethyl acetate separation is approxi-
mately 52%.
5. Conclusions
Acetonitrile was chosen as a suitable heterogeneous
entrainer for the separation of the azeotropic mixture
n-hexaneethyl acetate by heteroazeotropic batch distil-
lation from the comparison of the residue curve maps of
several candidate entrainers. We have confirmed the feasi-
bility of the heteroazeotropic batch distillation via simula-
tion and by experimentation in a bench batch distillation
column. Good agreement was obtained between simulation
and experimental results. The novel process with a single
heterogeneous distillation column shows a great potential
over a reported hybrid process with a homogeneous distilla-
tion column and a liquidliquid extraction column. Indeed,
the consumption of entrainer is low and the separation of
all three components can be performed in a single batch
distillation step using the defined reflux policy. The ethyl
acetate purity is high enough to enable its recycling in a
chromatography process whereas the n-hexane-rich phase
must be purified in a simple non azeotropic batch distillation
process.
References
[1] I. Rodriguez-Donis, V. Gerbaud, X. Joulia, Entrainer selection rules
for the separation of azeotropic and close boiling temperature mix-
tures by homogeneous batch distillation, Ind. Eng. Chem. Res 40
(2001) 27292741.
 I. Rodriguez-Donis et al. / Chemical Engineering and Processing 44 (2005) 131137
[2] R. Dssel, J. Stichlmair, Separation of azeotropic mixtures by batch
distillation using an entrainer, Comp. Chem. Eng. 19 (1995) s113
s118.
[3] J. Khler, H. Haverkamp, N. Schadler, Zur diskontinuierlichen Rekti-
fikation azeotroper Gemische mit Hilfsstoffeinsatz, Chem. Ing. Tech.
67 (1995) 967971.
[4] S. Skouras, S. Skogestad, Separation of ternary heteroazeotropic
mixtures in a closed multivessel batch distillation-decanter hybrid,
Chem. Eng. Proc. 43 (2004) 291304.
[5] I. Rodriguez-Donis, V. Gerbaud, X. Joulia, Heterogeneous entrainer
selection rules for the separation of azeotropic and close boiling
temperature mixtures by heterogeneous batch distillation, Ind. Eng.
Chem. Res. 40 (2001) 49354950.
[6] I. Rodrguez-Donis, V. Gerbaud, X. Joulia, Feasibility of heteroge-
neous batch distillation, AIChE J. 48 (2002) 11681178.
137
[7] I. Rodriguez-Donis, U. Juregui-Haza, E. Pardillo-Fondevila, Sepa-
ration of n-hexane-ethyl acetate mixtures by azeotropic distillation,
Latin Am. Appl. Res. 29 (1999) 119127.
[8] J. Gmehling, J. Menke, J. Krafczyk, K. Fischer, Azeotropic Data,
Verlag Chemie, Weinheim, 1994.
[9] J. Acosta, A. Arce, J. Martnez-Ageitos, E. Rodil, A. Soto, Vapour
liquid equilibrium of the ternary system ethyl acetate + hexane
+ acetona at 101.32 kPa, J. Chem. Eng. Data 47 (2002) 849
854.
[10] S.A. ProSim, 2000, http://www.prosim.net, Reference Manual.
[11] J. Gmehling, U. Onken, VapourLiquid Equilibrium Data Collection,
DECHEMA Chemistry Data Series, vol. 1, 12 parts, DECHEMA,
Frankfurt am Main, 1982.
 | |

Publication information: Chemical Engineering and Processing (ISSN 0255- Central and South America), including air speed delivery. Periodical postage paid
CHEMICAL ENGINEERING AND PROCESSING 2701). For 2005, volume 44 is scheduled for publication. Subscription prices are
available upon request from the Publisher or from the Regional Sales Office
at Jamaica, NY 11431.
USA POSTMASTER Send address changes to Chemical Engineering
nearest you or from this journals website (http://www.elsevier.com/ locate/cep). and Processing, Publications Expediting Inc., 200 Meacham Ave, Elmont, NY
Further information is available on this journal and other Elsevier products 11003.
Editors: through Elseviers website: (http://www.elsevier.com). Subscriptions are AIRFREIGHT AND MAILING in the USA by Publications Expediting Inc., 200
Prof. Dr.-lng. A. Gorak accepted on a prepaid basis only and are entered on a calendar year basis. Meacham Avenue, Elmont 11003, NY 11003.
Universitat Dortmund Issues are sent by standard mail (surface within Europe, air delivery outside
Lehrstuhl fur Thermische Verfahrenstechnik Europe). Priority rates are available upon request. Claims for missing issues Orders, claims, and journal enquiries: please contact the Customer Service
Fachbereich Chemietechnik should be made within six months of the date of dispatch. Department at the Regional Sales Office nearest you:
Emil-Figge-Strasse 70
D-44221 Dortmund Author enquiries
Phone: 49 721 608 2391 Orlando: Elsevier, Customer Service Department, 6277 Sea Harbor Drive,
Fax: 49 721 608 3490 Orlando, FL 32887-4800, USA; phone: (+1) (877) 8397126 [toll free number
Submissions for US customers], or (+1) (407) 3454020 [customers outside US]; fax: (+1) (407)
e-mail: gorak.CEP@ct.uni-dortmund.de For information regarding the submission of manuscripts to Chemical 3631354; e-mail: usjcs@elsevier.com
Engineering and Processing, please refer to the Instructions to Authors in the
Prof. Dr.-Ing. A. Wild back of this issue.
Laboratoire des Science du Genie Chimique Amsterdam: Elsevier, Customer Service Department, PO Box 211, 1000 AE
Groupe ENSIC Amsterdam, The Netherlands; phone: (+31) (20) 4853757; fax: (+31) (20)
These instructions are also available on the World Wide Web: http://
1 rue Grandville 4853432; e-mail: nlinfo-f@elsevier.com
www.elsevier.com/locate/cep
BP 451
F-54001 Nancy For detailed instructions on the preparation of electronic artwork, consult the Tokyo: Elsevier, Customer Service Department, 4F Higashi-Azabu, 1-Chome
France Elseviers author gateway at http://authors.elsevier.com Bldg, 1-9-15 Higashi-Azabu, Minato-ku, Tokyo 106-0044, Japan; phone: (+81) (3)
Phone: 49 231 755 2323 5561 5037; fax: (+81) (3) 5561 5047; e-mail: jp.info@elsevier.com
Fax: 49 231 755 3035 Other enquiries
e-mail: Gabriel.Wild@ensic.inpl-nancy.fr Visit Elseviers Author Gateway (http://authors.elsevier.com) for the facility Singapore: Elsevier, Customer Service Department, 3 Killiney Road, #08-01
to track accepted articles and set up e-mail alerts to inform you of when an Winsland House I, Singapore 239519; phone: (+65) 63490222; fax: (+65)
Founding Editor: articles status has changed. The Author Gateway also provides detailed artwork 67331510; e-mail: asiainfo@elsevier.com
Prof. Em. Dr.-lng. Dr. h.c. mult. E.-U. Schlunder, Universitat Karlsruhe, Germany guidelines, copyright information, frequently asked questions and more.
Advertising Information: Advertising orders and enquiries can be sent to:
Honorary Editor: Contact details for questions arising after acceptance of an article, especially Elsevier, 84 Theobalds Rd, London, WC1X 8RR, tel: +44 (0) 20 7611 4117;
Dr. lng. V. Gnielinski, Universitat Karlsruhe, Germany those relating to proofs, are provided after registration of an article for publication. fax: +44 (0) 20 7611 4463; e-mail: k.barton@elsevier.com

USA mailing notice: Chemical Engineering and Processing (ISSN 0225- 2701) Reprints
is published monthly by Elsevier B.V. (P.O. Box 211, 1000 AE Amsterdam, The Reprint orders can be sent to currentreprints@elsevier.com, tel: +44 (0) 20 7611
Editorial Board: Netherlands). Annual subscription price in the USA US$1,965 (valid in North, 4494; fax: +44 (0) 20 7611 4463
P. Aptel, Universite Paul Sabatier, France
P.I. Barton, Massachusetts Institute of Technology, Cambridge, Massachusetts, USA
P. Bekiaroglou, University of Thessaloniki, Thessaloniki, Greece
P. Burghardt, Instytut Inzynierii Chemicznei l, Gliwice, Poland
K.J. Carpenter, Syngenta Global Specialist Technology, Huddersfield, UK
E. Drioli, Universita Della Calabria, Italy
J. Gyenis, Hungarian Academy of Sciences, Veszprem, Hungary
K. Jelemensky, Slovak Technical University, Bratislava, Slovak Republic
J.T.F. Keurentjes, Eindhoven University of Technology, The Netherlands
B. Kraushaar-Czarnetzki, University of Karlsruhe, Karlsruhe, Germany
F. Larachi, Universite Laval, Canada
R. Marr, Technische Universitat Graz, Graz, Austria
D. Mewes, University of Hanover, Hanover, Germany
H. Muller-Steinhagen, German Aerospace Centre, Stuttgart, Germany
K.D.P. Nigam, Indian Institute of Technology Delhi, New Delhi, India
A. Olcay, Ankara University, Ankara, Turkey
L. Puigjaner, Universitat Politencnica de Catulunya, Barcelona, Spain
H. Schoenmakers, BASF AG, ZAT-L540, Ludwigshafen, Germany
S. Skogestad, Norwegian University of Science and Technology, Trondheim, Norway
A. Stankiewicz, DSM Research, The Netherlands
K. Sundmacher, Universitaet Magdeburg, Germany
J. Werther, Technische Universitat of Hamburg-Harburg, Hamburg, Germany
K.R. Westerterp, Twente University of Technology, Enschede, The Netherlands

Aims and Scope

Chemical Engineering and Processing is a journal intended for practising engineers in industry and academia. The aim of the published articles is to demonstrate
how novel discoveries, developments and theories may be used for improved analysis and design of chemical engineering equipment and processes. The Journal
offers space for articles on any branch of chemical engineering and is particularly concerned with mechanical, thermal, chemical and bioengineering processes and
also includes analyses relating to environmental chemical engineering and sustainable industrial development. Advanced knowledge on engineering fundamentals
and processes is presented in such a form that it can be readily implemented into practical application. From time to time, review papers will be published for
individual fields of activity to illustrate and summarise scientific and/or technological progress.

*
1 The paper used in this publication meets the requirements of ANSI/NISO Z39.48-1992 (Permanence of Paper).

Printed in The Netherlands

| |
 Chemical Engineering and Processing 44 (2005) 139140
Editorial
Special issue on conveying and handling
of particulate solids
Dear reader,
Already almost 1 year has flown away since the Fourth
International Conference on Pneumatic Conveying and Han-
dling of Particulate Solids took place in Budapest, Hungary,
2730 May 2003. This was the very first event that took
place in Europe followed by previous meetings held in Israel
(Herzilya 1995, Jerusalem 1997, Dead Sea 2000). In
addition to the habitually pleasant and friendly atmosphere
of these conferences, this latter was even greater from pro-
fessional point of view. During this 4 full days meeting,
more than 200 presentations were delivered by excellent sci-
entists. Among them, 20 plenary or keynote lectures, 120
oral presentations in four parallel streams of sessions, as
well as 60 posters gave detailed information on the research
and industrial practice carried out and applied recently in the
field. In addition, the marvelous conference center of Ho-
tel Inter-Continental Budapest and its surroundings with the
historical district of the Hungarian capital, one of the most
beautiful part of the worlds heritage, offered more than suit-
able milieu for the meeting and satellite workshops.
The objective of the event was to bring together re-
searchers, practicing engineers, industrialist and students
from all over the world. The rich scientific and technical
program have attracted great attention: more than 220 par-
ticipants joined to the conference from 28 countries, most
of them from Germany, Israel, United Kingdom, Poland,
USA, Hungary, Australia, Japan, Norway and France.
Full text papers have been published in two thick vol-
umes of conference proceedings, having altogether more
than thousand pages. However, to ensure broader publicity,
the International Scientific Council (ISC) and the Interna-
tional Organizing Committee (IOC) have decided to select
the most suitable papers for publication in several interna-
tionally respected periodicals, too.
Chemical Engineering & Processing was among the jour-
nals best suited to the topics of the majority of papers sub-
mitted to the conference. Selection of papers was the task
of ISC and invited guest editors on basis of quality and con-
tent. All papers submitted for publication in this special is-
sue were peer-reviewed by two independent referees. After
reviewing and final revisions, 27 papers were accepted for
publication. These papers give representative sample on the
0255-2701/$ see front matter 2004 Elsevier B.V. All rights reserved.
doi:10.1016/j.cep.2004.05.001
conference topics, especially from respect to the scope of
this journal.
The first, memorial paper written by Prof. David Mills
gives honour to the oeuvre of Prof. Predrag Marjanovic, one
of the most distinguished personalities in particle technol-
ogy, late member of ISC, who left us tragically not long ago.
The next three papers by M.J. San Jos et al., I. Descamps
et al., as well as J. Li et al. deal with gassolids systems and
disperse particle flow. Gas flow pattern in spouted bed, mod-
eling of particle take-off in turbulent flow, as well as solids
deposition during pneumatic conveying were discussed, re-
spectively. During pneumatic conveying various transforma-
tions of particles can also take place, as is shown in two
next papers: L. Frye and W. Peukert describe specific attri-
tion mechanism in dilute phase process, while I. Skuratovsky
et al. show the results of a numerical simulation study on
drying process during pneumatic conveying.
Particle production is an important process of particle
technology. The paper of J. Tth et al. discusses the results
of investigation on salting-out precipitation of glycine and
sodium chloride. In another paper, production of micronized
cocoa butter particles by supercritical process was reported
by J.-J. Letourneau et al.
Relatively big number of contributions has been selected
for this special issue dealing with processing particulate
materials. J. Fitzpatrick and L. Ahrn gave comprehensive
summary on food powder handling and processing. Microen-
capsulation of heat sensitive fine particles was reported by
H. Krber and U. Teipel in a novel fluidized-bed process
using supercritical carbon dioxide as solvent for coating ma-
terial. Plasma spheroidization of ceramic particles was de-
scribed by Z. Karoly and J. Szpvlgyi, while thermal plasma
processing was applied by I. Mohai and J. Szpvlgyi for
treatment of particulate metallurgical wastes. M.J.R. Aguado
et al. studied conical spouted bed reactor for pyrolysis of
plastic wastes using sand particles as fluidizing packing to
avoid agglomeration.
Energy demand of granulation process was studied by T.
Gluba, described in the next paper. Elimination of particle
aggregation during nanomilling was realized and described
by W. Peukert et al. Four more papers are dealing with size
reduction processes: milling, grinding, attrition and particle
140 Editorial / Chemical Engineering and Processing 44 (2005) 139140
degradation (R.K. Sahoo and D. Roach, A. Heim et al., V.
Mizonov et al., and P.B. Kis et al.). Separation and classi-
fication are inevitable processes in particle technology. M.
Shapiro and V. Galperin gave a review on various techniques
of separation of particles. The paper of W. McBride and S.
Keys have applied vertical Venturi separator to improve re-
cycling of automotive tyres.
A worldwide revival of research activities nowadays on
mixing, blending, homogenization processes of particulate
solids is evidenced by three papers. A new pulsated bottom
mixer and its investigation were described by M. Nemenyi
and A.J. Kovacs. Process dynamics and homogenization ef-
ficiency of mammoth silos were studied by D.L. Schott et
al. Assessment of homogeneity in powder mixtures by elec-
trical capacitance measurement was proposed N. Ehrhardt
et al., as a good tool to examine segregation during particle
flow or the efficiency of a mixer.
As a special topic, solids transport in a pilot-scale rotary
kiln was studied and outlined in a paper of N. Descoins et al.
Finally, two papers give examples on environmental aspects
of this field. A nice overview on dust control and explosions
in solids handling plants are given P. Wypych et al. with
description of some own experimental results. Application
of fluidized bed process for photo-catalytic degradation of
harmful gas components are described by T.H. Lim and S.D.
Kim.
We hope that this special volume offers a good cross-
section on the professional work of the conference for the
readers, giving useful information on the results achieved
recently in the field, and will also stimulate the participation
at the next event to be held in 2006 in Rome.
April 2004.
Janos Gyenis
Research Institute of Chemical and Process Engineering
University of Veszprem, P.O. Box 158
Veszprem 8201, Hungary
Corresponding author. Tel.: +36 88 425 206
fax: +36 88 424 424
E-mail address: gyenis@mukki.richem.hu
(J. Gyenis)
Avi Levy
Ben-Gurion University of the Negev
Beer-Sheva, Israel
Available online 19 June 2004
 Chemical Engineering and Processing 44 (2005) 141151

A review of the research work of Professor Predrag Marjanovic

David Mills
Pneumatic Conveying Consultant, Old Wives Lees, Canterbury CT4 8BQ, UK

Received 6 August 2003; received in revised form 18 December 2003

Available online 20 June 2004

Abstract

Predrag Marjanovic was born on 1 May, 1951 in Belgrade, Yugoslavia, where he lived until November 1992. He moved with his wife
Mara and daughter Nevena to Glasgow, Scotland, where he continued to reside throughout his time in the UK. I was supervisor of his Ph.D.
programme and colleague in Glasgow for about 5 years. Predrags involvement in research covered a diverse range of topics, including
hydraulic transport, fluid mechanics, bulk material characterisation and hopper design. This review will mostly feature the work undertaken
for his Ph.D. and that associated with pneumatic conveying. A definitive list of Predrags references, however, is appended. Predrag died
suddenly on 14 September, 2001.
2004 Elsevier B.V. All rights reserved.

Keywords: Fluid mechanics; Hopper design; Vertical pipeline; Pneumatic conveying; Inclined pipeline; Pipeline bends; Rubber hose; Dense phase

1. Belgrade 2. Ph.D. programnme

Predrag obtained a 1st class Honours degree in Mechan-
ical Engineering in 1975 from the University of Belgrade.
He then joined the Fluid Mechanics Department of the Uni-
versity of Belgrade as a teaching assistant, where he was
employed for 18 years, rising to the position of Associate
Professor. During this time he obtained an M.Sc. with dis-
tinction in Process Engineering. His M.Sc. was on the Ex-
perimental Determination of GasSolids Friction Factors. In
these early years, he also published a number of papers on
hydraulic conveying with Professor M. Sasic.
I met Predrag in Belgrade in October 1979, being in-
troduced by Dr. Z. Bukerov from the University of Novi
Sad. Predrag expressed the desire to undertake a Ph.D. pro-
gramme on pneumatic conveying and particularly wanted
to investigate conveying vertically up in dense phase. There
was the opportunity of undertaking such a programme of
research at Thames Polytechnic (now the University of
Greenwich) in London, where I was a Senior Lecturer in
the Department of Mechanical Engineering, and Professor
Stan Mason was Head of Department. Predrag obtained
sabbatical leave from the University of Belgrade and came
to London early in 1982.
Tel.: +44 1227 732493; fax: +44 1227 732504.
E-mail address: drdavid.mills@btinternet.com (D. Mills).
A high-pressure (7 bar gauge) top discharge blow tank
having a 1 m3 capacity was made available and a pipeline
was built on the outside wall of the building, which allowed
a 17 m vertical rise. With a need to re-circulate materials for
test purposes it was necessary to run the pipeline both up
and down the wall to complete the circuit. It was decided,
therefore, to investigate pneumatic conveying performance
in both the vertically up and vertically down sections of the
pipeline as it involved little additional effort in building and
instrumenting the test facility for the added data. There was
also a lot of interest at the time in the pneumatic conveying
of various materials down vertical mine shafts. A sketch of
the test pipeline facility is provided in Fig. 1.
Two pipelines, in fact, were built, one of 2 in. nominal
(53 mm actual) bore and another of 3 in. nominal (81 mm ac-
tual) bore. This was to allow the influence of pipeline bore to
be additionally investigated. The two pipelines ran alongside
one another and so the pipeline routing and geometry was
almost identical for the two pipelines. The pipelines were
built on the rear wall of the building and this had a fire escape
access with a large platform. As a consequence, because of
the convenience, it was decided to install sight glasses in the
two pipelines in both the vertically up and vertically down
sections. I had full video facilities available from the UK
Department of Trade and Industry for the research work.

0255-2701/$ see front matter 2004 Elsevier B.V. All rights reserved.
doi:10.1016/j.cep.2004.05.002
142 D. Mills / Chemical Engineering and Processing 44 (2005) 141151

Fig. 1. Test facility and pipeline employed.

In order to derive data for the individual sections of

pipeline in isolation from the total pipeline, all four sections Fig. 2. Pressure gradient results for a fine grade of pulverised fuel ash.
of vertical pipeline were provided with pressure tappings
along their entire length. In the vertically down sections of The discharge limit of the blow tank used was about
pipeline, there were seven sets of pressure tappings and in 26 tonne/h with the 53 mm bore pipeline and approximately
the vertically up sections, there were eight sets. A ring of 50 tonne/h with the 81 mm bore pipeline. Within this capa-
four pressure tappings was provided at every location and bility of the conveying facility, however, tests were carried
these were inter-connected. Every pressure tapping was fit- out with conveying line pressure drop values of well over
ted with a filter pad and provided with a high-pressure air two bar and the materials were all capable of being con-
purging facility, which was routinely operated after each veyed at solids loading ratios of well over one hundred. A
and every test run. Bearing in mind that this was 1982 and minimum of fifty individual tests were undertaken with ev-
that electronic pressure transducers were both in their in- ery material/pipeline bore combination, in order to draw the
fancy and very expensive, mercury manometers were used various families of curves required.
for all vertical pipeline pressure measurements. There was no lower limit on pressure drop, material flow
Two typical sets of pressure measurement data for the rate or solids loading ratio that the test facility could operate,
vertically down and vertically up sections of pipeline are
presented in Fig. 2. This shows the location of the pressure
tappings and their proximity to the various bends in the
pipeline. The data relate to the pneumatic conveying of a fine
grade of pulverised fuel ash. Five different bulk particulate
materials were investigated in the research programme, the
other four being barytes, bentonite, cement and fluorspar.
All five materials were capable of being conveyed in dense
phase and hence at low velocity.
Typical conveying data obtained for the total pipeline sys-
tems are presented in Figs. 36.
Several compressors, each capable of delivering 200 ft3 /
min (0.095 m3 /s) of free air at a pressure of 100 psig (7 bar
gauge) were available, together with a desiccant-type air
drier. For the 53 mm bore pipeline, one compressor was
used and for the 81 mm bore pipeline two were used.
Convergentdivergent choke flow nozzles were used to mea-
sure and control the flow rate of air used for conveying, as
well as to control the discharge rate of material from the blow
tank. The receiving hopper was mounted on load cells and
these were used to measure the conveyed material flow rate. Fig. 3. Conveying characteristics for bentonite in 53 mm bore pipeline.
 D. Mills / Chemical Engineering and Processing 44 (2005) 141151
Fig. 4. Conveying characteristics for fluorspar in 53 mm bore pipeline.
but Predrag chose to investigate dense phase conveying for
his Ph.D. study and he just did not have the time available
to extend the work down into the dilute phase range of con-
veying. As a consequence very few tests were undertaken
with a conveying line pressure drop below 08 bar.
3. Pressure gradient data
Data was recorded from the manometer banks from each
test and this was analysed in terms of a pressure gradient
in mbar/m. The resulting data was also plotted in the form
of conveying characteristics, and representative sets of data
obtained are presented in Figs. 710. In Figs. 7 and 8, data
obtained for vertically upward flow are shown, with cement
in the 53 mm bore pipeline in Fig. 7 and barytes in the 81 mm
Fig. 5. Conveying characteristics for cement in 81 mm bore pipeline.
143
Fig. 6. Conveying characteristics for pulverised fuel ash in 81 mm bore
pipeline.
bore pipeline in Fig. 8. As will be seen, almost the entire
dense phase conveying capability of the materials has been
covered in the programme of work.
Conveying data presented in this form clearly show
the capability of pneumatic conveying systems and the
inter-relating effects of pressure, material concentration
and pipeline bore, as well as air flow rate when designing
a system to convey a material at a given flow rate, over a
specified distance. Since there is generally a limit on air
supply pressure, a compromise has to be made between
solids loading ratio and pipeline bore.
Figs. 9 and 10 show data obtained for vertically downward
flow in the 53 mm bore pipeline, with cement in Fig. 9 and
barytes in Fig. 10. From these two sets of data, it will be
Fig. 7. Pressure gradient for cement in 53 mm bore vertically up pipeline.
144 D. Mills / Chemical Engineering and Processing 44 (2005) 141151
Fig. 8. Pressure gradient for barytes in 81 mm bore vertically up pipeline.
seen that when the materials are conveyed at a solids loading
ratio of about 35, there is no pressure drop associated with
the conveying of the materials. Results with the other three
materials also showed the same pattern. At higher solids
loading ratios, the pressure gradient is negative and hence
there is a pressure recovery in downward flow. At lower
values of solids loading ratio, however, the pressure gradient
is positive, and hence by inference over the entire dilute
phase range of conveying, there will be a pressure drop.
4. Subsequent collaboration
Following Predrags return to the University of Belgrade,
and the award of his Ph.D. in 1984, Predrag maintained
Fig. 9. Pressure gradient for barytes in 53 mm bore vertically down
pipeline.
Fig. 10. Pressure gradient for pulverised fuel ash in 81 mm bore vertically
down pipeline.
his collaboration with myself and published numerous pa-
pers on the many aspects of his Ph.D. research programme,
including dense phase conveying, pipeline bore, material
influences and blow tank performance. The collaboration
was continued with Predrag being involved in my contin-
uing research work, and this included papers on pipeline
orientation, pipeline material and the influence of bends in
pneumatic conveying pipelines. The nature of the work and
the range of topics will be seen from the list of date ordered
references and publications at the end of this paper.
5. Scaling parameters
With my work on horizontal conveying, and Predrags
work on vertical conveying, using many of the same materi-
als, it was a natural development of our joint work to com-
pare data and derive scaling parameters for vertical pipelines
in terms of horizontal pipelines. In Fig. 11, data are pre-
sented on pressure gradients for the conveying of barytes in
a horizontal pipeline of 53 mm bore. Predrag had obtained
identical data for conveying vertically up and so to provide
a comparison of the two sets of results a rectangular grid
was placed on both sets of curves and the value of the ra-
tio of the two pressure gradients was noted. Fig. 12 shows
the value of the ratios of the vertical line pressure gradient
divided by the horizontal line pressure gradient, determined
for every grid point.
It will be seen that the relationship obtained covers a very
wide range of conveying conditions, although it is unfortu-
nate, in this case, that the data could not be extended to cover
dilute phase conveying also. This exercise was also carried
out with a number of other materials and very similar results
were obtained. In all cases, the ratio between vertical and
horizontal pressure gradients was in a very narrow band, at
 D. Mills / Chemical Engineering and Processing 44 (2005) 141151 145

Fig. 11. Pressure gradient for barytes in 53 mm bore horizontal pipeline.

Fig. 13. Equivalent length of bends.

Data obtained with cement and analysed in terms of an

equivalent length of straight horizontal pipeline are pre-
sented in Fig. 13. This is for 90 bends having a bend diam-
eter D, to pipe bore d, ratio of 24:1 in 53 mm bore pipeline
in horizontal plane. Almost identical results were obtained
when a similar analysis was carried out for the conveying of
barytes. A simple correlation in terms of the conveying line
inlet air velocity was not expected, but it was not possible
to determine any effect of the position of the bends in the
pipeline. Data obtained with barytes and analysed in terms
of a pressure drop across a bend are presented in Fig. 14.
Once again, almost identical results were obtained for the
conveying of cement. These data are for the same bends re-
ported in Fig. 13.
Fig. 12. Ratio of vertical to horizontal pressure gradient for barytes in
53 mm bore pipeline.

a mean value of about 2, over the entire range of conveying

conditions considered.

6. Pipeline bends

In my work for the Department of Trade and Industry in

the UK, I had generated a considerable amount of data with
a number of different materials conveyed through pipelines
of approximately 100 m in length but having differing num-
ber of bends. Predrag worked with me to analyse the data
and isolate the influence of the straight pipeline so that the
effects of the bends could be determined. Recognising that
engineers design pneumatic conveying systems in different
ways, the data were analysed in order to provide values both
in terms of an equivalent length and in terms of an actual
pressure drop value. Fig. 14. Pressure loss for bends.
146 D. Mills / Chemical Engineering and Processing 44 (2005) 141151

7. Pipeline material

Another area of interest at this time was the transport of

cement and drilling mud powders, with a particular require-
ment for the transferring of these materials onto off-shore
drilling platforms. This involves the use of flexible hose for
a significant proportion of the pneumatic conveying pipeline
for the transport of these materials. There was, therefore, a
need to know how pressure drop through rubber hose com-
pared with the pressure drop through steel pipeline. Oil well
cement and barytes were the materials used for this partic-
ular investigation. 40 m of rubber hose of 53 mm bore was
available and so a 40 m long pipeline was built of 53 mm
bore steel pipeline and a full set of conveying characteris-
tics were obtained for each material. The rubber hose was
then strapped to the steel pipeline to create a pipeline hav-
ing exactly the same routing and number of bends and bend
geometries. Full sets of conveying characteristics were then
obtained for the two materials conveyed through this rubber
hose pipeline.
Fig. 16. Comparison of pressure drop data for steel and rubber hose lines
The conveying characteristics for the oil well cement con- for cement.
veyed through the steel pipeline are presented in Fig. 15.
Once again as wide a range of conveying conditions were
investigated as possible. Similar sets of conveying character- the percentage differences between the two pipeline mate-
istics were obtained for the barytes in the steel pipeline and rials follow a regular pattern. With increase in air flow rate
for both materials in the rubber hose pipeline. The rubber there is a significant increase in the pressure drop required
hose was rated to a 10 bar capability. for the rubber hose line. The lines drawn through the data
A rectangular grid was constructed on each set of convey- also approximate to lines of mean value of conveying air
ing characteristics, as in the above analysis, and the ratio of velocity through the pipeline.
pressure drops was evaluated at corresponding grid points It is believed that this can be related to the differences in
of material and air flow rates. In Fig. 16, a comparison of coefficient of restitution between the conveyed particles and
the pressure drop required to convey the cement through the pipeline walls. On impact with the rubber, the particles
the rubber hose, compared with that to convey the cement will be decelerated, since the rubber will absorb much of
through the steel pipeline, is presented. It will be seen that the energy of impact. As a consequence, the particles will
have to be re-accelerated back to their terminal velocity. The
coefficient of restitution of the particles against the steel
pipeline wall will be very much lower. This effect is clearly
magnified by increase in velocity and explains why there is
little difference between the two pipeline materials in low
velocity dense phase conveying, but differ by 50% in high
velocity dilute phase conveying. The results obtained with
the barite were very similar.

8. Glasgow

In 1988, I and a number of colleagues moved to Glasgow

Fig. 15. Conveying characteristics for cement in 40 m long steel pipeline.
College (now Glasgow Caledonian University) to join Pro-
fessor Stan Mason who had moved there the previous year
and was later appointed Principal and Vice Chancellor of
the University. A major teaching aim was to provide a post-
graduate Masters Degree course in Bulk Solids Handling
in Glasgow. To help in the preparation and development of
this new and innovative venture, and early teaching on the
course, Pregrad obtained sabbatical leave for a year from the
University of Belgrade in 1990/91.
 D. Mills / Chemical Engineering and Processing 44 (2005) 141151
Predrag joined Glasgow Caledonian University in 1992
as Senior Research Fellow and was appointed Professor in
1999. In Glasgow, Predrags research work expanded con-
siderably, with the supervision of Ph.D. programmes, con-
tract research and in the undertaking of consultancy work.
In 1998, for instance, his CV lists 13 publications. His
work in this period included major programmes on inclined
pipelines, bend losses, very high-pressure conveying sys-
tems, hopper design, on-site measurements, and material
degradation and segregation. I have concentrated on the first
three of these topics as they represent the largest proportion
of Predrags published output.
9. Inclined pipelines
The performance of inclined pipelines has long been a
topic of mystery and contention and so Predrag set up an
experimental facility in the laboratory at Glasgow Caledo-
nian University in order to study the problem. A 100 m long
pipeline of 81 mm bore was built, having a section in the
central area which was 8 m long and that could conveniently
be adjusted to provide inclinations ranging from 20 to
+90 . Material was fed by means of a rotary valve, and a
Roots-type blower capable of delivering up to 033 m3 /s of
free air was used, giving superficial air velocities from 10
to 20 m/s. A series of seven papers over a period from 1994
to 1997 were written specifically on this subject, with var-
ious distinguished co-authors, as will be seen from the list
of publications.
A major feature of the work was the analytical modelling
of the inclined pipeline situation using computational fluid
dynamics, and comparing the numerical predictions with the
experimental results. Test results obtained with 3 mm poly-
mer pellets are presented in Figs. 17 and 18. In Fig. 17, the
results are presented in terms of the difference between the
Fig. 17. Variation of pressure gradient difference with angle of inclination.
147
Fig. 18. Variation of normalised pressure gradient with angle of inclination.
total pressure gradient for an inclined pipe and that for a
horizontal pipe, under the same flow conditions, as a func-
tion of the angle of inclination. Data are presented for solids
loading ratios of 5, 10 and 20. It should be noted that the
pressure gradient for horizontal flow is also dependent upon
the solids loading ratio and, in addition, increases continu-
ously as the solids loading ratio increases.
In Fig. 18, the results are presented in terms of a nor-
malised pressure gradient, which is the ratio of the pressure
gradient for a particular angle of inclination divided by that
for the horizontal. Once again, solids loading ratios of 5, 10
and 20 were considered.
10. Pipeline bends
Pipeline bends are another area of particular interest
to pneumatic conveying engineers on which Predrag has
worked. A series of six papers over a period from 1995
to 2000 were written specifically on this subject, working
mostly with the same group of co-authors as above. For
the majority of the work, very little new experimental work
was undertaken, for once again the main thrust of the work
was on modelling, and for this purpose two sets of existing
data were used.
One of the sets of data used was that for cement, obtained
by the author in a 50 m long pipeline of 53 mm bore that
contained 99 bends having a D/d ratio of 24:1. The other
set of data was for polyethylene pellets conveyed through a
50 m long pipeline of 80 mm bore that contained six ninety
degree bends having a D/d ratio of 85:1 and obtained later
from the laboratory in Glasgow. The modelling employed a
solids friction coefficient and the authors attempted to model
the experimental data.
The model was capable of separating the constituent ele-
ments of the total pipeline pressure drop and so the authors
148 D. Mills / Chemical Engineering and Processing 44 (2005) 141151

Fig. 19. Cement in 53 mm bore pipeline with nine bends.

Fig. 21. High-pressure conveying facility and pipeline.

high-pressure blow tank and into a reception vessel main-

Fig. 20. Polyethylene in 80 mm bore pipeline with six bends.
were able to demonstrate the proportion of the total pres-
sure drop that could be attributed to the bends. The results
for the cement are presented in Fig. 19 and those for the
polyethylene pellets in Fig. 20. Both materials were capable
of being conveyed in dilute as well as dense phase, although
at low velocity, the cement was conveyed in a sliding bed
mode of flow and the pellets in plug flow. Influence of bends
on pressure drop required to convey materials in 50 m long
pipelines is shown in Figs. 19 and 20.
tained at pressures of up to 20 bar gauge. The blow tank
feeding the conveying system was rated at 25 bar and the
system was tested to 32 bar. The filter unit was located in-
side the pressurised reception vessel.
In the work undertaken by Predrag, crushed coal was
conveyed, and so nitrogen was used as the conveying gas.
Modelling of system performance was a major part of the
programme once again. The conveying pipeline employed
was 78 m long, of 49 mm bore and included eight 90 bends.
A sketch of the test facility and pipeline is given in Fig. 21.
A number of sets of test results obtained for different back
pressures and line pressure drops are presented in Fig. 22.
Minimum conveying air velocity was also determined dur-
ing the test programme. It was found that this velocity is in-
fluenced by system pressure, with its value decreasing from
10.5 m/s at atmospheric pressure to 6 m/s at a pressure of
20 bar gauge. This effect was expected due to a significant
increase of gas density at higher pressures.

11. High-pressure conveying

Very high-pressure conveying is a subject of particular in-

terest of mine, and this was another topic on which Predrag
worked. A series of five papers over a period from 1995 to
1997 were written specifically on this subject. The test fa-
cility used was designed and built in the laboratory at Glas-
gow Caledonian University under my supervision, with test
work carried out in which material was conveyed from the Fig. 22. Conveying characteristics for coal at different back pressures.
 D. Mills / Chemical Engineering and Processing 44 (2005) 141151
12. What could have come
We are naturally left wondering what might have been
achieved with so many more years that should have been his,
to spend on the loves of his life: his family and his research
work, and not forgetting the golf that he discovered in his
new home in Scotland. Thank you for being such a valued
colleague and friend. You will be long remembered by us
all.
Further reading
References and publications
[1] P. Marjanovic, M. Sasic, Hydraulic transport of ash in thermo-
power stations, Proceedings of the 4th Yugoslav Thermal Symposium,
Belgrade, Yugoslavia, 1975 (in Serbian).
[2] M. Sasic, P. Marjanovic, On the methods for calculation of hydraulic
transport and their reliability in practice. Part 1, Proceedings of the
5th International Conference on Hydraulic Transport of Solids in
Pipes, vol. 1, Hanover, Germany, 1978, pp. A5:61A5:76.
[3] M. Sasic, P. Marjanovic, On the methods for calculation of hy-
draulic transport and their reliability in practice. Part 2, 5th Interna-
tional Conference on Hydraulic Transport of Solids in Pipes, vol. 2,
Hanover, Germany, 1978, pp. X2X3.
[4] P. Marjanovic, Experimental determination of gassolids friction fac-
tor, M.Phil. Thesis, University of Belgrade, Yugoslavia, 1978 (in
Serbian).
[5] M. Sasic, P. Marjanovic, Hydromechanics of gassolid systems, Jnl
Tehnika-Masinstvo, no. 6, Belgrade, Yugoslavia, 1979, pp. 19 (in
Serbian).
[6] P. Marjanovic, J. Jovanovic, R. Askovic, Sur lecoulement dun
liquide conducteur dans la region dentree un tube circulaire en
presence dun champ magnetique. Publ. of Mathematical Institute
SANU, new series, vol. 3, no. 11, Belgrade, Yugoslavia, pp. 7580.
[7] M. Sasic, P. Marjanovic. One more general approach for investigation
of hydro-mechanics of two-component systems, J. Powder Bulk
Solids Tech. 5 (1) (1981) 1319.
[8] M. Sasic, P. Marjanovic, Non-isothermal compressible flow in pipes,
ZAMM 62 (1982) 226228.
[9] P. Marjanovic, A comparative study of performance characteristics for
horizontal and vertical pneumatic conveying in pipelines, Proceedings
of the Pneumatech 1. PAC Conference Stratford-upon-Avon, UK,
1982.
[10] D. Mills, J.S. Mason, P. Marjanovic, Pneumatic conveying vertically
down. Proceedings of the 8th Powder and Bulk Solids Conference,
Atlanta, USA, May 1983, pp. 546557.
[11] D. Mills, J.S. Mason, P. Marjanovic, An analysis of the dense phase
pneumatic conveying of cement in vertical pipelines. Proceedings of
the 9th Powder and Bulk Solids Conference, Chicago, USA, May
1984, pp. 125147.
[12] D. Mills, J.S. Mason, P. Marjanovic, The influence of pipeline
bore on dense phase pneumatic conveying in vertical pipelines.
Proceedings of the 16th Yugoslav Congress on Theoretical and
Applied Mechanics, Becici, Yugoslavia, May/June 1984, pp. 261
269.
[13] P. Marjanovic, An investigation of the behaviour of gassolids mix-
ture flow properties for vertical conveying in pipelines, Ph.D. Thesis,
CNAA, Thames Polytechnic, London, UK, 1984.
[14] D. Mills, J.S. Mason, P. Marjanovic, The influence of product type
on dense phase pneumatic conveying in vertical pipelines, Proceed-
ings of the Pneumatech 2, Powder Advisory Centre Conference,
Canterbury, UK, September 1984, pp. 193210.
149
[15] D. Mills, J.S. Mason, P. Marjanovic, A comparison of pressure drops
in horizontal and vertical dense phase pneumatic conveying, Proceed-
ings of the 3rd International Conference on Pneumatic Conveying,
Pecs, Hungary, March 1985.
[16] D. Mills, P. Marjanovic, J.S. Mason, An analysis of line pressure
gradient for dense phase vertical gassolids flow, Proceedings of the
GAMM Congress, Dubrovnik, Yugoslavia, April 1985.
[17] P. Marjanovic, D. Mills, J.S. Mason, An investigation of high con-
centration vertical pneumatic conveying, J. Procesna Tehnika, vol. 1,
no. 2, Belgrade, Yugoslavia, December 1985, pp. 3949 (in Serbian).
[18] P. Marjanovic, On the choking velocity for vertical pneumatic con-
veying, Proceedings of the 17th Yugoslav Congress of Theoretical
and Applied Mechanics, Zadar, Yugoslavia, 1986 (in Serbian).
[19] P. Marjanovic, The design of vertical pneumatic conveying using
high-pressure blow tank system, Proceedings of the International
Colloquium on Modern Trends in the Development of Pneumatic
Conveying, Sarajevo, Yugoslavia, June 1986, pp. 4564.
[20] P. Marjanovic, D. Mills, J.S. Mason, A method of calculating
flow parameters for pneumatic conveying in pipelines, Proceedings
of the Pneumatech 3, PAC Conference, Jersey, UK, March 1987,
pp. 143167.
[21] P. Marjanovic, I. Vuskovic, L. Bodiroga, Hydraulic Transport of
granular activated carbon in water refinery plant, Symposium on
Water Supply, Belgrade, Yugoslavia, 1987 (in Serbian).
[22] P. Marjanovic. Bends in gassolids mixture flow in pipes a view to
the prediction of pressure loss, GAMM Congress, Vienna, Austria,
1988.
[23] P. Marjanovic, Analysis of the change of flow parameters in dis-
continuous pneumatic conveying systems, Proceedings of the 18th
Yugoslav Congress of Theoretical and Applied Mechanics, Vrnjacka
Banja, Yugoslavia, 1988, pp. 8992 (in Serbian).
[24] P. Marjanovic, Determination of performance characteristics of dilute
phase pneumatic conveying system, Proceedings of the 3rd Sympo-
sium of Process Industry Applications, Belgrade, Yugoslavia, 1988,
pp. 215223 (in Serbian).
[25] P. Marjanovic, A comparison between pneumatic and hydraulic con-
veying, Proceedings of the 19th Yugoslav Congress of Theoretical
and Applied Mechanics, Ohrid, Yugoslavia, 1990, pp. 6166 (in Ser-
bian).
[26] P. Marjanovic, D. Mills, J.S. Mason, The influence of bends on the
performance of a pneumatic conveying system, Proceedings of the
15th Powder and Bulk Solids Conference, Chicago, USA, June 1990,
pp. 391399.
[27] P. Marjanovic, D. Mills, J.S. Mason, The influence of pipeline ma-
terial on the performance of pneumatic conveying systems, Proceed-
ings of the Pneumatech 4, Glasgow, UK, June 1990, pp. 453464.
[28] N. Barbalic, P. Marjanovic, E. Dzaferovic, Z. Mujcinovic, A. Cam-
para, Measurement of P.F. ash pneumatic conveying system parame-
ters in thermo-power station Ugljevik-1, J. Procesna Tehnika, vol.
7, no. 1, Belgrade, Yugoslavia, 1991, pp. 2326 (in Serbian).
[29] P. Marjanovic, M. Stanojevic, B. Todorovic, A. Vlajcic, The influ-
ence of variable operating conditions on the design and exploitation
of air-slide systems in thermo-power stations, Proceedings of the
RelPowFlo 2, EFChE Publication Series no. 96, Oslo, Norway, 1993,
pp. 659671.
[30] P. Marjanovic, Modelling the transient behaviour of blow tank pneu-
matic conveying system. Powder handling and processing, Trans.
Tech. Pub. 5 (3) (1993) 219226.
[31] P. Marjanovic, Pneumatic conveying in an inclined pipeline: inappro-
priate, unfortunate or wrong concept? Proceedings of the 19th Powder
& Bulk Solids Conference, Chicago, USA, May 1994, pp. 235251.
[32] P. Marjanovic, D.J. Mason, Gassolid flows in an inclined pipeline,
Proceedings of the 1st International Particle Technology Forum, vol.
3, Denver, USA, August 1994, pp. 466471.
[33] P. Marjanovic, Assessing the flow properties of powdered materials,
IMechE Seminar on Hoppers and Silos, London, UK, 1994.
150 D. Mills / Chemical Engineering and Processing 44 (2005) 141151
[34] P. Marjanovic, V. Djordjevic, On the compressible flow losses through
abrupt enlargements and contractions, J. Fluids Eng. Trans ASME
116 (1994) 756762.
[35] P. Marjanovic, D. Geldart, J.L.R. Orband, T. Mooney, A comparative
analysis of two hopper design methods, Proceedings of the Interna-
tional Congress for Particle Technology PARTEC 95, Nrnberg,
Germany, March 1995, pp. 6978.
[36] B. Armstrong, M.G. Jones, P. Marjanovic, G. Welford, P.J. Blenkin,
R.G. Holder, An evaluation of the effects of high system pressure
on the performance of pneumatic coal conveyors, Proceedings of the
20th International Conference on Coal Utilization & Fuel Systems,
Clearwater, USA, 1995, pp. 393403.
[37] T. Mooney, P. Marjanovic, Bend pressure loss in a pneumatic con-
veying system, Proceedings of the 20th Powder & Bulk Solids Con-
ference, Chicago, USA, May 1995, pp. 6173.
[38] P. Marjanovic, G. Welford, M.G. Jones, Pneumatic conveying of coal
against high back pressures, Proceedings of the 12th International
Pittsburgh Coal Conference, Pittsburgh, USA, September 1995.
[39] M.G. Jones, P. Marjanovic, G. Welford, The prediction of pneumatic
conveying performance when conveying into back pressures up to
20 bar, Proceedings of the 1st International Conference BULK ASIA,
Singapore, 1995, pp. 107115.
[40] R.J. Hitt, P. Marjanovic, A computer technique to develop a model
of vertical dense phase gassolids flow from experimental data.
Proceedings of the 1st International Symposium on Two-Phase Flow
Modelling and Experimentation, vol. 1, Rome Italy, October 1995,
pp. 6774.
[41] P. Marjanovic, M.G. Jones, Assessment of a new technique for
measuring flow properties of powdered materials stored in hoppers,
IMechE J. Process. Mech. Eng. 210 (1996) 18.
[42] P. Marjanovic, Theoretical and practical silo design, Seminar on
Storage and Transportation of Bulk Solids, Moreton-in-Marsh, UK,
March 1996.
[43] M.G. Jones, P. Marjanovic, Maintaining product quality in pneumatic
conveying, Proceedings of the Conveyorex 96 Seminar, Harrogate,
UK, April 1996.
[44] P. Marjanovic, G. Welford, M.G. Jones, Pneumatic conveying of
solids at high system pressures, Proceedings of the 21st Powder &
Bulk Solids Conference, Chicago, USA, May 1996, pp. 9198.
[45] M.G. Jones, P. Marjanovic, The modification of material properties
for improved flow characteristics, Proceedings of the Seminar on
Solving Problems in Hopper and Silo Systems, IMechE, London,
UK, June 1996.
[46] P. Marjanovic, D.J. Mason, T. Mooney, The Performance of a pneu-
matic conveying system which incorporates an inclined pipeline sec-
tion, Proceedings of the 1st International Conference on Pneumatic
and Hydraulic Conveying Systems, Florida, USA, 1996.
[47] D.J. Mason, P. Marjanovic, A. Levy, A simulation system for pneu-
matic conveying systems, Proceedings of the 1st International Con-
ference on Pneumatic and Hydraulic Conveying Systems, UEF Con-
ference, Florida, USA, 1996.
[48] P. Marjanovic, D.J. Mason, The transient flow conditions during
feeding pneumatic conveying system using high-pressure blow tank,
Proceedings of the 12th International Congress of Chemical and
Process Engineering. Prague, Czech Republic, August 1996.
[49] D.J. Mason, A. Levy, P. Marjanovic, Modelling the influence of bends
on the flow of gas-solids mixture through pipelines, Proceedings of
the 12th International Congress of Chemical and Process Engineering,
Prague, Czech Republic, August 1996.
[50] A. Levy, P. Marjanovic, D.J. Mason, A Comparison of analytical
and numerical models for gas-solid flow through straight pipe of
different inclinations with experimental data, Proceedings of the 12th
International Congress of Chemical and Process Engineering, Prague,
Czech Republic, August 1996.
[51] P. Marjanovic, Determination of bulk solids indices using the Johan-
son indicizer system, Proceedings of the Bulk 96 Design Seminar,
Manchester, UK, December 1996, pp. 233238.
[52] T. Mooney, A. Levy, P. Marjanovic, D.J. Mason, An investigation
of gassolids flow through inclined pipes, Proceedings of the 1997
Jubilee Research Event, vol. 1, Nottingham, UK, 1997, pp. 425428.
[53] P. Marjanovic, M.G. Jones, G. Welford, Pneumatic conveying of
solids at high system pressures, J. Powder/Bulk Solids Technol. 1 (1)
(1997) 37.
[54] P. Marjanovic, M. McGarvey, R.B. McKay, Development of lab-
oratory methodology to determine flow properties of organic pig-
ment powders, Proceedings of the 2nd Israel Conference for Con-
veying and Handling of Particulate Solids, Jerusalem, Israel, 1997,
pp. 2.282.33.
[55] A. Levy, T. Mooney, P. Marjanovic, D.J. Mason, Analytical, Nu-
merical and experimental investigations for gassolid flow through
straight pipe of different inclinations, Proceedings of the 2nd Is-
rael Conference for Conveying and Handling of Particulate Solids,
Jerusalem, Israel, 1997, pp. 4.304.35.
[56] D.J. Mason, A. Levy, P. Marjanovic, The influence of bends on the
flow of gas-solids mixtures through pipelines, Proceedings of the
2nd Israel Conference for Conveying and Handling of Particulate
Solids, Jerusalem, Israel, 1997, pp. 4.364.41.
[57] P. Marjanovic, D.J. Mason, Gas solids flow in an inclined pipeline.
Powder handling and processing, Trans. Tech. Pub. 9 (3) (1997)
217220.
[58] A. Levy, T. Mooney, P. Marjanovic, D.J. Mason, A comparison of
analytical and numerical models with experimental data for gassolid
flow through a straight pipe at different inclinations, Powder Technol.
93 (1997) 253260.
[59] D.J. Mason, P. Marjanovic, A. Levy, A simulation system for pneu-
matic conveying systems, Powder Technol. 95 (1998) 714.
[60] D.J. Mason, A. Levy, P. Marjanovic, The influence of bends on the
performance of pneumatic conveying systems, Adv. Powder Technol.
9 (3) (1998) 197206.
[61] D.J. Mason, P. Marjanovic, Re-Visit of the fundamental definitions
of fluid-solids flow properties in freight pipelines, Proceedings of
the 9th International Symposium on Freight Pipelines, Monterrey,
Mexico, April 1998.
[62] P. Marjanovic, A. Levy, D.J. Mason, An investigation of the flow
structure through abrupt enlargement of circular pipe, Proceedings
of the 9th International Symposium on Freight Pipelines, Monterrey,
Mexico, April 1998.
[63] P. Marjanovic, M. McGarvey, R.B. McKay, Development of labora-
tory methodology to determine flow properties of organic pigment
powders. Powder handling and processing, Trans. Tech. Pub. 10 (2)
(1998) 151154.
[64] M.G. Jones, P. Marjanovic, The influence of rotary valve air leakage
on pneumatic conveying system performance, Proceedings of the
23rd Powder & Bulk Solids Conference, Chicago, USA, May 1998,
pp. 231239.
[65] M.G. Jones, P. Marjanovic, The influence of rotary valve air leakage
on pneumatic conveying system performance, J. Powder/Bulk Solids
Tech. 2 (2) (1998) 310.
[66] M.G. Jones, P. Marjanovic, The optimisation of an existing pneumatic
conveying system, J. Powder/Bulk Solids Technol. 2 (2) (1998) 12
16.
[67] P. Marjanovic, D. Geldart, J.L.R. Orband, Techniques for assess-
ing powder flowability a comparison, Proceedings of the World
Congress on Particle Technology 3, Brighton, UK, July 1998.
[68] D.J. Mason, A. Levy, P. Marjanovic, The influence of bends on
the flow of gas-solids mixture trough pipelines, Proceedings of the
World Congress on Particle Techn 3, Brighton, UK, July 1998.
[69] M.G. Jones, P. Marjanovic, D. McGlinchey, R. McLaren, Segregation
in handling processes of blended industrial coal, Proceedings of the
6th International Conference on Bulk Material Storage, Handling
and Transportation, Wollongong, Australia, September 1998.
[70] P. Marjanovic, M. McGarvey, R.B. McKay, Development of labora-
tory methodology to determine flow properties of organic pigment
 D. Mills / Chemical Engineering and Processing 44 (2005) 141151
powders, Hopper & Silo Discharge: Successful Solutions. IMechE,
London, UK, November 1998.
[71] E.A. Knight, M.G. Jones, P. Marjanovic, On-site measurement for
troubleshooting pneumatic conveying systems, Proceedings of the
Bulk Design Seminar, Sutton Coldfield, UK, December 1998.
[72] D.J. Mason, P. Marjanovic, Re-visit to the fundamental definitions
of fluid-solids flow properties in conveying pipelines, Proceedings
of the 2nd International Conference on Pneumatic and Hydraulic
Conveying Systems. UEF Conf. Davos, Switzerland, June 1999.
[73] P. Marjanovic, E. McGee, Determination of powder flow properties
using different shear cells, Proceedings of the RelPowFlo 3, Pors-
grunn, Norway, August 1999, pp. 151158.
[74] M.G. Jones, P. Marjanovic, D. McGlinchey, D. Morrison, Investiga-
tion of the discharge pattern of industrial coals from wedge shaped
hoppers, Proceedings of the RelPowFlo 3, Porsgrunn, Norway, Au-
gust 1999, pp. 543550.
[75] P. Marjanovic, A. Levy, D.J. Mason, An investigation of the flow
structure through abrupt enlargement of circular pipe, Powder Tech-
nol. 104 (1999) 296303.
[76] P. Marjanovic, M. McGarvey, R.B. McKay, Determination of the
influence of surface coating and particle size on flow properties of
organic pigment powders, Proceedings of the 3rd Israel Conference
for Conveying and Handling of Particulate Solids, Dead Sea, Israel,
vol. 1, May 2000, pp. 3.393.44.
[77] M.G. Jones, P. Marjanovic, D. McGlinchey, An investigation of
degradation and segregation in typical coal handling processes, Pro-
ceedings of the 3rd Israel Conference for Conveying and Handling
of Particulate Solids, vol. 1, The Dead Sea, Israel, May 2000, pp
8.558.60.
[78] D.J. Mason, J. Li, P. Marjanovic, Numerical simulation of solids
feeding in a gas-solids pneumatic transport system, Proceedings of
the 3rd Israel Conference for Conveying and Handling of Particulate
Solids, vol. 2, The Dead Sea, Israel, May 2000, pp. 10.9310.101.
[79] D. McGlinchey, P. Marjanovic, M. G. Jones, S. Cook, Particle seg-
regation in pneumatic conveying lines, IMechE Conference Trans-
actions, June 2000, pp. 331340.
151
[80] M.G. Jones, A.G. Mason, P. Marjanovic, E. A. Knight, Bend Ef-
fects in Pneumatic conveying, IMechE Conference Transactions, June
2000, pp. 351361.
[81] R.O. Ansell, P. Marjanovic, Hazardous substances COSHH revis-
ited, Powder Rep. 3 (4) (2000) 1314.
[82] P. Marjanovic, E.A. Knight, J.R. Pugh, On-site measurement of
pneumatic conveying system performance: why, when, what and how,
Proceedings of the 16th International Conference on Material Flow,
Machines and Devices in Industry, Belgrade, Yugoslavia, December
2000, pp. 2025.
[83] J.S. Xiang, P. Marjanovic, Modelling and simulation for pyroflow
type circulating fluidised bed boiler, Proceedings of the Interna-
tional Congress for Particle Tech PARTEC 2001, Nrnberg, Ger-
many, March 2001.
[84] P. Marjanovic, Technology transfer between academia and industry
two-way street leading to mutual benefit and success, MHEA
Bulk 2001 Technical Awareness Seminar, Manchester, UK, April
2001.
[85] P. Marjanovic, E.A. Knight, J.R. Pugh, Benchmarking, optimising
and uprating of discontinuous pneumatic conveying system perfor-
mance through on-site measurement, Proceedings of the 6th World
Congress of Chemical Engineering, Melbourne, Australia, September
2001.
[86] J. Li, P. Marjanovic, J.S. Xiang, E.A. Knight, An experimental tech-
nique for the analysis of plugs travelling through pneumatic pipelines
using pressure measurements, Proceeding of the 7th International
Conference on Bulk Materials Storage, Handling and Transportation,
Newcastle, Australia, October 2001, pp. 5367.
[87] E. McGee, P. Marjanovic, L. Bates, Modifying flow behaviour in
hoppers using inserts and novel wall profiles, Proceedings of the 7th
International Conference on Bulk Materials Storage, Handling and
Transportation, Newcastle, Australia, October 2001.
[88] J.S. Xiang, P. Marjanovic, Hydrodynamic model of gas-solid flow
in circulating fluidised bed, Proceedings of the 7th International
Conference on Bulk Materials Storage, Handling and Transportation,
Newcastle, Australia, October 2001, pp. 825832.
 Chemical Engineering and Processing 44 (2005) 153157

Influence of the particle diameter and density in the

gas velocity in jet spouted beds
Mara J. San Jos , Sonia Alvarez, Alvaro Ortiz de Salazar, Martn Olazar, Javier Bilbao
Departamento de Ingeniera Qumica, Universidad del Pas Vasco, Facultad de Ciencias, Apartado 644, 48080 Bilbao, Spain

Received 30 September 2003; received in revised form 25 February 2004; accepted 19 March 2004
Available online 20 June 2004

Abstract

A gas flow pattern in the regime of jet spouted bed (or dilute spouted bed) has been analyzed in conical spouted beds using materials of
different density and particle diameter. The experimental results have been fitted to an empirical equation for calculation of the interstitial
velocity at the contactor axis in jet spouted beds. The general equation proposed relates this property to the geometric factors of the contactor
(angle, ; and inlet diameter, Do ) and to the operating variables (particle size, dp ; particle density, s ; particle shape, ; stagnant bed height,
Ho ; and air relative velocity referred to that minimum spouting, uo /uoms ).
2004 Elsevier B.V. All rights reserved.

Keywords: Spouted beds; Jet spouted beds; Gas flow

1. Introduction sions, by the characteristics of the inlet, by the particle

Markowski and Kaminski [1] identified the hydrodynamic
characteristics of the regime of jet spouting obtained by
expansion of the spouted bed in conical contactors: high
velocity; high average bed voidage, over 0.75 depending on
the operating conditions; systematic cyclic movement of the
particles; a hydrodynamic behaviour different to that of the
conventional spouted bed.
The jet spouted bed, Fig. 1, has a vigorous movement for
the solid and a uniform gas-solid contact, which is necessary
in operations with sticky solids and in highly exothermal
reactions. In the same way, it allows for working in a close to
uniform way with a wide particle size distribution. In short,
the hydrodynamic differences between the jet spouted bed
and the conventional conical contactors are:
Non-existence of hysteresis in the pressure drop versus
velocity curves.
Deviation of the experimental results of minimum velocity
for which the correlations deduced for the spouted bed
are not applicable.
Lesser instability problems, despite the fact the hydrody-
namic regime is greatly affected by the contactor dimen-
Corresponding author. Tel.: +34 94 6015362; fax: +34 94 6013500.
E-mail address: iqpsaalm@lg.ehu.es (M.J. San Jose).
diameter and by other operation and design parameters.
The design of the jet spouted bed requires the rigorous
definition of the gas flow pattern in order for the residence
time distribution to be considered. In previous papers, the
regime of jet spouted bed and its hydrodynamics correlations
have been defined [28]. The minimum jet spouting velocity
is calculated by the following correlation [7].
(Reo )mj = 6.891 Ar0.35 (Db /Do )1.46 [tan(/2)]0.53 (1)
The geometric factors and the design requirements for
stability have been delimited and the operational conditions
of the jet spouted bed have been compared with those of the
other gas-solid contact techniques [2,3].
The good behaviour of this regime has been proven in a
satisfactory way in the obtaining of thermally stable poly-
benzyls by polymerization of gaseous benzyl alcohol on
solid acid catalyst [9,10]. The jet spouted bed has also
been successfully used in coal gasification [1113]. The jet
spouted bed allows for working with short gas residence
times, by using a simpler and more economical device than
that of other innovative design reactors, such as the cyclonic
reactors [1416], spout-fluid bed reactors with draft tube
[17], spouted bed reactors with draft tube [18], or the im-
pinging stream reactors [19,20]. Consequently, they have a

0255-2701/$ see front matter 2004 Elsevier B.V. All rights reserved.
doi:10.1016/j.cep.2004.03.011
154 M.J. San Jose et al. / Chemical Engineering and Processing 44 (2005) 153157

Fig. 2. Experimental equipment.

Fig. 3. They have made to the following dimensions: column

Fig. 1. Solid circulation in jet spouting regime.
potential interest for fast reactions in which selectivity is the
factor for optimization.
A model for gas flow using glass beads has been pro-
posed in previous papers [21]. In order to describe the gas
flow in jet spouted bed regime in conical contactors. In this
paper, the vertical component of the gas interstitial velocity
has been calculated in jet spouted beds and the experimental
results have been fitted to an empirical equation for calcu-
lation of the interstitial velocity at the contactor axis in jet
spouted beds.
diameter Dc , 0.36 m; contactor base diameter Di , 0.62 m;
cone angles, , 33, 36, and 45 ; height of the conical section
Hc , 0.50, 0.45, and 0.36 m; gas inlet diameter Do , 0.03, 0.04,
0.05 and 0.06 m; stagnant bed height Ho , between 0.05 and
0.30 m.
The solids used correspond to group D and B of Geldart
classification [22,23] and they are set out in Table 1.
The velocity readings are carried out by four probes whose
radial and longitudinal position can be established at will
inside the contactor by means of a displacement device
controlled by computer, which established the coordinates

2. Experimental

The pilot plant scale experimental unit, Fig. 2, described

in previous papers [2,3] is provided with a blower that sup-
plies a maximum air flowrate of 300 Nm3 /h at a pressure of
1500 mm of water column. The flow measurement is carried
out with two rotameters, used in the ranges from 2.5 to 30
and from 30 to 250 Nm3 /h. In the range from 0 to 4 Nm3 /h a
gas volume meter provided with paddles is used. A system
of valves allows for choosing the suitable rotameter for the
desired flowrate. This flowrate is fixed by closing or opening
a butterfly valve.
Five contactors of poly(methyl methacrylate) have been
used, whose geometric characteristics are definided in the Fig. 3. Geometric factors of the contactors.
 M.J. San Jose et al. / Chemical Engineering and Processing 44 (2005) 153157 155

Table 1
Properties of the solids used
Material s dp o Geldart
(kg/m3 ) (mm) classification
Glass beads 2420 1.0 1 0.322 B
2420 2.0 1 0.328 D
2420 3.0 1 0.345 D
2420 3.5 1 0.351 D
2420 4.0 1 0.355 D
2420 5.0 1 0.358 D
Polypropylene 890 3.5 0.90 0.390 D
Extruded polystyrene 1030 1.0 0.80 0.395 B
1030 2.0 0.80 0.385 D
1030 3.5 0.80 0.380 D
Expanded polystyrene 65 3.5 0.95 0.507 D
Sawdust 242 1.0 0.90 0.324 B
242 1.5 0.90 0.328 D
242 2.0 0.90 0.330 D Fig. 4. Evolution with the dimensionless position in the bed, *, of the
vertical component of the gas interstitial velocity at the axis, uj (0), in jet
spouting regime. = 36 , Do = 0.04 m, Ho = 0.23 m for the different
materials of particle diameter dp = 3.5 mm.

of the point to be measured by each probe following a given

regression coefficient of r2 = 0.97 and the standard devia-
sequence.
tion of 3%.


2.78 102
uj (0) = 2.14 105 + H 8
o
3. Results Do2

0.17

dp K
u (3)
A wide experimental study of the vertical component of 1+K
the gas interstitial velocity in the contactor has been carried
out with the aim of obtaining a correlation that relates this
0.1
property to the geometric factors of the contactor (angle, , s
K = Ho0.81
and inlet diameter, Do ) and to the operating variables (parti- v
cle size, dp ; particle density, s ; particle shape, ; stagnant (0.85 0.22Do0.5 + 0.75dp0.5 + 0.21 0.5
bed height, Ho ; and air relative velocity referred to that min-
0.248 1.17 106 s )
imum spouting, uo /uoms ). (4)
The general equation proposed to calculate the vertical
component of the interstitial velocity at any position of the The expression proposed for the vertical component of
contactor is a Gaussian curve, in which the velocity in each velocity near the wall, uW , which has been chosen taking
position is a function of radius, r; of the maximum velocity into account simplicity and because it gives a satisfactory
at the axis of the contactor, uj (0) and of the velocity along fitting for all the experimental data is only function of the
the flow at the wall, uW : dimensionless spherical radius, :

uj = (uj (0) uW ) exp[0.5(r/)2 ] + uW (2) uW = 202 exp(1.42 ) (5)

From the vertical components of points velocity in the

The evolution of the maximum velocity, uj (0), with the contactor and by using the continuity equation, the radial
dimensionless longitudinal position at the axis, , has a component, ur , and consequently, the modulus and direction
similar tracing for all the systems, the velocity decreases as of velocity vector at each point in the contactor are calcu-
the longitudinal position at the axis increases, Fig. 4. These lated.
velocity profiles are more pronounced than those of conical The continuity equation applied to differential volume el-
spouted beds [24]. ement in cylindrical coordinates is:
The maximum velocity at the axis of the contactor, uj (0), 1 u
is a function of the geometric factors and operating condi- (rur ) + =0 (6)
tions. The experimental results have been fitted to an em- r r z
pirical equation for calculation of the interstitial velocity In Fig. 5, the map of velocities and of fluid direction is
at the contactor axis in jet spouted beds. The fitting has a shown as an example, which corresponds to the system =
156 M.J. San Jose et al. / Chemical Engineering and Processing 44 (2005) 153157

ing this decreasing more pronounced as the solid density in-

creases. Besides these velocity profiles are more pronounced
than those of conical spouted bed regime.
Empirical correlations proposed in this paper have been
proven to be valid for calculating the interstitial velocity
at any position of the contactor. These equations takes into
account, as well as the geometric factors of the contactor
the operating variables such as particle size, particle density
and particle shape.
The calculated average velocity in the spout at each level
are very similar to the velocity at the axis for beds of low
density solids. Nevertheless, for beds of solids of higher
density, the velocity profiles in the spout zone are more
pronounced.

Acknowledgements

This work was carried out with the financial sup-

port of the University of the Basque Country (Project
9/UPV00069.310-13607/2001).

Fig. 5. Map of gas interstitial velocity vector in jet spouting regime. Appendix A. Nomenclature
= 36 , Do = 0.04 m, Ho = 0.23 m and extruded polystyrene of particle
diameter dp = 3.5 mm.
Ar Archimedes number gdp3 g (s g )/2
As cross-sectional area of the spout (m2 )
36 , Do = 0.04 m, Ho = 0.23 m and extruded polystyrene dp particle diameter (m)
dp = 3.5 mm. The length of the arrows is proportional to Db , D c , D i , D o diameter of the upper level of the
the value of the corresponding velocity vector modulus. stagnant bed, of the column, of the bed
From the results it has been proven for all the experimen- base and of the inlet, respectively (m)
tal systems studied that the flow rate is constant along the g gravity constant (m s2 )
streamtubes defined in the annular zone, such as has been Hc , Ho height of the the conical section of the
proposed in the model. contactor, and of the stagnant bed,
In order to solve the model conservation equations, only respectively (m)
the knowledge of the average interstitial velocity in the r, z radial and longitudinal coordinates (m)
spout zone at each level, us , is required. This is calculated R contactor radius at z height (m)
as: (Reo )mj Reynolds modulus of minimum jet
 
1 1 Do /2 spouting referred to Do
us = u dAs = 2ur dr (7) u, uj (0), uj , uw vertical component of interstitial velocity,
As As 0
maximum velocity at the axis, at any
The values of us calculated using Eq. (7) are very similar position and at the contactor wall (m s1 )
(differ in less than 3%) to the values of velocity at the axis ur , us radial component of interstitial velocity
uj (0) for beds of sawdust and expanded polystyrene. Conse- and vertical component of the average
quently, the velocity profile in the spout zone is almost flat interstitial velocity in the spout at z level,
in beds of these solids. Nevertheless, for beds of glass beads respectively (m s1 )
and extruded polystyrene the velocity profiles in the spout u ratio between velocity and minimum
zone are more pronounced. spouting velocity, dimensionless

Greek letters
4. Conclusions o bed voidage of the static bed,
dimensionless
In jet spouted bed regime in conical spouted bed, the gas particle shape factor
interstitial velocity at the axis, uj (0), decreases as the dimen- contactor angle (deg)
sionless longitudinal position at the axis, , increases; be- viscosity (kg m1 s1 )
 M.J. San Jose et al. / Chemical Engineering and Processing 44 (2005) 153157
, N , o longitudinal spherical coordinates,
maximum and minimum longitudinal
spherical coordinates (m)
g , s , v density of the gas, of the solid and of
the glass beads, respectively (kg m3 )
dimensionless longitudinal spherical
coordinate, ( o )/(N o )
parameter of Eq. (2) (m).
References
[1] A. Markowski, W. Kaminski, Hydrodynamic characteristics of jet
spouted beds, Can. J. Chem. Eng. 61 (1983) 377381.
[2] M. Olazar, M.J. San Jos, A.T. Aguayo, J.M. Arandes, J. Bilbao,
Stable operation conditions for gassolid contact regimes in conical
spouted beds, Ind. Eng. Chem. Res. 31 (1992) 17841792.
[3] M. Olazar, M.J. San Jos, A.T. Aguayo, J.M. Arandes, J. Bilbao,
Design factors of conical spouted beds and jet spouted beds, Ind.
Eng. Chem. Res. 32 (1993) 12451250.
[4] M.J. San Jos, Operation regimes in conical spouted beds. Stability
conditions and hydrodynamics, Ph.D. Dissertation, University of the
Basque Country, Bilbao, Spain, 1991.
[5] M.J. San Jos, M. Olazar, A.T. Aguayo, J.M. Arandes, J. Bilbao,
Design and hydrodynamics of conical jet spouted beds, Rcents
Progrs en Gnie des Procds, La Fluidisation 5 (1991) 146153.
[6] M.J. San Jos, M. Olazar, A.T. Aguayo, J.M. Arandes, J. Bilbao, in:
D. Behrens (Ed.), Gassolid contact regimes in conical spouted beds,
Proceedings of the Fourth World Congress of Chemical Engineering,
Fluidization Session, Chemische Technik und Biotechnologie e.V.,
Frankfurt am Main, Germany, 1991, pp. 9.513.
[7] M.J. San Jos, M. Olazar, A.T. Aguayo, J.M. Arandes, J. Bilbao,
Hydrodynamics correlations of conical jet spouted beds, in: O.E.
Potter, D.J. Nicklin (Eds.), Fluidization VII, Engineering Foundation,
New York, 1992, pp. 831838.
[8] M.J. San Jos, M. Olazar, A.T. Aguayo, J.M. Arandes, J. Bilbao,
Expansion of spouted beds in conical contactors, Chem. Eng. J. 51
(1993) 4552.
[9] J. Bilbao, M. Olazar, A. Romero, J.M. Arandes, Design and operation
of a jet spouted bed reactor with continuous catalyst feed in the benzil
alcohol polymerization, Ind. Eng. Chem. Res. 26 (1987) 12971304.
157
[10] J. Bilbao, M. Olazar, A. Romero, J.M. Arandes, Optimization of
the operation in a reactor with continuous catalyst circulation in the
gaseous benzyl alcohol polymerization, Chem. Eng. Commun. 75
(1989) 121134.
[11] O. Uemaki, T. Tsuji, in: K. Ostergaard, A. Sorensen (Eds.), Flu-
idization V, Engineering Foundation, New York, 1986, p. 497.
[12] O. Uemaki, T. Tsuji, in: Proceedings of the 41st Canadian Chem-
ical Engineering Conference, Vancouver, BC, CSChE Publications
Department, Ottawa, Canada, 1991, pp. 117.
[13] T. Tsuji, T. Shibata, K. Yamaguchi, O. Uemaki, in: Proceedings of
International Conference on Coal Science, vol. I, 2327 October
1989, Tokyo, p. 457.
[14] J. Lede, H.Z. Li, J. Villermaux, Le Ciclon reactor. Partie I: measure
directe de la distribution des temps de sjour de la phase Gaseuze-
Lois dExtrapolation, Chem. Eng. J. 42 (1989a) 3755.
[15] J. Lede, K.B. Mathur, J. Villermaux, Le Ciclon reactor. Partie II:
measure directe de la distribution des temps de sjour de la phase
Solide-Lois dExtrapolation, Chem. Eng. J. 42 (1989b) 103117.
[16] T.A. Gauthier, C.L. Briens, M.A. Bergougnou, P.A. Galtier, Distri-
bution des temps de sejour du gaz dans un cyclone a co-courant,
Rcents Progrs en Gnie des Procds La Fluidisation 5 (1991)
185192.
[17] R.K. Stoecker, A. Rastogi, L.A. Behie, W.Y. Svreck, M.A. Bergoug-
nou, A computer simulation of propane cracking in a spout-fluid bed
reactor with draft tube, in: K. Ostergaard, A. Sorensen (Eds.), Flu-
idization V, Engineering Foundation, New York, 1986, pp. 465472.
[18] R.K. Stoecker, J.H. Eng, W.Y. Svreck, L.A. Behie, Ultrapyrolysis
of propane in a spouted-bed reactor with a draft tube, AIChe J. 35
(1989) 16171624.
[19] D. Sonnet, S. Afara, C.L. Briens, J.F. Large, M.A. Bergougnou, 1988.
Micromixing of particles in an ultra-rapid fluidized (URF) reactor,
in: P. Basu, J.F. Large (Eds.), Circulating Fluidized Bed Technology
II, Pergamon Press, Oxford.
[20] A. Tamir, Process and phenomena in impinging-stream reactors,
Chem. Eng. Prog. 86 (9) (1989) 5361.
[21] M.J. San Jos, M. Olazar, F.J. Peas, A.T. Aguayo, J.M. Arandes,
J. Bilbao, A model for gas flow in jet spouted beds, Can. J. Chem.
Eng. 71 (1993) 189194.
[22] D. Geldart, Types of fluidization, Powder Technol. 7 (1973) 285292.
[23] D. Geldart, Gas Fluidization Technology, John Wiley, New York,
1986.
[24] M.J. San Jos, M. Olazar, F.J. Peas, A.T. Aguayo, J.M. Arandes, J.
Bilbao, Correlation for calculation of the gas dispersion coefficient
in conical spouted beds, Chem. Eng. Sci. 50 (1995) 21612172.
 Chemical Engineering and Processing 44 (2005) 159166

Taking-off model of particles with a wide

size distribution
Isabelle Descamps, Jean-Luc Harion , Bernard Baudoin
Dpartement Energtique Industrielle, Ecole des Mines de Douai, 941 rue Bourseul, B.P. 838, 59508 Douai, France

Received 23 June 2003; received in revised form 4 April 2004; accepted 15 April 2004
Available online 2 July 2004

Abstract

The aim of this work is to estimate more accurately fugitive dust emissions due to wind erosion of exposed aggregate storage piles. The
model constructed allows to quantify the temporal evolution in the mass flux when a bed of materials is exposed to a turbulent flow. It is based
on the interaction between particles take-off and wall turbulence. The model allows as well to take into account materials having a wide size
distribution, which is typical of materials such as coal or ore. Some experimental studies have shown a temporal decrease in the mass flux as
the bed contains a wide range of particle sizes. This decrease, due to large particles, is predicted by the model. The rate depends on the flow
velocity and the characteristics of the particles.
2004 Elsevier B.V. All rights reserved.

Keywords: Fugitive dust emissions; Wind erosion; Particle take-off; Mass flux

1. Introduction
Atmospheric diffuse dust emissions constitute a pollution
source, which is difficult to be quantified. In fact, a diffuse
emission is defined as a polluting atmospheric flux which
is spread out over a large area or which is not concentrated
[1]. In the case of steelwork sites, diffuse dust emissions can
represent many as 20% of airborne particle pollution, they
result from conveyances and handling of materials and from
wind erosion of exposed aggregate storage piles. This latter
category of emission source is widely predominant.
The aims of the investigations presented in this paper are
to estimate accurately fugitive dust emissions resulting from
wind erosion. Thus, a model which allows to quantify emis-
sions resulting from an exposed particle bed to a turbulent
flow has been developed. Its originality is to take into ac-
count the wide particle size distribution of materials used,
as example, at steelwork sites. Typically, the finer particles
can have a size of about 10 m and the larger running to
centimetres. This characteristic is very important. In fact,
Meunier [2] observed for various experimental tests carried
Corresponding author. Tel.: +33 3 2771 2379; fax: +33 3 2771 2915.
E-mail address: harion@ensm-douai.fr (J.-L. Harion).
out in a wind tunnel with coal and ore, that the emitted mass
flux of particles decreases with time. He remarked the very
fast take-off of fine particles staying at the bed surface, then
the increasingly difficult take-off of the particles being be-
low. This can be explained by the presence of coarse par-
ticles close to the surface, which dont take-off because of
their inertia. These large particles constitute a cover and this
phenomenon is called the pavement. The pavement of the
bed, which is inherent to the wide size distribution of parti-
cles (especially to the presence of non-erodible particles at
the bed surface), generates a temporal decrease in emitted
mass flux.
The temporal decrease in emitted mass flux in the case of
particles with a wide size distribution has been shown by a
few experimental studies [35]. These authors assume that
the temporal decrease in emitted flux is due to modifica-
tion of soil surface during wind erosion. These experimen-
tal studies underline the importance to consider the tempo-
ral evolution of the emitted mass flux in order to accurately
estimate particles emissions with a wide size distribution.
The first part of the present work describes the wind ero-
sion model of a horizontal particle bed. A simple test of the
model is presented in the second part, which reports the in-
fluence of a wide size distribution on the emitted mass flux
of particles.

0255-2701/$ see front matter 2004 Elsevier B.V. All rights reserved.
doi:10.1016/j.cep.2004.04.007
160 I. Descamps et al. / Chemical Engineering and Processing 44 (2005) 159166
2. Wind erosion model of a particle bed
2.1. Aerodynamic entrainment modelling
The modelling of aerodynamic entrainment is based on
the close link between particles take-off and turbulent co-
herent structures above the surface. In fact, some authors
[6,7] have experimentally observed that a particle take-off
can be associated to the ejection of fluid from the wall re-
gion due to the presence of streamwise counter rotating vor-
tices. If it is assumed that the presence of two streamwise
counter rotating vortices produces only one ejection, each
pair of streamwise vortices is considered as a possibility that
a particle takes-off. Thus, for each of these possibilities, a
take-off criterion is tested.
The vortex structures appear with statistical spatial peri-
odicities, whose values are dependent on the flow properties
and velocity, and on the presence or not of particles. Jimenez
and Moin [8] have obtained, by numerical simulations, the
values of average spatial periodicities of appearing in the
spanwise and streamwise directions of the flow which are
respectively, z + 100 and x + 250300.
The periodicities are adimensionalized with the two scale
parameters of the inner layer of the turbulent boundary, the
kinematic viscosity and the friction velocity u , by the
equation:
u domly from its probability density function given by Zimon
+ = (1)
Although, these values are established in the case of a
smooth wall, they will be considered in the first approxima-
tion like the spanwise and streamwise spatial periodicities
of the coherent structures over the particle bed. The parti-
cle bed is thus subdivided into boxes having spanwise and
streamwise sizes given by average spatial periodicities of
the turbulent coherent structures.
Fig. 1 shows the number of taking-off possibilities over
the bed surface if it is assumed that each box, which contains
an ejection, is associated with a possibility of taking-off. The
numbers of possibilities of taking-off along the spanwise
and streamwise directions of the bed are equal respectively
to l/z and L/x . Consequently, they depend on the flow
velocity.
The frequency of the take-off criterion, which is tested for
each box, is supposed to be equal to the burst frequency in
Fig. 1. Number of taking-off possibilities at the bed surface.
the wall region of the turbulent boundary layer. Blackwelder
and Haritonidis [9] have shown that the average periodicity
of occurence is given by TB + 250 whose expression is:
TB u2
TB+ = (2)
So the time step of the model is equal to TB .
2.2. The take-off criterion
Before the take-off criterion, a size range of representa-
tive diameter Dp is selected randomly from the total number
of size ranges of the particle size distribution. The take-off
criterion, which is established from the balance of forces
exerted on a particle lying at a flat surface, requires the
lift force from the flow to exceed the forces holding the
particle down, namely the adhesive forces and the particle
weight. The expression of the lift force used in the model
is given by the experimental expression of Mollinger and
Nieuwstadt [10]. They have developed a new technique for
measuring the lift force and its fluctuations. Consequently,
the instantaneous lift force value is equal to the sum of the
 ,
average lift force Faero and the fluctuating lift force Faero
which is selected randomly from its probability density func-
tion which has been measured by Mollinger and Nieuwstadt
[10]. In the same way, the adhesive force is selected ran-
[11].
The aerodynamic entrainment model is based on the com-
parison between the distributions of forces holding the par-
ticles on the bed and of the lift force exerted on the particles
by the flow. The expression of the take-off criterion is given
for each size range by:
 
Faero (Dp ) + Faero (Dp ) P(Dp ) + Fad (Dp ) + Fad (D)
Particle take-off
The take-off test of the model developed in the present
study allows to simulate the stochastic nature of the phe-
nomenon. The establishment of the take-off criterion is de-
tailed more precisely in the thesis of Descamps [12].
One of the characteristic parameters of particle take-off
is the threshold flow velocity, at which particles start to take
off. The threshold flow velocity can be defined from the
take-off criterion.
 I. Descamps et al. / Chemical Engineering and Processing 44 (2005) 159166 161

Fig. 2. Take-off curve [13].

Fig. 2 shows the take-off curve obtained by Foucaut [13]. The values of Dpref and Uref are, respectively, (2 / p g)1/3
It traces out the adimensionalized threshold friction velocity and ( p g)1/3 , where p = (p )/ [13]. The take-off
u according to the adimensionalized particle diameter Dp . curve allows to estimate by knowing the adimensional-
These two parameters are given by the following equations: ized friction flow velocity the granulometric slice liable to
take-off. In fact, the diameter noted Dpcrit1 is the diameter
Dp u
Dp = u = (3) from which the particles take-off and the diameter Dpcrit2
Dpref uref

Fig. 3. Integration of the effect of the pavement by diminishing by graduated steps the lift force exerted on a particle.

Fig. 4. Principle of the model of simulation of the emitted mass flux.

162 I. Descamps et al. / Chemical Engineering and Processing 44 (2005) 159166

is the diameter from which the particles dont take-off be- The increment is given by:
cause of their inertia. Thus, if the value of Dpcrit2 is known,
the proportion of non-erodible particles can be determinate. Max pdf(Faeroinitial ) Min pdf(P + Fad )
F aero (Dp ) =
Nerosion (Dp )
2.3. Modelling of the pavement (7)

The temporal decrease in emitted mass flux can be ex- Fig. 3 shows schematically the modelling of the bed pave-
plained by the phenomenon of pavement, which is linked to ment.
the strong inertia particles, which cannot take-off and which This formulation allows to include in the model the lift
form a protection, called cover. It is supposed that these force influence decrease for particles which are in a deeper
non-erodible particles remain at the bed surface, and conse- position in the bed. When the erosion depth is attained, the
quently, overlay it completely from a certain depth, which take-off criterion can no more be respected. Indeed, the val-
is called erosion depth, Eerosion . From this depth, the vortex ues of the probability density functions of the instantaneous
structures bump only particles which they are not able to
entrain. No more take-off is possible.
In order to determinate the erosion depth, we assume that
the particles have a spatial uniform distribution along the
height H of the bed. For each size range, the number of layers
Nc (Dp ) including into the height of the bed is calculated by
dividing the height of the bed by the particle diameter. The
number of particles Npart (Dp ) of each size range, given by
the size distribution of beds particles, is distributed equally
on each layer of the bed:
Npart (Dp )
Np (Dp ) = (4)
Nc (Dp )

Now, the depth from which all the non-erodible particles

cover the bed surface has to be determined. Kurose and
Komori [14] define the rate of cover like the ratio between
the surface equal to the projected area of particles on the wall
and the total area of the considered surface. Consequently,
the rate of cover is determined for each size range by:
Np (Dp ) Dp2 /4
TC (Dp ) = (5)
Ll
Thus, the bed surface covered by the non-erodible parti-
cles is calculated by successive summation on all the size
ranges of the non-erodible particles in the increasing order
of diameters. For a y depth of the bed given, the surface
covered by the non-erodible particles can be determinate
since the number of layers, included by y, is known for
each size range. The evolution of the surface, covered by the
non-erodible particles, according to the depth can be estab-
lished. The depth from which the covered surface is equal
to the bed surface, namely the value of the erosion depth
Eerosion , is deduced from this evolution. We suppose that, if
there are not non-erodible particles in the bed, the erosion
depth is equal to infinity. The erosion depth is then divided
by the diameter Dp of each size range. The number of layers
likely to be affected by erosion is given by:
Eerosion
Nerosion (Dp ) = (6)
Dp

For each size range, the value of the lift force is then de-
creased as soon as all the particles of a layer have taken-off. Fig. 5. Algorithm of the model.
 I. Descamps et al. / Chemical Engineering and Processing 44 (2005) 159166
lift force are always smaller than the sum of the adhesion
and weight forces. In fact, the maximum of the probability
density function of the lift force is lower than the minimum
of the sum of the probability density function of the adhe-
sion force and the weight force.
2.4. Synthesis and algorithm of the model
This part describes concisely the model presented in this
paper. The modelling is constituted of three parts (Fig. 4):
the aerodynamic entrainment of particles, which is based
on the interaction between the coherent structures appear-
ing at the bed surface and the particles take-off;
the definition of the take-off criterion which depends on
the particles diameters and density for a given flow veloc-
ity;
the pavement modelling as a result of the presence of
coarser particles at the bed surface.
The principle of the model is to scan the bed surface,
which is subdivided into boxes whose the width and the
length are equal respectively to the spanwise and streamwise
statistical periodicities of appearing of the coherent struc-
tures. In fact, some authors have shown that the phenomenon
of ejection in a turbulent boundary layer could be connected
with the particles take-off.
For each box, namely for each ejection, the criterion for
taking-off is tested. Previously, it is necessary to determine
which particle, that is to say, which size range will be tested.
The size range of representative diameter Dp of the particle
likely to be carried away by the flow is selected randomly
from the size distribution. If all the particles of this size range
have taken-off, then an other size range is randomly selected.
For each take-off, a particle is substracted from its size
range then it is counted in the emitted mass flux. Propor-
tionately to these takes-off, the flow has more and more dif-
ficulty to carry away particles, this difficulty is taken into
account by the modelling of the bed pavement.
The model algorithm is showed on Fig. 5.
Fig. 6. Size distributions tested by the model.
163
The input variables are:
the exposition time to the turbulent flow, dexpo ;
the flow velocity, U ;
the particles density, the representative diameters of the
size distribution and their mass percentages, the number
of size ranges and the value of their depth;
the adimensionnalized values of the streamwise, spanwise
and temporal appearing periodicities of coherent struc-
tures;
the sizes of the bed (height, width and length);
the kinematic viscosity and the density of the flow.
Afterwards, the modelling is divided into various parts:
the calculation of Npart (Dp ), Nc (Dp ) and Np (Dp );
the establishment of Eerosion and Nerosion (Dp ),
the calculation of Faero (Dp ) and of the average lift force,
the standard deviation of the probability density function
of the lift force, the average adhesion force, the standard
deviation of the probability density function of the adhe-
sion force, the weight force,
the bed is scanned along width and length. The iterations
number is equal to the number of coherent structures ap-
pearing along the width multiplied by the number of co-
herent structures appearing along the length multiplied by
the ratio between the exposition period and the occurence
period of the vortex structures. The take-off criterion is
tested for each iteration and the modelling of the pave-
ment is taken into account.
The results obtained by the model are the temporal varia-
tion of the emitted mass and of the mass flux as well as the
temporal evolution of the surface size distribution and the
size distribution of eroded particles.
3. Numerical results
This part aims at showing that the model allows to
estimate the temporal decrease of the emitted mass flux
164 I. Descamps et al. / Chemical Engineering and Processing 44 (2005) 159166
observed in presence of non-erodible (relatively coarse)
particles.
A test has been investigated with a simple size distri-
bution constituted by a blend of erodible (60160 m)
and non-erodible sand particles (540640 m). The rate of
non-erodible particles, which is the cover rate noted tdc,
varies from 0 to 80%. Fig. 6 shows the size distributions
which are tested by the model.
The input variables are:
the exposition time to the turbulent flow, dexpo = 60 s,
the flow velocity, U = 10 m/s,
the particles density, p = 2500 kg/m3 , the representative
diameters of each size range, Dp , and the mass percentages
corresponding, Pmass , which are given by Fig. 6,
the statistical spatial and temporal periodicities of appear-
ing of ejections x + =300, z + =100 and TB + = 250,
the sizes of the bed H = 2.5 cm, L = 48.5 cm and l = 8.7
cm.
Fig. 7 shows the results obtained by the model. The figure
traces out the evolutions of the mass flux in kg/m2 s according
to the time in s for a velocity of the flow equal to 10 m/s
and for various cover rates, tdc = 0, 20, 40, 60 and 80%.
Several remarks can be made from the numerical results.
If the particles bed does not contain any non-erodible parti-
cles, then the emitted mass flux is constant with time. When
the particles bed contains non-erodible particles, the emitted
mass flux decreases with time.
Fig. 7 shows that the rate of the temporal decrease is
higher when the initial percentage of large particles in-
creases, which corresponds to experimental observations.
Moreover, the initial flux of particles decreases with the
presence of non-erodible particles.
Other tests, which are not represented in this paper, have
shown that, when the velocity is higher, the rate of the tem-
Fig. 7. First results obtained by the model in the case of a wide size distribution: U =10 m/s; p =2500 kg/m3 ; tdc = 0, 20, 40, 60 and 80%.
poral decrease and the initial percentage of the emitted mass
flux increase.
All these results can be deduced from the modelling of
the wind erosion of the bed. The decrease rate goes higher,
when the cover rate, noted tdc, increases. This can be ex-
plained by the modelling of the pavement, described in part
2.3. The reduction increment of the lift force is inversely
proportional to the erosion depth (Eq. (6)). According to the
definition of the erosion depth, the higher is the proportion
of non-erodible particles, the smallest is the erosion depth.
Consequently, when the proportion of non-erodible particles
is increased, the reduction increment becomes greater and
the temporal decrease is faster.
When the mass percentage of non-erodible particles
is greater, the emitted mass flux is lower. In fact, be-
fore the take-off criterion, a size range is randomly se-
lected from the size distribution. So, the proportion of
non-erodible particles is taken into account. When cover
rate is increased the chances that the particle is erodible are
weaker.
As flow velocity is increased, the emitted mass flux in-
creases. In fact, the number of taking-off possibilities equals
the number of boxes at the bed surface times the number
of time step of the model. Moreover, the number of boxes
at the bed surface multiplied by the number of time step is
proportional to U 4 .
When the takes-off are more numerous, the layers of par-
ticles are blown off faster. Consequently, the decrease of
emitted mass flux is greater.
Fig. 7 also shows a temporal decrease in mass flux in
graduated steps. The bound from one step to another is due
to the passage of one layer to another in the numerical model
and consequently to the decrement of the probability density
function of the lift force. This is not directly representative of
the reality. However, a general trend of the temporal decrease
 I. Descamps et al. / Chemical Engineering and Processing 44 (2005) 159166 165

Fig. 8. Interpolation of the temporal decrease in emitted mass flux obtained by the model: U =10 m/s; p =2500 kg/m3 ; tdc= 20, 40, 60 and 80%.

in the mass flux can be obtained by interpolating the data
by a negative exponential function.
In Fig. 8 it is observed that the temporal decrease of emit-
ted mass flux can be interpolated by an exponential negative
function:
This study constitutes a first step on studying the mod-
elling of wind erosion of multiple size grains bed.
Acknowledgements

flux(t) = flux(tinitial )exp(kd t) (8) This work was carried out with the financial support
of ADEME (the French Agency for Environment and
kd is the decrease rate and flux(tinitial ) is the emitted mass Energy Management), ARCELOR Dunkerque and Fos-
flux at tinitial 0 s. sur-Mer (steelworks in France) and European Community
(FEDER).

4. Conclusions
Appendix A. Nomenclature
The pavement modelling allows to introduce into the
model the temporal evolution of the size distribution of Dp particle diameter (m)
materials at the bed surface. By a progressive decrease of Eerosion erosion depth (m)
the probability density function of the lift force, this model Fad adhesion force (N)
successfully predicts the temporal decrease in mass flux Faeroinitial initial instantaneous lift force (N)
that occurs with the presence of coarse particles at the sur- Faero mean lift force (N)
face. The rate of this decrease depends on the flow velocity F aero fluctuating lift force (N)
and the characteristics of the particles. In order to improve H bed height (m)
the accuracy of the estimation of fugitive particle emissions l bed width (m)
with a wide size distribution, it is necessary to take into L bed length (m)
account this temporal decrease. P particle weight (N)
An experimental study, whose aim is to measure the tem- Pmass mass percentage (%)
poral decrease in mass flux, is on running and will allow to t time (s)
compare the results obtained by the model with the measure- TB average periodicity of burst
ments. Thus, it will allow to adjust the numerical results. occurence (s)
Various improvements could be brought to the model. tdc cover rate: mass percentage of non-erodible
Another study, whose objective is to measure the statistical particles (%)
periodicities of the coherent structures on a rough wall, will TC size range cover rate (%)
allow to adjust the results obtained by the model with the u friction velocity (m s1 )
experimental results. U flow velocity (m s1 )
166 I. Descamps et al. / Chemical Engineering and Processing 44 (2005) 159166
Greek letters
x + adimensionalized average streamwise
periodicity of appearing of vortex structure
z + adimensionalized average spanwise
periodicity of appearing of vortex structure
kinematic viscosity (m2 s1 )
References
[1] R. Bouscaren, Emissions diffuses and fugitives in Journe dtudes
du CITEPA, 2002.
[2] M.A. Meunier, Envol et transport de particules en couche limite
turbulente instationnaire, Ph.D. Thesis, University of Valenciennes,
France, 1999.
[3] L. Li, L.W. Martz, Aerodynamic dislodgement of multiple size grains
over time, Sedimentology 42 (1995) 683694.
[4] M.V. Lpez, Wind erosion in agricultural soils: an example of limited
supply of particles available for erosion, Catena 33 (1998) 1728.
[5] L.Y. Liu, P.J. Shi, X.Y. Zou, S.Y. Gao, H. Erdon, P. Yan, X.Y. Li,
Z.B. Dong, J.H. Wang, Short terms dynamics of wind erosion of
three newly cultivated grasslands soils in Northern China, Geoderma
115 (2003) 5564.
[6] Schet, P., Contribution ltude des structures cohrentes en tur-
bulence de paroi, De leur influence sur le transport des sdiments
dans le cas du charriage, Ph.D. Thesis, INPT, France,. 2000.
[7] Y. Nio, M.H. Garcia, Experiments on particle-turbulence interactions
in the near-wall region of an open channel flow: Implications for
sediment transport, J. Fluid Mech. 326 (1996) 285319.
[8] J. Jimenez, P. Moin, The minimal flow unit in near wall turbulence,
J. Fluid. Mech. 225 (1991) 213240.
[9] R.F. Blackwelder, J.H. Haritonidis, Scaling of the bursting frequency
in turbulent boundary layers, J. Fluid. Mech. 316 (1983) 285
306.
[10] A.M. Mollinger, F.T.M. Nieuwstadt, Measurement of the lift force
on a particle fixed to the wall in the sublayer of a fully devel-
oped turbulent boundary layer, J. Fluid. Mech. 316 (1996) 285
306.
[11] A. Zimon, Adhesion of dust and powder, Plenum Publishing Corpo-
ration, 1982.
[12] I. Descamps, Erosion olienne dun lit de particules large spectre
granulomtrique, Ph.D. Thesis, University of Valenciennes France,
2004.
[13] J.M. Foucaut, M. Stanislas, Take-off threshold friction velocity of
solid particles lying under a turbulent boundary layer, Exp. in Fluids.
20 (1996) 377382.
[14] R. Kurose, S. Komori, Turbulence structure over a roughness, Int. J.
Mult. Pha. Flow. 27 (2001) 673683.
 Chemical Engineering and Processing 44 (2005) 167173

Solids deposition in low-velocity slug flow pneumatic conveying

J. Li a, , C. Webb a , S.S. Pandiella a , G.M. Campbell a ,
T. Dyakowski a , A. Cowell b , D. McGlinchey b
a Department of Chemical Engineering, Satake Centre for Grain Process Engineering, UMIST, P.O. Box 88, Manchester M60 1QD, UK
b Centre for Industrial Bulk Solids Handling, School of Engineering, Science and Design, Glasgow Caledonian University, Glasgow G4 0BA, UK

Received 26 August 2003; received in revised form 4 December 2003; accepted 11 February 2004
Available online 25 September 2004

Abstract

Solids deposition in the horizontal pipeline of a pneumatic conveying system was studied both mathematically and experimentally. Math-
ematically modelled results using the coupled discrete element method (DEM) and computational fluid dynamics (CFD) approach have
demonstrated an intensive exchange of particles between the stationary layer (deposited particles) and the moving slug and a variation of
solids concentration and pressure and velocity distributions across the slug. Slug flows were also visualised experimentally through a glass
section and analysed by a high-speed video camera. The amount of particle deposition in the pipeline after a conveying was calculated
by controlling the solids feeding rate using a rotary valve and by monitoring the solids flow out of the system using dynamic load cells.
Experimentally generated data have quantitatively shown a tendency of more solids deposition with lower gas mass flow rate in slug flows
except that, below a certain amount of solids mass flow rate, the deposition becomes independent of gas flow rate.
2004 Published by Elsevier B.V.

Keywords: Pneumatic conveying; Slug flow; Solids deposition; CFD model; DEM simulation

1. Introduction
Low-velocity slug flow pneumatic conveying has become
more and more common in many industrial sectors ranging
across chemical, pharmaceutical and food industries, due to
its advantages over conventional high-velocity suspension
flow in preventing product degradation and plant wear and
in delivering high throughput and efficient power utilisation
[1]. One of the distinguishing characteristics in such an op-
eration is the gravitational deposition of solid particles in
pipelines, which is due to the low gas velocity used, usu-
ally below the solids saltation velocity. A layer of particles
at the bottom of the horizontal conveying line is commonly
seen in these systems and this solids deposition significantly
affects the transition and flow of slugs along the pipeline.
This paper presents mathematical and experimental stud-
ies of solids deposition characteristics in the horizontal
pipeline of a dense phase slug flow pneumatic conveying
Corresponding author. Tel.: +44 1236 878448;
fax: +44 1236 872837.
E-mail address: jintang.li@devro-casings.com (J. Li).
system. To obtain an insight into the slug flow and solids de-
position characteristics, a coupled discrete element method
(DEM) and computational fluid dynamics (CFD) approach
was used to simulate the solids and gas flows. Solids de-
position in the horizontal section of the conveying system
was visualised through a glass section and recorded by a
high-speed video camera. The amount of particle deposition
was calculated by closely monitoring the solids flow into
and out of the system. Mathematically and experimentally
generated data were analysed in order to develop a useful
tool for the future prediction of solids deposition and slug
flow behaviours in low-velocity pneumatic conveying.
2. Characteristics of slug flow - DEM modelling
Although low-velocity slug flow pneumatic conveying has
been studied intensively in the past two decades [18], due
to the practical restrictions of applying modern measuring
tools to gas and solids multiphase flows there is still a lack
of understanding of how gases and particles interact inside a
moving slug and how these interactions affect the formation

0255-2701/$ see front matter 2004 Published by Elsevier B.V.

doi:10.1016/j.cep.2004.02.011
168 J. Li et al. / Chemical Engineering and Processing 44 (2005) 167173
Fig. 1. Initial conditions used in the simulation of slug flow in pipes.
and flow of slugs. A recent study by Klinzing [8] demon-
strated that the slugging behaviours of solids in a pipeline
are complex, involving intensive interchanges of particles
between the stationary layer and the moving slug, i.e. pick-
ing up particles at the front of a slug and dropping off at
the tail. He also stated that a thin layer of particles in the
pipeline was essential for the development of stable slugs. To
obtain this realistic experimental condition several pre-runs
were conducted in his experiments to let particles lay down
in the pipeline. Obviously, a quantitative description of this
particle layer, which is equivalent to the amount of particle
deposition after a conveying, would bridge the understand-
ing of solids slugging behaviour and the performance of the
transport system.
To understand the mechanisms of solids slug flows, a
two-dimensional coupled DEM/CFD numerical model was
built to simulate the motion of a pre-formed slug (ca. 0.3 m
long) in a 1 m long horizontal 50 mm bore pipe as shown in
Fig. 1. The pipe was initially filled with a layer of particles,
approximately 15 mm thick at the bottom. (The thickness
of this stationary layer was determined based on experience
from previous experiments and computer test runs).
In the model, the solids phase was modelled as discrete
particles using DEM and the gas phase was modelled as a
continuum by CFD. The influence of solid particles on the
gas flow was considered as a source of mass, momentum
and energy, via the change to its volume fraction, the aero-
dynamic forces exerting on the gas and the heat transfer with
the gas. The effect of the gas flow on particle motion was
calculated via the fluid drag forces. (Detailed description of
the model is found in a previous publication of the authors,
Li and Mason [9]). This method allows individual particle
trajectories and interstitial gas flows to be traced, and lo-
cal gas-particle, particle-particle and particle-wall interac-
tions to be computed with the evolution of the slug flows.
The basic advantage of this coupled approach over conven-
tional continuum techniques is that DEM simulates effects
at the particle level. There is less need for global assump-
tions and the assembly response is a direct output from the
simulation [10]. DEM simulations have been used to model
different granular systems by a number of researchers and
have reported positive results: e.g. Tsuji et al. [11] on a flu-
idised bed, Langston et al. [12] on hopper flow, Frank et al.
[13] on channel flow, Cleary [14] on ball mills and Stewart
et al. [15] and Kuo et al. [16] on solids mixing. The main
disadvantage of DEM modelling is that it requires a large
amount of resource to compute increased number of parti-
cles, which makes it difficult to model fine particles or in
three-dimensional geometries. Therefore, discrepancies are
expected when comparing the current two-dimensional sim-
ulations with experiments.
Particles used in this simulation were 5 mm spherical
polyethylene pellets with a particle density of 880 kg/m3 .
The gas was introduced into the pipe inlet with a constant
superficial gas velocity of 2 m/s and all particles in the solids
slug and the stationary layer were assigned a low initial ve-
locity at 0.01 m/s to represent a dynamic starting condition.
The simulation started with an ambient pressure in the whole
pipe and this ambient pressure was kept constant at the pipe
outlet while the pressure rose in the pipeline when gas was
introduced from the inlet.
Fig. 2 shows six snapshots of slug flows along the pipe
from the DEM/CFD modelling at a time interval of 0.2 s for
a total time length of 1.0 s. This simulation demonstrates a
similar picture as described by Klinzing [8], in terms of par-
ticle motion in the slug. That is, particles in the stationary
layer (dark coloured particles at the bottom of the pipeline)
ahead of the slug wave are drawn into the slug (light coloured
particles) and form part of it when being moved along the
pipe by the pressurised gases. The slug wave actually com-
presses the particle layer just ahead of it, pushing some of
the particles up from the layer into the wave. It is also inter-
esting to see that at the tail of the slug particles fall into the
bottom part of the pipeline and form the stationary layer. The
exchange of particles between the stationary layer and the
slug has shown a unique slugging mechanism in slug flows,
which explains why particles move slower than slugs [8].
A statistical analysis shows that, after the slug flow, the
number of particles deposited in the pipe is about 90% of the
number of particles originally placed in the stationary layer.
Taking into account the entrance effect (much less particle
deposition in the front section from 0 to 0.3 m), this shows a
reasonable comparison between the assumed layer thickness
(15 mm) and the modelled particle deposition, though an
uneven particle layer has been produced after the slug flow.
This also suggests that the uneven particle layer from the
current model should be adopted as initial input for further
modelling.
It can be clearly seen from Fig. 2 that there is a variation of
solids concentration across the slug with a looser front and
a denser back. Some particles are even pushed ahead of the
slug wave at the front and this group of particles behave in a
very dynamic and suspended fashion. This phenomenon has
been confirmed by analysing video footages from experi-
ments in slug flows using similar particulate materials (3 mm
spherical polymer pellets), as shown in Fig. 3. Although the
general slugging behaviour is determined by the aeration,
permeability and de-aeration characteristics of the particu-
late material, the local variation of solids concentration in
the slug is very much dependent upon the actual interstitial
gas flow through the slug and the pressure distribution along
the pipe. Fig. 4 shows a plot of pressure drops along the
pipeline at both top and bottom of the pipe cross-section at
 J. Li et al. / Chemical Engineering and Processing 44 (2005) 167173
Fig. 2. Solids slug flow and deposition DEM/CFD simulation (time interval between each picture 0.2 s and total time 1.0 s).

169
170 J. Li et al. / Chemical Engineering and Processing 44 (2005) 167173

is due to the blockage of gas passage by particles in the pipe

Fig. 3. Snapshots of slug flow from video footages recorded in experi-
ments.
an instantaneous time of 0.4 s. A clear response of pressure
build-up at the back of and across the slug is shown and
a pressure difference between top and bottom of the pipe
cross-section at the back of the slug is observed, both of
which have been confirmed by experiments using accurate
pressure measurement [17]. Obviously the pressure build-up
cross-section and the gas retention characteristics in front
of the slug plays an important role in determining solids
slugging behaviours. The presence of particles has severely
altered the gas flow field in the vicinity of the slug. Figs. 5
and 6 show the simulated gas velocity profiles horizontally
along the pipeline and vertically in the pipe cross-section
at different locations, respectively. In general the axial gas
velocity is much higher at the top of the pipe cross section
than at the bottom and the highest occurs at the back of the
slug (around 0.4 m), which follows the profile of the solids
assembly at the back of the slug. However, gas velocity pro-
files become much flatter in the middle of the slug as shown
in Fig. 5 at 0.5 m. It is also interesting to see the change
of gas flow direction in the pipe cross-section as shown in
Fig. 6 (positive value indicating an up-flow and negative a
down-flow). An up-flow of gas is observed at the front of
the slug (0.6 and 0.7 m) and the far back of the slug (0.3 m)
and a down-flow at the back of the slug (0.4 m) but a com-
plex picture in the middle of the slug (at 0.5 m). In general,
these cross section velocity profiles comply with the axial
velocity profiles in terms of mass balance in the gas phase.

Fig. 4. Pressure drop along the pipeline DEM/CFD simulation (at 0.4 s).

Fig. 5. Gas velocity profile along the pipe DEM/CFD simulation (at 0.4 s).
 J. Li et al. / Chemical Engineering and Processing 44 (2005) 167173 171

Fig. 6. Gas velocity profile in pipe cross section DEM/CFD simulation (at 0.4 s).

The gas velocity profiles may explain why a loose slug
front is generated. That is particles in the front of the slug
wave are pushed by the slug body and dragged ahead of
the slug by the high local gas velocity at the top pipe cross
section while being agitated through severe interactions in-
curred by the rising particles from the stationary layer be-
neath the slug front and floating over them. The agitated
particle layer underneath the slug front is actually acting as
multiple fluidised rolling wheels for the slug, which greatly
reduces the frictional forces comparing with a direct inter-
action with the pipe bottom wall. In some cases, even an
air pocket or a particle swirling flow is created between the
suspended loose slug front and the bottom particle layer, of
which the resulting mechanisms and its consequence on slug
flow need further studies.
3. Experimental analysis
Experiments were conducted to determine quantitatively
the amount of particle deposition in the pipeline of a spe-
cially designed pneumatic conveying system [17,18]. A
schematic layout of the rig is shown in Fig. 7. Briefly
it comprises of a feed vessel, a receiving vessel, a 13 m
long conveying pipeline with interchangeable 5.1, 7.6 and
10.2 cm (i.e. 2, 3 and 4 in.) nominal bore pipes, a bank of air
supply nozzles, a variable speed rotary valve with dropout
box, an air filter unit, and connecting pipes and control
valves. In the system, both gas and solids mass feeding
rates are closely controlled through the choked nozzle bank
and the variable speed rotary valve. A set of three load cells
is mounted underneath the receiving vessel to monitor the
weight of solids conveyed. This particular design allows the
desired solids mass flow rate to be selected by adjusting
the rotary valve speed while the actual conveying rate is
calculated simultaneously from the load cell readings. Fig.
8 shows a typical diagram of a non-suspension wave-like
flow. Each step in the mass collected history is caused by
the arrival of a slug of solids. It is noticed that there is a
difference between the mass of solids flowing into and out
of the system and the difference in the final masses is the
deposition of solids in the system, which usually forms a
particle layer in the pipeline at the end of conveying.
Different combinations of gas and solids flow rates were
tested in a 7.6 cm bore pipe with a common industrial gran-
ular material - polyethylene pellets. The pellets were spher-

Table 1
Solids deposition in a pneumatic conveying system
Ug (ca.) Mg (kg/s) Ms1 Ms2 Ms3 Ms4
(m/s)
Fixed Deposition Fixed Deposition Fixed SLR5 Deposition Fixed SLR10 Deposition
(kg/s) (kg) (kg/s) (kg) (kg/s) (kg) (kg/s) (kg)
25 0.1534 1.2270 0.00 0.6135 0.00 0.7669 0.00 1.5337 0.00
8 0.0491 1.2270 3.94 0.6135 9.35a 0.2454 11.35a 0.4908 9.51
6 0.0368 1.2270 6.28 0.6135 7.29 0.1840 9.18a 0.3681 5.43
4 0.0245 1.2270 6.05 0.6135 8.03 0.1227 8.33a 0.2454 8.05
2.6 0.0160 1.2270 9.79 0.6135 8.92 0.0798 11.70 0.1595 12.88a
a Large difference ( > 2 kg) of load cell readings occurred between repeated runs, which may be due to the unstable flow of slugs in these flow
conditions.
172 J. Li et al. / Chemical Engineering and Processing 44 (2005) 167173

Fig. 7. Schematic layout of the pneumatic conveying rig.

ical in shape with an average diameter of 3 mm, a particle
density of 880 kg/m3 and an approximate bulk density of
550 kg/m3 . The experiments were started initially with high
superficial gas velocity at 25 m/s, in which very little particle
deposition occurred (This was confirmed with visualisation
via a 3 m glass section in the conveying line). When using
low gas velocity (< 10 m/s), a particle layer was built at the
bottom of the pipeline after a slug passing and at the end of
each test case. This solids deposition was assumed equiva-
lent to the difference of the final load cell readings between
the tested cases and that obtained using a high gas velocity
(25 m/s).

Fig. 8. Mass trace and solids deposition in slug flows.

 J. Li et al. / Chemical Engineering and Processing 44 (2005) 167173
Table 1 summarises solids depositions in the four groups
of tests conducted: Ms1 - fixed mass flow rate of solids
with a higher rate at 1.2270 kg/s, Ms2 - fixed mass flow
rate with a lower rate at 0.6135 kg/s, Ms3 - fixed solids
loading ratio (ratio between solids and gas flow rates) SLR
= 5, and Ms4 - fixed SLR = 10, totalling 20 experimen-
tal cases. From Ms1 , a strong tendency for more solids de-
position with a lower gas mass flow rate is shown. How-
ever, this effect does not appear to be true for either the
lower solids mass flow rate Ms2 , or the two groups with
fixed SLR, Ms3 and Ms4 (also with lower mass flow rate
< 0.5 kg/s). It shows that below a certain amount of solids
mass flow rate the deposition of solids becomes independent
of gas flow rate, which can be explained by the fluidised
motion of particles before a full slug moves and the grad-
ual build-up of long slugs in the down stream pipeline as
discussed previously in Li et al. [18]. Further experiments
with a wider range of flow conditions using advanced mea-
suring tools such as tomographic image analysis [19] are
required to validate the model and to build up a quantita-
tive description of solids deposition for different particulate
materials.
4. Conclusions
Gravitational solids deposition in a horizontal pipeline
of a pneumatic conveying system was studied both math-
ematically and experimentally. Mathematically modelled
results using the DEM/CFD simulation have demonstrated
an intensive exchange of particles between the slug body
and the stationary layer as slugs move along the pipeline,
and a large variation of solids concentration and pres-
sure and velocity distributions across the slug. A slug
wave actually compresses the particle layer just ahead of
it, pushing some of the particles up from the layer into
the wave. On the other hand, particles at the tail of the
slug fall into the bottom part of the pipeline and form a
stationary particle layer behind the slug. These phenom-
ena have been confirmed by analysing recorded video
footages.
Experimentally generated data have quantitatively shown
a tendency of increased solids deposition at lower gas mass
flow rate in slug flows except that, below a certain amount of
solids mass flow rate, the deposition becomes independent
of the gas flow rate, which is due to particle fluidised motion
before a slug flow commences. Further work will focus on
experiments with a wider range of flow conditions using
sophisticated measuring tools to validate the model and to
build up a quantitative protocol for the description of solids
deposition in conveying pipes.
173
Acknowledgements
Funding for this project was provided in part by the Satake
Corporation of Japan.
References
[1] K. Konrad, Dense-phase pneumatic conveying: a review, Powder
Technol. 49 (1986) 135.
[2] T. Ramakrishnan, K. Ramakoteswara Rao, M.A. Parameswaran, S.
Sivakumar, Experimental investigation of a dense-phase pneumatic
transport system, Chem. Eng. Process. 32 (1993) 141147.
[3] A.J. Jaworski, T. Dyakowski, Investigations of flow instabilities
within the dense pneumatic conveying system, Powder Technol. 125
(2002) 279291.
[4] R. Pan, P.W. Wypych, Pressure drop and slug velocity in low-velocity
pneumatic conveying of bulk solids, Powder Technol. 94 (1997)
123132.
[5] Y. Tomita, K. Tateishi, Pneumatic slug conveying in a horizontal
pipeline, Powder Technol. 94 (1997) 229233.
[6] O. Molerus, A. Burschka, Pneumatic transport of coarse-grained
materials, Chem. Eng. Process. 34 (1995) 173184.
[7] A. Levy, Two-fluid approach for plug flow simulations in horizontal
pneumatic conveying, Powder Technol. 112 (2000) 263272.
[8] G.E. Klinzing, Dense phase (plug) conveying observations and pro-
jections, in: A. Levy, H. Kalman (Eds.), Handbook of Conveying and
Handling of Particulate Solids, vol. 10, Elsevier, 2002, pp. 291301.
[9] J. Li, D.J. Mason, A computational investigation of transient heat
transfer in pneumatic transport of granular particles, Powder Technol.
112 (2000) 273282.
[10] J.P.K. Seville, U. Tuzun, R. Clift, Processing of Particulate Solids,
Blackie Academic & Professional, London, 1997, p. 372.
[11] Y. Tsuji, T. Kawaguchi, T. Tanaka, Discrete particle simulation of
two-dimensional fluidized bed, Powder Technol. 77 (1993) 7987.
[12] P.A. Langston, U. Tuzun, D.M. Heyes, Distinct element simulation
of granular flow in 2D and 3D hoppers: dependence of discharge rate
and wall stress on particle interactions, Chem. Eng. Sci. 50 (1995)
967987.
[13] T.H. Frank, K.-P. Schade, D. Patrak, Numerical simulation and ex-
perimental investigation of a gas-solid two-phase flow in a horizontal
channel, Int. J. Multiphase Flow 19 (1993) 187198.
[14] P.W. Cleary, Predicting charge motion, power draw, segregation and
wear in ball mills using discrete element methods, Miner. Eng. 11
(1998) 10611080.
[15] R.L. Stewart, J. Bridgwater, Y.C. Zhou, A.B. Yu, Simulated and
measured flow of granules in a bladed mixer a detailed comparison,
Chem. Eng. Sci. 56 (2001) 54575471.
[16] H.P. Kuo, P.C. Knight, D.J. Parker, Y. Tsuji, M.J. Adams, J.P.K.
Seville, The influence of DEM simulation parameters on the particle
behaviour in a V-mixer, Chem. Eng. Sci. 57 (2002) 36213638.
[17] D.J. Mason, J. Li, A novel experimental technique for the inves-
tigation of gas-solids flow in pipes, Powder Technol. 112 (2000)
203212.
[18] J. Li, S.S. Pandiella, C. Webb, T. Dyakowski, M.G. Jones, Analysis
of gas-solids feeding and slug formation in low-velocity pneumatic
conveying, Part. Sci. Technol. 21 (2003) 5773.
[19] T. Dyakowski, J.F.C. Jeanmeure, A. Jaworski, Applications of elec-
trical tomography for gas-solids and liquid-solids flows a review,
Powder Technol. 112 (2000) 174192.
 Chemical Engineering and Processing 44 (2005) 175185

Identification of material specific attrition mechanisms

for polymers in dilute phase pneumatic conveying
Lars Frye a, , Wolfgang Peukert b
a Technische Universitt Muenchen, Lehrstuhl fr Feststoff und Grenzflchenverfahrenstechnik (LFG),
Boltzmannstr. 15, 85748 Garching, Germany
b Institute of Particle Technology, Friedrich-Alexander-Universitt Erlangen-Nrnberg, Cauerstr. 4, 91058 Erlangen, Germany

Received 22 July 2003; received in revised form 25 February 2004; accepted 19 March 2004
Available online 27 August 2004

Abstract

In the approach presented, attrition of bulk solids in pneumatic conveying is regarded as a result of a process function (stress conditions)
and a material function (influence of material properties). For dilute phase conveying the process function was determined by employing
computational fluid dynamics (CFD). Opposed to what was expected initially, the numerical simulations indicate that sliding friction apparently
is of importance in dilute phase conveying. This conclusion can be drawn from the low calculated impact angles in a pipe bend (rB /Dn = 5;
Dn = 80 mm) which lie between 5 and 35 . Consequently, the tangential impact velocity components ranging from 33 to 44 m/s are considerably
higher than the normal ones (525 m/s). These results were at first endorsed by observations made in experiments to determine the material
function. The relative attrition behavior of four different polypropylenes (PP) under pure sliding friction conditions closely resembles that
observed in attrition experiments carried out in a pipe bend of the above geometry, while differences are observed for normal impact conditions.
But further experiments with polymethylmethacrylates (PMMA) and polystyrenes (PS) showed that here the attrition behavior encountered
under normal impact conditions is similar to that observed in the pipe bend. With dynamic mechanical analysis (DMA) it was possible to
show that PP on one hand and PMMA and PS on the other are attrited by different mechanisms and that the glass transition temperature of
the polymers is a key factor in the determination of the prevailing attrition mechanism.
2004 Elsevier B.V. All rights reserved.

Keywords: Attrition; Pneumatic conveying; Polymers; Process function; Material function; Dynamic mechanical analysis (DMA)

1. Introduction and approach ing importance of engineered products possessing carefully

Pneumatic conveying belongs to the well established
processes for transporting a vast variety of bulk solids.
Especially in the handling of granular polymers it is used
extensively. Nevertheless, the problem of product degra-
dation due to attrition is unresolved to date. This is partly
due to the fact that research was focused on different topics
like bend erosion or pipe wear [1,2]. But with the increas-
Abbreviations: DMA, dynamic mechanical analysis; DSC, differential
scanning calorimetry; He, helium; PE, polyethylene; PMMA, polymethyl-
methacrylate; PP, polypropylene; PS, polystyrene; St-bend, steel disc with
topography similar to that of pipe bend; St-sb, steel disc with sandblasted
surface
Corresponding author. Tel.: +49 89 289 15656;
fax: +49 89 289 15674.
E-mail address: l.frye@lfg.mw.tum.de (L. Frye).
designed material properties, even smallest amounts of at-
trition are intolerable. Thus, focus has to be and is laid on
the investigation of bulk solids attrition [38].
In the present paper, the approach chosen at our institute
to study the phenomenon of attrition is briefly presented.
In the results section, the link between bulk solids mate-
rial properties and attrition results is discussed. The findings
document the suitability of our approach for identifying ma-
terial specific attrition mechanisms.
The approach chosen goes back to the work of Rumpf [9],
Krekel and Polke [10] and Peukert and Vogel [11]. Rumpf
was the first who analyzed comminution processes by dis-
tinguishing between machine and material parameters. He
stated that the main parameters affecting the comminution
result are stress mode, stress intensity and number of stress
events as well as particle size and shape and material com-
position. This approach in which a mill is not regarded as a

0255-2701/$ see front matter 2004 Elsevier B.V. All rights reserved.
doi:10.1016/j.cep.2004.03.012
176 L. Frye, W. Peukert / Chemical Engineering and Processing 44 (2005) 175185

veying pipelines and a material function summarizing the

Fig. 1. Interpretation of attrition formation as a result of a process function
and a material function.
black box, but where the stress conditions of different mills
are used as characteristic features to distinguish between
them was later also applied by Krekel and Polke [10]. They
introduced the term process function which describes the
combined effect of process or machine parameters and ma-
terial properties on the chemical and physical product prop-
erties.
Peukert and Vogel [11] divided the process function into
a machine and a material function. The machine function
describes the connection between design and machine spe-
cific parameters and the driving potential for the engineering
operation, i.e. the stress conditions in a mill. The material
function summarizes the reaction of the materials used on
this driving potential. This concept was transferred to the
problem of attrition in pneumatic conveying. Since no sin-
gle part of machinery can be directly associated with this
process, the machine function was renamed as process or
conveying function in order to obtain an unambiguous termi-
nology. The process function thus differs from what Krekel
and Polke [10] have defined. The term material function is
used as introduced by Peukert and Vogel [11].
Consequently, attrition formation in pneumatic conveying
is interpreted as the result of a process function describing
the stress conditions the particles are subjected to in the con-
material specific response to the process function in terms of
intrinsic material properties. This approach is schematically
shown in Fig. 1. The process function of course depends on
the mode of conveying. For dilute phase conveying attrition
will be mainly caused by particle wall and inter-particle
impacts, whereas in plug flow conveying friction is the
dominating stress mode. Stress intensities as well as the
number of stress events will vary greatly likewise. For the
material function, it is unknown which properties can be
used to at least qualitatively predict the material reaction
to the stress conditions. Both issues are discussed in detail
posterior to the description of the materials used for the
model experiments. Special attention is given to the deter-
mination of the material function while the process function
is only discussed in short. Both functions are of course
not fully independent, but can in our opinion be separated
satisfyingly for the investigation of attrition processes.
2. Material
Since pneumatic conveying is largely applied to trans-
port granular polymers and on the other hand even smallest
amounts of attrition of these solids cannot be tolerated the
results presented are focused on these materials. Polymers
of four chemically different polymer classes were examined.
Polypropylene (PP) and polyethylene (PE) belong to the
semicrystalline polymers, which possess both, an amorphous
phase and a crystalline phase. The polymethylmethacry-
lates (PMMA) and polystyrenes (PS) are fully amorphous.
Some material properties of the polymers are summarized
in Table 1.
The particle shape was determined by visual inspec-
tion of the particles. To determine the particle diameter
xp , the volume of individual particles was measured in a
He-pycnometer and the diameter of the sphere possessing
the same volume was calculated. The solid density p was
measured in the He-pycnometer as well. The Vickers hard-
ness HV was determined in a microhardness tester while
Youngs modulus E and yield stress y were given by
the manufacturer. Finally, the J-integral value JQd which

Table 1
Overview of different material properties
Name Shape xp (mm) p (kg/m3 ) HV (MPa) E (MPa) y (MPa) JQd (kJ/m2 )

PP 1040 N Elliptical 4.04 869.9 95.5 2,000 40 2.04

PP 1100 RC Elliptical 4.00 869.7 85.0 1,500 34 2.09
PP 1148 RC Elliptical 4.11 867.8 90.0 1,650 35 1.92
PP 2500 H Elliptical 4.06 896.3 68.0 1,100 23 n.a.
PE 2420 H Elliptical 3.28 919.3 61.5 260 11 n.a.
PE 5031 L Elliptical 3.46 946.3 84.9 1,000 26 4.00
PMMA G7 Cylindrical 3.20 1,191.8 208.9 3,200 n.a. 0.77
PMMA G55 Cylindrical 3.03 1,203.9 178.0 3,100 n.a. 0.65
PS 144 C Cylindrical 3.46 1,053.1 177.7 3,300 n.a. 1.54
PS 158 K Cylindrical 3.46 1,059.8 178.8 3,300 n.a. 1.99
 L. Frye, W. Peukert / Chemical Engineering and Processing 44 (2005) 175185
corresponds to the critical energy release rate Gc for materi-
als with a nonlinear deformation curve was measured at the
Institute of Polymeric Materials, Martin-Luther University
Halle-Wittenberg by employing a modified Charpy notched
impact test.
3. Comments on the process function
Since a detailed description of the determination of the
process function for dilute phase conveying has been given
elsewhere [12] only the main findings are presented here.
The stress conditions were calculated by employing compu-
tational fluid dynamics (CFD). The fluid phase was modeled
by using the Eulerian approach, i.e. the Reynolds averaged
conservation equations for mass and momentum in conjunc-
tion with the k- turbulence model while for the particulate
phase the Lagrangian particle tracking as expressed in the
BassetBoussinesqOseen equation was applied. The parti-
cles were modeled as being spherical, particle wall impacts
as fully elastic and particle lift forces were neglected.
The computational domain consisted of a straight pipe
with a diameter Dn of 80 mm and a length le of 2400 mm.
Proceeding the pipe is a bend. For the simulations different
rB /Dn ratios of 2, 5 and 10 corresponding to short, standard
and long radius bends were used. Details on the geometry
can also be found in the experimental section, since an in-
stallation possessing the above geometry of the straight pipe
in conjunction with a pipe bend of rB /Dn = 5 was con-
structed. Due to the fact that the simulations were intended
to be as closely related to the experimental conditions as
possible, the gas and particle velocities measured in this in-
stallations were used as boundary conditions for the simula-
tions. Thus the average gas entrance velocity normal to the
inlet face was set to vf,i = 40.77 m/s. The particle entrance
velocities vp,i were intended to be equal to the gas velocity,
however due to experimental limitations it was not possible
to exactly match them with the gas velocity. Consequently,
Fig. 2. Cumulative number distributions for wall impact angles and normal and tangential impact velocity components for one polymer of each class.
177
depending on the material vp,i ranged from 40.22 to 45 m/s.
The material properties required for the simulations were
taken from Table 1.
In Fig. 2 the cumulative number distributions Q0 of im-
pact angles and impact velocity components are shown for
selected polymers. It can be observed that with 535 the
calculated angles are comparatively small. This results in
low normal (525 m/s) and high tangential (3344 m/s) im-
pact velocity components which leads to the conclusion that
not only impact stresses contribute to attrition in dilute phase
conveying but that sliding friction stresses might play an im-
portant role as well. The differences in the distributions of
the tangential velocity component can be mainly attributed
to the already mentioned slight variations in the initial par-
ticle velocities.
Since the results were unexpected, the particle wall im-
pact locations were plotted from the data generated in the
simulations for a total of 2000 individual particles. They are
shown in Fig. 3 from two different angles. Due to the large
particle diameters resulting in high inertia forces the parti-
cles do not follow the gas stream in the pipe bend, but pro-
ceed directly to the outer bend wall. Only a slight sinking
caused by the gravity force can be observed. Since this be-
havior can be attributed to the high initial particle velocities,
simulations varying these have been carried out. They lead
to changes in the respective number distributions of impact
angle and velocity components, but did not have any effect
on the maximum angle. It can thus be concluded that not
only the bend geometry is of influence as simulations with
varying rB /Dn ratio have shown but that the curvature of
the pipe plays an important role for the impact conditions
as well. Although Fig. 3 clearly shows that the results are
reasonable an experimental validation of the simulations is
nevertheless still necessary.
For reasons of completeness, results for the determina-
tion of the process function for plug flow conveying shall
be stated briefly as well. Since in this case a two phase flow
with high particle concentrations is regarded the options to
178 L. Frye, W. Peukert / Chemical Engineering and Processing 44 (2005) 175185

Fig. 3. Visualization of particle wall impact locations for 2000 particles of PP 1040 N with a diameter xp = 4 mm and initial gas and particle velocities
vx,i 41 m/s.

simulate this flow are very limited. Therefore, experimental
results obtained by Mi and Wypych [13,14] were used to
obtain at least a rough estimate of the stress conditions the
particles are subjected to. If it is assumed that the particles
in a plug are at least for short times fixed relative to each
other, the highest relative velocities will occur between the
particles forming the outer layer of the plug and the pipe
wall. The stress mode in this case is sliding friction. From
measurements of the pressure exerted by a plug on the pipe
wall and by applying the Hertzian equations for elastic con-
tact, a contact pressure between a single particle and the
pipe wall of roughly 4 MPa is obtained.
4. The material function
4.1. Concept for determination
It is generally known that different bulk solids might ex-
hibit widely varying attrition rates even if conveyed in the
same mode, q.v. [3]. Since under these circumstances the
stress conditions are almost identical the significance of the
material function becomes obvious. Unfortunately it is still
unknown which material properties determine the attrition
behavior. In the case of polymers the identification is dif-
ficult due to their viscoelastic nature. Thus the concept for
the determination of the material function shown in Fig. 4
was developed.
The crucial point is to simulate the previously identified
basic stress modes impact and friction under well defined
stress conditions. This way, material properties can be re-
lated to attrition caused by these stress modes and the re-
spective attrition mechanisms in effect. For this purpose it
was chosen to perform single particle experiments in simple
experimental setups to realize the defined stress conditions.
Details on these setups are given in the next section.
Ideally, the stress conditions obtained from the process
function are used as input parameters for the single particle
experiments. But due to practical reasons this is not always
possible, since a compromise between stress conditions

Fig. 4. General concept and experimental devices to determine the material function for attrition processes.
 L. Frye, W. Peukert / Chemical Engineering and Processing 44 (2005) 175185
closely related to the conveying process and conditions
where the attrition rates can be measured reliably has to
be found. The knowledge gathered from the single particle
experiments will be broadened by and compared to results
of the defined stressing of a particle ensemble. For this
purpose, experiments in a pipe bend are carried out to in-
vestigate the polymers under conditions closely related to
those in the conveying pipelines.
The results of the attrition experiments are compared to
various material properties in order to identify those rele-
vant for attrition. The long term goal is to develop a phys-
ical model of the attrition process which makes it possible
to predict the amount of attrition that will be encountered
in a pneumatic conveying installation by knowledge of the
process and material functions.
4.2. Experimental setup and parameters
The experimental setups of two out of the three installa-
tions used to date are shown in Fig. 4. The installation on the
left-hand side is used to stress single particles under normal
impact loads. It was originally developed by Schnert and is
described in Marktscheffel and Schnert [15]. The particles
are fed one by one into the center of a rotor (2) by means
of vibration (1). In the rotor, the particles are accelerated in
radial channels and finally hit the impact ring (3) under an
impact angle of 90 . The impact velocity is determined by
the number of revolutions of the rotor. Particles and attri-
tion debris are collected in the impact chamber (4) and can
be discharged through a tube at the bottom of this chamber.
The experiments were carried out under vacuum conditions
to eliminate any effects due to viscous drag.
For the attrition experiments approximately 25 g of poly-
mer particles were used. After three, six and nine consec-
utive impacts with a velocity of 40 m/s, the attrition rate A
was determined as the relative loss of mass (see Eq. (1)).
This procedure was necessary since the attrition rates for
some polymer classes were not measurable with satisfying
accuracy for lower impact numbers. Since the attrition pro-
cess is highly statistical, each experiment was repeated three
times. This also holds for the other experimental setups. In
the following diagrams, the median values of these three re-
peated experiments are plotted with the standard deviations
as error bars. For all experiments, the attrition rate A was
calculated according to Eq. (1), where Mi denotes the initial
particle mass. Ma is the mass of particles at the initial size
after the attrition experiment.
Mi Ma
A= (1)
Mi
The installation to simulate sliding friction shown on the
right-hand side of Fig. 4 was developed at our institute. In
design it resembles a classic pin-on-disc tribometer. Here,
a rectangular polymer particle cut from tensile bars with a
length of 10 mm, a width of 4 mm and a thickness of 4 mm
is pressed onto a rotating disc with a defined normal force
179
F. Due to its shape a constant contact area of 16 mm2 and
thus a defined contact pressure pc is maintained throughout
the experiment. Due to sample geometry, the velocity over
the contact face varied by a maximum of 2.5% from the
median value. Through the number of revolutions n of the
disc and the distance r between the particle and the disc axis,
the sliding velocity vs is adjusted. The sliding distance s
results from the sliding velocity and the contact time tc . The
disc simulating the wall material can be changed to account
for different materials and surface topographies conveying
pipelines might exhibit.
The experimental parameters for the sliding friction ex-
periments, were chosen to simulate dense phase conveying
conditions. Basically, two different sets of experimental pa-
rameters were used. The polypropylenes were tested against
different wall materials. Here, steel possessing different sur-
face roughness values was used. Since a sandblasted surface
lead to high attrition rates, a low contact pressure of 61 kPa
as well as a long sliding distance of 1500 m were used. The
sliding velocities ranged from 4 to 8 m/s. Based on the find-
ings from the process function for dense phase conveying,
a higher contact pressure of 0.61 MPa in conjunction with
a lower sliding distance of 200 m was used for further ex-
periments. To accomplish similar conditions to the experi-
ments in the pipe bend installation, here a steel disc with
approximately the same roughness value as the pipe bend
was installed. The parameters, especially the comparatively
high sliding distances had to be chosen in such a way, to
make sure that reproducibly measurable attrition rates were
obtained. As already mentioned, here the attrition rate was
determined according to Eq. (1) by weighing the single par-
ticles on a high precision scale.
The third installation was constructed to conduct single
particle attrition experiments under well defined stress con-
ditions which are closely related to conditions in industrial
conveying installations. For this purpose a pipe bend was
chosen. The exact setup is given in Fig. 5. As it has already
been discussed, the geometry of this setup was used for the
simulations to determine the process function in dilute phase
conveying.
The installation was constructed as a negative pressure
system with the blower being located downstream of the fil-
ter. This way a high flexibility in feeding the particles was
Fig. 5. Design and dimensions of the pipe bend installation.
180 L. Frye, W. Peukert / Chemical Engineering and Processing 44 (2005) 175185
realized. For the experiments presented, the particles were
fed into the pipe along its center axis by a specially designed
feeder in which they are accelerated with pressurized air to
approximately 41 m/s prior to entering the pipe. The veloc-
ity of the particles was measured by photo-electric guards.
The pipe has a diameter Dn of 80 mm and possesses the di-
mensions already presented in the process function section.
For separating the particles from the gas stream, a filter bag
was used. Particles and attrition debris are discharged at the
bottom of the filter housing. With the blower, gas velocities
in the pipe of up to 41 m/s are reached. They are measured
and monitored with a Prandtls tube.
Similar to the normal impact experiments, the attrition
rates were determined after three, six and nine impacts. Here
it is assumed that only the first particle wall impact in the
pipe bend contributes significantly to attrition formation and
that further impacts are of minor importance. Again 25 g
of the granular polymers were used. In order to maintain
conditions close to those in conveying processes the particle
velocity was set to be approximately equal to the gas velocity
of 41 m/s.
4.3. Results of the single particle experiments
The first results of the single particle experiments carried
out for different polypropylenes have shown that the relative
attrition behavior changes between the normal impact and
the sliding friction tests [12]. This was taken as a chance
to identify which attrition mechanism dominates in the pipe
bend installation and thus presumably in dilute phase con-
veying. In Fig. 6, a comparison of the results obtained for
the normal impact experiments, the sliding friction experi-
ments and the pipe bend installation is given. The normal
impact velocity was 40 m/s. For sliding friction a vanadium
steel disc possessing a sandblasted surface was used. The
contact pressure was 61 kPa, the sliding distance 1500 m and
the sliding velocities ranged from 4 to 8 m/s. The particle
Fig. 6. Comparison of the attrition rates for different polypropylenes as obtained from experiments in the normal impact, sliding friction and pipe bend
experiments.
and gas entrance velocities in the pipe bend installation were
approximately 41 m/s.
In the middle diagram, the error bars were omitted for
reasons of clarity. Although the attrition rates cannot be
compared directly due to the different stress conditions, it
can be observed that the relative attrition behavior of the
polypropylenes in the sliding friction experiments is iden-
tical to that observed in the pipe bend experiments, i.e. in
both cases PP 2500 H is the least attrited material, while
PP 1148 RC is the most attrited one. For normal impacts,
considerable deviations from the results in the pipe bend are
found. This leads to the conclusion that attrition formation
can be mainly attributed to sliding friction. This is in good
agreement with the findings from the process function for
dilute phase conveying discussed earlier. In case of PMMA
and PS a different behavior is observed (see Fig. 7). Here,
the relative attrition behavior found for the normal impact
experiments corresponds to that in the pipe bend.
These findings clearly show the significance of the ma-
terial function, since the material reaction to similar stress
conditions in the pipe bend can vary greatly. In contrast to
PP, PMMA and PS apparently are attrited by the same mech-
anisms in effect under normal impact stress conditions. This
also clarifies that it is only possible to deduce the stress mode
from the process function but that without knowledge of the
material function no conclusions concerning the governing
attrition mechanisms can be drawn.
4.4. Correlation between single particle experiments and
material properties
One of the most important tasks in the determination of the
material function is to identify those material properties rele-
vant to attrition formation. A comparison of the above results
to classic mechanical material properties, e.g. Youngs mod-
ulus E or Vickers hardness HV as given in Table 1 suggests
that these are not suitable to describe attrition formation.
 L. Frye, W. Peukert / Chemical Engineering and Processing 44 (2005) 175185 181

Fig. 7. Comparison of the attrition rates for different PMMA and PS granules as obtained from normal impact, sliding friction and pipe bend experiments.

This is due to the fact that the respective measurement pro-
cedures for their determination do not take into account the
dynamic nature of the attrition process. Therefore, in the
present study dynamic mechanical analysis (DMA) was em-
ployed.
For the measurements, the single cantilever flexure mode
was chosen. For this, rectangular polymer samples of
32 mm 6 mm 2 mm had to be produced. This was done
by injection molding of the original polymers. Fig. 8 shows
the setup of a DMA-experiment in the single cantilever
configuration.
Prior to the experiments, the polymer sample is fixed be-
tween two clamps and the sample dimensions are measured
carefully. One of the two clamps is fixed, while the other
can be moved up and down. To this clamp, a sinusoidal de-
formation with an amplitude A and a frequency is ap-
plied and the corresponding force is measured, which can be
transferred into the stress A with the sample dimensions.
From these two quantities the complex modulus of elastic-
ity E is measured in dependence of sample temperature T
and oscillation frequency .
A 140 C. After an isothermal period of 1 min, a frequency
|E (, T)| = (2)
A
The viscoelastic behavior of the polymers is reflected in
the phase lag between the deformation signal and the cor-
responding stress signal. For a fully elastic material, would
be zero whereas it would be 90 for a fully plastic material.
With this phase lag, which is measured as well, the com-
plex modulus of elasticity can be divided into the storage
modulus E and the loss modulus E .
E (, T) = |E | cos (3)
E (, T) = |E | sin (4)
The storage modulus is proportional to the amount of
energy which is stored in the material elastically, whereas
the loss modulus corresponds to the energy that is dissipated
during one load cycle. Both quantities are combined in the
damping factor tan which is defined as
E
tan = (5)
E
After checking the influence of various experimental pa-
rameters as well as the reproducibility of the measurements,
a standard procedure was developed. The polymer samples
are cooled quickly with liquid nitrogen to a temperature of
sweep is carried out for 1, 3, 5, 10, 30, 40, 50 and 60 Hz with
a deformation amplitude of 20 m. Afterwards the temper-
ature is increased by 5 C and after another isothermal pe-
riod of 1 min, the next frequency sweep is carried out. This

Fig. 8. Setup of a DMA-experiment in the single cantilever configuration.

182 L. Frye, W. Peukert / Chemical Engineering and Processing 44 (2005) 175185

Fig. 9. Storage (E ) and loss (E ) modulus functions for different PP, PMMA and PS granules.

procedure is repeated until the value of the storage modu-
lus drops down to 10 MPa which was set as the criterion
to end the measurement in time before melting of the sam-
ple starts. The obtained storage and loss modulus functions
for the polymers are depicted in Figs. 9 and 10. The values
given were obtained for a frequency of 10 Hz.
From the measurements the typical viscoelastic behavior
of the polymers becomes obvious. The polymer chains are
fixed relative to each other at low temperatures. Thus the
amount of energy that is stored elastically is high which is
reflected in a high value of the storage modulus. The corre-
sponding values of the loss modulus are low. The mobility
of the polymer chains increases with increasing tempera-
ture and consequently the storage modulus decreases while
the loss modulus increases. In the glass transition regime
the maximum of energy dissipation is reached (maximum
of loss modulus) and the storage modulus falls rapidly.
The structural differences between the semicrystalline
polypropylenes/polyethylenes and the amorphous poly-
methylmethacrylates/polystyrenes are reflected as well.
After the glass transition, the storage modulus reaches a
plateau for PP and PE due to the crystalline structures which
do not change in stiffness significantly until melting. The
amorphous polymers on the other hand fully soften during
glass transition leading to very low storage modulus values.
In Table 2 the glass transition temperatures Tg as obtained
from the storage and loss modulus values are given. For the
storage modulus they were determined by using the inflec-
tion point method whereas for the loss modulus the peak
maximum was used [16].
It can be observed that the agreement of the correspond-
ing temperatures is quite well. As Ehrenstein et al. [16] point
out, this deviation is normal and usually Tg (E ) Tg (E )
Tg (DSC) holds. The materials where a second temperature
value is given in italics do not only possess a glass transi-
tion but also show a second order transition. For PP 2500 H,
PMMA G7 and G55 these second order transitions which
are most likely caused by increased mobility of side-chains

Fig. 10. Storage (E ) and loss (E ) modulus functions for different PP and PE granules.
 L. Frye, W. Peukert / Chemical Engineering and Processing 44 (2005) 175185 183

Fig. 11. Comparison of the attrition rates obtained from sliding friction experiments for 6 m/s, 0.61 MPa and 200 m against St-bend with storage (E ) and
loss modulus (E ) values measured at 10 Hz and 25 C.

or chain segments lie below their glass transition tempera- possible explanation for this behavior can be found in
tures. PE 2420 H and PE 5031 L on the other hand show the microscopic attrition mechanisms known from tribol-
their respective second order transitions above their glass ogy and was briefly discussed in [17]. As Fig. 11 shows,
transition temperatures (see Fig. 10). Here it is not quite PE 5031 L fits well into this tendency while PE 2420 H
clear what causes them. again does not. From the materials which are attrited by
If the ambient temperature at which the single particle ex- an impact mechanism, PMMA G7 and G55 also follow
periments were carried out is regarded as the macroscopic the above trend while the two polystyrenes do not. It is
stress temperature, it can be observed with respect to the currently under investigation how this observation can be
experimentally obtained attrition rates, that the polypropy- explained.
lenes which are attrited by sliding friction are stressed in If the above comparison is made for the attrition results
their glass transition regimes whereas the glass transition from the pipe bend installation, again a decrease in attrition
temperature of PMMA and PS lies well above the stress rate for the polypropylenes corresponds to decreasing E and
temperature. Obviously, the relation between the values of E values. Of course, the scatter gets larger since in this
the macroscopic stress temperature and the glass transi- installation although sliding friction is the most prominent
tion temperature can be used to evaluate which material attrition mechanism, impact also marginally contributes to
specific attrition mechanism impact or sliding friction overall attrition. Furthermore, PE 5031 L, PMMA G7 and
is dominating in attrition formation. This conclusion is PMMA G55 do not fit quite as well as for the sliding friction
also valid for PE 5031 L, while PE 2420 H which is a experiments while PE 2420 H, PS 144 C and PS 158 K
very soft material (see Youngs modulus in Table 1 or stor- shows the same discrepancies as before.
age modulus in Fig. 10) does not fit. In this case, addi- The results presented show that three levels have to be dis-
tional effects like adhesion are likely to play an important tinguished when investigating attrition processes. The first
role. one is the stress mode as derived from the process function
For PP it was furthermore found that the attrition rates which is essential to know if the attrition process is to be
observed in the sliding friction experiments decrease with simulated successfully in a simple experimental setup. The
decreasing values of both storage and loss modulus. A second point is the material reaction to this stress mode,
i.e. the material function which varies depending on mate-
rial properties like storage and loss modulus as measured
Table 2 by DMA. Finally, the microscopic attrition mechanisms (see
Glass transition temperatures as obtained from DMA measurements ( =
10 Hz)
[18] for impact and [19,20] for sliding friction) describing
the formation of attrition on a microscopic scale constitute
Name Tg (E ) ( C) Tg (E ) ( C) the bottom level.
PP 1040 N 17.54 15.08
PP 1100 RC 15.04 15.10
PP 1148 RC 15.16 15.08 5. Conclusions
PP 2500 H 37.33/12.43 34.88/9.99
PE 2420 H 122.32/12.02 114.82/9.62
PE 5031 L 107.31/45.03 109.81/55.12 In the present research project the problem of attrition
PMMA G7 45.07/107.69 29.90/110.20 in pneumatic conveying was addressed by regarding it as
PMMA G55 90.12 25.05/95.12 the result of a process and a material function. The process
PS 144 C 92.13 95.17 function summarizes the influence of all process parame-
PS 158 K 107.84 110.20
ters and as a result provides the stress conditions in terms
184 L. Frye, W. Peukert / Chemical Engineering and Processing 44 (2005) 175185

of stress mode, intensity and number of stress events. The Acknowledgements

material function on the other hand contains material prop-
erties which govern the attrition behavior of individual bulk This project is sponsored by the German Federal
solids and as a result supplies an attrition function. With Ministry of Economics and Technology, grant AiF-No.
knowledge of these two functions it should be possible to 13252 N/1. The authors appreciate the financial sup-
predict the amount of attrition that is encountered in pneu- port.
matic conveying.
For dilute phase conveying numerical simulations with a
commercial computational fluid dynamics code were carried
Appendix A. Nomenclature
out. The analysis of particle wall impact conditions in a pipe
bend showed that they take place under low wall impact an-
A attrition rate (%)
gles of 535 which results in low normal (525 m/s) and
Dn pipe diameter (m)
high tangential (3344 m/s) impact velocity components.
E Youngs modulus (Pa)
These findings lead to the conclusion that not only normal
E complex modulus (Pa)
stresses caused by the impacts are important in dilute phase
E storage modulus (Pa)
conveying but that sliding friction stresses play an important
E loss modulus (Pa)
role as well.
F force (N)
This was corroborated by experimental results obtained in
G energy release rate (kJ/m2 )
single particle experiments carried out for polypropylenes. It
H microhardness (Pa)
was possible to show that the relative attrition behavior under
pure sliding friction stresses was identical to that observed JQd J-integral (kJ/m2 )
in a pipe bend whereas deviations to that under pure normal L, l length (m)
impact conditions were found. For polymethylmethacrylate M mass (kg)
and polystyrene on the other hand, the relative attrition be- n number of revolutions (s1 )
havior in the pipe bend corresponds to that of the normal p pressure (Pa)
impact experiments. Q heat flux (W/s)
An explanation for this was deduced from material prop- r radius (m)
erties obtained by dynamic mechanical analysis. It was s sliding distance (m)
found that the glass transition temperature or glass tran- t time (s)
sition regime is a key factor in the determination of the T temperature ( C)
mechanism the polymer particles are attrited by. If the stress v velocity (m/s)
temperature lies below the glass transition temperature, an xp particle diameter (m)
impact-like mechanism is observed (PMMA, PS) while in x, y, z cartesian coordinates (m)
the other case friction is the dominating mechanism (PP,
PE). It is thus possible to identify the dominating attrition Greek symbols
mechanism by knowledge of the glass transition and the angle ( )
macroscopic stress temperature. Furthermore the results in- phase lag ()
dicate that both a low storage modulus as well as a low loss deformation (m)
modulus are feasible for low attrition rates if sliding friction frequency (Hz)
is the dominating attrition mechanism. This correlation was density (kg/m3 )
also observed for the PMMA granules but not for PS. An stress (Pa)
explanation for this has still to be found.
The investigated polypropylenes are difficult to catego- Subscripts
rize. While PE 5031 L fits in the above findings and appar- a after attrition
ently is mainly attrited by sliding friction as expected from A amplitude
analysis of the DMA measurements, PE 2420 H does not fit B bend
into this scheme. Here different mechanisms like adhesion c critical, contact
might play an important role, which can be neglected for e entrance
the other polymers. f filter, fluid
Nevertheless, the presented results indicate that if it is g glass transition
carefully distinguished between the process function (stress i initial
mode), the material function (material specific attrition p particle
mechanisms) and microscopic attrition mechanisms (attri- s sliding
tion formation on the microscale) in the analysis of attrition V Vickers
processes, significant progress in the understanding of the x, i initial component in x-direction
complex phenomenon of attrition can be made. y yield
 L. Frye, W. Peukert / Chemical Engineering and Processing 44 (2005) 175185
References
[1] D. Mills, J.S. Mason, Particle size effects in bend erosion, Wear 44
(1977) 311328.
[2] P.W. Wypych, P.C. Arnold, Minimizing wear and particle damage
in pneumatic conveying, Powder Handl. Process. 5 (1993) 129
134.
[3] H. Kalman, Attrition of powders at various bends during pneumatic
conveying, Powder Technol. 112 (2000) 244250.
[4] H. Kalman, M. Hubert, E. Grant, Fatigue behavior of impact com-
minution and attrition units, in: Proceedings of the 10th Euro-
pean Symposium on Comminution (ESC), Heidelberg, Germany,
2002.
[5] D. Papadopoulos, M. Ghadiri, Impact breakage of polymethyl-
methacrylate (PMMA) extrudates. I. Chipping mechanism, Adv. Pow-
der Technol. 7 (1996) 183197.
[6] A.D. Salman, M.J. Hounslow, A. Verba, Particle fragmentation in
dilute phase pneumatic conveying, Powder Technol. 126 (2002) 109
115.
[7] A.D. Salman, C.A. Biggs, J. Fu, I. Angyal, M. Szab, M.J. Houn-
slow, An experimental investigation of particle fragmentation us-
ing single particle impact studies, Powder Technol. 128 (2002) 36
46.
[8] A.D. Salman, J. Fu, D.A. Gorham, M.J. Hounslow, Impact breakage
of fertiliser granules, Powder Technol. 130 (2003) 359366.
[9] H. Rumpf, Die Einzelkornzerkleinerung als Grundlage einer tech-
nischen Zerkleinerungswissenschaft, Chemie-Ingenieur-Technik 37
(1965) 187202.
185
[10] J. Krekel, R. Polke, Qualittssicherung bei der Verfahrensentwick-
lung, Chemie-Ingenieur-Technik 64 (1992) 528535.
[11] W. Peukert, L. Vogel, Product Engineering am Beispiel des Zerklein-
erns von Kunststoffen, Chemie-Ingenieur-Technik 73 (2001) 327
331.
[12] L. Frye, W. Peukert, Attrition of bulk solids in pneumatic conveying:
mechanisms and material properties, Part. Sci. Technol. 20 (2002)
267282.
[13] B. Mi, P.W. Wypych, Pressure drop prediction in low-velocity pneu-
matic conveying, Powder Technol. 81 (1994) 125137.
[14] B. Mi, P.W. Wypych, Investigations into wall pressure during slug
flow conveying, Powder Technol. 84 (1995) 9198.
[15] M. Marktscheffel, K. Schnert, Liberation of composite particles by
single particle compression, shear and impact loading, in: Preprints
of 6th European Symposium on Comminution, Nuremberg, Germany,
1986, pp. 2945.
[16] G.W. Ehrenstein, G. Riedel, P. Trawiel, Praxis der Thermischen
Analyse von Kunststoffen, Carl Hanser, Mnchen, second ed., 2003,
p. 269.
[17] L. Frye, W. Peukert, Neue Anstze zum Studium des Frdergutabriebs
bei pneumatischer Frderung, Schttgut 7 (2001) 435442.
[18] J.A.S. Cleaver, M. Ghadiri, Impact attrition of sodium carbonate
monohydrate crystals, Powder Technol. 76 (1993) 1522.
[19] R. Kaneko, T. Miyamoto, E. Hamada, Microwear, in: B. Bhushan
(Ed.), Handbook of Micro/Nano Tribology, CRC Press, Boca Raton,
1995, pp. 183221.
[20] H. Czichos, K.-H. Habig, Tribologie Handbuch: Reibung und
Verschlei, Vieweg, Braunschweig, 1992, pp. 100144.
 Chemical Engineering and Processing 44 (2005) 187192

Two-dimensional numerical simulations of the

pneumatic drying in vertical pipes
I. Skuratovsky, A. Levy , I. Borde
Department of Mechanical Engineering, Ben-Gurion University of the Negev, Beer-Sheva 84105, Israel

Received 5 August 2003; received in revised form 16 December 2003; accepted 11 February 2004
Available online 25 June 2004

Abstract

The Two-Fluid model has been used for modeling the flow of particulate materials through pneumatic dryer. The model was solved for
a two-dimensional steady-state condition and considering axial and radial profiles for the flow variables. A two-stage drying process was
implemented. In the first drying stage, heat transfer controls evaporation from the saturated outer surface of the particle to the surrounding
gas. At the second stage, the particles were assumed to have a wet core and a dry outer crust; the evaporation process of the liquid from a
particle assumed to be governed by diffusion through the particle crust and by convection into the gas medium. As evaporation proceeds, the
wet core shrinks while the particle dries.
Our developed model was applied to simulate drying process of wet sand in a pneumatic dryer. The numerical solutions were compared
with the results of independent numerical and experimental investigations. These comparisons are quite encouraging.
2004 Published by Elsevier B.V.

Keywords: Pneumatic drying; Two-fluid model

1. Introduction phases are formulated. Often the flow in a pneumatic dryer

Pneumatic drying is a widely used process in chemical
industries and includes simultaneous conveying, and heat
and mass transfer between the particles and the hot gas.
The large surface area for heat and mass transfer and the
high convective heat and mass transfer coefficients, which
appear in these units, are leading to high drying rates and
as a consequence to high drying capacity. The increase in
the use of this unit operation requires reliable mathematical
models to predict processes in the industrial facilities.
Three types of theoretical approaches can be used for
modeling the gas-particles flows in the pneumatic dry-
ers, namely: Two-Fluid Theory [1], Eulerian-Granular [2]
and the Discrete Element Method [3]. Traditionally the
Two-Fluid Theory was used to model dilute phase flow.
In this theory, the solid phase is being considering as a
pseudo-fluid. It is assumed that both phases are occupy-
ing every point of the computational domain with its own
volume fraction. Thus, macroscopic balance equations of
mass, momentum and energy for both the gas and the solid
Corresponding author.
E-mail address: avi@bgumail.bgu.ac.il (A. Levy).
can be classified as dilute.
One-dimensional mathematical models are widely applied
to calculate drying processes, which can predict average
values of different properties of the phases in cross-sections
of the dryer [49].
Rocha and Paixo [10] presented a pseudo two-dimensional
mathematical model for a vertical pneumatic dryer that
was based on the two-fluid approach. Axial and radial
profiles for gas and solid velocity, water content, porosity,
temperatures and pressure were considered. Their predic-
tions were not validated with experimental results. Another
two-dimensional numerical investigation was conducted by
Silva and Nebra [11] for drying of particulate material in
cyclone. They used the two-fluid approach and presented
steady-state drying model that considered incompressible,
axi-symmetric, turbulent gassolids flow, slip condition
of particles on the wall, particleswall heat transfer and
particles shrinkage. Gravity force effect on the spherical
particle was neglected. The predictions of their numerical
simulations were compared with experimental results and
it was claimed that the parameters most influencing the
predictions are the particles slip conditions on the walls and
material shrinkage during the drying process.

0255-2701/$ see front matter 2004 Published by Elsevier B.V.

doi:10.1016/j.cep.2004.02.012
188 I. Skuratovsky et al. / Chemical Engineering and Processing 44 (2005) 187192
Blasco et al. [12] proposed two-dimensional mathemati-
cal model for the drying process of dense phase pneumatic
conveying. However, heat and mass transfer were not consid-
ered and therefore their model may be used for dense phase
pneumatic transport only. In their paper, both experimental
and numerical predictions for axial and radial profiles for
gas and solid velocity, axial profiles for solid concentration
and pressure drop were presented.
Contrary to traditional one-dimensional models, two-
dimensional models are required for taking into consid-
eration the effects of the radial distribution of the most
influential thermophysical properties. As can be seen from
the above literature survey, not many studies have adopted
the two-dimensional approach for simulation the drying
process in a vertical tube. In addition, their predictions were
not validated with experimental data and in some cases only
the momentum transfer was taken into account.
In the present study, two-dimensional Two-Fluid Eulerian
model was used to describe the steady state, dilute phase flow
of a wet dispersed phase (wet solid particles) in a continuous
gas phase through a pneumatic dryer. The predictions of
the numerical solutions were compared successfully with
the results of other one-dimensional numerical solutions and
experimental data of Baeyens et al. [5] and Rocha [13]. Axial
and the radial distributions of the characteristic properties
were examined.
2. Theoretical model
A two-dimensional, Two-Fluid model has been developed
for modeling the steady-state flow of particulate materi-
als through vertical pneumatic dryer. Two-dimensional flow
field interpretations provide information about properties of
the continuous and dispersed phases at every point of the
flow field. Axial and radial profiles were obtained for the
flow variables.
The gas phase was treated as a mixture of gas and liquid
vapor. The wet particle is made of porous solid structure
that contains saturated liquid. The drying model considered
a two-stage drying process. In the first drying stage, heat
transfer controls evaporation from the saturated outer surface
of the particle to the surrounding gas. In the second stage,
the particles are assumed to have a wet core and a dry outer
crust. The evaporation of the liquid from a particle assumed
to be governed by diffusion through the particle crust and by
convection into the gas medium. As evaporation proceeds,
the wet core shrinks as the particle dries.
The developed model was solved numerically and applied
to simulate drying of wet sand in a laboratory-scale pneu-
matic dryer.
The drying model is based on the following assumptions:
Steady-state flow.
The flow in the dryer is vertical, two-dimensional,
non-rotational and axi-symmetrical.
Non-slip and non-penetrating wall conditions.
Dispersed phase behavior like continuous phase.
The particles are made from a substance containing solid
porous matrix and liquid or gas with liquid vapor.
The specific densities of the liquid and the solid, which
compose the particle, are constants.
Mass, momentum and heat transfer occur only between
the two phases and not between the particles themselves.
The gas phase is a mixture of water vapor with air.
The gas behaves as ideal gas.
Influence of gravity on gas phase is neglected.
Heat transfer can occur between the pipe wall and the
continuous gas phase.
The acting forces per unit volume between the phases are
due to drag and virtual-mass.
At the first drying stage, the particle is isothermal and at
the second drying stage the particle wet core temperature
is different from that of dry crust.
The influence of the pressure gradient on the inertia of the
solid particles is negligible in comparison to that of the
drag force.
Electrical and surface tension forces are neglected.
Based on the above-mentioned assumptions, the mass,
momentum and energy balance equations for the gas and the
dispersed phases in two-dimensional, two-phase flow were
developed [14]. In order to solve the mass, momentum and
energy balance equations, several complimentary equations,
definitions and empirical correlations were required. These
were presented by [14]. In order to obtain the water vapor
distribution the gas phase the water vapor diffusion equa-
tion was added. During the second drying period, the model
assumed that the particle consists of a dry crust surround-
ing a wet core. Hence, the particle is characterized by two
temperatures i.e., the particles crust and core temperatures.
Furthermore, it was assumed that the heat transfer from the
particles crust to the gas phase is equal to that transferred
from the wet core to the gas phase i.e., heat and mass cannot
be accumulated in the particle crust, since all the heat and
the mass is transferred by diffusion through the crust from
the wet core to the surrounding gas. Based on this assump-
tion, additional heat balance equation was written.
The mass transfer source term was obtained by multiply-
ing the evaporation rate from a single particle by the total
number of particles per unit volume. Two-stage drying pro-
cess was applied for calculating the mass transfer from a wet
particle to the gas phase [6]. In the first drying period, the
gas phase resistance controls the evaporation rate. When the
liquid evaporated from the particle surface and fills all the
voids inside the porous particle, the second drying period
starts. During this drying period, the dry crust causes addi-
tional resistance to heat and mass transfer. This resistance,
which occurs between the outside diameter of the particle
and the diameter of the wet core is governed by a diffu-
sion process. The evaporation rate from a particle with a dry
crust was expressed by Stephan-type diffusion rule [15]. The
 I. Skuratovsky et al. / Chemical Engineering and Processing 44 (2005) 187192 189

Fig. 1. Comparison between the predictions of the present pneumatic drying model, DryPak model, Rocha in [20] and Levy and Borde [8] models and
the experimental data for (a) gas temperature, (b) solid temperature, (c) gas humidity and (d) particles moisture content along the dryer length.

mass transfer coefficient was calculated in analogy to the
heat transfer coefficient from the Sherwood number, which
is equivalent to the Nusselt number. It should also be noted
that both the gas and the dispersed phases are mixtures, and
hence all their thermodynamic properties were calculated
using the mixture theory.
3. Numerical simulations
The pneumatic drying model was solved numerically for
the drying processes of sand particles. The numerical proce-
dure includes discretization of the calculation domain into
torus-shaped final volumes, and solving the model equa-
tions by implementation of the semi-implicit method for
pressure-linked equations (SIMPLE) algorithm [16]. The nu-
merical procedure also implemented the Interphase Slip Al-
gorithm (IPSA) of [17] in order to account the various cou-
pling between the phases. The simulation stopped when the
moisture content of a particle falls to a predefined value or
when the flow reaches the exit of the pneumatic dryer.
In order to validate the theoretical model, the predic-
tions of the numerical simulations were compared with
experimental data. Comparison between the results of the
present two-dimensional model and the results of experi-
mental data and one-dimensional models requires a repre-
sentation model for presenting the two-dimensional data in
one-dimensional terms. Thus, an averaging technique was
used to obtain the average values of the various solution
properties at the dryer cross-section. The average value of
a property, , at the dryer cross-sectional area was obtained
by mass weighted method and an average temperature of
a phase was evaluated by the mean bulk temperature across
the tube.

R
R
2 0 urdr 2 0 ucp Trdr
=
R ; Tave =
R (1)
2 0 urdr 2 0 ucp rdr
4. Numerical results and discussion
The predictions of the numerical simulations with
Baeyens et al.s [5] correlation for heat transfer were also
compared with the experimental results of Rocha [13] (pre-
sented by Silva and Correa [18]) that were obtained in a
4-m high pneumatic dryer with diameter of 5.25 cm. In this
study, 380-m sand particles having density of 2622 kg/m3
and solid mass flow rate of 4.74 103 kg/s were dried
190 I. Skuratovsky et al. / Chemical Engineering and Processing 44 (2005) 187192

Fig. 2. The predictions of the pneumatic drying model for (a) gas phase velocity, (b) dispersed phase velocity, (c) gas phase temperature, (d) dispersed
phase temperature, (e) gas humidity, (f) particles moisture content and (g) particles diameter along stream lines and the dryer length.
 I. Skuratovsky et al. / Chemical Engineering and Processing 44 (2005) 187192
with 3.947 102 kg/s air mass flow rate. Comparison
between the predictions of the numerical simulations and
the experimental data for changes of gas temperature, solid
temperature, gas humidity and particles moisture content
with length under known wall temperature operating condi-
tions is presented in Fig. 1ad, respectively. When known
wall temperature operating conditions were simulated, it
was assumed that in average the pipe wall temperature is
just about the outlet air temperature and it is falling linearly
from 360 K at the inlet to 354 K at the outlet. In this way,
heat losses from the dryer walls are considered indirectly.
In these figures, the star symbols represent the experimen-
tal data of Rocha [13] that were published by Silva and
Correa [18] and the other curves and symbols represent the
predictions of various models and numerical simulations.
Form these figures, it can be seen that the predictions of the
two-dimensional numerical simulations fit fairly well with
the experimental data and both one-dimensional DryPak
[19] and Levy and Borde [8] models. Detailed comparison
between the various one-dimensional models can be found
at Levy and Borde [20]. Since the theoretical pneumatic
drying model and the numerical simulations done in the
course of this study predicted successfully the most impor-
tant flow parameters (gas temperature, solid temperature,
gas humidity and particles moisture content) of the drying
process, it was concluded that the presented model could
be used for further investigations. The distributions of the
various flow properties along the dryer length for various
stream lines (i.e., fixed r/R) are presented in Fig. 2.
Form these figures, it can be seen that a fully developed
flow was obtained after about 3 m from the dryer inlet. In
addition, various thicknesses of boundary layers were ob-
tained for the various phases properties. This may influence
the final product properties.
5. Conclusions
Two-Fluid, two-dimensional model was developed for
modeling the dilute flow of particulate materials through
pneumatic dryer. The model was solved for a steady-state
condition. Axial and radial profiles for the flow variables
were obtained. The developed model was applied to sim-
ulate the drying process of wet sand in a laboratory-scale
pneumatic dryer. The numerical solutions were compared
with the results of other numerical and experimental in-
vestigations. The successful comparisons validate the theo-
retical and the numerical models. As demonstrated by the
two-dimensional numerical simulations, the radial distribu-
tion of the flow characteristics causes an uneven product
quality. Thus in order to improve the drying process, a re-
liable two-dimensional model is required. Further investi-
gations are needed in order to validate the two-dimensional
distribution of the flow characteristic and to improve the
model to account for particle circulation in the investigated
domain. The present investigation significantly extends the
191
ability to simulate and to predict drying processes in pneu-
matic dryer.
Appendix A. Nomenclature
cp (J/kg K) specific heat of the gas
R (m) radius of the pipe
Tave (K) average temperature
T (K) temperature
u (m/s) velocity
Greek symbols
density
dummy property
References
[1] R.M. Bowen, in: A.C. Eringen (Ed.), Theory of Mixtures in Con-
tinuum Physics, Academic Press, NY, USA, 1976.
[2] D. Gidaspow, Multiphase Flow and Fluidization, Academic Press,
Boston, USA, 1994.
[3] P.A. Cundall, O.D. Strack, A discrete numerical model for granular
assemblies, Geotechnique 29 (1979) 4765.
[4] J. Andrieu, R. Bressat, Experimental, experimental & theoretical
study of a pneumatic dryer, Proc. 3rd Int. Drying Symp. 2 (1982)
1019.
[5] J. Baeyens, D. van Gauwbergen, I. Vinckier, Pneumatic drying: the
use of large-scale experimental data in a design procedure, Powder
Technol. 83 (1995) 139148.
[6] D. Levi-Hevroni, A. Levy, I. Borde, Mathematical modelling of
drying of liquid/solid slurries in steady sate one-dimensional flow,
Drying Technol. 13 (5-7) (1995) 11871201.
[7] A. Levy, D.J. Mason, I. Borde, D. Levi-Hevroni, Drying of wet solids
particles in a steady-state one-dimensional flow, Powder Technol. 85
(1988) 1523.
[8] A. Levy, I. Borde, Steady-state one-dimensional flow for a pneumatic
dryer, Chem. Eng. Proc. 38 (1999) 121130.
[9] Z. Mindziul, A. Kmiec, Modeling gas-solid flow in a pneumatic-flash
dryer, Drying96 Proc. 10th Int. Drying Symp. A (1996) 275
282.
[10] S.C.S. Rocha, A.E.A. Paixo, Pseudo two-dimensional model for a
pneumatic dryer, Drying96 Proc. 10th Int. Drying Symp. A (1996)
340348.
[11] M.A. Silva, S.A. Nebra, Numerical simulation of drying in a cyclone,
Drying Technol. 15 (6-8) (1997) 17311741.
[12] R. Blasco, R. Vega, P.I. Alvarez, Pneumatic drying with superheated
steam: bi-dimensional model for high solid concentration, Drying
Technol. 19 (8) (2001) 20472061.
[13] S.C.S. Rocha, Contribution to the study of pneumatic drying: simula-
tion and influence of the gas-particle heat transfer coefficient, Ph.D.
Thesis, So Paulo University, So Paulo, Brazil, (in Portuguese),
1988.
[14] I. Skuratovsky, A. Levy, I. Borde, Two-fluid, two-dimensional model
for pneumatic drying, Drying Technol. 21 (9) (2003) 16491672.
[15] N. Abuaf, F.W. Staub, Drying of liquidsolid slurry droplets, Drying
86 Proc. 5th Int. Drying Symp. 1 (1987) 227248.
[16] S.V. Patankar, Numerical Heat Transfer and Fluid Flow, Hemisphere
Publishing Corporation, New York, USA, 1980.
[17] D.B. Spalding, Developments in the IPSA procedure for numeri-
cal computation of multiphase-flow phenomena with interphase slip,
192 I. Skuratovsky et al. / Chemical Engineering and Processing 44 (2005) 187192

unequal temperatures, etc. in: Numerical Properties and Method-
ologies in Heat Transfer, Hemisphere, Washington, DC, 1983,
pp. 421476.
[18] M.A. Silva, J.L.G. Correa, Using DryPak to simulate drying process,
in: Proceedings of 11th International Drying Symposium Drying98,
vol. A, 1998, pp. 303310.
[19] Z. Pakowski, DryPak v.3, Program for psychometric and drying
computation, 1996.
[20] A. Levy, I. Borde, Two-fluid model for pneumatic drying of
particulate materials, Drying Technol. 19 (8) (2001) 1773
1788.
 Chemical Engineering and Processing 44 (2005) 193200

The formation of fine particles by salting-out precipitation

Judit Tth , Andrea Kardos-Fodor, Susan Halsz-Pterfi
Research Institute of Chemical and Process Engineering, University of Veszprm, P.O. Box 125, 8201 Veszprm, Hungary

Received 25 August 2003; received in revised form 27 January 2004; accepted 11 February 2004
Available online 17 June 2004

Abstract

A systematic study of salting-out precipitation is carried out to obtain the operational limits within which this precipitation method can
be applied for the production of fines (mean particle size <10 m) with acceptable quality and productivity. The model substances: glycine
and sodium chloride are salted-out from their aqueous solutions by using ethanol as antisolvent. The main operational parameter is the initial
supersaturation of the solutions. It is shown that the smallest particles can be produced at the limits of the metastability domain determined by
three optional process parameters: the initial solution concentration, the equilibrium solubility and the operational time. The product quality
(crystallinity, polymorphic states, aggregation) and productivity considerably change with the operational conditions.
2004 Elsevier B.V. All rights reserved.

Keywords: Salting-out precipitation; Weimarn laws; Metastability; Initial supersaturation; Particle size

1. Introduction mental findings [3], it is possible to reduce or even elim-

Formation of micron and submicron size particles has
become particularly important in the material-research and
processing. The small size, high surface and surface activity
of these crystals and nanocrystals determine their various
applications as catalysts, ceramics, electronics, pharmaceu-
ticals, cosmetics etc. and this may explain the world-wide
increasing research activities to find the suitable crystalliza-
tion modes and techniques for processing them [1,2].
Chemical precipitation is a well known method to produce
fines being a rapid solidification process of sparingly solu-
ble substances. Comparatively less known about the other
type of precipitation called salting-out or antisolvent pre-
cipitation although it is also widely used, especially in the
pharmaceutical industry owing to the available high yield,
purity and the fast, simple operation. In salting-out precip-
itation the necessary supersaturation is achieved by the ad-
dition of an antisolvent or electrolyte to the solution to be
crystallized. The produced crystals are small but often ag-
glomerated thus questioning the applicability of the pro-
cess for producing fines. Recently the salting-out precipi-
tation has got new perspectives. According to new experi-
Corresponding author. Tel.: +36 88 405 265; fax: +36 88 424 424.
E-mail address: toth@mukki.richem.hu (J. Toth).
inate the agglomeration in the presence of protective col-
loids or surfactants and/or by controlling the precipitation
kinetics.
In this work, batch salting-out experiments were per-
formed with the objective to know more about the limits and
possibilities on the formation of fine particles. As model ma-
terial the zwitterionic glycine was investigated within a wide
range of initial supersaturations. Considering the polymor-
phic nature of glycine, the non-polymorphic, ionic sodium
chloride was also investigated as reference system.
2. The precipitation theory
Not long ago, the main crystallization trend was to pro-
duce well-developed, bigger crystals in proper size and habit.
Accordingly, the research activities focused on the growth
process and the role of habit modifying additives. Nowa-
days, when extra fine particles are claimed by several in-
dustrial areas, increased attention has been paid to the early
stages of nucleation by applying even extreme conditions
for the particle formation. This area is the typical field of
the colloidal state which is not yet enough utilized for the
industrial practice but it is getting more and more impor-
tant due to the increasing demands on the production of
fines.

0255-2701/$ see front matter 2004 Elsevier B.V. All rights reserved.
doi:10.1016/j.cep.2004.02.013
194 J. Toth et al. / Chemical Engineering and Processing 44 (2005) 193200
In the colloid chemistry practice the antisolvent precip-
itation is called Weimarns method. Weimarn published
his famous work The precipitation laws early in the past
century [4] but his statements seem to be still valid and
useful for reconsidering. His main conception was that the
mean size of precipitates may change with the supersatura-
tion through a maximum. The maximum shifts up and left
with increasing time until the system reaches its stable state
where the precipitation process is completed. The principle
of the Weimarn laws on precipitation is schematically shown
in Fig. 1. According to the Weimarn laws, fine particles can
be produced both at low and high supersaturation depending
on the time of crystallization and the solubility of the pre-
cipitate. Weimarn based his statements on his chemical pre-
cipitation experiences. What is the situation for salting-out
precipitation? It is not necessarily the same because there
is a fundamental difference between the chemical- and salt-
ing out precipitation. In the first case the supersaturation
changes with the concentration of the reaction partners while
the equilibrium concentration (solubility) remains constant,
but at salting-out processes both the supersaturation and the
equilibrium solubility change with the solution/antisolvent
ratios, therefore, both can influence the particle size.
The relationship between particle size, solubility
and supersaturation is expressed by the well-known
GibbsThomson equation [5a]:


c(r) 2V
ln = ln S = , (1)
c kTr
where c(r) is the solubility of the particles of size (radius)
r, c* is the normal equilibrium solubility of the substance,
S is the supersaturation ratio, is the surface energy, V is
the molecular volume, k is the Boltzmann constant, T is the
absolute temperature. Eq. (1) does not consider the growth
process which follows the nucleation step. At precipitation,
in most cases, the growth process is rather an aggregation
step. When aiming fine crystals, the growth subprocess has to
be retarded, consequently, the precipitation has to be stopped
Mean size of crystalline individuals
stable state
unstable states
time
Supersaturation, S
Fig. 1. The Weimarn laws on precipitation (via [4]).
(quenching) before the particles could grow up their final,
stable size.
3. Experimental
In the experiments analytical grade chemicals: glycine
(purity 98.5%) or sodium chloride (99.5%), ethanol (99.8%)
and deionized water were used. The precipitation mode
was one-shot: at high supersaturations various amount (V0
= 0.00412 cm3 ) of saturated or undersaturated aqueous so-
lutions of glycine (c0 = 20.000.50 g/100 g solution) were
added in one portion to 12 cm3 ethanol at ambient temper-
ature. At low supersaturations 12 cm3 saturated solutions of
glycine were added to various amount (Ve = 0.112 cm3 ) of
ethanol. Depending on the solution/antisolvent ratios, series
of initial supersaturations were created. In each case con-
centrated ethanol, saturated at 298 K (0.0037 g glycine/100 g
solvent) was used. In the most cases the crystals appeared
at once after the shots but 10 min stirring was applied ad-
ditionally. No longer stirring time was allowed not to leave
time for the polymorphic change. Some precipitation runs,
together with the in-line particle size measurements, were
carried out directly in the 15 cm3 volume MALVERN cu-
vette. When the suspension densities were higher than al-
lowed in direct CSD measurements by MALVERN 2600
the samples were precipitated in an Erlenmeyer flask under
stirring. In these cases the CSD measurements were possi-
ble after the appropriate dilution of the slurry. The diluting
agent was the actual mother liquid or concentrated and satu-
rated ethanol. For characterizing the particle size the volume
mean diameter was used:
 4
nd
d =  3 , (2)
nd
where n is the number of particles of equivalent diameter d.
For characterizing the particle size distribution the so called
span was used:
D90 D10
span = , (3)
D50
where D10 , D50 , and D90 is the diameter below which 10,
50 and 90% of the volume of the particles lies.
The precipitates were separated by glass fibre filter,
washed by ethanol and air-dried. In some cases X-ray
diffraction (by Philips 3710) for polymorphism and SEM
analysis (by Philips XL 30 ESEM) for habit determination
were made.
The runs with sodium chloride were carried out by
the same techniques as detailed above but the operational
concentrations were different: c0 was changed between
26.55 and 7.40 g/100 g solution. A characteristic difference
between the two systems was their solubility in ethanol
(0.0037 g glycine/100 g solvent and 0.0649 NaCl g/100 g
solvent) at 298 K. Because of the slower precipitation abil-
 J. Toth et al. / Chemical Engineering and Processing 44 (2005) 193200 195

ity of sodium chloride, longer (60 min) additional stirring best fitted the whole range of solubilities have been found
was allowed. as follows:
for the glycinewaterethanol system:
y = 2.618 + 2.264 102 x
4. Experimental results and discussion
+ 23.17 exp(4.060 102 x), (5)
The main driving force of a salting-out precipitation pro- for the NaClwaterethanol system:
cess is the supersaturation generated by the different solubil-
ities of the solute in the solvent and antisolvent. The one-shot y = 24.91 4.348 101 x
mode generally means that the necessary amount of the an- + 18.66 exp[2.433 102 (100 x)], (6)
tisolvent is added in one portion to the initial solution (either where x is the mass percent of the ethanol, y is the equilib-
saturated or undersaturated) as it is schematically shown in rium concentration. In the range of 90100% ethanol con-
Fig. 2. In our cases the aqueous solution was added to the tent another type of equation was construed for the sake of
antisolvent. As a result of the antisolvent addition the orig- higher accuracy:
inal solubility of the substance will change until the new for the glycinewaterethanol system:
equilibrium solubility, ceq has been reached. To characterize
the supersaturation driving force in a precipitation system lg y = 2.656 9.476 102 x
the so called initial supersaturation, Si is used: + 1.363 103 x2 9.243 106 x3 , (7)
c0 for the NaClwaterethanol system:
Si = , (4)
ceq
lg y = 6.764 101 + 7.363 102 x 7.861 104 x2
i.e. it is a ratio between the initial concentration of the
(8)
solution to be crystallized (in this case the aqueous solu-
tion) and the equilibrium concentration (solubility) in the 25
solute-solvent-antisolvent system. In this way different val-
(a)
c0 (g glycine/100 g solution)

ues of Si can be chosen as experimental parameter by chang-

20
ing the concentration of the aqueous solution or the ratio of
the aqueous solution and the antisolvent. Although Si means
a certain simplification but it is still acceptable, because the 15
actual operational concentrations are very uncertain accord-
ing to the micromixing insufficiencies. The highest applied 10
initial supersaturations were Si 5000 for glycine and Si
400 for NaCl owing to their different solubilities in ethanol. 5
The solubility data of the model systems were taken from
the literature [6,7]. For the calculation of the missing data
0
correlation methods were elaborated. The equations which
0 20 40 60 80 100
Mass percent (g ethanol/100 g solvent), (%)

30
'one-shot' addition
C0- 25 (b)
c0 (g NaCl/100 g solution)

20
supersaturation driving
Solubility

forces in saturated and 15

C0-
undersaturated solutions
Ceq- 10

0
0 20 40 60 80 100 0 20 40 60 80 100
antisolvent
Mass percent of the antisolvent Mass percent (g ethanol/100 g solvent), (%)

Fig. 2. The scheme of the one shot salting-out precipitation from the Fig. 3. Solubilities and applied supersaturation programmes for the
saturated and undersaturated solutions. glycinewaterethanol (a) and NaClwaterethanol (b) model systems.
196 J. Toth et al. / Chemical Engineering and Processing 44 (2005) 193200
The solubility curves (at 298 K) as well as the ways of cre-
ating the different supersaturations are shown in Fig. 3, for
the glycinewaterethanol (a) and NaClwaterethanol (b)
model systems, respectively.
The experiments with saturated aqueous solutions show
remarkable differences concerning the two model systems
as it can be seen in Fig. 4, where the corresponding ini-
tial supersaturation, Si and average particle size, d data of
glycine (a) and NaCl (b) are plotted, separately. When only
a very small amount of glycine solution (0.004 cm3 ) was
added to 12 cm3 of antisolvent the alcohol mass percent
was very high: 99.96% but surprisingly the particle size was
not as small as expected, it was about 50 m. At somewhat
lower Si and alcohol ratio a minimum occurred (point A in
Fig. 4a). After the minimum the particle size approached
to a plateau (point B in Fig. 4a) then a peak of 100 m,
and finally decreased down to 20 m. These just mentioned
anomalies in the glycine particle size and Si relation can
be caused by several reasons (the change of the nucleation
mechanism or polymorphic states, etc.). The polymorphic
change of glycine itself is already known from the litera-
120
Average particle size, d (mm)
100
80
C
B solution. The same experiences were reported by Garti and
60
40
20
0 by one. The best result was found when 100 ml ethanol
1 10 100
Initial supersaturation, Si ()
120
Average particle size, d (mm)
100
80
60
40
20
0 the particles showed less and less agglomeration. Starting
1 10
Initial supersaturation, Si ()
Fig. 4. Relations between the initial supersaturation, Si and the av-
erage particle size, d (under the applied operational time) in the
glycinewaterethanol (a) and the reference NaClwaterethanol (b) sys-
tems.
ture [8] where the unstable form is usually reported for
waterethanol systems.
SEM micrographs were made on the representing samples
belonging to the points A, B, C in Fig. 4a, at the same time
XRD investigations were also carried out from these sam-
ples. At point A the polymorphic form is mainly (97%),
the particles seem rather dendritic (Fig. 5). At point B the
particles are pure with the signs of aggregation growth
(Fig. 6), while at point C the pure form was produced
(Fig. 7). As a remarkable new experience: it was possible to
produce the pure form by precipitation.
Comparing the precipitation data of glycine and the refer-
ence non-polymorphic sodium chloride, the latter shows def-
inite maximum in the Si d relation. It has to be mentioned
that fine particles (d < 10 m) were produced even from sat-
urated NaCl solutions (Fig. 4b). It is interesting since NaCl is
better soluble in ethanol and longer operational time (60 min)
was allowed for its crystallization. But comparing the growth
abilities of the two substances the linear growth rates are dif-
ferent at least in one magnitude: 108 ms1 for NaCl [5b]
and for glycine 107 ms1 was found [9]. This case well
demonstrates the importance of the material properties.
The experiences with saturated solutions did not result in
acceptable small particle size for glycine and the particles
were more or less agglomerated. Not much improvement
(a) was achieved when some agglomeration retarding additives
were used, i.e. gelatine, polyvinylpyrrolidone, Tween 80 etc.
By our experiences these additives seem to be ineffective in
cases when very rapidly high supersaturation generates in the
Zour [10]. In the region of the lower supersaturations certain
size-reduction effect was observed (6030 m) but not to the
A expected extent. The most significant effect was achieved
by the application of foreign ions, electrolytes. In glycine
precipitation NaCl, CH3 COOK and CaCl2 were tested one
1000 10000 containing 5 g CaCl2 was used as antisolvent resulting in
gel-like precipitates. Unfortunately, these type additives can
hardly be applicable in the food- or pharmaceutical industry.
The real turning-point in our precipitation strategy ap-
(b)
peared when less and less saturated aqueous solutions were
used. Even keeping the ethanol content at 99.6% (see point
F in Fig. 8a) but starting the salting-out precipitation with
more and more undersaturated aqueous solutions (i.e. de-
creasing the values of c0 ), the initial supersaturation, Si will
be also decreasing according to its definition, Si = c0 /ceq
(in this case ceq = constant). With the decreasing initial
solution concentrations the average particle sizes also de-
creased (see points f1 f5 in Fig. 8a) down to 6 m and
100 1000 from other equilibrium points it means another alcohol
contents the tendencies were the same but the average
particle size was over 10 m. In this way, it was possible
to determine experimentally the whole operational areas
both for glycine (Fig. 8a) and sodium chloride (Fig. 8b). In
Fig. 8a and b the upper boundaries of the operational areas
 J. Toth et al. / Chemical Engineering and Processing 44 (2005) 193200 197

Fig. 5. SEM micrograph on the glycine sample (mainly form) precipitated at high alcohol content and initial supersaturation (see point A in Fig. 4a).

represent the saturated initial solutions, while the lowest
ones belong to the less saturated solutions. At the same time
the lowest boundaries show the limits under which there
was no crystallization within the applied operational time
(10 min for glycine and 60 min for NaCl). It means that the
metastable limit must be not at a certain point but along a
rather broad operational area.
By determining the boundaries of the whole operational
field the precipitation area of the expected particle size can
be chosen within these limits. Fig. 8a and b show the pos-
sibilities for the investigated model systems. Regarding the
formation of fine particles (d< 10 m) the following ten-
dencies have been found: in the case of glycine, under the
before mentioned experimental conditions, the desired small
size was available only when undersaturated aqueous solu-
tions (c0 = 3.442.56 g/100 g solution) but high initial su-
persaturations (Si 800600) were applied. Because of the
low solubility at high ethanol content (at 99.6% ethanol ceq
= 0.0041 g glycine/100 g solution) it was possible to cre-
ate high supersaturations even in the originally undersat-

Fig. 6. SEM micrograph on the glycine sample ( form) precipitated at medium alcohol content and initial supersaturation (see point B in Fig. 4a).
198 J. Toth et al. / Chemical Engineering and Processing 44 (2005) 193200

Fig. 7. SEM micrograph on the glycine sample ( form) precipitated at low alcohol content and initial supersaturation (see point C in Fig. 4a).

urated solutions. In the case of NaCl the expected small the so called span see Eq. (3) as the measure of the
size was already obtained from saturated aqueous solution width of the particle size distribution. Having checked all
(c0 = 26.55 g NaCl/100 g solution) also at high supersatu- the particle size analysis data the characteristic value of span
ration (Si = 394). Decreasing the initial solution concentra- changed between 0.9 and 1.6 for glycine and between 0.9
tion down to 7.4 g NaCl/100 g solution (Si = 107, at 99.6%
ethanol content) the average particle size also decreased
down to 1 m.
When the goal is the production of fine particles it is
important to save the primary size of the crystals as they
appear first in the solution, i.e. the nucleation has to be pro-
moted over the growth and aggregation steps. In the case
of undersaturated, weak initial solutions the precipitation
takes place near the metastable region where the kinetic
processes are rather slow. For example, the induction time
which was necessary for crystallization in the weakest NaCl
solutions approaches to 60 min. Repeating the precipita-
tion (where the ethanol content was the same 99.6%) with
saturated aqueous solution there was no measurable induc-
tion time and the particle size changed considerably within
the applied 60 min operational time: the d was 4.41 m
after 10 min, 8.86 m at 20 min and finally16.27 m at
60 min. It is obvious that in the latter case not the small-
est available size was measured after 60 min, but for the
sake of comparison the same operational time had to be
applied.
Beside the average particle size, the size distribution is
also an important parameter of the product. Comparing the
obtained cumulative undersize distribution of the precipi-
tates it has been found that the ratio of the smallest parti-
cles (below 10 m) in the distribution was more than 50%
when d 10 m. In the other cases (d > 10 m) the ratio
was less than 50% and showed decreasing tendency with
increasing size. The Malvern 2600 size-analyzer calculates Fig. 8. The operational areas of the glycine (a) and NaCl (b) precipitation.
 J. Toth et al. / Chemical Engineering and Processing 44 (2005) 193200 199

120 strongest effect on the product particle size as it was ex-

(a) pected. Contrary to the Weimarns precipitation model, fine
Average particle size, d (mm)

100 particles (d < 10 m) were produced only at high supersat-

urations while at the low values (at low ethanol content),
80 where the solubility of the substances were rather high, it
was not possible to get the desired size. The idea to work
60
with weak, undersaturated initial solutions, but then creat-
ing high supersaturation proved to be useful and has led to
40
the determination of the lower metastable boundary of the
20
operational field. Consequently, the finest and agglomera-
tion free particles can be gained near the metastable region.
0 In the other part of the operational field it is possible only if
0.00 0.02 0.04 0.06 0.08 0.10 0.12 the particles are quickly removed from the solution before
Q (gcm-3) they could grow up to their final, stable size and when the
substance has only moderate growing ability like NaCl. Of
120
(b)
course, it means consequences on the available productivity.
Average particle size, d (mm)

100
The productivity of a crystallization process is usually
expressed as the amount of the produced crystals in unit
80 volume of solution and unit time. The maximum available
amounts of the crystals produced from unit volume of solu-
60 tions, Q (g cm3 ) were calculated on the bases of the solu-
bility differences and shown as function of the average size
40 in Fig. 9a for glycine and Fig. 9b for NaCl. Although it is an
approximation especially for the weakest slurries, where the
20 measuring error would be also high, but in other cases it is
acceptable because the measured particle sizes were close to
0 their final, stable size. Taking the produced amounts of both
0 0.02 0.04 0.06 0.08 model materials as well as the operational time (10 min for
Q (gcm-3) glycine and 60 min for NaCl) into account the available pro-
Fig. 9. The relation between the average particle size, d and the maximum ductivities are between 1.66 106 1.82 104 g cm3 s1 ,
available amount of crystals, Q produced in unit volume of solution in or expressed in the usual industrial unit 6654 kg m3 h1 .
the glycinewaterethanol (a) and the NaClwaterethanol (b) systems The relatively low productivity is quite common in colloid
( refers to saturated initial solutions, refer to undersaturated systems or in nanoparticle formation. It is always a question
initial solutions).
of consideration to decide about what productivity is still
acceptable and on what expense, although there is a strong
and 1.8 for NaCl samples. These values are typical for the rule in the crystallization practice: the particle size and pro-
finest precipitates, too. ductivity cant be changed independently of each other.
Summarizing the experiences on the formation of fines In Fig. 10 the operational fields of the two model systems
by salting-out precipitation: the supersaturation has the are plotted together to make a comparison possible and assist
in the explanation of the experimental findings.
120
Average particle size d, (mm)

100 5. Conclusion
NaCl
80 A systematic study of the salting-out precipitation pro-
cess was made to obtain characteristic particle size and pro-
60 ductivity data within wide range of the initial supersatu-
rations. The appropriate supersaturations were created by
40 changing two independent parameters: the initial solution
glycine
metastability concentration and the solubility. The operational time was
20
domain fixed at a given system but in limited cases the effect of
the operational time on the particle size was also investi-
0
gated.
1 10 100 1000 10000
Initial supersaturation, Si () It has been concluded that the salting-out precipitation is
a flexible and acceptable process for producing fines but the
Fig. 10. The operational areas of the two systems studied. material properties (solubility, nucleation and growth ability)
200 J. Toth et al. / Chemical Engineering and Processing 44 (2005) 193200
may considerably limit the possibilities as it is well seen in
Fig. 10.
The supersaturation particle size relations show definite
maximums just following the Weimarns statement in this
respect. The metastable limit below which there is no crys-
tallization within the applied batch time exists along the
lower boundary of the operational field. The finest and ag-
glomeration free particles can be produced at the extremes,
i.e. at the edges of the metastability domain determined by
changing three optional parameters: the initial solution con-
centration, the solubility and operational time. Since the
particle size depends on the relative rate of the nucleation
and aggregation growth kinetic processes the application of
growth retarding techniques (quenching) can be useful. The
polymorphic state may change at the extremes: for example
pure -glycine form was precipitated at low alcohol content
although the form is more typical at salting-out precipita-
tion.
Relatively low productivity can be expected when fines are
produced by salting-out process although it also depends on
the substance properties. The post-crystallization processing
(separation, concentration etc.) of very weak slurries may
go according to the routines of colloid chemistry practice.
Acknowledgements
The authors are very grateful to the University of
Veszprm, Department of Silicate Chemistry and Material
Sciences, Hungary for the XRD and SEM facilities.
Appendix A. Nomenclature
c(r) solubility of the particles of size (radius) r
(kmol m3 )
c* normal equilibrium solubility of the substance
(kmol m3 )
c0 initial concentration of aqueous solutions
(g solute/100 g solution)
ceq solubility of the substance (g solute/100 g solution)
CSD crystal size distribution
d equivalent diameter of the particles (m)
d average particle size (m)
Dx diameter below which x% of the volume of the
particles lies (m)
k Boltzmann constant (J K1 )
n the number of particles of equivalent diameter d ()
Q maximum available amount of crystals produced
from unit volume of solution (g cm3 )
r particle radius (m)
S supersaturation ratio ()
SEM scanning electron microscopy
Si initial supersaturation ()
T absolute temperature (K)
V molecular volume (m3 )
V0 volume of aqueous solutions (cm3 )
Ve volume of ethanol antisolvent (cm3 )
XRD X-ray diffraction
surface energy (J m2 )
References
[1] The 15th Symp., Ind. Cryst., Sorrento, Italy, 2002.
[2] The 8th Conf., Colloid Chem., Keszthely, Hungary, 2002.
[3] P.H. Karpinsky, Maneuvering through the Complexities of Batch
Precipitation, Proc. Int. Symp. Ind. Cryst., Tokyo, Japan, 1998,
pp. 1322.
[4] P.P. von Weimarn, The precipitation laws, Chem. Rev. 2 (1926) 217
242.
[5] (a) J.W. Mullin, Crystallization, third ed., Butterworh-Heinemann,
London, 1993. p. 102.;
(b) J.W. Mullin, Crystallization, third ed., Butterworh-Heinemann,
London, 1993, p. 222.
[6] A. Seidell, Solubilities of Organic Compounds, D. van Norstrand
Company Inc., New York, 1941, p. 124.
[7] H. Stephen, Solubilities of Inorganic and Organic Compounds, Perg-
amon Press, 1964, pp. 102103.
[8] W. Hubig, ber die drei Formen des Glycins, Z. Naturforsch. 13b
(1958) 633638.
[9] S. Halsz-Pterfi, Controlled Precipitation, CSc Thesis, Hung. Ac.
Sci., Budapest, 1994.
[10] N. Garti, H. Zour, The effect of surfactants on the crystallization
and polymorphic transformation of glutamic acid, J. Cryst. Growth
172 (1997) 486498.
 Chemical Engineering and Processing 44 (2005) 201207

Micronized cocoa butter particles produced by a supercritical process

J.-J. Letourneau a, , S. Vigneau a , P. Gonus b , J. Fages a
a Chemical Engineering Laboratory for Particulate Solids, UMR CNRS 2392, cole des Mines dAlbi, Campus Jarlard, 81013 Albi, France
b Nestl Product Technology Centre York, UK

Received 25 July 2003; received in revised form 15 February 2004; accepted 19 March 2004
Available online 24 June 2004

Abstract

A process using supercritical CO2 has been developed to produce form V crystals of cocoa butter. In this process, solid cocoa butter is put
in an autoclave in which supercritical carbon dioxide is then introduced. At high pressure, two liquid phases, one CO2 -rich and one cocoa
butter-rich are formed. Both of them can be sent to a depressurisation nozzle leading to particle formation in an expansion vessel. Moreover,
particles can also be formed in the first autoclave undergoing slower depressurisation. Both particle populations (upstream and downstream
the nozzle) have been characterised using calorimetric method (differential scanning calorimetry, DSC) and environmental scanning electron
microscopy (ESEM). The polymorphism of the particles produced as well as their size and morphology show that they are potentially usable
in the tempering step of the manufacturing process of chocolate.
2004 Elsevier B.V. All rights reserved.

Keywords: Cocoa butter; Supercritical; CO2

1. Introduction
1.1. Polymorphism of cocoa butter and chocolate tempering
Cocoa butter, the fat of chocolate, is responsible for the
pleasant melt in the mouth sensation given by chocolate.
It is a complex mixture of triacylglycerol (TAGs), which
melting points are in the good range (3135 C) to give
this feeling [1].
The most abundant are monounsaturated TAGs, the un-
saturated fatty acid (very often oleic acid) being in the
2-position. 1-Palmitoyl-2-oleyl-3-stearoylglycerol (P-O-S),
P-O-P and S-O-S account for 3638, 1416 and 2328%,
respectively of the total TAGs. It is these TAGs which are
mainly responsible for the behaviour of chocolate when un-
dergoing crystallisation [2].
There are six different ways for these TAGs to arrange
themselves when they crystallise. Six different polymorphs
noted Ithe less stable with the lowest melting point 290 K
(17 C)to VIthe most stable with the highest melting
point 308 K (35 C)can be obtained. The relative amount
of these six polymorphs will condition several features of the
Corresponding author. Tel.: +33 563 493088; fax: +33 563 493025.
E-mail address: letourne@enstimac.fr (J.-J. Letourneau).
URL: http://wwww.enstimac.fr.
end product. If the wrong type of fat is present the chocolate
may exhibit a white powdery surface called fat bloom which
is to be avoided because of its unpleasant appearance. The
most desirable form is known as form V (or 2 ) having a
melting temperature of 305307 K (3234 C). The transfer
to the even more stable form VI is not desirable since it
might lead to fat bloom as do the less stable forms.
The processing conditions determine which poly-
morph is formed. To obtain predominantly form V a
pre-crystallisation step also called tempering has been intro-
duced in the chocolate manufacturing process. It is aimed
at the formation of form V seed crystals which in turn
will force the bulk of the remaining fat to crystallise in
this stable form. Two main ways are used: (i) starting with
hot liquid chocolate 318323 K (4550 C) a succession
of cooling to 300 K (27 C) in order to form both stable
and unstable crystals followed by re-heating to 303305 K
(3032 C) to melt only less stable crystals will lead to form
V crystals which, in turn, will seed the mass of chocolate
and (ii) seeding the mass of untempered chocolate by finely
divided form V crystals of cocoa butter.
1.2. Crystallisation with supercritical processes
Supercritical fluid (SCF) technology presents a new and
interesting route for particle formation, which avoids most of

0255-2701/$ see front matter 2004 Elsevier B.V. All rights reserved.
doi:10.1016/j.cep.2004.03.013
202 J.-J. Letourneau et al. / Chemical Engineering and Processing 44 (2005) 201207
the drawbacks of the traditional methods of milling/crushing
or liquid crystallisation. Supercritical processes are known
to give micro- or even nano-particles with narrow size distri-
bution, and may allow the control of crystal polymorphism.
In addition, several other advantages inherent to these pro-
cesses must be emphasised: they are single-step processes
with simplified downstream processing and they are envi-
ronmentally desirable since they avoid the use of organic
solvents. Moreover, they allow the processing of thermola-
bile molecules and give high purity products [3].
Although most of the applications of supercritical particle
formation are known in the pharmaceutical industry, some
examples can be found with foodstuff. Cocoa butter a rela-
tively high value product has been successfully micronised
using the PGSS (particles from gas-saturated solutions) pro-
cess [4]. In this paper we will investigate the production of
fine cocoa butter crystals in order to use them for the seed-
ing of chocolate.
2. Material and methods
2.1. Experimental apparatus
The experiments were carried out in a versatile pilot plant
(Separex, France). A schematic diagram of the apparatus is
shown in Fig. 1. Carbon dioxide is cooled and stored in a liq-
uid CO2 storage tank (2). It can be circulated at a maximum
flow of 25 kg h1 by a membrane pump (Lewa, Germany)
(3). Compressed CO2 passes then through a heat exchanger
(4) and becomes supercritical. Some of the devices (1, 8, 9,
13) shown in Fig. 1 are used for other experiments such as
SAS or RESS with co-solvent.
The vessel (5) (PARR Instrument, USA) is equipped
with three sapphire windows (16), a magnetic stirrer (15)
Fig. 1. Experimental set-up.
with a maximum torque of 1.8 Nm and a flexible mantle
heater.
The flow coming out this vessel is sprayed into the ex-
pansion vessel (7) by means of a nozzle (6) (Spraydry
SK-SIY80-SKY16, orifice insert diameter: 340 m, Spray-
ing System). The powder formed is collected in a porous bag
made of PTFE fibres. Then, CO2 is depressurised (10) and
can either be directed to the vent or purified and recycled.
2.2. Cocoa butter particle generation
Pieces of solid cocoa butter are loaded in the vessel (5)
equipped with a stirrer and this vessel is filled with CO2 .
Then, the operating temperature and the stirrer speed are
fixed by the controllers. Further addition of CO2 allows
reaching the operating pressure. The mixture thus obtained is
stirred during a long enough period of time (at least 5 mn at
constant pressure and temperature) in order to reach the ther-
modynamic equilibrium. In particular if two liquid phases
are present, each of them is assumed to be saturated.
The second vessel (7) is then filled with CO2 until reach-
ing the second operating pressure. This is the expansion
pressure at which the mixture is placed downstream of the
nozzle. This pressure is fixed during all the experiment with
a back-pressure regulator (10).
The temperature of the CO2 upstream from the stirred
vessel is controlled by the heating device (4) and fixed to
be the same as the temperature of the mixture in the stirred
vessel.
During the experiments the valve between the two ves-
sels is open. The saturated mixture sprayed into the expan-
sion vessel via the nozzle, is subjected to a sudden drop
in pressure, resulting in supersaturated conditions and the
crystallisation of the cocoa butter. This powder is collected
in the bag inside the second vessel. This process is known
 J.-J. Letourneau et al. / Chemical Engineering and Processing 44 (2005) 201207
Fig. 2. Phase diagram of pure CO2 (Peng-Robinsons model [6]).
as RESS process (rapid expansion of supercritical solutions)
[3].
Two major types of experiments have been achieved:
In the first one, the first vessel is let to slowly depres-
surise during its emptying through the nozzle. Therefore,
particles of cocoa butter can also be formed in this vessel.
In the second, the membrane pump (3) feeds the vessel
with CO2 in order to compensate the emptying via the
nozzle and to maintain a constant pressure.
Another refinement in the classification of types of ex-
periments depends on the initial load of cocoa butter. For
small amounts, the mixture appears in a homogeneous single
phase. On the other hand, when the initial amount increases,
two phases are present.
If the temperature is too low, the solid cocoa butter formed
this second phase, but with higher temperatures two fluid
phases are in equilibrium: one is rich in CO2 and the other
is rich in cocoa butter. Kokot et al. [5] showed that at 313 K
and 20 MPa, the mass fractions of CO2 for these two phases
are 0.44 and 0.98. When working with two fluid phases
and without stirring, the results of the experiments depend
on the position of the outlet pipe of the first vessel: top or
bottom. If the fluid is extracted from the top of the vessel,
the cocoa butter powder is produced from the richest phase
in CO2 . Conversely, it is the cocoa butter rich phase which is
extracted from the bottom. If the mixture is homogenised by
203
agitation these two phases are mixed and the resulting fluid
appears (as seen through the windows) to be homogeneous
just like an emulsion.
The drop in pressure undergone by the mixture in the noz-
zle, leads also the temperature to fall and crystals of CO2
may appear (CO2 triple point: T = 216.6 K, P = 0.52 MPa).
To prevent this solid CO2 formation, the pressure P7 in the
expansion vessel must be kept beyond the triple point pres-
sure. But this does not always prevent the CO2 liquefaction
if the temperature upstream the nozzle is not high enough.
As an example shown in Fig. 2 (path A), the final tempera-
ture of pure CO2 at T = 333 K submit to an isenthalpic de-
pressurization from P = 10 to 1 MPa is T = 233 K, which
is a point within the boundary of the liquidvapour region.
The presence of two phasesliquid and gasaround the
particles in formation could change the growth of the crys-
tals and therefore their final morphology. The heating device
(17) is used to prevent the temperature T7 in the expansion
vessel to be lower than the dew temperature of pure CO2 at
P7 . The theoretical path for the experiment #3 of Table 1 is
shown in Fig. 2 (path B). For this experiment boiling water
has been used in the heating device and the final tempera-
ture of the mixture at T = 373 K submit to a depressuriza-
tion from P = 20 to 1 MPa is T = 288 K. As it can be seen
this depressurization is almost isenthalpic.
The flowrate of CO2 is adapted to fit with the operating
conditions. It mainly depends on the characteristics of the
204 J.-J. Letourneau et al. / Chemical Engineering and Processing 44 (2005) 201207

Table 1
Experimental operating conditions
Experiment number Series Diam. (mm) M0 (g) V5 (cm3 ) P5 (MPa) T5 (K) P7 (MPa) T7 (K) Duration (min)
1 E-I 0.34 20.8 2000 20 313 0.6 288 23
2 E-I 0.34 30.6 2000 25 373 0.3 288 28
3 E-I 0.41 18.6 2000 20 318 1 288 22
4 E-I 0.41 36 2000 20 323 0.1 288 21
5 E-I 0.41 19.9 2000 30 313 0.9 288 21
6 E-II 0.41 20.6 2000 30 303 0.1 291 10
7 E-II 1.50 20.2 2000 25 303 0.1 293 6
8 E-III 0.41 19.9 2000 30 313 0.1 289 4
9 E-III 1.50 102.3 2000 31 298 0.1 292 3
10 E-IV 0.34 99.7 2000 20 293 0.7 291 7
11 E-IV 0.34 106.4 2000 27 305 0.4 288 24
12 E-IV 0.41 100.6 2000 20 313 0.2 288 14
13 E-IV 0.41 94.4 2000 32 313 0.3 288 15
14 E-V 0.34 128 1200 20 303 4 283 5
15 E-V 0.34 135 1200 25 308 2 283 5
Diam.: diameter of the nozzle orifice; M0 : initial mass of cocoa butter; V5 , P5 , T5 : volume, pressure and temperature of the tank (5), respectively; P7 ,
T7 : pressure and temperature of the tank (7), respectively.

nozzle and on the pressure difference P5 P7 between the 3. Results and discussion
two vessels. A typical flowrate value for the experiences
listed in Table 1 is about 20 kg h1 . 3.1. Particle formation experiments

The first series of experiments (E-I) was done with small

2.3. Powder characterisation
Differential scanning calorimetry (DSC) has been used to
characterise the samples in a DSC-7 Perkin-Elmer. The DSC
patterns of the samples (25 mg) were obtained between 283
and 323 K at a heating rate of 5 K min1 in a N2 gas stream.
The picture presented in Figs. 4 and 5 have been made
with an Environmental Scanning Electron Microscope
(XL30 ESEM FEG, FEI Philips, The Netherlands).
In order to use these powders for chocolate seeding, a
characterisation of their mechanical and rheological proper-
ties is needed and this will be done in future work.
2.4. Tempering experiments
Chocolate tempering experiments were carried out by
seeding with various type of cocoa butter powder. The de-
gree of pre-crystallisation (temper) was measured using a
computerised tempermeter (Exotherm 7400 from Systech
Analytics). This measurement is basically a cooling curve
where the change in temperature is measured against the
time. The chocolate sample is taken at processing tempera-
ture in a special container and placed in the instrument cell
at 281 K (8 C). In a typical tempering curve, the exothermic
process of crystallisation generates an inflexion point. The
slope at this point is used to determine whether the chocolate
is under-, well- or over-tempered. Usually a well-tempered
curve shows a horizontal plateau. More details on tempering
curve measurements can be found in Bolliger et al. [7].
The process of cocoa butter seeding is usually carried out
at 303 K (30 C) where a mass fraction of 0.52% of ordinary
cocoa butter seeds is mixed with the melted chocolate.
amounts of cocoa butter (single-phase system) and with con-
tinuous feeding in CO2 . The second series (E-II) was done
without feeding and top extraction: particles were collected
in both vessels. The third series (E-III) was done without
feeding and bottom extraction: particles were collected in
both vessels. The fourth series (E-IV) was done with larger
amounts of cocoa butter, continuous feeding and bottom ex-
traction. The fifth and last series of experiments (E-V) was
done with larger amounts of cocoa butter and homogeneous
mixing.
Two typical experiments have been selected for dis-
cussion: the experiment #5 which belongs to the first
series (rapid expansion of a single-phase mixture with
continuous feeding in CO2 ) and the experiment #15 of
the fifth series (rapid expansion of a two-phase stirred
mixture).
In Fig. 3, the DSC patterns for the original cocoa butter
and for experiments 5 and 15 are shown. Original cocoa
butter shows a peak at T = 307.1 K, an end-set point at
T = 309.6 K and a melting enthalpy of hfus = 131 J g1 ,
which are characteristics values for form VI crystals.
The powder obtained in experiment 15 was collected in
the bag of the expansion vessel. Its characteristics are: Tpeak
= 304 K, Tend-set = 306.3 K and hfus = 120.3 J g1 . This
form V powder was white, light, with regular individual
particles and was free-flowing. Nearly all experiments were
achieved with the following conditions: nozzle orifice di-
ameter 340 m, 298 T5 308 K, P5 20 MPa and
0.5 P7 4 MPa led to the same kind of form V powder
collected in the expansion vessel (7). A typical example of
this kind of particle is shown on Fig. 4. It is made of an
agglomeration of smaller particles.
 J.-J. Letourneau et al. / Chemical Engineering and Processing 44 (2005) 201207 205

Fig. 3. Typical DSC curves.

The powder obtained in experiment 5 was collected in particles obtained in this vessel are made of a mixture of both
the vessel (5) and has the following characteristics: Tpeak forms IV and V. The faster the drop in pressure in this vessel,
= 300.5 K, Tend-set = 302.5 K for the first peak and: Tpeak the larger the quantity of form V powder obtained. A typical
= 303.7 K, Tend-set = 305.0 K for the second peak. All the example of this kind of particle is shown in Fig. 5. This

Fig. 4. ESEM picture of a cocoa butter particle obtained with experiment #15.
206 J.-J. Letourneau et al. / Chemical Engineering and Processing 44 (2005) 201207
Fig. 5. ESEM picture of a cocoa butter particle obtained with experiment #5.
powder is generally white, extremely light and sometimes
pieces of original yellow cocoa butter can be found among
the particles.
3.2. Seeding of chocolate
The cocoa butter powder obtained by supercritical process
was able to seed using the following conditions.
At 306 K (33 C) and above, the seeding produced an
under-tempered chocolate; at 304 K (31 C) and below, the
chocolate was quickly over-tempered. However, at 305 K
(32 C), good tempering conditions were obtained with less
than 0.3% of cocoa butter seeds. This is at least 2 C higher
than common seeding and several degrees higher than most
industrial tempering step. Therefore, the chocolate is very
much thinner.
The powder samples described in series E-I and E-IV
(obtained with the smallest nozzle) gave the best results. The
seeding was carried out at 305 K (32 C) with a mass fraction
of 0.10.2% and produced a well-tempered chocolate. At
this high temperature, the stirrer velocity is a critical point
as an excess of mechanical energy could melt the seeds,
but this has to be tested using an industrial scale tempering
machine (batch or continuous).
4. Perspectives and conclusion
The best cocoa powders in terms of appearance, DSC,
reproducibility and shelf-stability are those of the E-V series.
They have not been tested yet to seed chocolate and this will
be part of a future work.
High shear stresses are known to favour the most stable
forms of the cocoa butter. The same phenomenon is probably
also appearing in the supercritical generation of cocoa butter
particles: when the particles are generated in the expansion
vessel with a huge drop in pressure through a small nozzle,
they are exclusively made of form Vwhen generated in the
first vessel with a slow drop in pressure form IV is present
and even predominant. This is a track for further studies.
A major interest in using this new process [8] is the value
of the temperature at which seeding can be done with the
cocoa butter particles supercritically generated. At 305 K
the chocolate is less viscous, this is advantageous in term
of flowability and energy consumption. An additional ad-
vantage lies in the fact that the time needed to obtain a
well-tempered chocolate is shorter (about 5 min). Moreover,
the overall quantity of seeds to temper the chocolate is very
small. Therefore, a scaling-up of this process to an industrial
scale appears feasible.
 J.-J. Letourneau et al. / Chemical Engineering and Processing 44 (2005) 201207 207

References [5] K. Kokot, Z. Knez, D. Bauman, SLG (solidliquidgas) phase

transition of cocoa butter in supercritical CO2 , Acta Alimentaria 28
(1999) 197208.
[1] S.T. Beckett, The Science of Chocolate, RSC Paperbacks, 2000.
[6] D.-Y. Peng, D. Robinson, A new two-constant equation of state, Ind.
[2] M. Lipp, E. Anklam, Review of cocoa butter and alternative fats for
Eng. Chem. Fundam. 15 (1976) 5964.
use in chocolate. Part A. Compositional data, Food chem. 62 (1998)
[7] S. Bolliger, B. Breitschuh, M. Stranzinger, T. Wagner, E.J. Windhab,
7397.
Comparison of precrystallization of chocolate, J. Food Eng. 35 (1998)
[3] J. Fages, H. Lochard, E. Rodier, J.-J. Letourneau, M. Sauceau, La
281297.
Gnration de Solides Diviss par Fluides Supercritiques, Can. J.
[8] J. Fages, J.-J. Letourneau, Procd de production de particules, French
Chem. Eng. 81 (2003) 161175.
Patent FR 2847486 (2002).
[4] Z. Knez, E. Weidner, Precipitation of solid with dense gases, in:
A. Bertucco, G. Vetter (Eds.), High Pressure Process Technology:
Fundamentals and Applications, Elsevier, 2001, pp. 587611.
 Chemical Engineering and Processing 44 (2005) 209214

Food powder handling and processing: Industry problems, knowledge

barriers and research opportunities
John J. Fitzpatrick a, , Lilia Ahrn b
a Department of Process Engineering, University College Cork, Cork, Ireland
b SIK, The Swedish Institute for Food and Biotechnology, Gothenburg, Sweden

Received 26 August 2003; received in revised form 1 March 2004; accepted 19 March 2004
Available online 17 July 2004

Abstract

This paper highlights some of the outcomes from a EU Accompanying Measure focussing on food powders. The main goals of the work
were to outline the major industrial problems, knowledge barriers, and research challenges and opportunities in relation to food powders, and
to promote the creation of a sustainable network of excellence in the area of food powders. Food powders are powders first and foremost,
thus many of the research challenges and opportunities are similar to handling and processing of other powders. As a result, much can be
learned from other industries that deal with powders. What makes food powders different from many other powders is their composition,
which is mainly of biological origin, and that they are eventually consumed by people and animals. Consequently, stability of food ingredient
functionality and prevention of contamination are major issues right through from powder production to final application of the powder, which
is usually in the form of a wet formulation.
2004 Elsevier B.V. All rights reserved.

Keywords: Food powder research; Powder handling; Powder processing

1. Introduction further targeted research initiatives;

1.1. Background and objectives
This paper originates from a 1-year project funded by the
EU Commission through an Accompanying Measure, which
was completed in April 2003. The focus of the project was
to highlight industry problems, knowledge barriers and re-
search and development opportunities in relation to food
powders. It was also envisaged that this would act as a nec-
essary step in fostering the potential for future research ideas
and collaborative research.
The specific objectives of the project were
to outline research and development needs and opportu-
nities in terms of
industrial problems with handling, processing and pro-
duction of safe, high quality food powders that require
Corresponding author. Tel.: +353 21 4903089;
areas requiring further science and engineering knowl-
edge so that this can be applied in the develop-
ment of new and more efficient processes to produce
added-value food powders that are safe and of high
quality;
new and emerging food powder technologies requiring
research and development;
to promote the creation of a sustainable network of excel-
lence in the area of food powders.
The main steps in implementing the objectives of the
project consisted of
holding of a workshop on food powders;
creation of a strategic document for research on food pow-
ders;
dissemination of the output from the project.

fax: +353 21 4270249. This paper summarises some of the key findings from the
E-mail address: j.fitzpatrick@ucc.ie (J.J. Fitzpatrick). strategic document for research in food powders [1].

0255-2701/$ see front matter 2004 Elsevier B.V. All rights reserved.
doi:10.1016/j.cep.2004.03.014
210 J.J. Fitzpatrick, L. Ahrne / Chemical Engineering and Processing 44 (2005) 209214
1.2. Food powders what is special about them
The development of formulation engineering concepts in
food manufacturing and the demand for diversity in food
products has driven a substantial market increase for food
ingredients. Most ingredients are supplied in powdered form
and therefore powder technology is an increasingly impor-
tant issue both to food ingredient manufacturers and food
producers. The two major aspects about food powders that
make them special are biological origin and the fact that
they are ultimately consumed by humans and animals.
The major reason for production in powder form is sim-
ply to prolong the shelf-life of the ingredient by reducing
water content, otherwise the ingredient will be degraded and
broken down in its natural biological environment. So, the
major function of the powder form is to maintain the sta-
bility of the ingredient functionality until it is required for
utilisation, which is usually in some sort of wet formulation.
The major functionalities of food ingredients can be broadly
classified as
physical/chemical: for example, gelation, emulsification,
foaming, pH control;
nutritional: for example, vitamins, nutraceuticals;
organoleptic: for example, colour, taste, smell, texture.
There are a multitude of food ingredients available with a
whole variety of different functions, and there is huge scope
for R&D into investigating new ingredients, and new and
improved functionalities. As the role of the powder form is
mainly to preserve this functionality over time and to deliver
it when required, the focus of this work deals with the pow-
der issues and not the different ingredient functionalities.
The major powder issue with regard to food ingredient pow-
ders is maintaining the stability of ingredient functionality
from production right through to final powder application.
Most food powders are not directly consumed by humans
and animals, they are usually mixed with water and other
liquids to produce wet formulations, which are further pro-
cessed to produce products that are directly consumed. As a
result of being ultimately consumed by humans and animals,
prevention of contamination of the powders with microor-
ganisms, viruses and undesirable chemical components is a
huge issue for food powder manufacturers and the food in-
dustry in general. One major case of food poisoning linked
to a contaminated powder can be very detrimental to a pro-
ducer.
Other powder issues are also important in delivering the
food powder, and these include the ability to handle and
transport ingredient powders, dust problems, dust fire and
explosion hazards, allergy problems, creation of desirable
powder particle properties, and the ability to dissolve these
powders when required. Many of these issues are in com-
mon with other powders and much can be learnt from other
industries dealing with powders.
1.3. Identification of problems, knowledge barriers and
research challenges and opportunities
When looking at the processes involved in going from
powder production to its final application, they can be
broadly divided into the following two categories in Table 1.
Category A processes have the dominant influence in cre-
ating the powder properties that influence the powders
final application. Category B processes may influence the
powder properties by degrading them or by requiring cer-
tain properties to allow effective handling and transport. To
identify problems, knowledge barriers and research chal-
lenges and opportunities, it may be useful to look at these
categories separately and to generically investigate how
research possibilities may come about as a result.
For example, for category A processes, it may be useful
to firstly look at the applications of the powder (how it is
finally applied) and work backwards to the production of
the powder. There are not many food powders that are di-
rectly consumed by the final consumer in powder form only.
Most are incorporated as ingredients into some sort of wet
formulation somewhere along the chain before being finally
ingested by the consumer. As a result, an important area is
research into the technologies and processes involved in ad-
dition and mixing of powders to produce these wet formu-
lations. Allied to this, is the creation of powder properties
that enable these processes to function and that enhance pro-
cess performance. From this, a major area of research is the
investigation of how powder properties influence their roles
in forming wet formulations. Some of these properties may
be difficult to define and measure, thus definition and mea-
surement may become research areas that must be tackled
before effective investigation of how the properties affect
application.
A further step back from application to powder produc-
tion is the processes that create the desired powder proper-
ties. Thus, another major area of research involves the study
of these processes, and investigating how raw material vari-
ables and process input variables affect the engineering of
the desired powder properties. Finally, as mentioned above,
a very important aspect of research for food powders is
component stability, all the way from powder production,
Table 1
Classification of powder processes
A. Processes that give B. Handling and
powder its properties transport processes
Powder production processes Storage (e.g. silos, IBCs, bags, cans)
Drying Transport (e.g. feeders and conveyers)
Comminution Packaging equipment
Crystallisation/precipitation
+ drying
Mixing
Separation
Agglomeration/granulation
Coating/encapsulation
 J.J. Fitzpatrick, L. Ahrne / Chemical Engineering and Processing 44 (2005) 209214
through storage, handling and processing right through to fi-
nal application of the powder. There are also holistic issues
of importance for food powders, including health and food
safety, fire and explosion hazards, energy efficiency and cost
of manufacture especially for basic food powders.
As a result of the above considerations, the following six
sections were created to gain a more focused insight into
typical industrial issues and problems, knowledge barriers
and research opportunities in areas that have importance to
food powders in particular, and which also are relevant to
powders in general
Quality and safety
Powder production processes
Mixing and agglomeration
Other added-value technologies
Powder storage and transport
Characterisation of powders
These areas are explored in detail in the strategic docu-
ment for research in food powders. Some of the major is-
sues that surface from these sections are presented in the
following two sections.
2. Powder issues of practical importance to food
powders
2.1. Maintaining or stabilising the functionality of
ingredients
This is a critical theme for food powder ingredients as
many of them will be added to form wet formulations and
they will be expected to perform. For example, will the egg
white have the same functional performance when dried,
handled and rehydrated as it did before it was dried? How
can polyunsaturated fats be dried, stored and applied with-
out being readily oxidised and producing noxious odours?
What lactic acid culture activity will remain after it has been
dried, stored and rehydrated after 1 year? Thus, research into
maintaining the stability of ingredient functionality during
formulation prior to drying, drying, storage, handling, pro-
cessing and rehydration is key to food powder ingredients.
Some examples of general research topics include
Research into formulation technology prior to drying and
the role of excipients in maintaining and enhancing com-
ponent stability.
Research into the mechanisms of small molecule stabili-
sation of biopolymers during the removal of water during
drying.
Research into improved and alternative drying technolo-
gies that maintain the stability of components (e.g. low
temperature drying).
Research into factors that affect the destabilisation of the
functionality of ingredients in powder form during stor-
age, handling and rehydration.
211
Research into agglomeration, coating, encapsulation, and
barrier technologies, to maintain component stability and
to provide smart composite powders for triggered re-
lease and targeted delivery.
2.2. Prevention of contamination
Contamination of a food powder with undesirable bio-life
forms and chemical components is a major concern due
to direct impact on human and animal health. As a conse-
quence, one high profile case can cause major damage to
an individual company and a whole industrial sector. Be-
cause of its importance, a lot of work has already been done
to prevent contamination. There are plenty of methods and
procedures available to eliminate most contamination risks,
however the key to success requires constant monitoring and
strict implementation of current good manufacturing prac-
tices (cGMP). The EU hygienic engineering design group
(EHEDG) has produced documentation on the hygienic de-
sign of food equipment, including those dealing with dry
particulate solids [2]. Problems still exist with dust forma-
tion leading to powder settling on equipment, which may
provide a growth environment for microbes when contact
with moisture exists.
2.3. Dust prevention and control
Dust generation can lead to a number of problems for food
powders. These can be summarised as follows: (i) health
problems, in particular allergy problems; (ii) contamination
and plant hygiene issues due to dust settling and sticking
onto equipment; and (iii) fire/explosion hazards. As a result,
improvement in dust measurement, prevention and control
is a major issue when dealing with food powders.
Due to their biological origin, there is always the pos-
sibility that a food powder may contain biological active
components, such as enzymes, that may be harmful and
may produce allergy problems. Consequently, this may
eventually lead to the imposition of stricter dust exposure
limits. Self-ignition of food powders that stick onto the
drying chamber is still a big problem. The powder may
be experiencing temperatures below its minimum igni-
tion temperature, however the temperature and moisture
content may be high enough to favour other exothermal
reactions, that may cause the product to heat up over time
until it reaches a temperature that can sustain combustion
reactions. At this stage, it will start to smoulder and then
it can start a fire or a dust explosion. This self-ignition
scenario can also occur in other equipment, such as filter
bag houses. There is a need for a greater understand-
ing of these self-ignition mechanisms and how to prevent
their occurrence. There is also a need for in-line sensors
that can detect volatiles coming from these reactions and
give early warning of self-ignition. Sensors for measuring
carbon-monoxide have been developed and are already in-
dustrially implemented in the dairy industry but sensors for
212 J.J. Fitzpatrick, L. Ahrne / Chemical Engineering and Processing 44 (2005) 209214
measuring other volatiles where CO is not given off are
required.
2.4. Stickiness
Occasionally during spray drying of many different prod-
uct formulations, there may be one formulation whereby the
powder starts sticking onto the wall of the drying chamber
and coating it. In powder mixing where a small amount of
liquid ingredient is sprayed onto a powder, the powder may
stick to the mixing chamber surface and form a crust over
time. Both these examples may appear mundane, however
they cause major problems when they occur, as the chamber
now needs to be cleaned. There is a need for testing that can
predict the likelihood of a powder sticking onto a surface,
and how this is affected by the composition of the pow-
der and processing conditions. Furthermore, there is a need
for determining approaches to overcoming these stickiness
problems.
2.5. Segregation of food ingredient mixes
Many food ingredients are supplied as food ingredient
mixes, which, in some cases, can contain many ingredients.
If the ingredients have different particle properties, in partic-
ular, particle size, this may lead to segregation or demixing
after the mixing process. This problem can seriously limit
the potential for creating ingredient mixes. Industry people
would like to have tools for measuring segregation tendency
and for predicting the maximum difference in particle prop-
erties tolerable before segregation becomes a problem. Fur-
thermore, it is desirable to have a systematic approach to
assist in the development of non-segregating mixes.
2.6. The value of food powders
Food powders are usually perceived to have a low quality
and thus low value. As a result, low cost production drives
powder production and thus there is little investment avail-
able to stimulate innovation and change, so the technology
tends to stagnate and change little over time. From this, there
is a need for investment in powder science and technology in
the food industry to help solve recurring problems, improve
process efficiency and engineer added-value properties into
food powders. Solving problems and improving efficiency
will reduce cost and added-value properties will increase the
value of food powders. Some of this cost saving and addi-
tional revenue can then be ploughed back in to continue the
cycle of product and process innovation. Simultaneously,
there is a challenge for scientists and engineers to continue
to demonstrate that powder education and R&D in the food
industry is worth the investment.
Powder storage and transport operations are perceived not
to add value to the powder, thus there is little investment
into dealing with and solving storage and transport prob-
lems. Processors appear to simply live with the problems
and do not appear to realise that solving these problems can
save a lot of headache and more importantly, bottom-line
cost. A major underlying problem is education insofar as
many working in the food industry do not have an adequate
knowledge of powder science and technology to be able to
apply it effectively in the food industry.
2.7. Particle engineering
This involves the engineering of food powder ingredient
particles that can maintain ingredient stability and function-
ality, that do not present dust formation problems, that do
not present difficult handling and transport problems, and
can be readily rehydrated to deliver their required functional
performance in a wet formulation. The major limitation to
this type of approach is the low value of food powders and
the low cost driven nature of food powder product that stunts
this type of innovation. On the other hand, particle engineer-
ing is the key to the higher added value of food powders,
e.g. by producing products with excellent instant properties.
3. Issues of importance to powders in general,
including food powders
Food powders are powders first and foremost, thus many
issues that are relevant to powders in general, are also very
relevant to food powders. These are summarised as follows.
3.1. Education
Most of the industry people dealing with food powders
come from a food science and technology background or an
engineering background. Powder science and technology is
seldom taught in engineering courses or food science and
technology courses. The subject is too important in the prac-
tical life of an engineer or scientist for this to be omitted
from curricula. Thus, there is a need for incorporation of
powders into existing undergraduate and postgraduate edu-
cation. The world of particle and powder technology should
be as important as fluids.
The educational scenario is more complex for practic-
ing engineers and scientists. For those wishing to obtain a
greater knowledge of powders, there are a number of short
courses available conducted mainly by industry consultants
and some academic institutions. The question is: Is this suf-
ficient or is there a need for more educational opportunities?
There may be potential for a European or even world-wide
distance learning programme dealing with aspects of pow-
der science and technology. Allied to this is the problem that
many practising engineers and scientist do not realise their
deficiency in powders or are simply not aware of the edu-
cational opportunities or are not given the time and support
to pursue them. This can only be overcome by promoting a
greater awareness of the importance of particle and powder
 J.J. Fitzpatrick, L. Ahrne / Chemical Engineering and Processing 44 (2005) 209214
science and technology, which is definitely making progress
with time.
3.2. Transfer of existing knowledge into application in
industry
Much more attention should be paid to transfer of exist-
ing knowledge from universities, research centres and con-
sultancies into industry. In most cases, knowledge transfer,
mainly in the form of innovative equipment and system de-
sign, is introduced to industry via powder equipment sup-
pliers. Consequently, there is potential for greater synergy
between these suppliers and centres of research.
Non-multinational companies/SMEs, in particular, can
use much more knowledge to escape from the trial and error
approach. There is an enormous amount of wasted time and
effort because the same mistakes and problems appear time
and time again throughout the process industries. There is
a lot of wisdom scattered about that can be applied to solve
many of these problems and prevent many of these mistakes.
However, the limiting steps are education and knowledge
transfer, although the situation is gradually improving with
time and with more and more individuals promoting the
importance of particle/powder science and technology.
3.3. Standardisation of powder property definition and
measurement
In the world of particles and powders, there are many
properties and there is a multitude of property definitions
and measuring techniques, which may even vary from one
industry to another. There is a need to standardise the defi-
nition and measurement of these properties. This will help
streamline definitions and measurement techniques to those
that are most applicable. It will help in obtaining greater
reproducibility and also enable greater comparison of work
undertaken by different groups.
3.4. Fundamental science, modelling and design
Description and modelling of the fundamental mecha-
nisms that can be applied in the design of most powder pro-
cesses is still very much at an early stage. It is comparable
to the pre McCabe-Thiele era in distillation technology. This
is particularly true for many common operations like dry
powder mixing, mixing of powders and liquids, granulation
and coating. Many of them have similar mechanisms taking
place, such as particle motion and particle/liquid contact,
thus there is potential for cross-fertilisation. There is much
scope for progress in this area. Modelling techniques, such
as DEM and CFD, are trying to model particle/particle and
particle/fluid motions and interactions, however it may take
many years before these techniques are useful for design in-
volving real powders, and some are dubious whether they
will ever be useful for real cohesive powders. The first step
in the modelling of spray-drying has already been taken.
213
In addition, there is potential for looking at alternative
mechanisms that may result in more effective ways of doing
things that could lead to the development of new equipment
based on these new concepts. For example, the concepts
behind most present-day dry mixers have been around for
decades. Maybe, there are alternative approaches, such as
the application of chaos in mixing that could lead to more
effective mixing with less energy requirement. Similar com-
ments could be made for other processes, such as granula-
tion, whereby a greater understanding of the fundamental
science may lead to alternative approaches to granulation
process design.
3.5. On-line and in-line monitoring and control
This is very much in its infancy as applied to powder pro-
cesses, and as a result, there is a lot of scope for work in
this area. The advantages of improved monitoring and con-
trol are early warning of problems, reduced costs and more
consistent product quality. Presently, there are a number of
sensors that have potential for use on- or in-line, however
there is scope for the development of on-/in-line versions
of existing sensors and totally new sensors. The major chal-
lenges for the application of existing sensors are either one
or a combination of the following: (i) reducing sensor cost;
(ii) improvement of sensor robustness for real-time opera-
tion in a powder environment, and (iii) development of data
analysis technology that can filter the sensor output and con-
vert it into useful information for control.
3.6. Nanoparticle technology
Nanotechnology is gaining more and more popularity
throughout industry. As a result, nanoparticle technology
represents a new promising field of application to food pow-
ders, although current research and development is very
much in its infancy.
3.7. Energy efficiency
Most powder processes, and in particular drying and com-
minution, are not very energy efficient. An improvement in
the fundamental science, modelling and application to de-
sign coupled with improved on- and in-line monitoring (e.g.
model predictive control) will give opportunities to improve
energy efficiency while ensuring product quality.
Acknowledgements
We would like to acknowledge the five group leaders
in the project for their valued contributions: Prof. Koen
Dewettinck, Ghent University, Belgium; Richard Farnish,
Wolfson Centre for Bulk Solids Handling Technology, UK;
Prof. Peter Lillford, University of York, UK; Dr. Sivert Ose,
POSTEC, Norway; and Prof. Karl Sommer, TU-Munich,
214 J.J. Fitzpatrick, L. Ahrne / Chemical Engineering and Processing 44 (2005) 209214

Germany. We also acknowledge Prof. Fernanda Oliveira, References

UCC Ireland and Prof. Thomas Ohlsson, SIK Sweden for
their assistance and support during the creation and im-
plementation of the project. Finally, we acknowledge all
those who participated in the workshop and the many oth-
ers who made contributions to the creation of the strategic
document.
[1] Strategic document for research in food powders. http://www.
foodpowders.net, March 2003.
[2] G. Hauser, K. Mager, R.R. Maller, K. Masters, G.M.H. Meesters, W.
Rumpf, G. Schleining, EHEDG document no. 22, general hygienic
design criteria for the safe processing of dry particulate materials,
Trends Food Sci. Technol. 12 (2002) 296301.
 Chemical Engineering and Processing 44 (2005) 215219

Microencapsulation of particles using supercritical carbon dioxide

H. Krber , U. Teipel
Fraunhofer Institute for Chemical Technology, Joseph-von-Fraunhofer-Str. 7, D-76327 Pfinztal, Germany

Received 1 August 2003; received in revised form 25 December 2003; accepted 11 February 2004
Available online 19 June 2004

Abstract

In this contribution a novel fluidized-bed coating process is introduced to encapsulate heat-sensitive materials with particle sizes below
100 m. Supercritical carbon dioxide is used as solvent for the coating material as well as carrier fluid for the core material. The behaviour
of the high pressure fluidized-bed was investigated for different process parameters and materials. It is shown that the fluidization starts at
lower fluid velocities if the pressure is increased and it was possible to fluidized particles with a mean size below 10 m. The coating of glass
beads with stearyl alcohol was carried out and layers with a thickness of 18 m were achieved.
2004 Elsevier B.V. All rights reserved.

Keywords: Microencapsulation; Fluidized-bed; Supercritical; Fluids; RESS process

1. Introduction
Solid particles are encapsulated for use in various prod-
ucts, including pharmaceuticals, processed food and fer-
tilizers, and energetic materials. To encapsulate particles,
a fluidized-bed coater equipped with a spray nozzle such
as a Wurster Coater is commonly used [1]. Conventional
coating processes involve layering a liquid coating mate-
rial usually in the form of atomized droplets through the
nozzle onto a fluidized-bed of particles. After the atomized
droplets of coating material come into contact with the
core particles, the solvent must evaporate quickly. If excess
droplets are present in the bed, agglomeration takes place,
leading to particle growth and defluidization. Therefore,
proper droplet atomization and drying are essential to avoid
unplanned agglomeration of solid particles. In case of fine
particles (<100 m), however, agglomeration becomes a
serious problem because of their strong cohesive forces.
The feed liquid acts as a binder for the fine particles.
A novel fluidized-bed coating process using the rapid ex-
pansion of supercritical solutions (RESS) is described for
the encapsulation of fine particles [2,3]. This process ex-
ploits the capability of supercritical fluids to act as a selec-
tive solvent. Supercritical fluids are noteworthy in that their
Corresponding author. Tel.: +49 7214 640 350;
fax: +49 7214 640 581.
E-mail address: hartmut.kroeber@ict.fraunhofer.de (H. Krber).
density is like that of a fluid (thus they are effective solvents)
while their viscosity and diffusion coefficient are like that
of gases (thus they exhibit good mass transport behaviour).
These properties make supercritical fluids effective as an
extractant. Via the RESS process, particles and droplets in
the submicron range can be produced by atomizing a super-
critical solution through a nozzle. These fine droplets make
it possible to improve the deposition onto the particles, so
that a thinner and more even coating layer can be achieved
[4,5]. In addition, because the cohesion and adhesion forces
are small compared to organic solvents, the capillary forces
are considerably reduced, making it possible to encapsulate
particles smaller than 100 m without forming agglomer-
ates. In this process, the supercritical fluid is used both as
solvent for the coating material and as carrier fluid for the
formation of the fluidized-bed.
2. Experimental
2.1. High pressure fluidized-bed device
The experiments were conducted in a laboratory scale de-
vice. A schematic view of the high pressure fluidized-bed
coating apparatus is shown in Fig. 1. The apparatus consists
essentially of three subsections: the CO2 supply, the extrac-
tor (in which the CO2 becomes saturated with coating) and
the high pressure reactor (where the coating takes place).
The maximum allowable pressure in the device is 30 MPa at

0255-2701/$ see front matter 2004 Elsevier B.V. All rights reserved.
doi:10.1016/j.cep.2004.02.014
216 H. Krber, U. Teipel / Chemical Engineering and Processing 44 (2005) 215219

Table 1
Properties of the used core materials
Material Density (kg/m3 ) Mean particle size (m)

Glass beads 1 2485 7.39

Glass beads 2 2483 69.9
Glass beads 3 2487 124.6
Glass beads, hollow 1124 13.4

Fig. 2. Phase equilibrium of the system stearyl alcohol-CO2 for different

concentrations (c = mst /mCO2 ).

of the stearyl alcohol was 328 K. The different glass beads

Fig. 1. A schematic view of the high pressure fluidized-bed apparatus.
a maximum temperature of 373 K. The reactor has a volume
of 500 cm3 and an internal diameter of 60 mm. The diameter
can be reduced by different glass tubes. The fluidized-bed
can be observed through two windows in the high pressure
vessel.
In each experiment, the fluidized-bed reactor was filled
with the core material (20 g). The bed was fluidized by
adding pure carbon dioxide via the line identified in Fig. 1
as carrier fluid. The reactor floor consisted of a perfo-
rated plate. When the reactor reaches thermal stability, the
microencapsulation process is initiated. CO2 flows through
the extractor, saturated by the coating material, and then ex-
pands through a nozzle into the fluidized-bed reactor. The
expansion produces a fine aerosol of droplets that deposit
onto the core particles, spread out as a film and then harden,
yielding a solid layer of coating on the particles. At the end
of each experiment the connection between the extractor and
nozzle was flushed by opening the bypass, at which time the
product was subjected to post-drying.
2.2. Materials
The first experiments were conducted using stearyl alco-
hol (C18 H38 O, obtained from Merck) as the coating material
and different core materials (see Table 1). The melting point
have a very high sphericity.
The phase equilibrium of the binary system stearyl
alcohol/CO2 was measured in a high pressure view cell (own
design) according to the synthetic method by Teipel et al.
[6] and presented in Fig. 2. The curves represent the border
between the homogeneous and two phase region; above the
curve a homogeneous solution is present, below two phases
exist. The measurements were carried out for different con-
centrations of stearyl alcohol in CO2 represented by the
different curves. It is seen that 2.1 wt.% of stearyl alcohol
are complete soluble in sc-CO2 at 333 K and 20 MPa.
3. Calculation of layer thickness
For smooth encapsulation of monodisperse and spherical
particles, the layer thickness Sw as a function of process pa-
rameters can be calculated according to the following equa-
tion.

 
t P mtw
Sw = rP,0 3
1+ 1 (1)
w mP,0
where P and w are the density of the encapsulated par-
ticles and the density of the coating material, respectively,
rP,0 the radius of the core particles and mP,0 the total mass
of the fluidized-bed material [4]. The total mass mw of the
deposited encapsulation substance can be calculated as func-
tion of the collection efficiency Eg of the fluidized-bed and
as a function of time:
 t
t
mw = Eg mw dt (2)
0
 H. Krber, U. Teipel / Chemical Engineering and Processing 44 (2005) 215219 217

Assuming homogenous expansion of the fluidized-bed the for the calculation of the single particle collection efficiency
collection efficiency Eg may be calculated as a function of may be found in literature [4,15,16].
process parameters, for example the size of the collector
particle dP or the height of the fluidized-bed z as well as
the effective single particle collection efficiency and the 4. Results
porosity of the fluidized-bed :
  4.1. Fluidization
mdep min mout 3 1
Eg = = = 1 exp z
min min 2 dP Before starting with the microencapsulation the fluid dy-
(3) namic of the high pressure fluidized-bed was investigated.
Fig. 3 shows the porosity (bed voidage; = Vspace /Vtotal
The collection efficiency Eg can be determined by mea- = 1Vparticle /Vtotal ) of the fluidized-bed (glass beads 3 are
suring the total mass of the coating deposited on the core used) at different pressures (corresponding to different fluid
particles (mdep ) by thermal gravimetric analysis (TGA) and densities). As expected the porosity increases if the fluid ve-
calculating the total mass of the coating reaching the reactor
(min ) from the phase equilibrium in the extractor (pextr , extr )
and from the mass flow of CO2 through the nozzle. This
mass flow rate can be calculated [7] and depends strongly
on the conditions before the nozzle (pN , N ) and the nozzle
diameter (dN ).
A successful coating process requires that the encapsu-
lation substance adheres sufficiently to the surface of the
fluidized-bed material. Of course, the adhesion also affects
the collection efficiency. In literature [4,8,9], the impact of
adhesion to the coating process is taken into consideration
by introduction of the adhesion coefficient . Thus effective
single particle collection efficiency may be formulated:
Fig. 3. Bed voidage of glass beads (mean particle size: 124.6 m) vs.
= (4) fluid velocity at different operating pressures (R = 328 K).

Herein, is the single particle collection efficiency. Ac-

cording to literature [9] for liquids or very soft substances
the adhesion coefficient has a value near to one. Another
study indicates that for solid aerosols having a diameter in
the range of 0.7 m the adhesion coefficient may vary from
0.001 to 0.01 [4].
There have been extensive studies focusing on the cal-
culation of the single particle collection efficiency [10,11].
Generally, the calculation of the single particle collection
efficiency requires the calculation of the aerosol stream line
around the particle. Therefore transport mechanisms like dif-
fusion and inertia as well as deposition mechanisms like in- Fig. 4. Influence of pressure on fluidization velocity for different glass
terception have to be taken into account. For example, the beads (R = 328 K).
stochastic random motion of aerosol particles is likely to in-
crease the probability of aerosol deposition on the collector.
According to the literature [12] diffusion is the dominating
deposition mechanism for aerosols having diameters below
0.1 m at a flow velocity below 0.1 m/s. Studies indicate
that in most cases the different deposition mechanism are
independent of each other [13,14]. Hence, the single particle
collection efficiency may be calculated by adding the sin-
gle particle collection efficiency maintained for the different
transport and deposition mechanisms.
=D + R + T + E (5)
Fig. 5. Thickness of the coating in dependency of coating time for different
where the indices D, R, T and E indicate diffusion, inertia pre-expansion pressures (d50 = 70 m, pR = 8 MPa, R = 328 K, N
as well as interception and electrostatic forces. Correlation = 353 K, dN = 100 m).
218 H. Krber, U. Teipel / Chemical Engineering and Processing 44 (2005) 215219
Fig. 6. SEM pictures of untreated glass beads (left) and with stearyl alcohol coated particles (right).
locity is increased. At low velocities a sharp increase of the
porosity is seen. This point represents the transition between
the fixed-bed and fluidized-bed regime. The minimum ve-
locity which is necessary to fluidize the particles decreases
if the pressure is risen. At 2 and 8 MPa this velocity is 8 and
2.9 cm/s, respectively.
The fluidization of smaller particles started at lower
velocities and the influence of the pressure on the min-
imum velocity was reduced. Using the glass beads 1
this velocity was 0.95 cm/s at 2 MPa and 0.67 cm/s at
6 MPa. Fig. 4 shows the influence of pressure on the
minimum fluidization velocity for the glass beads 1
and 2.
4.2. Microencapsulation
In first coating experiments glass beads with a mean par-
ticle size of 70 m were encapsulated with stearyl alcohol.
The layer thickness was measured at different coating times
by thermal gravimetric analysis. Fig. 5 presents these re-
sults for different pressures before the nozzle. The condi-
tion inside of the fluidized-bed reactor were constant (R
= 328 K, pR = 8 MPa). As nozzle we used a 100 m Laval
nozzle which was heated up to 353 K. The extraction condi-
tions were equal to them at the nozzle (353 K; 1619 MPa).
The velocity of the carrier fluid (carbon dioxide) through the
fluidized-bed was twice of the minimum fluidization veloc-
ity. The coating thickness varies from 1 to 8 m, depend-
ing on the coating time, and that, at long times, the thick-
ness approaches a constant value. It is seen that at higher
pre-expansion pressure the layer thickness increases. One
reason for that is the higher solubility of stearyl alcohol
if extraction pressure increases so that more coating mate-
rial reaches the reactor. This tendency is supported by an
increase of the mass flow rate of carbon dioxide through
the nozzle at elevated pressures. Furthermore, Krber et al.
showed that the droplet size of the aerosol of stearyl alcohol
behind the nozzle is independent of the pressure before the
nozzle [7].
SEM pictures of the encapsulated particles show that the
coating layer is uniform and smooth. In Fig. 6 coated glass
particle is presented and compared to untreated material.
5. Conclusions
A new fluidized-bed coating process (particle encapsula-
tion with supercritical fluids; PESF process) was introduced
to encapsulate fine and heat-sensitive particles. It was shown
that the fluidization of particles under sub- or supercritical
conditions is different from that under atmospheric pressure.
With increasing pressure the minimum fluid velocity which
is necessary to start the fluidization decreases. It was possible
to fluidize glass beads with a mean particle size of 7.4 m.
Stearyl alcohol was used for first coating experiments with
glass particles of a mean particle size of 70 m. Complete
coatings with a layer thickness between 1 and 8 m can be
achieved depending on the coating time and process condi-
tions.
Further investigations of the particle coating in a high
pressure fluidized-bed will be carried out. Especially, vari-
ations of the particle and coating material as well as coat-
ing conditions like pressure and temperature will be per-
formed. The encapsulation of particles smaller than 10 m
with very thin layers without agglomeration is a challenge of
this project. Therefore, the hydrodynamic of the high pres-
sure fluidized-bed has to be characterized by experiments as
well as calculations.
An estimation of the single particle collection efficiency
should be possible by measuring the collection efficiency
Eg and the layer thickness. Using this estimation the time
which is necessary to obtain a particular layer thickness can
be predicted.
 H. Krber, U. Teipel / Chemical Engineering and Processing 44 (2005) 215219 219

References [8] E. Schweers, Einflu der Filterstruktur auf das Filtrationsverhal-

ten von Tiefenfiltern, PhD-Thesis, University Karlsruhe, Germany,
1993.
[1] D. Wurster, Air-suspension technique of coating drug particles, J.
[9] K.V. Thambimuthu, Gas filtration in fixed and fluidized beds,
Am. Pharm. Assoc. 48 (1959) 451454.
PhD-Thesis, University Cambridge, UK, 1980.
[2] D.W. Matson, J.L. Fulton, R.C. Petersen, R.D. Smith, Rapid expan-
[10] F. Hhner, Trgheitsbedingte Abscheidung von Aerosolpartikeln an
sion of supercritical fluid solutions: solute formation of powders,
Einzelkugeln und Kugelverbnden Ph.D.-Thesis, University Kaiser-
thin films and fibers, Ind. Eng. Chem. Res. 26 (1987) 22982306.
slautern, Germany, 1995.
[3] H. Krber, U. Teipel, H. Krause, Manufacture of submicron particles
[11] W.E. Ranz, J.B. Wong, Impaction of dust and smoke particles on
via expansion of supercritical fluids, Chem. Eng. Technol. 23 (2000)
surface and body collectors, Ind. Eng. Chem. 44 (1952) 6.
763765.
[12] K. Lee, B. Liu, Theoretical study of aerosol filtration by fibrous
[4] M. Niehaus, Mikroverkapselung von Partikeln in einer Wirbelschicht
filters, Aer. Sci. Technol. 12 (1981) 79.
unter Verwendung von verdichtetem Kohlendioxid, Ph.D.-Thesis,
[13] R. Pfeffer, E. Gal, An experimental evaluation of the rotating fluidized
University Karlsruhe, 1999, Wissenschaftliche Schriftenreihe des
bed filter, In: Proceedings of the World Filtration Congress III,
Fraunhofer ICT, Band 23.
Dowington, PA, USA, 1982.
[5] A. Tsutsumi, S. Nakamoto, T. Mineo, K. Yoshida, A novel
[14] E. Schmidt, Theoretische und experimentelle Untersuchungen zum
fluidized-bed coating of fine particles by rapid expansion of super-
Einflu elektro-statischer Effekte auf die Nassentstaubung, J. Air
critical fluid solutions, Powder Technol. 85 (1995) 275278.
Pollut. Control Assoc. 28 (1978) 2, 143.
[6] U. Teipel, P. Gerber, H. Krause, Characterization of the phase equi-
[15] F. Lffler, Staubabscheiden, Georg Thieme Verl., Stuttgart, first ed.,
librium of the system trinitrotoluene/carbon dioxide, Propellants Ex-
1988.
plosives Pyrotechnics 23 (1998) 8285.
[16] M.M. El Halwagi, Mathematical modeling of aerosol collection in
[7] H. Krber, U. Teipel, Mikroverkapselung feiner Partikel in einer
fluidized bed filters, Aer. Sci. Technol. 13 (1990) 102.
berkritischen Wirbelschicht, Chem. Ing. Technol. 74 (2002) 1715
1717.
 Chemical Engineering and Processing 44 (2005) 221224

Plasma spheroidization of ceramic particles

Z. Kroly , J. Szpvlgyi
Institute of Materials and Environmental Chemistry, Chemical Research Center,
HAS, P.O. Box 17, Budapest 1525, Hungary

Received 30 August 2003; received in revised form 18 December 2003; accepted 11 February 2004
Available online 19 June 2004

Abstract

Formation of spherical, micron-sized ceramic particles was investigated in an RF thermal plasma reactor. It has been concluded that a
wide size distribution of feedstock powders gives rise to either excessive evaporation of smaller grains or insufficient melting of bigger ones.
Ceramic spheres with more or less voids inside can also be prepared starting from powders pretreated in special ways.
2004 Elsevier B.V. All rights reserved.

Keywords: Plasma reactor; Ceramic particles

1. Introduction 2. Experimental

Spherical particles proved to be superior in several ap-
plications owing to their favorable properties. Thus, they
are used in thermal spraying for their excellent flowabil-
ity, in powder metallurgy because of their excellent repro-
ducibility in manufacturing parts with controlled porosity
and as a filler material, as well. Metal microspheres can
be easily produced by melt atomization. Similar method
in the case of ceramics is impractical. Micron-sized ce-
ramic particles, however, can be smelted by thermal plas-
mas that provide exceptional conditions for spheroidization
due to its high temperature. In terms of purity and resi-
dence time of the particles in the hot temperature core, RF
plasmas provide better conditions as compared to arc plas-
mas.
The present report describes the preparation of spherical
ceramic particles of different kind and of different size in an
inductively coupled RF thermal plasma reactor. The effect
of plasma gases and the type of the feedstock materials on
the microstructure of the formed particles was studied in the
experiments.
Corresponding author. Tel.: +36 1 325 7933x415;
fax: +36 1 325 7892.
E-mail address: karoly@chemres.hu (Z. Karoly).
The experimental apparatus consisted of a TEKNA-type
induction plasma torch (PL-035LS) with a quartz confine-
ment tube of 25 mm and a water cooled steel chamber con-
nected to a cyclone. The plasma plate power of 21 kW was
provided by a four turn, water cooled induction coil from an
RF generator operating at an oscillator frequency of 3 MHz.
High purity argon was used both as plasma and sheath gas
with flow rates of 20 and 60 l min1 , respectively. In order
to raise the low enthalpy and heat conductivity of the argon
plasma gas, hydrogen was also mixed into the sheath gas
with a proportion of 10% (v/v).
The starting powders were fed with a PRAXAIR powder
feeder through an injection probe to the top of the plasma
flame by argon carrier gas (3 l min1 ) with a constant feed
rate of 10 g min1 . The outlet of the quartz injection tube of
an inner diameter of 2 mm, was located 10 mm below the
top of the induction coil.
Commercial alumina (ALO-Ex30 and ALO-GB1 from
AJKA Alumina Co. Ltd., Hungary), aluminum-hydroxide
and silica powders were investigated. Impurity levels of the
powders were below 0.5%. In some tests, minor amount of
NaNO3 and Na-silicates were, however, blended into the
silica aggregates. The sodium-silicates were used as binding
agents of primary particles, while NaNO3 was applied to
release gas (O2 and N2 ) at temperatures higher than 720 C.
Both the raw materials and the produced powders were
characterized for density, particle size distribution, specific

0255-2701/$ see front matter 2004 Elsevier B.V. All rights reserved.
doi:10.1016/j.cep.2004.02.015
222 Z. Karoly, J. Szepvlgyi / Chemical Engineering and Processing 44 (2005) 221224
surface area, phase composition (XRD) and microstructure
(SEM). Density was determined by means of picnometry.
The particle size distribution by volume of grains was de-
termined by a Malvern Master Particle Sizer. The specific
surface areas were measured with a conventional volumetric
instrument by BET method. A Philips Xpert XRD appara-
tus was used for X-ray diffraction measurements. Standard
X-ray powder techniques were used with Cu K radiations
generated at 40 kV and 30 mA. The microstructural (SEM)
investigations were performed on a JEOL JSN50A appara-
tus. Cross-section images of the particles were made by em-
bedding them into resin that was followed by their cutting
by diamond blade.
3. Results and discussion
According to the scanning images of particles spheroidiza-
tion took place effectively (compare Figs. 1 and 2). A
spheroidization rate as high as 90% could be achieved.
It was previously shown [1] that increasing feed rate en-
tail declining spheroidization efficiency. This phenomenon,
however, could not be experienced with increasing mean
Fig. 1. SEM image of alumina feedstock particles.
Fig. 2. SEM image of alumina particles after spheroidization.
particle size of the feedstock powder. Besides spheroidiza-
tion melting usually caused an increase in density that fell
close to the theoretical one ( = 2.32 g cm3 ) at silica pow-
ders. However, the higher was the size of the feedstock par-
ticles, the density of products remained increasingly below
the theoretical one. This can be attributed to the fact, that
above a certain size particles have not got enough time to
be melted on the whole. Thus, the core remains unaffected.
Considerable changes in the size of silica after processing
cannot be observed (Fig. 3), whereas the specific surface
area of products was increased and it was higher with de-
creasing mean particle size of the feedstock. Particles below
the mean size are prone to be evaporated completely to be
condensed again in the nanosize range in the cooler parts
of the plasma reactor. These nano-sized particles usually
cover the surface of bigger ones. Decreasing the power or
reducing the residence times would decrease evaporation
rate, however, it would also jeopardize the spheroidization
of bigger particles.
Spherical alumina can also be formed from commercial,
low cost aluminum-oxides or even from aluminum-hydr-
oxides. In the latter case energy of the plasma should provide
not only the enthalpy of melting but that of dehydration and
subsequent phase transformations of alumina as well. Under
the aforementioned conditions particles below 45 m have a
good chance to be spherodized. Presumably the wide particle
size distribution of starting gibbsite powder accounts for the
less spheroidization rate of 70%.
Application of agglomerated alumina powders as feed-
stock material gives rise to spheroidization too. Density of
product particles falls much below the theoretical one (th ,
Al2 O3 = 3.95 g cm3 ), while significant increase in the grain
size can be observed in case of alumina (Fig. 4). Scanning
images of these powders, however, show they mainly con-
sist of hollow particles. Formation of hollow particles and
the size enlargement can be attributed to the expansion of
gases located in the interior of melted droplets. The gases
Fig. 3. Main characteristic of silica product powders (S denotes specific
surface area, represents apparent density).
 Z. Karoly, J. Szepvlgyi / Chemical Engineering and Processing 44 (2005) 221224 223

Fig. 4. Comparison of the particle size distribution of feedstock material (straight line) and product (dashed line) ( represents apparent density).

are hindered from leaving after the development of a coher-

ent outer melted layer. The final size of these cavities are de-
termined by the freezing temperature of the oxide melt that
is approximately TF = 0.8TM [2,3], where TM is the melt-
ing point of particular oxide. The overpressure in the voids
can be calculated by this formula: P = (2/r), where
denotes the surface tension of melt and r denotes the radius
of the hollow. From the above formulas it can be calculated,
that the molar quantity of gas in the resulted voids are in the
same order of magnitude than that of the gas in the porous
structure of the starting grains. It is the case if the raw ma-
terial is made from primary particles of several microns by
spray drying or agglomeration. When the goal is to prepare
hollow spheres some gas-blowing agents can be blended to
the grains in order to provide substantial gas volumes. These Fig. 6. SEM image of a hollow microsphere of thick wall.

supplementary materials facilitate bubble generation by re-

leasing gas at a certain temperature by decomposition. This
process was applied to make hollow silica particles in ex-
periment 4 (Fig. 5). Density of particular powder is a mean
value that does not characterize each particle. The wall of
individual particles may differ in considerable extent, as this
can be recognized by comparing Figs. 5 and 6. While the
density of particles having thin shells is less than 1 g cm3 ,
that of the solid particles can reach even the theoretical value.

4. Crystallinity

Freezing of the melted particles takes place rapidly be-

Fig. 5. SEM image of a hollow microsphere. cause of the extremely high temperature gradients being
224 Z. Karoly, J. Szepvlgyi / Chemical Engineering and Processing 44 (2005) 221224
Fig. 7. XRD pattern of spheroidized alumina powders.
characteristic to thermal plasmas. According to XRD pat-
terns, majority of silica particles were transformed into
amorphous state on plasma treatment, only 12% remained
quartz.
Several papers report [4] that liquid alumina solidifies not
in the thermodynamically most stable phase of -Al2 O3 ,
but rather in the form of -Al2 O3 . This is attributed to the
fact that the solidified phase structure is basically deter-
mined by the relative critical free enthalpies of nucleation of
alternative crystal structures. Consequently, not surprising,
that considerable part of spheroidized particles composed
of -Al2 O3 and other metastable phases (such as , ) of
alumina (Fig. 7). The latter were formed from the phase
according to the usual route of phase transformation on cal-
cinations, assuming that the temperature history of given
particle makes it possible. For bigger particles (>30 m) the
heat effect of crystallization may promote additional phase
changes.
5. Conclusions
The RF thermal plasmas are suitable tools for making ox-
ide ceramic microspheres, either dense or hollow. The mi-
crostructure of the product can principally be influenced by
the structure of feedstock materials and to a less extent by
the plasma operating conditions. Highly porous raw materi-
als and/or the presence of blowing agents facilitate formation
of hollow spheres. By varying the process conditions, one
can primarily affect the thermal history of particles, hereby
the degree of evaporation.
References
[1] N.M. Dignard, M.I. Boulos, Sphericity analysis method for the in-
duction plasma processing of powders, in: Proceedings of the ISPC
14, Prague, 1999, pp. 22112217.
[2] R. McPherson, J. Mater. Sci. 8 (1973) 851.
[3] P. Florian, D. Massiot, B. Poe, I. Farnan, J.-P. Coutures, Solid State
Nucl. Mag. 5 (1995) 233.
[4] T. Ishigaki, Y. Bando, Y. Moriyoshi, M.I. Boulos, J. Mater. Sci. 28
(1993) 4223.
 Chemical Engineering and Processing 44 (2005) 225229

Treatment of particulate metallurgical wastes in thermal plasmas

I. Mohai , J. Szpvlgyi
Institute of Materials and Environmental Chemistry, Chemical Research Center, Hungarian Academy of Sciences,
P.O. Box 17, Budapest H-1525, Hungary

Received 27 August 2003; received in revised form 25 March 2004; accepted 15 April 2004
Available online 19 June 2004

Abstract

Steelmaking and plating industry generates large quantities of dusts and sludge. They are regarded as hazardous wastes because of the
leaching out of their toxic components. Therefore, particular wastes are stockpiled at high cost. Our work aimed at RF thermal plasma processing
of a steelmaking flue dust and sludge from the hot plating of steel products. Waste samples were treated in reducing atmosphere, respectively, to
produce fine metallic powders and in neutral/oxidizing atmosphere to synthesize zinc ferrite nanopowders. Starting powders and products were
characterized by their chemical (ICP-AES) and phase (X-ray diffraction, XRD) composition, particle size and morphology (scanning electron
microscopy, SEM). It was found that iron and zinc content of the particular wastes can be reduced to metals in the presence of hydrogen. In
neutral/oxidizing conditions spinel ferrites of variable composition were formed even during the very short residence time in the reactor.
2004 Elsevier B.V. All rights reserved.

Keywords: Metallurgical wastes; Thermal plasmas; Reduction; Ferrites

1. Introduction Co. the carbothermic reduction of steelmaking dusts con-

A plasma, considered as the fourth state of matter, consists
of a mixture of electrons, ions and neutral species, although
overall it is electrically neutral. Hot or equilibrium plas-
mas can be characterized by the approximate equality be-
tween heavy particle and electron temperature. Such plas-
mas are known as thermal plasmas. During the last decades
much work has been done to develop new processes based
on plasma technology. One option is to recycle metallurgical
wastes of valuable metal content by thermal plasma treat-
ment. PLASMADUST technology of SKF, which combines
a plasma gas heater with a coke-filled shaft furnace, pro-
duces alloyed molten iron, and metallic zinc and lead vapour
from steelmaking dusts [1]. More recently, Retech Inc., a
leading company in the application of thermal plasmas for
environmental applications, developed the Plasma Arc Cen-
trifugal Treatment (PACT) technology to process hazardous
metallurgical and military wastes [2]. The PLASMAREC
plant [3], one installation of the PACT technology, can easily
be mobilized and transported to eliminate local contamina-
tions. In the transferred arc plasma developed by Tetronics
Corresponding author. Tel.: +36 1 325 7933; fax: +36 1 325 7892.
E-mail address: mohaiti@chemres.hu (I. Mohai).
taining mainly iron and zinc resulted in non-toxic slag and
metallic Zn, as products.
In comparison to the conventional waste destruction pro-
cesses, thermal plasma treatment is inherently more expen-
sive in most cases [4]. Therefore, research activities in this
area have been restricted to special toxic waste destruction
problems, where thermal plasmas offer unique advantages,
such as (i) high temperatures and energy densities allowing
fast and complete decomposition of wastes, (ii) very high
quenching rates allowing the formation of non-equilibrium
compositions and preventing undesirable recombination,
(iii) easy control over the chemistry of processing. Compet-
itive new plasma technologies generate valuable products,
as well.
Contrary to the above mentioned technologies, which are
based on arc plasma furnaces, a radiofrequency (RF) plasma
system can process fine powders without granulation in
a continuous operation. This possibility, together with the
advantageous features of the thermal plasmas mentioned
above, offer great perspectives for the synthesis of special
ceramic powders such as spinel ferrites [5]. The RF plasma
treatment produces nanosized metal and/or oxide powders
depending on the parameters of processing. In this paper
application of an RF thermal plasma system for the treat-

0255-2701/$ see front matter 2004 Elsevier B.V. All rights reserved.
doi:10.1016/j.cep.2004.04.008
226 I. Mohai, J. Szepvlgyi / Chemical Engineering and Processing 44 (2005) 225229

ment of wastes from the ferrous metallurgy in reducing and Table 1

neutral/oxidative conditions is investigated. Bulk chemical composition of CFD and HPS samples
Sample Fe Zn Pb Ca Mg Si Cl

Concentration (wt.%)
2. Experimental CFD 65.0 4.22 0.98 1.96 0.15 1.02 0
HPS 17.5 39.5 0.23 2.76 0.29 2.22 6.99
Two wastes, a converter flue dust (CFD) from the steel-
making process and a sludge from the hot plating of steel
products (hot plating sludge, HPS), were treated in a RF (SEM) analysis was performed by a JEOL JSN50A appara-
thermal plasma reactor (35 MHz, max, 35 kW plate power). tus.
Power of the generator was coupled to a TEKNA PL-35
torch which was connected to a water cooled plasma cham-
ber, a cyclone and a filter (Fig. 1). Temperature field in the 3. Results and discussion
torch and the reactor was calculated by a 2D mathemati-
cal model based on the continuity, momentum, energy and 3.1. Reduction of the converter flue dust and the hot
species conservation equations [6]. The highest gas temper- plating sludge
ature at the 24 kW plate power case was 11,500 K near to the
coil region [7]. The mean temperature of the plasma flame Both CFD and HPS samples had broad particle size dis-
was about 5000 K. tribution between 1 and 100 m with a mean particle size
Argon was used as the central plasma gas and as the of 29.5 and 37.2 m, respectively. The HPS had to be dried
sheath gas, as well. Reagent (H2 ) or auxiliary (O2 ) gases prior to the analysis and the plasma treatment. Its weight
were mixed to the sheath gas. Powders were injected axially loss on drying at 105 C was 60.8%.
into the hottest region of the plasma by a PRAXAIR powder Chemical compositions of the industrial samples are
feeder through a water cooled probe. Both the raw materials shown in Table 1. Iron and zinc have the highest concen-
and the products were characterized in terms of particle size, tration in both samples. Chlorine content of HPS (6.99%)
chemical and phase compositions. In each run products were is also worth mentioning. In addition to the data in Table 1
collected from the reactor wall (R), from the reactor bottom both CFD and HPS samples contained other components,
(RB) and the cyclone (C). such as Al, Cr, Ni, Mn, Cu, Cd, P and C in an amount of
Particle size distribution was measured by a Particle Size less than 1%.
Analyser (Malvern 2600C). For bulk chemical compositions The main crystalline phase of the CFD sample was mag-
samples were digested in diluted HNO3 solution using a netite (FeOFe2 O3 ) (Fig. 2). Traces of hematite (Fe2 O3 )
CEM microwave digestion system. Dissolved samples were were also detected. Although carbon content of the sample
analysed by ICP-AES technique (Labtest PSX7521). Phase was rather low (1.2 wt.%) intense reflection of the graphite
conditions were determined by X-ray diffraction analysis could be observed on the diffractogram. This can be ex-
(XRD, Philips PW 1710). Scanning electron microscopic plained by the strong orientation of the graphite particles to
the direction of the X-ray.
Main diffraction peaks of the HPS sample (Fig. 3) can be

Fe

G Sample C
Intensity (a.u.)

M
M W Sample RB

Fe
W W Sample R

M
M M+G Sample CFD
H

20 30 40 50 60
2 ()
Fig. 2. X-ray diffractogram of the converter flue dust (CFD) and the
plasma treated samples produced therefrom: sample Rfrom the reactor
wall, sample RBfrom the reactor bottom, sample Cfrom the cyclone.
Assignation of peaks: Fe, metallic iron; M, magnetite; W, wustite; H,
Fig. 1. Scheme of the experimental setup. hematite; G, graphite.
 I. Mohai, J. Szepvlgyi / Chemical Engineering and Processing 44 (2005) 225229 227

SI Fe particles are ready to form agglomerates with strong cohe-

SI ZnO W
M sion forces. This is why the mean grain size data obtained
Sample C
from the laser diffraction method are much bigger. Agglom-
Intensity (a.u.)

SI G ZnO ZnO erates above 1015 m cannot warm up to the temperature

Sample RB
SI SI ZnO Zn Fe
Sample R SI SI M Zn W
SI
SI SI SI
SI SI
Sample HPS
5 15 25 35
2 ()
Fig. 3. X-ray diffractogram of the hot plating sludge (HPS) and the
plasma treated samples produced therefrom: sample Rfrom the reactor
wall, sample RBfrom the reactor bottom, sample Cfrom the cyclone.
Assignation of peaks: SI, simonkolleite; Fe, metallic iron; M, magnetite;
W, wustite; ZnO, zincite; G, graphite.
assigned as simonkolleite (Zn5 (OH)8 Cl2 H2 O). Iron com-
pounds are most probably present in the original HPS sam-
ple as amorphous hydroxides.
Typical plasma parameters of the reduction tests with hy-
drogen can be seen in Table 2. Feed rate of the starting
powder was changed by the flow rate of the carrier gas. The
specific energy was calculated as the ratio of plate power re-
lated to the powder feed rate. Hence, this parameter consid-
ers two process variables simultaneously. In each run three
samples were collected from different regions of the plasma
system: (i) from the reactor wall, (ii) from the bottom of
the reactor, and (iii) from the cyclone. Particles deposited to
different places of the plasma reactor and the cyclone were
subjected to different thermal effects, so they probably have
different physical and chemical characteristics, as well.
X-ray diffraction patterns of plasma-treated CFD samples
(Fig. 2) indicated a significant decrease in the intensity of
the magnetite peaks with a simultaneous increase of metal-
lic Fe peak. Extent of reduction, which is proportional to
the intensity of the Fe100/magnetite100 diffraction lines, is
almost complete in case of sample C. In sample R most of
the magnetite content reacted with hydrogen. However, re-
duction was not complete. Some FeO (wustite) remained in
the product. Powder collected from the reactor bottom (RB)
was hardly reduced: only traces of Fe and some FeO could
be detected by XRD. Explanation of this phenomenon can
be given from scanning electron micrographs of the starting
powder and the products (Fig. 4). Although the primary par-
ticle size of sample CFD is near to the submicron size, these
where reduction is favoured from thermodynamic point of
view. On the other hand, particles and agglomerates below
this size can evaporate easily with a simultaneous reduction.
Particles and agglomerates of different size pass along tra-
jectories with different residence time in the hot zone of the
plasma flame. Big agglomerates fall through along the axis
of the plasma practically without chemical and morpholog-
45
ical changes. These agglomerates can be collected from the
reactor bottom. Samples deposited to the reactor wall and
the cyclone, respectively, have much smaller and more uni-
form particle size. This indicates that these products were
formed by melting, evaporation and subsequent condensa-
tion of the starting powder. Extent of agglomeration can be
estimated from the distribution of products on the different
areas of the plasma equipment.
Sample CFD consisted of mostly agglomerated particles,
because about 7075% of the products could be collected
from the reactor bottom. Quantity of C samples is almost
negligible. Agglomeration could be suppressed by the addi-
tion of carbon powder to the sample. Although carbon could
not reduce oxides without hydrogen, in the presence of car-
bon, hydrogen reduction was more effective and the ratio of
nonreacted powder (sample RB) decreased from 7075 to
40%.
XRD measurements of plasma treated HPS samples in
the presence of hydrogen indicated a partial reduction of the
starting material (Fig. 3). Reflections of the main crystalline
phase of the starting powder (simonkolleite, SI) could be
observed in all samples. Samples collected from the reac-
tor wall consisted of nonreacted simonkolleite, metallic Fe,
FeO (wustite), magnetite, metallic Zn and ZnO. Diffraction
peak of metallic Zn was quite surprising, because in ther-
modynamic equilibrium Zn vapours are ready to form ox-
ide during condensation in the presence of H2 O vapours.
This phenomenon can be explained by the high cooling rate
in the plasma reactor which caused a kinetic hindrance of
re-oxidation. Phase compositions of samples C were simi-
lar to those of samples R, but without metallic Zn. In sam-
ples RB extent of reduction was negligible. Heat effect of
the plasma flame could be observed in the formation of
magnetite from the amorphous iron hydroxides and ZnO
from the zinc-hydroxide-chloride (simonkolleite). Relative
amount of RB products was rather small (about 25%). It

Table 2
Conditions of reduction tests
Waste powder Flow rate of gases (l/min) Power (kW) Feed rate (g/min) Espec (kWh/g)

Plasma Sheath Carrier

CFD 13 (Ar) 45 (Ar) 7.5 (H2 ) 6.3 25.5 4.8 0.09

HPS 15 (Ar) 45 (Ar) 7.0 (H2 ) 5.5 23.4 5.4 0.07
228 I. Mohai, J. Szepvlgyi / Chemical Engineering and Processing 44 (2005) 225229
Fig. 4. Scanning electron micrographs of sample CFD and its reduction products.
indicated a lower agglomeration of the HPS particles com-
paring to the CFD starting material. High extent of the non-
reacted simonkolleite in the plasma treated samples can be
traced back to the volatile content of the HPS sample: due
to thermal effect simonkolleite releases HCl and H2 O. A
considerable part of the plasma energy may be lost on the
dissociation of these gases.
3.2. Synthesis of zinc ferrites
For the synthesis of zinc ferrite (ZnFe2 O4 ) of stoichio-
metric composition CFD and HPS samples were mixed
in an appropriate ratio. From the chemical analysis the
Fe/Zn molar ratio of the mixture was 2.18. To override the
problem of the cooling effect of volatiles on the plasma
(mentioned above), prior to mixing HPS sample was pre-
heated at 300 C to reduce its volatile (H2 O) content. Con-
ditions of the plasma treatments can be seen in Table 3.
In order to improve heat transport between the hot gases
Table 3
Conditions of the ferrite synthesis
Waste powder Flow rate of gases (l/min)
Plasma Sheath
CFD+HPS 11 (Ar) 20 (Ar) 20 (O2 )
CFD+HPS+Ni salt 10 (Ar) 20 (Ar) 20 (O2 )
and the particles O2 was mixed to the sheath gas. Al-
though during conventional ferrite production the starting
oxide mixture is heated for several hours above 1000 C,
in the RF plasma some milliseconds of residence time was
enough to produce fine ferrite powders from the waste
mixture.
Both powders collected from the reactor wall and from the
reactor bottom showed similar diffraction patterns (Fig. 5.)
indicating the presence of spinel as the dominant phase
(amount of sample C was negligible in these experiments).
Composition of the spinel phase was estimated from the
lattice parameters (a), assuming, that the increase of lat-
tice parameter due to Zn incorporation is proportional to
the Zn concentration. From these calculations the composi-
tions of sample R and sample RB were Zn0.7 Fe2.3 O4 and
Zn0.4 Fe2.6 O4 , respectively. Consequently, some of the Zn
content of the starting mixture could not build into the spinel
structure and left the system through the exhaust in the form
of very fine ZnO powder.
Power (kW) Feed rate (g/min) Espec (kWh/g)
Carrier
2 22.3 2.4 0.16
3 24.5 1.9 0.22
 I. Mohai, J. Szepvlgyi / Chemical Engineering and Processing 44 (2005) 225229
ZF
Intensity (a.u.)
ZF
ZF
20 25 30 35 40 45
2 ()
Fig. 5. X-ray diffractogram of the zinc ferrite (ZF) synthetised from the
mixture of CFD and HPS samples.
NiZn ferrites could also be synthesized in RF plasma
reactor from the CFD and HPS powders by admixing
Ni3 H4 CO7 4H2 O to the waste mixture. Molar ratio of the
metals was set for Ni/Zn/Fe = 0.4/0.6/2. XRD diffrac-
tograms of the products showed the complete transfor-
mation of the starting mixture to spinel phases. However,
in sample R spinel phase consisted of the mixture of
Ni0.2 Zn0.5 Fe2.3 O4 and Ni0.6 Zn0.3 Fe2.1 O4 , while in sample
RB for the composition of the spinel (in accordance with the
chemical analysis) Ni0.2 Zn0.5 Fe2.3 O4 could be estimated
from the lattice parameter.
4. Conclusions
Metallurgical waste powders containing iron and zinc ox-
ides and hydroxides can be reduced to metals in RF thermal
plasma in the presence of hydrogen. However, if the particles
are agglomerated they cannot be reduced due to the short
residence time in the hot plasma region. Thus, to achieve
229
high reduction non-agglomerated powders with a mean par-
ticle size of less than 20 m should be introduced into the
plasma reactor. Spinel ferrites of variable composition were
formed with actually complete conversion, even during the
very short residence time. This is quite surprising, because
conventional ferrite synthesis needs heat treatment of 46 h
at temperatures above 1000 C.
50
Acknowledgements
The authors are grateful for the financial support from the
National Scientific Research Fund OTKA T032272.
References
[1] A.L. Hare, in: J. Feinman (Ed.), Plasma Technology in Metallurgical
Processing, ISS, Warrendale, PA, 1987, pp. 175187.
[2] R.C. Eshenbach, R.E. Haun, in: Proceedings Workshop on Industrial
Application of Plasma Chemistry, Drexel University Press, Minneapo-
lis, MN, 1995, p. 9.
[3] W. Hoffelner, R. Burkhard, V. Haefeli, H. Sun, in: Proceedings of the
13th International Symposium on Plasma Chemistry, Peking University
Press, Beijing, China, 1997, p. 1915.
[4] E. Pfender, Thermal plasma technology: where do we stand and
where are we going, Plasma Chem. Plasma Proc. 19 (1999) 1
31.
[5] I. Mohai, J. Szpvlgyi, I. Bertti, M. Mohai, J. Gubicza, T. Ungr,
Thermal plasma synthesis of zinc ferrite nanopowders, Solid State
Ionics 141-142 (2001) 163168.
[6] A. Gaillat, R.M. Barnes, P. Proulx, M.I. Boulos, Computer modelling
of enclosed inductively coupled plasma discharges, Spectrochim. Acta
B-At. Spectrosc. 50 (10) (1995) 11871205.
[7] Z. Karoly, Project report F 029731 supported by the National Scientific
Research Fund, Hungarian, 2002.
 Chemical Engineering and Processing 44 (2005) 231235

Defluidization modelling of pyrolysis of plastics in a conical

spouted bed reactor
Roberto Aguado, Rubn Prieto, Mara J. San Jos , Sonia Alvarez,
Martn Olazar, Javier Bilbao
Departamento de Ingeniera Qumica, Facultad de Ciencias, Universidad del Pas Vasco, Apartado 644, Bilbao 48080, Spain

Received 3 October 2003; received in revised form 29 January 2004; accepted 11 February 2004
Available online 17 June 2004

Abstract

Thermal pyrolysis for upgrading plastic wastes is one of the better methods for recycling plastics in terms of its perspectives for industrial
implementation. The conical spouted bed reactor proposed in this paper may be a solution to the problems arising in fluidized beds handling
sticky solids, as particle agglomeration phenomena, which can cause defluidization. In order to avoid defluidization, experiments have been
carried out in batch mode in the temperature range of 450600 C. A good performance of the reactor is proven under the conditions of
maximum particle stickiness.
2004 Elsevier B.V. All rights reserved.

Keywords: Defluidization; Pyrolysis; Conical spouted bed reactor

1. Introduction In order to avoid defluidization in bubbling fluidized beds,

The high consumption of plastics in developed countries
(over 50 million tons in the year 2001 and an increase of
4% year1 is expected) explains the great interest for ex-
ploitation of post-consumer plastics. Taking into account that
the environmental regulations prohibit their exploitation by
combustion, thermal decomposition or pyrolysis is one of
the procedures for recycling plastic with best perspectives
for obtaining feedstock and fuel.
In spite of the major effort in this field in the last 30 years,
the development at industrial scale of post-consumer plas-
tic pyrolysis has considerable uncertainties concerning the
selection of the more suitable technology. The more devel-
oped technology in the literature is the bubbling fluidized
bed reactor [15] where the fused plastic coats the inert par-
ticles (sand). Nevertheless, the operation at large scale in
this reactor presents problems of defluidization, due to par-
ticle agglomeration provoked by fusion of particles coated
with plastic [4].
Corresponding author. Tel.: +34 94 6015362; fax: +34 94 6013500.
E-mail address: iqpsaalm@lg.ehu.es (M.J.S. Jose).
the thickness of the fused plastic that coats the inert parti-
cles must be minimized by using sand/plastic ratio, which
means a serious limitation in the yield of the reactor and a
high-energy requirement in order to fluidized and then heat
the large amount of sand.
In this paper, the performance of a conical spouted bed
reactor in the pyrolysis of polyolefins is studied, and delim-
itation of the capacity of this original reactor is made for
minimizing the stickiness problems under the conditions of
maximum difficulty. In a previous paper, the good perfor-
mance of this reactor for the kinetic study of the pyrolysis of
polyolefins [6] and polystyrene at high temperature was ver-
ified. The good behaviour of the conical spouted bed reactor
is due to the fact that it has the characteristics of conven-
tional spouted beds and, furthermore, it has those provided
by the conical geometry. The main characteristic of these
reactors is their versatility in gas flow rate, which can be in-
creased from the spouted bed regime to the dilute spouted
bed (or jet spouted bed). In this way, operation can be carried
out within a wide range of bed voidage and, consequently,
of vigorousness in the gassolid contact [7]. In spite of the
increase in gas and particle velocities, the cyclic movement
characteristic to spouted bed is maintained.

0255-2701/$ see front matter 2004 Elsevier B.V. All rights reserved.
doi:10.1016/j.cep.2004.02.016
232 R. Aguado et al. / Chemical Engineering and Processing 44 (2005) 231235

2. Experimental
The components of the pilot plant are described in a previ-
ous paper [6]. The hopper contains plastic particles of 1 mm,
which are driven forward its end by paddles attached to a
shaft rotated by a motor. The hopper is sealed and has an N2
entry that allows for attaining the pressure required for in-
troducing the feed into the reactor by means of a worm. The
feed flow rate is regulated by means of a frequency shifter
that actuates on the rotation speed of the worm. The tube for
introducing the feed into the reactor, of 20 mm i.d., is made
up of two concentric tubes, and water circulates along the
annular zone for cooling the feed and, consequently, avoid-
ing the melting of the plastic and the ensuing plugging of
the reactor inlet. The feeding system is provided with free
vertical and horizontal movements, which permits the feeder
to be placed at the desired location on the upper surface of
the reactor. Previous experiments have shown that the more
suitable locations for feeding the plastic are those over the
annular zone of the bed, preferably intermediate positions
between the spout and the wall.
Fig. 1 shows a scheme of the reactor where its characteris-
tic dimensions are defined [8]: HT = 0.34 m; Hc = 0.205 m,
= 28 ; Dc = 0.123 m; Di = 0.02 m; Do = 0.01 m.
When the inert gas flow rate is increased, the gassolid
contact regime evolves from the conventional spouted bed
to a dilute spouted bed (or jet spouted bed, which is char-
acteristic of a conical contactor and in which bed voidage
is uniform and higher than 0.90) [9] and the gas residence
time decreases from a few seconds to approximately 20 ms.
As the plastic material is introduced into the reactor, it
melts onto the sand particles and coats them. The vigorous
cyclic movement of the particles can prevent agglomeration
problems. Temperature is measured by means of three ther-
mocouples placed at different radial positions in the reactor
and provided with free vertical movement. Bed isothermic-
ity is noteworthy, which is attained due to the vigorous par-
ticle circulation.
Reaction temperature is attained by means of two electric
resistances covered with ceramic material. One resistance is
located within the tube for heating the inert gas prior to entry
into the reactor and the second one surrounds the conical
section of the reactor. Temperature in both these positions
is measured by two thermocouples. The two heating zones
are thermically insulated.
The runs have been carried out at 400, 500, 550 and 600 C
by feeding 1 g of plastic material (with a particle size of
approximately 1 mm) and using 30 g of sand (with a size
between 0.63 and 1 mm).
The materials used for pyrolyzing are: low density
polyethylene (LDPE), high density polyethylene (HDPE)
and polypropylene (PP) (supplied by Dow Chemical, Tar-
ragona, Spain). The properties of these materials are set out
in Table 1.

3. Results

3.1. Hydrodynamics and particle velocity

The minimum spouting velocity is determined, as a func-

tion of the geometrical parameters of the particlereactor
system [8]:

1.68   
0.39 Db 0.57
(Re)ms = 0.126 Ar tan (1)
Do 2

where Db is the diameter of the upper level of the stagnant

bed:
 
Db = Di + 2 Ho tg (2)
2
The average cycle time (characteristic to particle trajectories
[10]) is




dp 0.68 Do 1.31 u 1.14
t c = 0.85 0.48
Di Di ums

0.86
0.11
Ho s
0.51 (3)
Di v

The vertical solid velocity in the spout is calculated by the

Fig. 1. Scheme of the reactor. equation proposed by Epstein and Grace [11] valid at the
 R. Aguado et al. / Chemical Engineering and Processing 44 (2005) 231235 233

Table 1
Properties of the materials studied
Material Molecular weight (g/mol) Polydispersity Density s (kg/m3 ) Heating value (MJ/kg) Pyrolysis heat (kJ/kg)

LDPE 92200 5.13 923 43 543.6

HDPE 46200 2.89 940 43 429.8
PP 5000090000 2.00 890 44 581.9

minimum spouting velocity and at velocities 20 and 30%

above this value:

2 
r
z = z (0) 1 (4)
rs

The calculation of velocity along the axis has been carried

out by solving a force balance for a spherical particle arising
along the spout:
dz (0) 3 CD g [uz (0) z (0)] (s g )g
z (0) = (5)
dz 4 s dp s
The particulate phase in the annular zone of a spouted bed
can be described as an isotropic, incompressible, rigid plas-
tic, non-cohesive Coulomb powder. Assuming that this ma-
terial is in a quasi-static critical condition, the stress field
can be described by equations developed for a static mate-
rial element.
The solid flow is coaxial, so that in cylindrical coordinates,
the following applies:
z /r + r /z
tan(2) = (6)
z /z r /r
where is the stress angle (the angle between the direction
of the z axis and the direction of the major principal stress
axis).The mass conservation for this zone is
1
[(1 )s z ] + [r(1 )s r ] (7)
z r r
Fig. 2. Particle velocities.
By solving simultaneously Eqs. (4)(7), the horizontal and
vertical components of particle velocity will be determined.
Fig. 2 shows the results calculated for the experimental sys- softens, its material is distributed by coating the surround-
tem used. ing sand particles. If the thickness of the layer that coats
the particles is lower than a critical value, the sand particles
3.2. Sand coating with fused plastic do not fuse. This situation corresponds to a low flow rate
of plastic in the feed, and beyond this value, agglomerates
In the deterioration of the particle mixture and evolu- grow irreversibly, and instead of sand coating, total blockage
tion to bed blockage, the mechanism of the initial step of of the bed or defluidization is the result. This mechanism
coating the inert particle (sand) with fused plastic is cru- of growth depends on the properties of the plastic material
cial. Subsequently, the tendency of the particles to agglom- [12].
erate depends on their stickiness, momentum and contact The good behaviour of the conical spouted bed reactor is
surface. quantified on the basis of the critical thickness of the fused
Concerning the first step of coating, when the particles of plastic that may be handled. The critical thickness is the
polyolefin are introduced into the bed, their external surface value above which the particles fuse when they collide and is
becomes softer but their inside is cold, due to their low ther- a function of the thickness (viscosity) and of the momentum
mal conductivity. The softened surface is responsible for the of the colliding particles. From the hypothesis of the critical
formation of agglomerate made up of a plastic particle sur- thickness model of Arena and Mastellone [12], the following
rounded by sand particles. When the whole plastic particle expression has been deduced for the critical thickness for
234 R. Aguado et al. / Chemical Engineering and Processing 44 (2005) 231235
which the defluidization of the bed can be avoided:


2.50 ums .
r m0
crit = exp 1 (8)
2 r2
Eq. (8) shows the dependency of the critical thickness
with mass m, radius rp , and velocity o of the inert parti-
cles coated with plastic and with the viscosity of the mate-
rial that coats the sand . For the experimental conditions
(total mass of sand, 30 g; mass of plastic in the feed, 1 g;
sand particle diameter, between 0.6 and 1.2 mm; sand den-
sity 2600 kg m3 , average velocity of the particles in the an-
nular zones, 0.25 m s1 [7], the critical thickness predicted
by Eq. (8) is 250 m. In this calculation, the viscosity of the
plastic has been taken at its fusion temperature, 16 poises,
because it is in the initial step of fusion when the stickiness
of the plastic is maximum point.
In the fluidized bed, the velocity of the particles is of a
few cm s1 and the critical thickness of the fused plastic is
between 4 and 7 mm. A fact that should be taken into account
is that, although gas velocity is increased, particle velocity
hardly changes because most of the excess gas rises through
the bed in the bubble phase. Nevertheless, in the conical
spouted bed, particle velocity is increased by raising gas
velocity up to the dilute spouted bed, in which bed voidage
is uniform (between 0.9 and 0.99) and the particle descent
velocity is almost equal to the ascent velocity [9].
3.3. Limit operating conditions for the equipment
Eqs. (4)(8) have been used to calculate the maximum
amount of plastic that can be fed (corresponding to the
critical thickness) without bed agglomeration, for different
amounts of sand in the reactor. Fig. 3 shows these results
for different nitrogen flow rates. When the nitrogen flow
rate is increased, the bed is in the regime of spouted bed
(beds made up of high amounts of sand). For intermediate
amounts of sand, the bed is the transition between bed and
Fig. 3. Limit operating conditions.
dilute spouted bed, as velocity is increased from 1.05 to
In Fig. 3 it is observed that, for a given amount of sand
in the reactor as gas velocity is increased, the amount of
plastic that may be fed increases almost linearly, which is a
consequence of particle velocity increase. Consequently, the
ratio of plastic/sand in the feed may be as high as 0.03 for
the experimental reactor, which may treat 100 g of sand in a
spouted bed regime and corresponds to 62 mm of stagnant
bed height. Likewise, and taking into account the scaling
capacity of the conical spouted bed, a reactor with a stagnant
bed height of 430 mm (33.3 kg of sand) may treat 1 kg of
plastic in batch operation.
4. Conclusions
The conical spouted bed reactor is especially suitable for
avoiding agglomeration problems in the pyrolysis of poly-
olefins, even when the operation is carried out under the
conditions of maximum stickiness. In this reactor, the cyclic
movement of sand particles allows for their uniform coat-
ing with fused plastic. Furthermore, the high velocity of
the particles generates collisions whose energy suffices to
avoid their agglomeration. The critical thickness of the plas-
tic layer is an order of magnitude higher than that corre-
sponding to the fluidized bed, which gives way to a higher
yield by reactor volume unit.
Acknowledgements
This work was carried out with the financial sup-
port of the University of the Basque Country (Project
UPV 069.310-13607/2001) and of the Ministry of Sci-
ence and Technology of the Spanish Government (Project
PPQ2001-0780).
Appendix A. Nomenclature
Ar Archimedes number g dp3 g (s g )2
Db , Dc , D i , Do diameter of the upper level of the
stagnant bed, of the column, of the bed
base and of the inlet, respectively (m)
dp particle diameter (m)
rp particle radius (m)
Ho height of the stagnant bed (m)
k kinetic constant for pyrolysis of PP (s1 )
m particle mass (kg)
rs spout radius at level z (m)
(Reo )ms Reynolds number at the minimum
spouting velocity
t, tc time and average cycle time (s)
u, ums gas velocity and gas velocity at the
minimum spouting (m s1 )
 R. Aguado et al. / Chemical Engineering and Processing 44 (2005) 231235 235

ur , u z , horizontal and vertical components of [2] J.A. Conesa, R. Font, A. Marcilla, Comparison between the pyrolysis
gas velocity, respectively (m s1 )
of types of polyethylenes in a fluidized bed reactor, Energy Fuels
11 (1997) 126136.
Wp plastic mass (kg) [3] W. Kaminsky, Possibilities and limits of pyrolysis, Makromol. Chem.
Macromol. Symp. 57 (1992) 145160.
Greek symbols [4] W. Kaminsky, Chemical Recycling of Mixed Plastics by Pyrolysis,
stress angle Adv. Polym. Technol. 14 (1995) 337344.
[5] P.T. Williams, E.A. Williams, Fluidised bed pyrolysis of low density
crit critical thickness of the plastic layer (m) polyethylene to produce petrochemical feedstock, J. Anal. Appl.
shape factor Pyrolysis 51 (1999) 107126.
cone angle, radians [6] R. Aguado, M. Olazar, B. Gaisn, R. Prieto, J. Bilbao, Kinetic study
viscosity of the melted plastic of polyolefins pyrolysis in a conical spouted bed reactor, Ind. Eng.
(kg m1 s1 ) Chem. Res. 41 (2002) 45594566.
[7] M. Olazar, M.J. San Jos, S. Alvarez, A. Morales, J. Bilbao, Mea-
o average particle velocity in the annulus surements of particle velocities in conical spouted beds using an
(m s1 ) optical fibre probe, Ind. Eng. Chem. Res. 37 (1998) 45204527.
s , v density of the particle and of glass, [8] M. Olazar, M.J. San Jos, A.T. Aguayo, J.M. Arandes, J. Bilbao,
respectively (kg m3 ) Stable operation conditions for gas-solid contact regimes in conical
spouted beds, Ind. Eng. Chem. Res. 31 (1992) 17841792.
o , z horizontal and vertical components
[9] M.J. San Jos, M. Olazar, S. Alvarez, J. Bilbao, Local bed voidage
of particle velocity, respectively (m s1 ) in conical spouted beds, Ind. Eng. Chem. Res. 37 (1998) 25532558.
z (0) component of particle velocity in the [10] M.J. San Jos, M. Olazar, S. Alvarez, M.A. Izquierdo, J. Bilbao,
z direction along the axis of the Solid cross-flow into the spout and trajectories in conical spouted
contactor (m s1 ) beds, Chem. Eng. Sci. 53 (1998) 35613570.
[11] N. Epstein, J.R. Grace, Spouting of particulate solids, In: L. Otten,
M.E. Fayed (Eds.), Handbook of Powder Science and Technology,
2nd ed., Van Nostrand Reinhold, New York, 1997, pp. 532567
References (Chapter 10).
[12] U. Arena, M.L. Mastellone, Defluidization phenomena during the
[1] D.S. Scott, S.R. Czernik, J. Piskorz, A.G. Radlein, Fast pyrolysis of pyrolysis of two plastic wastes, Chem. Eng. Sci. 55 (2000) 2849
plastic wastes, Energy Fuels 4 (1990) 407411. 2860.
 Chemical Engineering and Processing 44 (2005) 237243

The energy of bed processing during drum granulation

Tadeusz Gluba
Faculty of Process and Environmental Engineering, Technical University of Lodz, ul. Wolczanska 213/215, 90-924 Lodz, Poland

Received 30 July 2003; received in revised form 10 December 2003; accepted 11 February 2004
Available online 19 June 2004

Abstract

Results of investigations of the effect of particle size composition of the raw material and the amount of wetting liquid supplied to the
bed on the energy needed for growth and consolidation of granules during wet drum granulation process are presented in the paper. Energy
consumption at subsequent stages of the granulation process was determined on the basis of the instantaneous values of torque recorded every
1 s on the granulator shaft.
2004 Elsevier B.V. All rights reserved.

Keywords: Drum granulation; Agglomeration; Energy consumption; Power demand; Porosity

1. Introduction the power demand for drum granulation to the properties of

An important aspect of each technological process is the
tendency to decrease energy demand necessary to produce a
product of required properties. This can be accomplished by
both a selection of proper processing method and a choice
of appropriate parameters that would ensure correct run of
the process. To achieve advantageous conditions of agglom-
eration in rotary drums, it is necessary to ensure a deter-
mined nature of feed motion that is strictly connected with
its dynamics, and also the conditions for agglomerate forma-
tion and growth, dependent, among the other things, on the
properties of media that participate in the process and their
quantitative interrelations. In horizontal drum granulators
dynamics of the granular bed are main factor contributing
to the torque measured on the granulator shaft, and conse-
quently, on the drive engine power [1,2]. The assessment of
power demand at particular stages of the batch granulation
process and relating it to feed motion and granulated product
properties changing in time is an important problem, both
theoretical and practical. Results of research carried out so
far for model or real materials to some limited extent, only
partly, explain the dependence of torque and power during
granulation on the bed behaviour in the drum [3]. In drum
granulation, there have been only a few attempts to relate
Fax: +48 42 636 3069.
E-mail address: gluba@wipos.p.lodz.pl (T. Gluba).
newly formed product [4]. On the other hand, power demand
for granulation in agitation granulators and high-shear mix-
ers has been described rather extensively in the literature.
Power consumption of the impeller motor during granula-
tion in high-shear mixers and agitation granulators is widely
used for endpoint determination and scale-up [59] because
the measurement is inexpensive and is well correlated with
granule growth [10].
In the mixing process, changes in torque on the blades
and power consumption occur as a result of a change
in the cohesive force or the tensile strength of the ag-
glomerates in the moistened powder bed. Processing of
a fine-grained material to granulated form with desirable
physical properties requires each time a determined process
duration, which in connection with power demand deter-
mines the energy necessary to produce a specific material.
The relation between the energy input during drum gran-
ulation and product properties has been rarely discussed
in literature [11]; hence, further studies in this area seem
justified.
2. Aim of the study
The aim of this study was the assessment of the effect
of the particle size distribution of raw material and feed
moisture content on the energy required for the transfor-
mation of powder material into granulated product of de-

0255-2701/$ see front matter 2004 Elsevier B.V. All rights reserved.
doi:10.1016/j.cep.2004.02.017
238 T. Gluba / Chemical Engineering and Processing 44 (2005) 237243

fined particle diameters and porosity in wet drum granu-

lation.

3. Experimental

3.1. Materials

Fine-grained dolomite flour was used in the experiments.

Raw material for testing was prepared on the basis of five
commercially available size fractions of this material clas-
sified as <10, <15, <60, <100 and <250 m. The parti-
cle size distribution of each fraction was determined using
a laser particle size analyser Analysette 22 and described
by the statistical moments.
On the basis of these data, starting materials for granula-
tion experiments were prepared by blending mass fraction
of the initial raw materials. The particle size distributions of
raw materials to be granulated were calculated numerically
(by mixing input fractions of the original raw materials) as-
suming that they were geometrically similar. In the calcula-
tions, the following were assumed constant: variation coef-
ficient /dm = 1.07, and asymmetry coefficient 1 = 2.00,
for five values of mean particle size dm ranging from 10.6
to 34.9 m. The size distributions and mass mean size of
the resultant blends (denoted by symbols D1D5), which
are raw materials for the granulation process, are shown in
Fig. 1.
Such a selection of particle size distributions (constant
values of /dm and 1 ) enables an evaluation of the impact
of mean particle diameter on the granulation process and on
the properties of the product obtained.
For each raw material, basic physical properties such as
specific and bulk densities, and porosity were determined.
For bulk density, two values were determined each time: for
loose material bl and for material condensed to its minimum
volume on a riddle bc .
On this basis, the mean bulk density b and mean porosity
(Eqs. (1) and (2)) were calculated:
bl + bc
b =
2 was continued until the moment when the granulated mate-
b rial started to stick to the drum walls because water had been
=1
Fig. 1. Comparison of particle size distributions of raw materials.
Physical properties of the materials used in the experi-
ments are shown in Table 1.
3.2. Experimental set-up and methods
The process of granulation was carried out batch-wise
in a drum of diameter 0.5 and 0.4 m long at a constant
rotational speed n = 20 rpm and constant drum filling (
= 10%), for a range of moisture contents w [kg of water/kg
of dry material] selected for each raw material separately. A
schematic diagram of the experimental set-up is shown in
Fig. 2.
Wetting liquid (distilled water) was supplied by two pneu-
matic nozzles to the bed of fine particles tumbling in the
drum (the wetting stage). Operating parameters of the noz-
zles were constant:
water flow rate Qw = 3.33 106 m3 /s;
air flow rate Qa = 694.44 106 m3 /s; and
air pressure pa = 0.3 MPa.
After completing the wetting, the process of granulation
(1)
(2) pressed off (granulation stage). During this process, samples

Table 1
The physical properties of the raw materials
Material

D1 D2 D3 D4 D5

Mean particle size dm (m) 10.6 15.8 20.0 28.0 34.9

Bulk density of loose material bl [kg/m3 ] 842.5 939.0 891.4 989.6 1031.0
Bulk density of condensed material bc [kg/m3 ] 1235.5 1401.6 1348.2 1554.8 1585.0
Mean bulk density b [kg/m3 ] 1039.0 1170.3 1119.8 1272.2 1308.0
Mean porosity [] 0.629 0.582 0.600 0.546 0.533
Density (kg/m3 ] 2800
 T. Gluba / Chemical Engineering and Processing 44 (2005) 237243 239

5
Granulation 1: Granulation 2 (w=0.147):
y = -2E-12x4 + 2E-08x3 - 8E-05x 2 + 0.147x - 96.7
Nu [ W/kg ]

y = 7E-05x + 4.2

D2

Wetting segment 2:
Wetting
3 y = 0.0011x + 3.76 w=0.147
w=0.155
w=0.160
Wetting segment 1: w=0.165
y = 0.025x + 2.07

2
Fig. 2. Schematic diagram of the equipment for granulation tests: (1)
0 1000 2000 3000 t[s] 4000
granulator drum, (2) pneumatic nozzles, (3) engine, (4) inverter, (5) tripod,
(6) rotameters, (7) water tank, (8) compressor, (9) tachometer, (10) torque tg [ s ]
0 1000 2000 3000
meter, (11) reader, and (12) computer.
Fig. 3. Change of power demand during granulation.

representative of the whole feed were taken at determined
time intervals. On this basis, progressions of product prop-
erties (granule size, porosity) were obtained. The first sam-
ple was taken immediately after wetting (granulation time tg
= 0) and the last one in the moment when water pressed out
to the granule surface made them stick to the inner surface
of the drum preventing further processing in this way (tg
= tgmax ). This limiting (maximum) granulation time tgmax
depended on the mean particle size of the raw material dm
and feed moisture content w. For the whole process duration
instantaneous values of torque on the granulator shaft were
recorded every 1 s.
4. Results
The results of investigation show that the particle size
distribution of the raw material and wetting of the bed had
a significant effect on the granulation process and proper-
ties of the product obtained. Changes in the particle size
distribution of the raw material and feed moisture content
caused changes in dynamic parameters of the bed tumbling
in the drum, and in parallel, affected the rate of changes
in product properties (particle size distribution, porosity
of granules), which finally determined the total energy
demand for processing the raw material into granulated
product of specific characteristics. The energy of bed pro-
cessing was calculated taking instantaneous moments M(t)
measured on the granulator shaft and recorded in time in-
tervals equal to 1 s in the whole wetting and granulation
period.
On the basis of the values of torque M(t), the instantaneous
unit torque Mu (t) from Eqs. (3) and (4)
for the wetting stage:
M(t) Mi
Mu (t) =
ms + Q w w t
for the granulation stage:
M(t) Mi
Mu (t) = (4)
mb
and then unit power demand Nu (t) from Eq. (5).
Nu (t) = Mu (t) (5)
were calculated.
Diagrams representing changes in power demand in time
for the entire cycle of granulated material production (wet-
ting and granulation) are shown in Fig. 3.
The relationships Nu (t) have a characteristic run. During
feed wetting, a rapid linear increase of power demand in the
initial period and then a slow-down of the growth rate could
be observed. In the initial period of the granulation stage,
this relationship is linear too, while at the end some curving
is clearly visible. In the description of power changes on the
particular stages, linear and polynomial functions were used.
Specific energy values Eu (t) (i.e. energy per unit mass of
the feed) were calculated by integrating the fitted equations
over their respective time intervals (Eq. (6)).

t1
Eu (t) = Nu (t) dt (6)
t0
Fig. 4 shows examples (for raw material D2) of the ef-
fect of feed moisture content w on the energy necessary to
produce granulated material of specified mean particle di-
ameter.
It is apparent from Fig. 4 that the energy input required for
agglomerate growth decreases as the amount of liquid sup-
plied to the bed during wetting is increased (in a determined
range). Already a small, below 1%, increase of the liquid
supplied during wetting (particularly in the lower range of
moisture content) leads to a significant decrease of energy
demand (mainly due to a decrease of the required granula-
tion time).
A significant effect of the amount of added binding liq-
(3)
uid on the process of granulation was also shown by other
240 T. Gluba / Chemical Engineering and Processing 44 (2005) 237243
12
D2
dgm [ mm ]
w=0.147
8
w=0.155
w=0.160
w=0.165
0
0 4 8 12
Eu [ kJ / kg ]
Fig. 4. Energy input for the granule size growth.
authors [1216]. In the period of granulation after wetting,
Wauters et al. [12,13] distinguished two ranges: induction
stage during which the granules are gradually being com-
pacted and little or no growth occurs, and a stage of rapid
growth of granules that follows afterwards. Results of their
studies proved that the duration of the first period (induction
stage) is a function of the amount of added wetting liquid.
The higher is duration of this period the smaller is the ex-
cess of added liquid over some value of moisture content
called the critical moisture content. The value of this excess
determines also the granule growth rate in the second stage.
Occurrence of these two granulation periods and the im-
pact of feed moisture content on their duration are confirmed
by the present studies.
During bed wetting, comparable energy inputs were ob-
tained for different particle size distributions of the raw
material and moisture contents of the feed. Also, the mean
particle diameter of the feed after this period is very sim-
ilar and does not depend significantly on the particle size
distribution of the raw material and bed moisture content.
However, these two parameters have a great influence on
the proceeding of the second stage, i.e. granulation, and
on the rate of changes in the granulated feed properties.
They affect directly the relationships between individual
feed particles, and also interactions between the bed and
the granulator walls, which determine the nature of motion
of the wetted material tumbling in the drum. This is related
to changing properties of the wetted bed and granulated
product on which the character of bed motion and, con-
sequently, the moment necessary to incline the bed and to
cause its circulation necessary to induce the mechanisms of
agglomerate formation, growth and consolidation depend.
An example of energy input necessary for unit growth
of the mean agglomerate size for different feed moisture
content is shown in Fig. 5.
A comparison of energy input necessary for unit growth
of the mean agglomerate size on the stage of granulation
for subsequent particle size distribution of the raw ma-
terial, depending on the feed moisture content is shown
Fig. 5. Energy input for unit growth of mean granule size.
in Fig. 6. Experimental points for particular raw materi-
als can be approximated by a power function with good
accuracy.
The effect of mean particle size of raw material and feed
moisture content on energy input Eu related to unit incre-
ment of mean particle size dgm on the whole stage of gran-
ulation (from tg = 0 to tg = tgmax ) is described with good
accuracy (R = 0.96) by the power function of two variables:
Eu
= 1.26 107 dm 6.9 w18.5 (7)
dgm
Exponents in the obtained equation reveal a remarkable
effect of the feed moisture content and mean particle size
diameter of the raw material on energy input for the incre-
ment of granule size.
The advancement of drum granulation, beside the size in-
crement, is measured also by the change of the inner struc-
ture of the agglomerates being formed (thickening of par-
ticles), on which, among the other things, their mechanical
strength depends. The rate of these changes determines a
necessary process time and, as a result, affects the energy
of bed transformation. Results of the research provide ev-
idence on a significant effect of the two tested parameters
also on the rate of changes in the consolidation of particles
in the granules being formed (porosity of the granules), and
in this connection, on energy consumption.
The method of product porosity measurement depended
on the size of product particles. The porosity of granules for
the size fraction dg < 4 mm was determined on the basis of
bulk density measurement. The porosity of granules from a
given size fraction was calculated from the formula
gb
g = 1 (8)
(1 m )
where: m is intergranular porosity; gb , bulk density of
granules (kg/m3 ); and , density of fine-grained material
(kg/m3 ).The intergranular porosity was determined from
 T. Gluba / Chemical Engineering and Processing 44 (2005) 237243
Fig. 6. Energy input for unit growth of granule size.
the equation
gb0
m = 1 (9)
(1 g0 )
where: g0 , porosity of granules at the beginning of gran-
ulation (tg = 0); and gb0 bulk density of granules at the
beginning of granulation (kg/m3 ).
On the basis of the measurements, it was found that inner
porosity of granules obtained after wetting (tg = 0) did not
depend on the amount of added wetting liquid and for a
given raw material is similar to mean porosity of initial dry
material (Fig. 7).
This means that the density of particle packing in the
granules that were formed in the wetting period corresponds
to the density of particle packing in the raw material of
Fig. 7. Comparison of porosity of granules formed after wetting g0 with
mean porosity of raw material .
241
average bulk density b . After the wetting period, the bed
contains, beside granules, non-granulated material particles
whose mass depends on the amount of wetting liquid added.
Hence, spaces between the formed granules can be filled up
partly with non-granulated raw material and partly with air.
The investigations proved that bulk density of the bed after
wetting (in dried state) was often lower than the density of
loose raw material.
It was assumed that intergranular porosity m for grain
fraction i with narrow intervals did not change during
granulation and was constant for a given raw material:
mi = constant (10)
The porosity of granules of size 4 mm was determined
on the basis of mass and volume of a defined number of
granules from a given size fraction from the equation:
mg
g = 1 (11)
Vg
where: mg , mass of granules taken from a size fraction (kg);
and Vg = Vgi , total volume of taken granules (m3 )
The volume of single granules of shape close to spherical
was calculated on the basis of their dimensions in three main
directions, using the formula for ellipsoid. The volume of
granules of shape different from spherical was determined
by the pycnometric method.
On the basis of porosity determined for the granules from
particular size fractions of product samples taken from the
drum in a specific moment, mean porosity of granules was
calculated from the relation
i=max

g (t) = xi gi (t) (12)
i=1
where: g , mean porosity of granules at granulation time tg ;
and gi , porosity of granules from size fraction i at time tg .
242 T. Gluba / Chemical Engineering and Processing 44 (2005) 237243
Fig. 8. Energy input for granule porosity decrease.
Fig. 8 shows examples of unit energy input for the increase
of particle consolidation in the granules (a decrease of mean
granule porosity) for a selected material (D5), at variable
moisture content of the feed.
In Fig. 8, it is clearly visible that the energy required for
consolidation of particles in the newly formed agglomer-
ates decreases with an increase of the bed moisture content.
Porosity of the granules obtained at the end of the process is
similar for all moisture contents of the feed for a determined
particle size distribution of the raw material. Similar rela-
tionships were obtained for other particle size distributions
of the raw material. Fig. 9 illustrates energy input related
to a unit increment of particle consolidation ( = 1g ) in
the granules produced from particular raw materials during
granulation (a unit decrease of granule porosity), depending
on the feed moisture content.
The effect of mean particle diameter of the raw material
and feed moisture content on energy input as related to unit
Fig. 9. Energy input for the increase of particle condensation in granules.
decrease of agglomerate porosity during granulation is de-
scribed by the power function of two variables. As a result
of regression, the following equation at correlation coeffi-
cient R = 0.95 was derived:
Eu
= 5.2 105 dm 3.4 w15.4 (13)

5. Conclusions
The equations describing the energy input for the incre-
ment of agglomerate size and the rate of particle consolida-
tion in it during granulation prove that the greatest impact
has the amount of wetting liquid supplied to the bed dur-
ing wetting (feed moisture content), and slightly lowerthe
mean particle size of the raw material. Even a slight increase
of the quantity of wetting liquid added (in the range se-
lected for each particle size distribution of the raw material)
causes a significant decrease of energy input required for
the transformation of wetted powder material into a granu-
lated product. This refers both to the increase of agglomerate
size and the consolidation of its inner structure. These two
phenomena take place in parallel, so the energy inputs can-
not be distributed to each of them. Besides, not whole en-
ergy input is used for the processes of granules growth and
their consolidation. Part of it is used for destructive mech-
anisms occurring in the tumbling bed, and a part is trans-
formed into heat. However, by selecting for a given particle
size distribution of the raw material an adequate quantity
of the wetting liquid, much energy savings can be reached
in the process of transforming the powder into a granulated
product.
The main issue of every technological process is its
optimisation. Also in the process of fine-grained material
granulation, there is a problem of optimisation of outlays
necessary to form a product of defined, most desirable
properties such as granule size, particle composition, homo-
geneity of the composition, porosity, mechanical strength,
binding agent content, solubility, etc. In many cases, it is
related to the selection of appropriate wetting conditions,
in particular the amount of wetting liquid added. This is a
very complex problem, depending on specificity of a given
process and requiring an individual approach. As proven
by results of the presented research, the addition of a larger
amount of wetting liquid to the fine-grained material causes
a decrease of energy input necessary to produce the granu-
lated material. This increase, however, has also an influence
on changes in product properties that cause formation of
bigger particles with higher moisture content. If such a
product is to be dried, it becomes necessary to provide more
thermal energy. In this case, a reduction of energy input is
not obvious. An increase of the amount of wetting liquid
can also induce some disturbances in the process run re-
lated to feed overwetting (sticking to the drum walls). This
is particularly important for materials whose granulation
proceeds in a very narrow range of moisture content. Addi-
 T. Gluba / Chemical Engineering and Processing 44 (2005) 237243 243

tionally, an increase of liquid content in the granulated bed rotational speed of the granulator (1/s)
causes that the increment of granule size at the rapid growth mean particle consolidation in the granules
stage becomes more vehement, which can sometimes make
the process control difficult and consequently result in an
excessive growth of particles. As was shown in the investi- References
gations, an important indicator of the granulation process is
the torque measured on the apparatus shaft. The changes of [1] J. Mellmann, The transverse motion of solids in rotating cylinders
torque during the granulation can indicate advance of the forms of motion and transition behavior, Powder Technol. 118 (2001)
process and may facilitate its control. 251270.
[2] T. Gluba, A. Heim, A. Obraniak, Driving torque in the process of
drum granulation, In: Proceedings of the 14th International Congress
of Chemical and Process Engineering CHISA, Czech Republic, 2000.
Acknowledgements [3] A. Heim, T. Gluba, A. Obraniak, Bed dynamics during granulation in
rotating drums, In: Proceedings of the 7th International Symposium
The work was carried out under research project no. on Agglomeration, Albi, 2001.
[4] A. Heim, T. Gluba, A. Obraniak, Torque as an indicator of the
W-10/21/2003/B.W.
advance of drum granulation process, In: Proceedings of the 29th
International Conference of Slovak Society of Chemical Engineering,
Slovakia 2002.
Appendix A. Nomenclature [5] K. Terashita, M. Kato, A. Ohike, Analysis of end-point with power
consumption in high speed mixer, Chem. Pharm. Bull. 38 (1990)
19771982.
d particle size of raw material (m)
[6] J. B. Schwartz, C. E. Szymczak, Power consumption measurements
dg particle diameter of granulated product (mm) and the mechanism of granule growth in a wet granulation study,
dgm mean particle diameter of granulated product (mm) AAPS Meeting November, 1997.
di mean particle size of class i (m) [7] S. Watano, K. Terashita, K. Miyanami, Frequency analysis of power
dm mean particle size of raw material (m) consumption in agitation granulation of powder materials with spar-
ingly soluble acetaminophen, Chem. Pharm. Bull. 40 (1992) 269
Eu unit energy during granulation (kJ/kg)
271.
mb mass of wetted feed in the drum (kg) [8] H. Lauenberger, Scale-up of granulation processes with reference to
Mi the torque of idle run (at empty granulator) (Nm) process monitoring, Acta Pharm. Technol. 29 (1983) 274280.
ms mass of dry material in the drum (kg) [9] P. Holm, T. Schaefer, H.G. Kristensen, Granulation in high speed
Mu unit torque (Nm/kg) mixers. Part V. Power consumption and temperature changes during
granulation, Powder Technol. 43 (1985) 213223.
Nu unit power demand (W/kg)
[10] P. Holm, T. Schaefer, H.G. Kristensen, Granulation in high speed
t time (s) mixers. Part VI. Effects of process conditions on power consumption
tg granulation time (s) and granule growth, Powder Technol. 3 (1993) 286293.
tgmax limiting (maximum) granulation time (s) [11] T. Gluba, Drum granulation conditions for raw material with different
xi percentage of mass fraction particle size distributions. Handbook of Conveying and Handling of
Particulate Solids, Elsevier Science, B.V., 2001, pp. 717723.
w moisture content of the feed, kg water/kg
[12] P.A.L. Wauters, R.B. Jakobsen, J.D. Litster, G.M.H. Meesters, B.
dry material Scarlett, Liquid distribution as a means to describing the granule
1 asymmetry coefficient growth mechanism, Powder Technol. 123 (2002) 166177.
mean porosity of raw material [13] P.A.L. Wauters, R. van de Water, J.D. Litster, G.M.H. Meesters, B.
g porosity of granules (intra-granular porosity) Scarlett, Growth and compaction behaviour of copper concentrate
granules in a rotating drum, Powder Technol. 124 (2002) 230237.
m intergranular porosity
[14] S.M. Iveson, J.D. Litster, Growth regime map for liquid-bound gran-
density of fine-grained material (kg/m3 ) ules, AIChE J. 44 (1998) 15101518.
b mean bulk density of raw material (kg/m3 ) [15] S.M. Iveson, P.A.L. Wauters, S. Forrest, J.D. Litster, G.M.H.
bc bulk density of condensed raw material (kg/m3 ) Meesters, B. Scarlett, Growth regime map for liquid-bound granules:
bl bulk density of loose raw material (kg/m3 ) further development and experimental validation, Powder Technol.
117 (2001) 8397.
gb bulk density of granules (kg/m3 ) [16] S.M. Iveson, J.D. Litster, K. Hapgood, B.J. Ennis, Nucleation, growth
w density of water (kg/m3 ) and breakage phenomena in agitated wet granulation processes: a
standard deviation review, Powder Technol. 117 (2001) 339.
 Chemical Engineering and Processing 44 (2005) 245252

Control of aggregation in production and handling of nanoparticles

Wolfgang Peukert a, , Hans-Christoph Schwarzer b , Frank Stenger b
a Institute of Particle Technology, Universitt ErlangenNrnberg, Cauerstr. 4, 91058 Erlangen, Germany
b Institute of Particle Technology, Technische Universitt Mnchen, Boltzmannstr. 15, 85748 Garching, Germany

Received 22 July 2003; received in revised form 15 December 2003; accepted 11 February 2004
Available online 2 July 2004

Abstract

In product engineering of particulate systems, the property function relates the dispersity to the product properties, whereas the process
function shows how to produce the required dispersity. These principles are applied to the production of nanoparticles. Nanoparticles are
controlled by surface forces. Due to their high mobility nanoparticles are unstable and may coagulate rapidly if the particles are not stabilized.
Stabilization is achieved by tailoring the particulate surfaces, e.g. through repulsive double layer forces. Macroscopic properties are thus
controlled by microscopic control of the interfaces, i.e. we bridge the gap between the molecular level and material properties. These principles
are generally valid and are thus applied to precipitation and to nanomilling in stirred ball mills. The mean particle size in precipitation can be
controlled by either the mixing intensity or the surface charge density of the particles. In stirred media mills oxide particles as small as 10 nm
can be achieved by stabilizing the particles appropriately.
2004 Elsevier B.V. All rights reserved.

Keywords: Nanoparticle; Aggregation; Stabilization; Precipitation; Milling

1. Concept of product engineering ties (structureproperty correlation). Examples of property

The production of substances with well-defined product
properties under consideration of economical and ecological
boundary conditions is the objective of many processes. For
particulate materials, the product properties depend not only
on the chemical composition but also on the dispersity of the
material. The dispersity is characterized by the particle size
distribution, the particles shape and morphology as well as
their interfacial properties. This relation was called property
function by Rumpf [1]. The control of the property function
is known as product engineering or product design.
Property function :
Product property = f(dispersity, chemical composition)
where dispersity is characterized by particle size and shape
and their respective distribution, particle morphology and
particle surface properties.
The property function relates the particulate structure
(size, shape, morphology, surface) to the product proper-
Corresponding author. Fax: +49 9131 85 29402.
E-mail address: w.peukert@lfg.uni-erlangen.de (W. Peukert).
functions are the taste of chocolate, the color of pigments,
the strength of cements or the band gap of semiconductor
nanoparticles. Particle ensembles in form of agglomerates,
thin films or filter cakes are also included in this consider-
ation. The process function (processstructure correlation)
as defined by Krekel and Polke [2] relates the process
parameters to the product property.
Process function :
Dispersity = f(process parameters, educt concentrations)
Process parameters are the type of unit operations (e.g.
precipitation), their interaction in the process, process
conditions under which the unit operations are operated
(e.g. temperature, pressure, flow rates, etc.) and the ma-
terials processed. The structureproperty as well as the
processstructure correlations must be known in order to
run a process successfully and achieve the desired goal, i.e.
to produce well-defined product properties. In this paper, we
show how the property function, i.e. the state of aggregation
can be controlled by surface forces of the particles.

0255-2701/$ see front matter 2004 Elsevier B.V. All rights reserved.
doi:10.1016/j.cep.2004.02.018
246 W. Peukert et al. / Chemical Engineering and Processing 44 (2005) 245252
2. From particle surfaces to macroscopic properties
It is common to all nanoparticle applications that inter-
facial and surface properties of the particles play a central
role and that macroscopic properties can only be tailored by
microscopic design of the interfaces. The ratio of adhesion
forces to particle weight scales with particle diameter x2
and it is for instance at 1 m in the order of 106 . Thus, in
order to generate well-defined properties, particle interac-
tions have to be controlled carefully. Surface chemistry and
physics determine not only the particulate interactions with
fluid or solid phases. These types of interactions are van der
Waals forces, polar interactions, hydrogen or even chemical
bonds. Particle interactions also control particle and struc-
ture formation. For product engineering of nanoparticulate
systems, the conceptual starting point is the particle surface
which transports the respective particle interactions, thus
leading to the desired structure. Vice versa, structure forma-
tion can only be understood by considering the relevant in-
teractions which are determined by the particle surface. This
concept is illustrated in Fig. 1 for oxide particles in an aque-
ous solution where particle interactions can be understood
in the view of the well-known DLVO-theory as a superposi-
tion of van der Waals and electrostatic double layer forces.
This approach is based on some wide-ranging precondi-
tions. In order to bridge the gap between microscopic molec-
ular nature of a particle surface and macroscopic properties,
a multi-scale approach covering several orders of magni-
tude of space and time is needed. On the most basic level
quantum mechanics prevail. However, it is often possible by
using the HellmanFeynman theorem [3] to transfer the in-
trinsic quantum mechanical nature of surfaces to the physics
Fig. 1. From particle surfaces to macroscopic properties.
of molecular interactions, described by classical force laws.
This theorem states that once the electron density distribu-
tions have been determined the intermolecular interactions
can be calculated on the basis of classical electrostatics. The
contact value theorem [3] is analogous: the force between
two surfaces is determined by the density distribution of the
molecules and particles between them.
Depending on the application, particles have to be sta-
bilized against aggregation. In general, stabilization can be
achieved by tailoring the particleparticle interactions by
electrostatic or steric means. Whereas in the gas phase the
possibilities of stabilization are rather limited, a variety of
methods exists in the liquid phase. In liquids, particles of-
ten carry electrostatic charges which lead, in accordance
with the DLVO-theory, to repulsive interaction potentials be-
tween the particles since they carry the same charges and
thus to stabilization. Aggregation and stabilization may also
be tailored by polymers. The adsorption of polymers on the
particles surfaces may either influence the surface charge or
act as a steric barrier avoiding close contact of the particles
surfaces. The selection of the optimal polymer for a cer-
tain combination of particle and liquid is still accomplished
empirically, although some guidelines have been formulated
based on polymer physics and chemistry [4]. Special effects
like hydration forces, e.g. observed for silica [5], are not
considered in this paper.
3. Modeling of population dynamics
Aggregation kinetics depends on transport mechanisms
bringing the particles in contact and on particleparticle
 W. Peukert et al. / Chemical Engineering and Processing 44 (2005) 245252 247

interactions. Both aspects influence the kinetics (i.e. the re- above. Particleparticle interactions are not considered. As
sulting particle size distributions as a function of time) and particle density, the value of barium sulfate of 4500 kg/m3
the structure of the aggregates. The most important transport was used. It becomes evident from the results shown in Fig. 2
mechanisms are Brownian diffusion, laminar and turbulent that Brownian motion and turbulent shear at moderate shear
shear due to fluid motion and sedimentation. The follow- rates are the most important transport mechanisms not only
ing equations describe the collision kernels of these mech- for nanoparticles but also for submicron particles.
anisms: To additionally account for the effect of the aggregate
structure on the aggregation kinetics, a collision diameter xc
Brownian diffusion [6]:

 is introduced which can be calculated based on the fractal
2kT 1 1 dimension DF of the aggregate (Eq. (5)).
0 = (x1 + x2 ) + (1)

3 x1 x2 6m 1/DF
xc = (5)
Sedimentation [7]: P

  

x1 2  x2 2 F In this equation m represents the mass of the aggregate and
0 = g + 1 P the density of the aggregate material. Recent small angle
2 2 P
  
 neutron scattering (SANS) measurements have shown that
 x1 2  x2 2  2P
  (2) the fractal dimension of aggregates increases from 1.8 for
2 2  9 Brownian diffusion to values in the range of 2.2 when shear
Laminar shear [6]: forces prevail [9]. Thus, higher external forces lead to denser
aggregates (e.g. caused by compaction) and the aggregate
4  x1 x 2 3 structure depends on aggregation mechanism.
0 = + (3)
3 2 2 In order to take particleparticle interactions into account,
Turbulent shear [8]: a stability ratio W is included which relates the collision ker-

 nel 0 to the aggregation kernel agg . The stability ratio W
8  x1 x2 3 F 1/2 depends on the interaction potential (r) and is defined as
0 = + (4)
15 2 2 ratio of the aggregation rate without to the rate with inter-
actions additional to the omnipresent van der Waals forces.
In these equations, x1 and x2 denote the particle sizes of the For Brownian motion as dominant reason for collisions, the
colliding particles, T the absolute temperature, k the Boltz- stability ratio W can be calculated according to Eq. (6) taken
mann constant, the dynamic viscosity of the fluid, F the from Fuchs [10]. In case of shear as aggregation mechanism,
density of the fluid, P the density of the particle, the lo- the force d/dr relative to the friction force should rather be
cal shear rate and the local mass-specific power input due considered instead of the ratio of interaction energy relative
to turbulence. to thermal energy.
Fig. 2 compares collision kernels calculated for a 250 nm 0 0
particle as function of the collision partner size for Brown- agg = =  (6)
W x x (1/r )exp((r)/kT) dr
2
ian motion, laminar and turbulent shear flows as well as sed-
imentation at 25 C in water based on the equations given Particles can either be produced by bottom-up pro-
cesses (e.g. precipitation) or topdown approaches (e.g.
wet milling). In these processes particleparticle interac-
10-12 Brownian diffusion tions become relevant when the particle size is below 1 m.
sedimentation Engineering macroscopic product properties is then only
10-13 laminar shear (10 s-1)
collision kernel in m3s-1

possible through tailored surface and interfacial properties,

10-14 turbulent shear (10 W/kg)
no matter whether a bottomup process like precipitation
10-15 [11] or a topdown process such as milling in stirred media
10-16 mills [12] is studied. Aggregation is an important aspect in
10-17 both processes which are studied in the following.
In precipitation, particle formation is extremely fast due
10-18
to high supersaturations which in turn lead to fast nucleation.
10-19 At least in the beginning, size distributions are narrow with
10-20 particle sizes around one 1 nm. Nanomilling in stirred media
10-21
mills is characterized by relatively slow particle formation
1 10 100 1000 10000 kinetics, particle sizes ranging from several microns down
to 10 nm and high solids volume concentrations of up to
particle size of collision partner in nm 40%. Large particles may scavenge the fine fractions. The
Fig. 2. Comparison of collision kernels in water for a 250 nm particle as evolution of the particle size distribution can be described
function of the collision partner size. for both cases by population balance equations (Eq. (7)),
248 W. Peukert et al. / Chemical Engineering and Processing 44 (2005) 245252

which superposes the terms for nucleation, growth, grinding higher initial concentrations, smaller particles would result.
and aggregation. While nucleation and growth act as supersaturation sink,
mixing can be considered as supersaturation source. Con-
n(x) dn sequently, the time scale of mixing must fit the time scale
= Bhom (xn ) + G + Bagg (n, x) Dagg (n, x)
t 
 dx 

nucleation

of particle formation, i.e. mixing must be at least as fast as
aggregation
growth precipitation, and thus highly turbulent mixing is required
 xmax for nanoparticle generation.
S(x)n(x) + S(y)n(y)b(x, y) dy (7) The precipitation process was simulated based on the
x

 population balance equation (Eq. (7) without the grinding
grinding
terms) coupled via mass and species balances with a mixing
In this equation Bhom represents the nucleation rate at which model based on the Engulfment model of micromixing by
nuclei of size xn are formed. G is the linear growth rate and Baldyga and Bourne [15]. The modifications of the mixing
Bagg as well as Dagg are the well-known expressions for the model are described in [14]. The particle formation is mod-
birth and death rate of aggregation [7], given below. Within eled through homogeneous nucleation based on the classical
Eq. (7) S(y) is the selection function describing the relative description and diffusion-controlled growth (for details see
amount of particles with size y breaking within a time step [16]). The aggregation of particles is modeled by Eqs. (4), (8)
dt. The breakage function b(x, y) describes the number of and (9). As solver for the population balance equation, the
fragments with size x produced per breaking particle y. commercial software PARSIVAL by CiT GmbH was used.
The implemented algorithm fits the particle size distribu-
Bagg (n, x) tion by polynomials based on a finite-element type Galerkin
    h-p-method optimizing the number of variables (number of
 x agg
3
x 3 3 , n x 3 n()
3 3
x2 polynomials, order of polynomials) and applies an adaptive
= d time discretization of Rothes type. Details on the simula-
2 0 (x3 3 )2/3
(8) tion algorithm can be found in [17]. Besides computational

Dagg (n, x) = n(x) agg (x, )n() d
0 size distribution are necessary and that the overall accuracy
The kinetics of stabilization must match the kinetics
of aggregation. While ion diffusion and adsorption is fast
(electrostatic stabilization), the diffusion and adsorption of
polymers (steric stabilization) is slow and may be a limit-
ing factor which can only be overcome by applying high
polymer concentrations. The stability of the electrostatic
stabilized suspension in moderately fast processes can for
example be monitored and controlled in-line by modern
electro-acoustic techniques [13].
4. Stabilization during precipitation
In the conducted experiments, barium and sulfate solu-
tions were mixed continuously in a T-mixer resulting in a
supersaturated mixture and consequently the formation of
barium sulfate particles. Apart from educt concentrations it
was found that the intensity of mixing has a strong influ-
ence on the resulting particle size distribution [11,14] in the
nanometer range. The more intense the mixing, the smaller
the particles. The reason for this behavior lies in the differ-
ent dependencies of nucleation and growth rates on super-
saturation, the thermodynamic driving force of solid forma-
tion. With increasing supersaturation, the nucleation rate as
a highly non-linear function increases much stronger than
the growth rate. As a consequence, considerably more parti-
cles are formed at higher supersaturations. The increase is so
strong that even if higher supersaturations were achieved by
speed, the main advantages of the Galerkin h-p-method are
that no assumptions concerning the shape of the particle
(9)
can be estimated quantitatively from the mass defect during
simulation. Simulations of the precipitation process includ-
ing mixing are promising and the number-weighted mean
particle size can be predicted very well by taking into ac-
count mixing, nucleation, particle growth and aggregation.
It is currently worked at an advanced approach that com-
prises computational fluid dynamics (CFD) modeling cou-
pled with mixing and the solution of the population balance
accounting for nucleation, growth and aggregation in order
to improve the prediction of the width of the particle size
distribution.
Simulation results of a precipitation process as presented
in Fig. 3 clearly show the different time scales of mixing, par-
ticle formation and aggregation. Fig. 3 also shows the range
of aggregation for the mentioned conditions and thus the
possible range of particleparticle interactions. Depending
on stabilization conditions, i.e. for stability ratios W ranging
from 1 (no repulsive interaction) to infinity, very different
particle sizes (distributions) are obtained even at times in the
range of milliseconds. Thus, the aggregation of nanoparti-
cles is a challenge to be met on time-scales of milliseconds
when controlling or tailoring particle size distributions in
precipitation. This is also depicted in Fig. 4, showing sim-
ulation results of the volume-weighted particle size for dif-
ferent mixing and aggregation conditions. Since fast mixing
requires highly turbulent flows, it comes along with signif-
icant turbulent aggregation. In the presented calculations a
constant stability ration of W = 1 and aggregation only in
the mixer was considered (i.e. during the mean residence
 W. Peukert et al. / Chemical Engineering and Processing 44 (2005) 245252 249
mean (volume) particle size in nm

1200 120 residence time in ms

supersaturation particle size,
aggregation (W=1) 50 20 10 5 2 1 0.5
1000 100
supersaturation S

mean (volume) particle size in nm

1400 turbulent, W=1

800 80
mixing & no agglomeration residence
aggregation 1200
supersaturation Brownian, W=1 time
600 build-up 60
ra

1000
ng
e

particle
400 formation 40 800
particle size,
stabilization

no aggregation 600
200 20
range

(W)
400
0 0
10-6 10 -5
10 -4
10 -3
10 -2
10 -1 200

precipitation duration in s 0
100 101 102 103 104 105 106 107
Fig. 3. Time scales in precipitation calculated for BaSO4 precipitated from
0.5 m BaCl2 - and 0.33 m H2 SO4 -solutions at 25 C. To quantify mixing, mean specific power input in W/kg
a specific power input typical for the applied T-mixer ( = 103 W/kg)
was used. Fig. 4. Simulated influence of aggregation and stabilization in nanoparticle
precipitation.
time), which leads to a residence time effect (the faster the
flow through the mixer, the larger the specific power input cipitation. The theoretically calculated supersaturation (as-
but shorter the residence time). As can be seen, particle sizes suming ideal mixing) was kept constant in all experiments
well in the micrometer range can easily be reached within while the initial molar ratio R of barium to sulfate ions was
milliseconds and that the influence of turbulent aggregation altered. Operating parameters such as flow rate and mixer
can be controlled (besides stabilization) by the duration of geometry were kept constant. Thus, it can be assumed that
aggregation. Latter, however, only applies to turbulent aggre- mixing, the build-up of supersaturation and solid formation
gation. Aggregation due to Brownian motion must be con- were similar in all cases and that possible small variations
trolled by stabilization since the resulting aggregates would cannot account for the measured differences. Hence, the sig-
reach the micrometer range within seconds. nificant differences in particle size distributions are due to
Concerning aggregation kinetics and stabilization it was the altered particleparticle interactions and the resulting al-
found in case of barium sulfate that barium ions adsorb more tered stabilization. As the barium excess is increased, so
readily than sulfate ions [11]. Thus, increasing the ratio R of is the stability against aggregation. Thus, particle size and
barium to sulfate ions in the educt composition and thereby eventually aggregate structure of a precipitation process can
increasing the excess of barium ions after precipitation leads be adjusted to fit needs.
to high surface potentials and stabilization. Fig. 5 shows Recent investigations [18] to quantify stabilization were
measured particle size distributions of continuously precip- successful in correlating to observed stability of colloidal
itated barium sulfate, measured based on quasi-elastic light barium sulfate suspensions with the suspension compo-
scattering (UPA 150 by Microtrac) within 3 min after pre- sition and in particular the concentrations of Ba2+ - and

1.0
cumulative volume distribution

R = 10
R = 3.0
0.8 R = 2.0
R = 1.5
R = 1.2
0.6
R = 1.1
R = 1.0
0.4
excess of
barium ions out of
0.2
measurement
range
0.0
20 100 1000 6000
particle size in nm
Fig. 5. Measured particle size distributions of precipitated barium sulfate showing the influence of the barium to sulfate initial ratio R on stability and
aggregation, respectively.
250 W. Peukert et al. / Chemical Engineering and Processing 44 (2005) 245252
mean (volume) particle size in nm

400 1
R = 1.5 turbulent shear, = 400 W/kg
particle size x50,3 in m

D = 20
R = 1.3
300
D = 1.25
stability
0.1 factor
200 R = 2 R = 1.5
D = 80 D = 100 pH = 7.8
pH = 5.1
pH = 4.1
100 pure breakage
simulation
0.01
104 105
0 time in s
100 101 102 103 104 105 106
Fig. 7. Comparison of mean particle size obtained in a wet milling of
duration in s -alumina (cm = 0.2) with the results of a population balance model.

Fig. 6. Comparison of experimental data and simulation results predicting

the evolution of mean particle size for different suspension compositions
characterized by the initial molar ratio R prior to precipitation and the
dilution ratio D after precipitation.
H+ -ions. These ions adsorb on barium sulfate and thereby
determine the surface properties, i.e. the surface charge
which is relevant to aggregation kinetics, since the result-
ing repulsive particleparticle interactions lead to increased
stability against aggregation. Fig. 6 shows a comparison
of experimental data and the predicted evolution of the
volume-weighted mean particle size with time for vari-
ous compositions (characterized by the initial molar ratio
R and the dilution ratio D after precipitation), showing
a good agreement. Thus, the general concept presented
in Section 2 is found appropriate to describe, model
and control aggregation and stabilization in nanoparticle
precipitation.
5. Stabilization during wet milling
It has been shown recently in a cooperation with the
Schwedes group [12] that alumina can be ground to 10 nm
x0 + (y x0 )/3 full-filling the mass conservation equation.
In a first simulation step the pure grinding process was
modeled and the parameters of the selection function c and
K were adjusted for particles >1 m (K = 1500 s1 mc ,
c = 0.41). With the assumption of a constant grinding
behavior for particles <1 m, the system of differential
equations for grinding and aggregation was solved. Sim-
ulations for grinding and turbulent shear were carried out
to adjust a constant, size-independent stability ratio W to
the experimental data. Again, the system of differential
equations was solved by the above-mentioned software
PARSIVAL.
As can be seen the numerical simulations describe the
observed grinding behavior and the influence of different
suspension stability conditions reasonably well, although
quite stringent assumptions were made in our model. This
supports the theory that the sub-micron grinding process is
strongly influenced by suspension stability. In Fig. 8 a com-
parison of the cumulative mass distribution for the grind-
ing experiment conducted at pH 5.1 and the simulation re-
sults are presented. Again a reasonable agreement between
1.0
cumulative mass distribution Q3

particles if particle aggregation is carefully controlled by the pH = 5.1

principles discussed in this paper and appropriate milling pa- .
W = 2.9 10
7 0 min
0.8
rameters. Fig. 7 compares experimental results of milling ex- 15 min
periments of -alumina at a concentration of cm = 20 wt.% 20 min
in a continuously operated 1 l Drais mill with simulation 0.6 30 min
results. In the experiment -potential was characterized on- 60 min
line by electro-acoustic spectroscopy and particle stabil- 0.4 120 min
240 min
ity was controlled through pH-adjustment (10 M HNO3 ).
480 min
The simulation results were derived for a combined pop- 0.2
1620 min
ulation balance model for aggregation and breakage based 2280 min
on Eq. (7). 0.0
Within Eq. (7), the selection function S(y) was approx- 0.01 0.1 1 10 100
imated by S(y) = Kyc with adjustable parameters K and c particle size x in m
and the breakage function b(x, y) was simply modeled by
Fig. 8. Comparison of the cumulative mass distribution Q3 obtained by
a triangular shaped function between a minimal breaking wet milling of -alumina (cm = 0.2, pH 5.1) at different grinding times
particle size of x0 = 25 nm and x with a modal value at with simulation results.
 W. Peukert et al. / Chemical Engineering and Processing 44 (2005) 245252 251

simulation (lines) and experiment (symbols) is achieved es- T temperature (K)

pecially for short and very long grinding times. Differences t time (s)
at the edges of the distributions for intermediate grinding W stability ratio ()
times can be explained with the fact that the experimen- x particle size (diameter of volume-equivalent
tal particle size distributions could not be determined with sphere) (m)
one single method over the entire time of the experiment. x0 minimal fracture size (m)
While a static light scattering instrument (Malvern Master- xn size of nuclei (m)
Sizer 2000) was used for particle sizes >1.0 m smaller par- xc collision diameter of aggregates (m)
ticles were analyzed by the method of quasi elastic light xmax size of largest particles (aggregates) (m)
scattering (UPA 150 by Microtrac). y particle size (m)
z surface-to-surface distance of particles (m)

6. Conclusions Greek symbols

Nanoparticles produced by precipitation and by wet
milling must be stabilized against aggregation. This can be
achievedas shown in the presented examples by experi-
ment and simulationby microscopic control of particulate
interfaces, i.e. by the particle surface charge density and
the electrostatic repulsion. Thus, the overall concept of
controlling the microscopic properties in order to tailor the
macroscopic behavior was demonstrated. In this paper, we
show how generally applicable concepts derived from col-
loid and surface science can successfully be applied to a
variety of different unit operations. We thus transcend the
traditional view of the unit operations concept in Chemi-
cal Engineering in two ways: first, we introduce concepts
which are applicable to various unit operations and thus
open a new engineering point of view for the design of
such systems. Second, we include the material influence
based on a sound physical understanding of the underlying
molecular phenomena.
Appendix A. Nomenclature
Bagg birth rate of aggregation (m4 s1 )
nucleation rate (m3 s1 )
Bhom
b breakage function (m1 )
c constant in selection function ()
cm mass fraction ()
D dilution ratio ()
Dagg death rate of aggregation (m4 s1 )
DF fractal dimension of particles (aggregates) ()
G linear growth rate (m/s)
g gravitational constant, g = 9.81 m/s2 (m/s2 )
K constant in selection function (s1 mc )
k Boltzmann constant, k = 1.381 1023 J/K (J/K)
m particle mass (kg)
n number density concentration (m4 )
pzc point of zero charge
R molar ratio of barium to sulfate ions prior
to precipitation ()
r center-to-center distance of particles (m)
S breakage function (s1 )
0 collision kernel (m3 /s)
agg aggregation kernel (m3 /s)
shear rate (s1 )
mass-specific power input (W/kg)
dynamic viscosity (Pa s)
F fluid density (kg/m3 )
P density of particle material (kg/m3 )
particleparticle interaction potential energy (J)
0 surface charge density (C/m2 )
References
[1] H. Rumpf, ber die Eigenschaft von Nutzstuben, StaubReinhaltung
der Luft 27 (1967) 313.
[2] J. Krekel, R. Polke, Quality assurance in process-development, Chem.
Ing. Tech. 64 (1992) 528535.
[3] J. Israelachvili, Intermolecular and Surface Forces, second ed., Aca-
demic Press, London, 1991.
[4] D.H. Napper, Polymeric Stabilization of Colloidal Dispersions, Aca-
demic Press, London, 1983.
[5] L. Gnther, W. Peukert, The relevance of interactions in nanopar-
ticle systemsapplication to particulate thin films, Part. Part. Syst.
Charact. 19 (2002) 312.
[6] M. von Smoluchowski, Versuch einer mathematischen Theorie der
Koagulationskinetik kolloidaler Lsungen, Z. Phys. Chem. 92 (1917)
129168.
[7] A. Mersmann, B. Braun, L. Loeffelmann, Prediction of crystallization
coefficients of the population balance, Chem. Eng. Sci. 57 (2002)
42674275.
[8] P.G. Saffman, J.S. Turner, On the collision of drops in turbulent
clouds, J. Fluid Mech. 1 (1956) 1630.
[9] M. Sommer, N.J. Wagner, W. Peukert, J. Green, D. Spahr, Funda-
mental investigations on nanoparticle production by stirred media
milling, in: Proceedings of the AIChE Annual Meeting San Fran-
cisco, CD-ROM, 1621 November 2003, paper 54f.
[10] N. Fuchs, ber die Stabilitt und Aufladung der Aerosole, Z. Phys.
89 (1934) 736743.
[11] H.C. Schwarzer, W. Peukert, Experimental investigation into the in-
fluence of mixing on nanoparticle precipitation, Chem. Eng. Technol.
25 (2002) 657661.
[12] S. Mende, J. Schwedes, F. Stenger, W. Peukert, Mechanical pro-
duction and stabilization of submicron particles in stirred ball mills,
Powder Technol. 132 (2003) 6473.
[13] A.S. Dukhin, P.J. Goetz, Acoustic spectroscopy for concentrated
polydisperse colloids with high density contrast, Langmuir 12 (1996)
49874998.
252 W. Peukert et al. / Chemical Engineering and Processing 44 (2005) 245252

[14] H.C. Schwarzer, W. Peukert, Tailoring particle size through nanopar-
ticle precipitation, Chem. Eng. Comm. 191 (2004) 580606.
[15] J. Baldyga, J.R. Bourne, Turbulent Mixing and Chemical Reactions,
John Wiley, Chichester, 1999.
[16] A. Mersmann, Crystallization Technology Handbook, second ed.,
Marcel Dekker, New York, 2001.
[17] M. Wulkow, A. Gerstlauer, U. Nieken, Modeling and simulation of
crystallization processes using Parsival, Chem. Eng. Sci. 56 (2001)
25752588.
[18] H.C. Schwarzer, W. Peukert, Prediction of aggregation kinetics based
on surface properties of nanoparticles, Chem. Eng. Sci., in press.
 Chemical Engineering and Processing 44 (2005) 253261

Effect of different types of impact surface on coal degradation

R.K. Sahoo , D. Roach
Faculty of Engineering and Physical systems, Central Queensland University, Rockhampton, QLD, Australia

Received 1 August 2003; received in revised form 25 December 2003; accepted 11 February 2004
Available online 25 June 2004

Abstract

Degradation characteristics of South Blackwater and Ensham coal from the stockpiles of Gladstone Port Authority were tested by using
a drop tests apparatus to minimise the fines generation during transport and handling through the Port Authoritys ship loading facilities.
Non-cushioned coal sample of both types of coal dropped on different impact surfaces produced higher percentage of fines than the sample
of cushioned coal, initial 10% fines, initial 30% fines and initial 50% fines. The effect of different types of surfaces onto which the coal was
dropped with initial fines was determined by dropping onto a steel surface, conveyor surface and coal surface (stockpile). The results showed
that the tests with initial 30% fines with lump coal has no significant difference of fines generation in between the three impact surfaces. Both
types of coal with initial 10% fines dropped on steel surface produced more fines than conveyor and coal surface. The cushioning effect of fines
on coal impact surface was least compare to other two impact surfaces. The degradation model was used satisfactorily for South Blackwater
coal on different impact surfaces.
2004 Elsevier B.V. All rights reserved.

Keywords: Cushioning effects; Degradation; Mean particle size; Types of impact surface

1. Introduction power generation or pulverised coal injection (PCI) in a blast

South Blackwater and Ensham coal are shipped from
Gladstone port in QLD, Australia to the Jindal COREX
plant in India. The COREX process is a unique and modern
process of iron making that has many distinct advantages
over the conventional blast furnace iron making process.
The COREX process does not required coking coal and uses
only thermal coal for generating heat, reduction gas and ade-
quate permeability. The properties of South Blackwater and
Ensham coal used in the Jindal COREX plant are given in
Table 1.
The Mean particle size (MPS) of coal is an important
factor for the COREX Process because a decrease in the
MPS reduces the permeability of the char bed resulting in
gas channelling, which drops the hot metal temperature and
quality. To maintain consistency, the MPS of the coal should
be maintained within 2030 mm, which is a very important
factor for efficient plant operation. The COREX technology
coal should be sized at +16 mm with a top size as high as
7080 mm. Fines coal (16.0 mm) included with the sized
coal is separated at the COREX steel mill and used for either
Corresponding author. Tel.: +61 7 4930 9313.
E-mail address: r.sahoo@cqu.edu.au (R.K. Sahoo).
furnace.
Between the mine and end user, lump coal is subject to
a number of mechanical actions, which cause degradation.
These include drops at transfer points, screening, stockpil-
ing (which may involve dozing) reclaiming and ship loading.
In passing through various processing stages, the coal fre-
quently encounters a number of transfer points, i.e. transfer
from one conveyor belt to another, or transfer from a con-
veyor belt into a screen, a feed bin, a rail car, the hull of a
ship, or onto stockpiles. All of these transfer points provide
opportunities for some of the lumps to undergo breakage
into smaller pieces and produce fines. Large coal lumps fre-
quently command better price than fines so that degradation
during transportation and handling can represent a consid-
erable loss of revenue.
Degradation is defined by Goodwin and Ramos [1] and
Sahoo et al. [2,3] as the loss or gain of a given size fraction
to a smaller size fraction through the range of final products
expressed as a percentage of the whole.
In order to predict the total level of degradation of fines
for coal handling systems of the Gladstone port, a study was
commenced to examine the mechanism of lump degradation
and to reduce the fines generation. The effect of different
types of surface onto which the coal with initial fines was

0255-2701/$ see front matter 2004 Elsevier B.V. All rights reserved.
doi:10.1016/j.cep.2004.02.019
254 R.K. Sahoo, D. Roach / Chemical Engineering and Processing 44 (2005) 253261

Table 1 the distribution of this mass can be described by using an

Properties of South Blackwater and Ensham coal
Properties Values (%)
South Blackwater coal Ensham coal
Fixed carbon 63.0 61.6
Volatile matter 27.6 27.3
Moisture 8.0 9.0
Ash 9.4 11.1
Sulphur 0.6 0.7
dropped was determined by dropping onto a steel surface,
conveyor surface and coal surface (stockpile).
Degradation is not only confined to coal but also occurs
with other materials such as iron ore, coke, sinter, limestone
and pellets. In fact degradation is a problem in almost all
bulk materials handling facilities.
2. Literature review
Fagerberg and Sandberg [4] reviewed the literature on
the degradation of iron ore lumps in transportation. Their
investigation showed that, the disintegration of lump ores is
mainly caused by impact when the material was dropped and
to a lesser extends by compression and abrasion in gravity
flows through hoppers, bins and similar devices. Their results
suggested that a small number of large drops caused about
the same amount of degradation as many smaller drops that
add up to the same total height, which was contrary to a
report by Waters et al. [5] and Sahoo et al. [2].
Norgate et al. [6] has produced a mathematical model of
the degradation of ore lumps during transportation and han-
dling. It was reported that, factors such as drop height (from
0.9 to 32 m); ore type, ore size distribution, pre drop and
post drop handling cause the degradation of lump ore. The
probability of breakage is a statistical process that depends
on the size of particles present, their distribution of tensile
strength and the applied stress. The nature of stress applied
appropriate breakage distribution function of Austin et al.
[7]:



sj sj
BSj,Si = + (1 ) (2)
si si
where , and are parameters and BSj,Si is the mass frac-
tion of material of size sj or smaller resulting from breakage
of particles of initial size si .
Their analysis of experimental data shows that tensile
strength was the only parameter that varied as a function
of particle size. Model simulation indicate that larger lumps
were stronger than smaller lumps which is contradictory to
Waters et al. [8], Teo and Waters [9], and Griffith [10] the-
ory of fracture, which implies that larger particle are more
likely to contain larger cracks and hence be more suscepti-
ble to breakage.
Waters et al. [5] stated that the presence of iron ore fines
in the test sample at levels greater than 10% by mass had a
cushioning effect, which reduced lump breakage during drop
tests. They also observed that the amount of degradation
was not significantly different, when lump iron ore drop on
different impact surface.
3. Experimental procedure
A five-meter drop tower was used to drop the coal lumps
on different impact surfaces shown in Fig. 1. The differ-
ent impact surfaces are steel plate, conveyor belt and coal
stockpile. The details development of the drop tower was
40 cm
70 cm

to each individual particle depends on the height of the drop,

the orientation of each particle at impact and the number of
contact points between each particle and its nearest neigh-
bours i.e. coordination number of particle.
In order to keep the mathematical modelling in a tractable
form Norgate et al. [6], the individual processes were lumped
15 cm

into a semi-empirical equation, viz.:

g2
1/2 (h h0 )g1 si
i = (1)
i (CN)g3
Polythene envelope
where i is the mass of size si that is broken when unit mass
of particles of the size si is dropped from a height h. i and Steel plate (2 m 2 m 3 mm)
CN are the mean tensile strength and mean co-ordination
number for particles of size si respectively, and , h0 , g1, g2
and g3 are model parameters.
2m
For unit mass of particle of size si , the amount remaining
is (1i ). The broken mass i , appears in smaller sizes and Fig. 1. Line diagram of drop tower.
 R.K. Sahoo, D. Roach / Chemical Engineering and Processing 44 (2005) 253261 255

discussed in Sahoo et al. [2]. The drop tower was enclosed 50

South Blackwater Coal, Steel Surface
in plastic sheeting to protect the loss of coal fines due to
Cumulative Percentage of fines

external winds during drop. The coal was containing in the 40

released hopper with a bottom release that allowed the coal
Non-cushioned coal
to be dropped from various heights. 30
Cushioned coal
Samples of South Blackwater and Ensham coal of Initial 10% fines
+16.0 mm size were collected separately, from the stock- 20
Initial 30% fines
piles of the Gladstone Port. Before testing, both types of Initial 50% fines
10
coals were sieved to obtain a 76.2 + 50.8 mm sized sam-
ples for the drop tests. The sample was placed inside the 0
released hopper, which was raised by overhead crane. A 0 1 2 3 4 5 6
sample of 10 kg was dropped each time onto an enclosed Number of drops

horizontal steel plate, conveyor belt and coal stockpile. The
dropped coal was sieved to remove the fines (16.0 mm
size). In this way the percentage of fines produced at each
drop were determined. The calculation methodology for
fines percentage was given in Sahoo et al. [2].
Experiments were carried out to determine the reduction
of lump coal degradation due to the presence of coal fines
(16.0 mm) with the lumps at the different impact surfaces.
Cushioning effects of fines consists of five different types
of tests are as follows.
3.1. Non-cushioned coal
Non-cushioned coal is a coal sample having an initial
size range 76.2 + 50.8 mm, which was used for repeated
drops, and has the fines component (16.0 mm size fraction)
removed after each drop.
3.2. Cushioned coal
Cushioned coal is a coal sample having initial size range
76.2 + 50.8 mm, which was used for repeated drops, but
does not remove the fines component after each drop.
3.3. Initial 10% fines
It is a coal sample containing 90% mass of sample to be
dropped of size range 76.2 + 50.8 mm and a 10% mass of
sample of fines of 16.0 mm size fraction.
Fig. 2. Cumulative percentage of 16.0 mm fines vs. number of drops of
five types of cushioning tests of South Blackwater coal on steel surface.
4. Results and discussion
4.1. Cushioning tests of two types of coal on three different
impact surfaces
4.1.1. Steel surface
The cushioning effects of fines of South Blackwater and
Ensham coal tested on steel surface at 5 m drop height
were shown in Figs. 2 and 3. Data for the Fig. 2 are given
in Table 2. The result of these figures shows that, there
was a significant difference of fines percentage between the
non-cushioned and cushioned coal sample. Non-cushioned
coal sample of South Blackwater coal dropped on steel sur-
face produced 12% more fines than a cushioned coal sam-
ple, 13.3% more fines than a sample of initial 10% fines,
27.5% more fines than a sample of initial 30% fines and
30.9% more fines than a sample of initial 50% fines at 30 m
total drops (six drops at 5 m) shown in Fig. 2.
At 5 m drop height non-cushioned coal sample of Ensham
coal dropped on steel surface produced 11% more fines than
a cushioned coal sample, 15.7% more fines than a sample of
initial 10% fines, 26.3% more fines than a sample of initial
30% fines and 29.8% more fines than a sample of initial
50% fines at 30 m total drops shown in Fig. 3.
Fig. 2 shows that the cushioned coal sample of South
Blackwater coal dropped on steel surface produced 1.3%
50
Ensham Coal, Steel Surface
Cumulative percentage of fines

3.4. Initial 30% fines 40

30 Non-cushioned coal
It is a coal sample containing 70% mass of sample to be
cushioned coal
dropped of size range 76.2 + 50.8 mm and a 30% mass of 20 initial 10% fines
sample of fines of 16.0 mm size fraction. initial 30% fines
10
initital 50% fines

3.5. Initial 50% fines 0

0 1 2 3 4 5 6
Number of drops
It is a coal sample containing 50% mass of sample to be
dropped of size range 76.2 + 50.8 mm and a 50% mass of Fig. 3. Cumulative percentage of 16.0 mm fines vs. number of drops of
sample of fines of 16.0 mm size fraction. five types of cushioning tests of Ensham coal dropped on steel surface.
256 R.K. Sahoo, D. Roach / Chemical Engineering and Processing 44 (2005) 253261

Table 2
Cumulative percentage of 16.0 mm fines of five types of cushioning tests of South Blackwater coal dropped on steel surface from 5 m height
Number of drops Cumulative percentage of 16.0 mm fines of South Blackwater coal

Non-cushioned coal Cushioned coal Initial 10% fines Initial 30% fines Initial 50% fines

1 7.4 6.2 5.7 2.0 0.9

2 15.0 12.4 11.3 5.1 2.6
3 22.5 18.1 15.9 7.0 4.4
4 29.7 21.7 20.4 10.1 6.7
5 36.0 26.4 25.4 12.1 9.3
6 41.7 29.7 28.4 14.2 10.8

35 South Blackwater Coal, Conveyor Surface 35 Ensham Coal, Conveyor Surface

Cumulative Percentage of Fines

Cumulative Percentage of Fines

30
30
25
25
20 Cushioned coal
Initial 10% fines Cushioned coal
20
15 Initial 10% fines
Initial 30% fines
10 15 Initial 30% fines

5 10
0
0 1 2 3 4 5
Number of Drops
Fig. 4. Cumulative percentage of 16.0 mm fines vs. number of drops
of three types of cushioning tests of South Blackwater coal dropped on
conveyor surface.
more fines than a sample of initial 10% fines, 15% more
fines than a sample of initial 30% fines and 19% more fines
than a sample of initial 50% fines at 30 m total drops. The
cushioned coal sample of Ensham coal dropped on steel
surface produced 5% more fines than a sample of initial 10%
fines, 16% more fines than a sample of initial 30% fines and
19% more fines than a sample of initial 50% fines at 30 m
total drops shown in Fig. 3.
4.1.2. Conveyor surface
Fig. 4 shows that the cushioned coal sample of South
Blackwater coal dropped on conveyor surface produced
0.5% more fines than a sample of initial 10% fines and
8.3% more fines than a sample of initial 30% fines at 30 m
total drops. At 5 m drop height the cushioned coal sample of
Ensham coal dropped on conveyor surface produced 4.5%
more fines than a sample of initial 10% fines and 12% more
fines than a sample of initial 30% fines at 30 m total drops
5
6
0
0 1 2 3 4 5 6
Number of Drops
Fig. 5. Cumulative percentage of 16.0 mm fines vs. number of drops
of three types of cushioning tests of Ensham coal dropped on conveyor
surface.
shown in Fig. 5. The data for the Figs. 4 and 5 are given in
Table 3.
4.1.3. Coal surface
Fig. 6 shows that the cushioned coal sample of South
Blackwater coal dropped on coal surface produced 1.3%
more fines than a sample of initial 10% fines and 4.3% more
fines than a sample of initial 30% fines at 30 m total drops.
The cushioned coal sample of Ensham coal dropped on coal
surface produced 2% more fines than a sample of initial
10% fines and 4.5% more fines than a sample of initial 30%
fines at 30 m total drops shown in Fig. 7. The data for the
Figs. 67 are given in Table 4.
The results of cushioning effects of both types of coal in
Figs. 6 and 7 shows that the initial fines with lump coal does
not reduce much fines generation when dropped on coal

Table 3
Cumulative percentage of 16.0 mm fines of three types of cushioning tests of two types of coal on conveyor belt surface from 5 m height
Number of drops Percentage of 16.0 mm fines of South Blackwater coal Percentage of 16.0 mm fines of Ensham coal

Cushioned coal Initial 10% fines Initial 30% fines Cushioned coal Initial 10% fines Initial 30% fines

1 3.4 2.8 1.2 4.1 2.6 1.2

2 7.7 6.9 3.2 8.5 6.1 3.1
3 11.6 10.9 5.5 12.8 10.1 5.3
4 15.1 14.3 8.5 16.6 13.5 8.2
5 17.4 17.1 11.6 19.9 16.2 10.5
6 21.2 20.7 12.9 23.9 19.4 12.2
 R.K. Sahoo, D. Roach / Chemical Engineering and Processing 44 (2005) 253261 257

35 South Blackwater Coal, Coal Surface South Blackwater Coal, Cushioned Coal
40
Cumulative Percentage of Fines

Cumulative Percentage of Fines

30

25 30
Cushioned coal
20 Steel plate
Initial 10% fines
15 20 Conveyor belt
Initial 30% fines
Coal piles
10

5
10

0
0 1 2 3 4 5 6 0
0 1 2 3 4 5 6
Number of Drops
Number of Drops
Fig. 6. Cumulative percentage of 16.0 mm fines vs. number of drops
of three types of cushioning tests of South Blackwater coal dropped on Fig. 8. Comparison of percentage of fines generation of three impact
coal surface. surfaces of South Blackwater coal tested with cushioned coal sample.

Ensham Coal, Coal Surface

35
Cumulative Percentage of Fines

40 Ensham Coal, Cushioned Coal

Cumulative Percentage of Fines

30

25
30
20 Cushioned coal
Steel plate
Initial 10% fines
15 Conveyor belt
Initial 30% fines 20
10 Coal piles

5 10
0
0 1 2 3 4 5 6
0
Number of Drops 0 1 2 3 4 5 6
Number of Drops
Fig. 7. Cumulative percentage of 16.0 mm fines vs. number of drops of
three types of cushioning tests of Ensham coal dropped on coal surface. Fig. 9. Comparison of percentage of fines generation of three impact
surfaces of Ensham coal tested with cushioned coal sample.
impact surface. Hence, cushioning effects of fines on coal
surface is minimum compare to other two impact surfaces.
South Blackwater Coal, Initial 10% Fines
35

4.2. Comparison of three different types of impact surface

Cumulative Percentage of Fines

30
tested with three types of cushioning samples (Cushioned 25 Steel impact
coal, Initial 10% fines and Initial 30% fines) surface
20 Conveyor impact
surface
The comparison of three different impact surfaces tested 15 Coal impact
surface
with cushioned coal sample, initial 10% fines coal sample 10

and a sample of initial 30% fines of South Blackwater and 5

Ensham coal were shown in Figs. 813. The cushioned coal 0
sample of South Blackwater coal dropped on steel surface 0 1 2 3 4 5 6

produced 8.5% more fines than the drops onto a conveyor Number of Drops

surface and 13.4% more fines than a coal surface (coal stock- Fig. 10. Comparison of percentage of fines generation of three impact
pile) at 30 m total drops shown in Fig. 8. Similarly, En- surfaces of South Blackwater coal tested with initial 10% fines coal
sham coal produced 4.6% more fines on steel surface than sample.

Table 4
Cumulative percentage of 16.0 mm fines of three types of cushioning tests of two types of coal on coal stockpile surface from 5 m height
Number of drops Percentage of 16.0 mm fines of South Blackwater coal Percentage of 16.0 mm fines of Ensham coal

Cushioned coal Initial 10% fines Initial 30% fines Cushioned coal Initial 10% fines Initial 30% fines

1 2.7 2.4 1.0 2.7 2.1 1.0

2 6.2 5.5 2.5 6.2 5.1 2.3
3 9.0 8.1 5.4 8.9 7.7 4.9
4 11.4 10.8 8.0 11.4 10.5 7.8
5 14.1 13.0 10.5 14.0 12.8 10.1
6 16.3 15.0 12.0 16.2 14.6 11.7
258 R.K. Sahoo, D. Roach / Chemical Engineering and Processing 44 (2005) 253261

35 Ensham Coal, Initial 10% Fines Steel Surface

Cumulative Percentage of Fines

30
30
cumulative percentage of fines

25 Steel impact surface

South B.water
20 Conveyor impact 20 (initial 10% fines)
surface
15 Ensham (initial 10%
Coal impact surface
fines)
10
South B.water
10
5 (initial 30% fines)
Ensham (initial 30%
0
fines)
0 1 2 3 4 5 6
Number of Drops 0
0 1 2 3 4 5 6
Fig. 11. Comparison of percentage of fines generation of three impact Number of drops
surfaces of Ensham coal tested with initial 10% fines coal.
Fig. 14. Comparison of percentage of fines generation of two types of
coal tested with initial 10% fines and initial 30% fines on steel surface.

South Blackwater Coal, Initial 30% Fines

35
fines than coal surface shown in Figs. 10 and 11. The re-
Cumulative Percentage of Fines

30
sults of these figures revel that different impact surface has
25 Steel impact
surface
different degradation characteristics which is opposite to the
20 results of Waters et al. [5].
Conveyor impact
15 surface Figs. 12 and 13 shows that there were no significant dif-
10 Coal impact ference of fines percentage in between three impact surfaces
surface
5
tested with sample of initial 30% fines of South Blackwater
and Ensham coal at 30 m total drops. The results of these
0
0 1 2 3 4 5 6 figures reveal that at higher percentage of initial fines the
Number of Drops cushioning effects of fines on different impact surfaces are
same.
Fig. 12. Comparison of percentage of fines generation of three impact
surfaces of South Blackwater coal tested with initial 30% fines coal
sample. 4.3. Comparison of two types of coal tested on three
surfaces

35 Ensham Coal, Initial 30% Fines When two types of coal are compared, the coal sample of
initial 10% fines of south Blackwater coal dropped on steel
Cumulative percentage of fines

30
surface produced 5% more fines than Ensham coal at 30 m
25
Steel impact total drops shown in Fig. 14 and there was no significant
surface
20 difference of fines percentage between two types of coal
Conveyor impact
15 surface
tested with initial 30% fines sample on steel surface shown
Coal impact in Fig. 14.
10
surface
Figs. 15 and 16 shows that, there was no significant dif-
5 ference of fines percentage between two types of coal tested
0 with initial 10% fines and initial 30% fines on conveyor sur-
0 1 2 3 4 5 6
face and coal surface at 30 m total drops.
Number of drops

Fig. 13. Comparison of percentage of fines generation of three impact

surfaces of Ensham coal tested with initial 30% fines coal sample. Conveyor Surface
35
Cumulative Percentage of Fines

30 South B.water
initial 10% fines
conveyor surface and 12.3% more fines than coal surface 25
Ensham initial 10%
shown in Fig. 9. 20 fines
At initial 10% fines of South Blackwater coal, the steel 15 South B.water
initial 30% fines
surface produced 8% more fines than conveyor surface and 10
Ensham initial 30%
13% more fines than coal surface at 30 m total drops (six 5 fines

drops at 5 m) shown in Fig. 10. Fig. 11 shows that Ensham 0

coal with a sample of initial 10% fines, the steel surface pro- 0 1 2 3 4 5 6
Number of Drops
duced 4% more fines than conveyor surface and 9% more
fines than coal surface. The conveyor surface with a sample Fig. 15. Comparison of percentage of fines generation of two types of
of initial 10% fines of both types of coal produced 5% more coal tested with initial 10% and 30% fines on conveyor surface.
 R.K. Sahoo, D. Roach / Chemical Engineering and Processing 44 (2005) 253261 259

Coal Surface Table 5

The value of parameters for the degradation model for both types of coal
at 5 m drop heights
35
ij (%) ij (%)
Cumulative Percentage of Fines

Types of coal Drop mij (%) hij (%)

30 South B.water height (m)
initial 10% fines
25 South Blackwater 5 100 100 7.77 0.27
Ensham initial
20 10% fines
South B.water
15 initial 30% fines Steel Surface
Raw data
Cumulative Percentage of Fines

35
Ensham initial (Cushioned Coal)
10 30
30% fines D'Model data
(Cushioned Coal)
5 25
Raw data (Initial
20 10% Fines)
0
D'Model data
0 1 2 3 4 5 6 15
(Initial 10% Fines)
Number of Drops 10 Rawa data (Initial
30% Fines)
5 D'Model data
Fig. 16. Comparison of percentage of fines generation of two types of
0 (Initial 30% Fines)
coal tested with initial 10% fines and 30% fines on coal surface.
0 1 2 3 4 5 6
Number of Drops
4.4. Degradation model of South Blackwater coal on three
Fig. 17. Comparison of raw data and model data of South Blackwater at
impact surfaces 5 m drop height. (D Model data = Degradation Model data).

The degradation model of three samples (cushioned coal,

Conveyor Belt Surface
initial 10% fines and initial 30% fines) of South Blackwater
Cumulative Percentage of Fines

35 Raw data (cushioned)

coal is determined on three impact surfaces (steel plate, con- 30
veyor belt and coal stockpile). Degradation model equation 25
D'Model data
(cushioned)
can be written as follows: Raw data (initial 10%
D
N 20
fines)
    15
F= ij ij mij + hij cij sij (3) 10
D'Model data (initial
10% fines)
j i 5 Raw data (initial 30%
fines)
0
where, F is the percentage of fines produced as a results of 0 1 2 3 4 5 6
D'Model data (initial
30% fines)
the drops from different; heights, when dropping the differ- Number of Drops
ent types of coal of various sizes; ij , the mass percentage
of each feed size fraction i with each coal type j, ij , the Fig. 18. Comparison of raw data and model data of South Blackwater
coal at 5 m height.
mass percentage of each coal type j with each feed size frac-
tion i; mij , the fines percentage of each feed size fraction i
that would appear at a; particular drop height for coal type arate stockpiles through different handling circuits. Hence,
j; hij ; the percentage of fines produced during handling of ij = 100%, ij = 100%, N = 1, D = 1 and the value of
each feed size fraction i and coal type j; cij , the percentage other parameters are as indicated in Tables 56. The degra-
of fines reduced due to cushioning effects of each feed size dation model data (DModel data) were determined from
fraction i and coal type j; sij , the percentage of fines reduced above Eq. (3) by MatlabTM computer programming.
due to change of impact surfaces of each feed size fraction The results of the degradation model of South Blackwater
i and coal type j; N, number of feed size fractions dropped; coal dropped on steel plate, conveyor belt and coal stock-
D, number of coal types dropped. pile surfaces from 5 m heights are shown in Figs. 1719
Determination of fines parameter, handling, cushioning (Table 6). The raw data for the Figs. 1719 are given in
and surface parameters for degradation model of South Tables 24 and model data are given in Tables 79. There
Blackwater coal were presented in Sahoo and Roach [11]. was a close agreement between raw data and degradation
The percentage of fines generation varies in different types model data (DModel data) of both types of coal dropped
and sizes of coal. The number of handling events also in- on different surfaces. Within the experimental error most of
creases the fines generation in the handling circuits, which the raw data of the cushioning tests of the cushioned coal
is varying in coal types and its sizes. Cushioning and surface sample and initial 10% fines sample of South Blackwater
parameters are also varies with types and sizes of coal. coal dropped from 5 m height on steel surface are same with
In the present research, a single size fraction (76.2 the model data shown in Fig. 17. The model data points of
+ 50.8 mm) of one type of coal was dropped on different initial 30% fines sample of South Blackwater coal dropped
impact surfaces from 5 m heights because the Gladstone from 5 m height on steel surface differ from raw data points
Port Authority handles one type of coal at a time from sep- by 4% at 30 m total drops shown in Fig. 17.
260 R.K. Sahoo, D. Roach / Chemical Engineering and Processing 44 (2005) 253261

Table 6
The value of parameters for the degradation model for both types of coal at 5 m drop heights
Types of coal Drop height (m) Cushioned parameter = cij (%) Surface parameter = sij (%)

Cushioned coal Initial 10% fines Initial 30% fines Conveyor Coal

South Blackwater 5 2.08 2.26 4.62 1.4 2.2

Coal Stockpile Surface Table 9

30 Raw data (cushioned) Model data of 16.0 mm fines of three types of cushioning tests of South
Cumulative Percentage of Fines

Blackwater coal at 5 m height on coal surface

25 D'Model data
(cushioned) Number of Model data of 16.0 mm fines of South Blackwater coal
20 Raw data (initial 10% drops
fines) Cushioned Initial 10% Initial 30%
15 coal fines fines
D'Model data (initial
10% fines)
10 1 3.8 3.6 1.2
Raw data (initial 30%
fines) 2 7.4 7.0 2.4
5
D'Model data (initial
3 10.9 10.4 3.6
0 30% fines) 4 14.2 13.6 4.8
0 1 2 3 4 5 6 5 17.4 16.7 6.0
Number of Drops 6 20.5 19.6 7.1
Fig. 19. Comparison of raw data and model data of South Blackwater
coal at 5 m height.
from the model data points by 4% and the raw data points
of initial 30% fines curve differ from the model data points
Table 7 by 5% on 30 m total drops (six drops at 5 m) as shown in
Model data of 16.0 mm fines of five types of cushioning tests of South
Blackwater coal at 5 m height on steel surface
Fig. 19.

Number of Model data of 16.0 mm fines of South Blackwater coal

drops
Cushioned Initial 10% Initial 30% 5. Conclusion
coal fines fines
1 6.0 5.8 3.4 Non-cushioned coal sample of both types of coal dropped
2 11.6 11.2 6.7 on different impact surfaces produced higher percentage of
3 16.8 16.4 9.9
fines than the sample of cushioned coal, initial 10% fines,
4 21.8 21.2 13.0
5 26.5 25.7 16.0 initial 30% fines and initial 50% fines. The cushioning ef-
6 30.8 30.0 18.8 fects of South Blackwater and Ensham coal with initial 10%
fines dropped on steel surface produced more fines than
conveyor surface and coal surface. The cushioning effect of
Table 8 fines on coal surface was least compare to other two impact
Model data of 16.0 mm fines of three types of cushioning tests of South
surfaces. There was no significant difference of fines gen-
Blackwater coal at 5 m height on conveyor surface
eration in between the three impact surfaces tested with a
Number of Model data of 16.0 mm fines of South Blackwater coal sample of initial 30% fines of both types of coal. The degra-
drops
Cushioned Initial 10% Initial 30% dation model was used satisfactorily for South Blackwater
coal fines fines coal on different impact surfaces.
1 4.6 4.4 2.0
2 8.9 8.6 4.0
3 13.1 12.6 5.9 Acknowledgements
4 17.0 16.4 7.8
5 20.8 20.1 9.7
6 24.4 23.6 11.5 The authors wish to thanks to the management of Glad-
stone Port Authority for funding and using their coal sam-
ples for this research work. The authors are also grateful to
Within the experimental error, the raw data points of cush- Queensland Government to arrange this scholarship through
ioned and initial 10% fines curves of South Blackwater coal the Port Authority.
dropped on conveyor surface differ from model data points
by 3% and the raw data points of initial 30% fines curve dif-
fer from model data points by 1% on 30 m total drops (six References
drops at 5 m) shown in Fig. 18.
Raw data points of cushioned and initial 10% fines curve [1] P.J. Goodwin, C.M. Ramos, Degradation of sized coal at transfer
of South Blackwater coal dropped on coal surface differ points, BSH 7 (4) (1987) 517534.
 R.K. Sahoo, D. Roach / Chemical Engineering and Processing 44 (2005) 253261
[2] R.K. Sahoo, D.M. Weedon, D. Roach, Coal degradation Part-1:
factors affecting coal degradation during handling, BSH 22 (4) (2002)
282288.
[3] R.K. Sahoo, D.M. Weedon, D. Roach, Coal degradation, Part-2:
experimental study of several factors affecting Gladstone Port
Authoritys lump coal degradation, BSH 22 (5) (2002) 356361.
[4] B. Fagerberg, N. Sandberg, Degradation of lump ores in transport,
in: Proceedings of the Second International Symposium on Transport
and Handling of Minerals, Rotterdam, The Netherlands, vol. 2, 1973,
pp. 28156.
[5] A.G. Waters, A. Vince, M. Mahoney, An experimental investigation
of some factors influencing the degradation of lump iron ore, in:
Proceedings of the Third International Conference on Bulk Materials,
Storage, Handling and Transportation, Newcastle, 1989, pp. 8488.
[6] T.E. Norgate, D.F. Tompsitt, R.J. Batterham, Computer simulation of
the degradation of lump ores during transportation and handling. in:
261
Proceedings of the Second International Conference on Bulk Material
Storage, Handling and Transportation, Wollongong, Australia, 1986,
pp. 2025.
[7] L.G. Austin, R.R. Klimple, P.T. Luckie, The process engineering of
size reduction, Ball Milling, SEM/AIME, New York, 1984.
[8] A.G. Waters, A. Vince, C.S. Teo, A technique for determining the
resistance to shatter of lump materials, Proc. of CHEMECA, Mel-
bourne 2 (1987) 107.1107.8.
[9] C.S. Teo, A.G. Waters, The effect of drop height and particle size
on the extent of degradation of agglomerates, in: Proceedings of
Australian Biocentennial International Conference for the Process
Industries, CHEMECA88 Sydney, 1988 pp. 665670.
[10] A.A. Griffith, The phenomena of rupture and flow in solids, Philos.
Trans. R. Soc. London A221 (1920) 163198.
[11] R.K. Sahoo, D. Roach, Degradation model of the dropped coal on
different impact surfaces, BSH 23 (6) (2003) 6.
 Chemical Engineering and Processing 44 (2005) 263266

Using statistical moments to describe grinding in a ball

mill for industrial-scale process
Andrzej Heim, Tomasz P. Olejnik,1 , Agnieszka Pawlak1
Faculty of Process and Environmental Engineering, Technical University of Lodz, Wlczanska 213, 90-924 dz, Poland

Received 4 August 2003; received in revised form 17 December 2003; accepted 11 February 2004
Available online 2 July 2004

Abstract

The aim of the paper was to describe the process of grinding of raw materials used in the industrial-scale production of ceramic tiles, by
applying the theory of statistical moments. Grinding was performed in industrial ball mills in ceramic tile factories Ceramika Paradyz Ltd.
and Opoczno S.A. The ball mills operated in a batch mode. A mixture of feldspars and clay was comminuted. Its composition and fractions
depended on the conditions that should be satisfied by raw materials for the production of wall tiles (monoporosis and stoneware) and terracotta.
The ground material was subjected to a particle size analysis. Results of the analysis were used in the calculation of relationships applied in
the theory of statistical moments. The main parameters, i.e. zero moment of the first order and central moments of the third and fourth order
were determined. The values of central moments were used in the calculation of skewness and flatness coefficients. Additionally, changes of
mean particle size in time were determined.
2004 Elsevier B.V. All rights reserved.

Keywords: Ball mill; Industrial-scale; Moment theory; Grinding

1. Introduction Heim and Olejnik [13,9] proposed a simple mathemati-

Studies on the application of the theory of statistical mo-
ments in the description of grinding in ball mills have been
carried out in the Department of Process Equipment, dz
Technical University [13]. The research was carried out
in a laboratory scale for selected mineral materials. Results
obtained confirmed applicability of the theory of statistical
moments in the description of particle size distribution dur-
ing grinding.
Modelling of the grinding process covers two problems
[48]. The first one is a mathematical description of the
particle size distribution of material being ground, i.e. an
equation which describes in the best way the dependence
of particular size fractions on the characteristic dimensions.
The other one includes the equation of grinding kinetics that
describes the rate of changes in the particle size distribution
in time, depending on the properties of material being ground
and process parameters.
cal model based on the theory of statistical moments, whose
applicability was confirmed by the results of laboratory-scale
studies.
Laboratory tests were performed for gabbro, syenite,
sandstone and quartzite. Materials were ground in variable
process conditions, i.e. changing mass of grinding media,
frequency of mill revolutions and feed mass. Results of the
investigations were discussed in detail in [2,3]. It was con-
cluded that the proposed model could be used for grinding
in industrial ball mills. This would enable a modelling of
grinding in a large scale.
In this paper, the applicability of the model to grinding in
industrial ball mills was checked. This would enable mod-
elling of large-scale grinding with the use of data collected
during grinding in small mills taking into account the scale
effect.

2. Description of the model
Corresponding author. Tel.: +48 42 631 3713.
E-mail addresses: heim@wipos.p.lodz.pl (A. Heim),
tolejnik@mail.p.lodz.pl (T.P. Olejnik). In the classical theory of statistical moments they were
1 Tel.: +48-42-631-3713. determined by the formulae

0255-2701/$ see front matter 2004 Elsevier B.V. All rights reserved.
doi:10.1016/j.cep.2004.02.020
264 A. Heim et al. / Chemical Engineering and Processing 44 (2005) 263266

m

Table 2
m1 = di w i (1) Feed composition for stoneware and terracotta at Opoczno [10]
i=1 Stoneware Terracotta
M3 Solid components (kg) 20000 20000
K1 = 3/2
(2) Including feldspar (kg) 6000 10000
M2
Clay (kg) 14000 10000
M4 Liquid components (kg) 1500 1500
K2 = 3 (3)
M22 Including water (kg) 1422 1422
Flumix (kg) 20 20
where the central moment of the k-th order is Water glass (kg) 68 68

m
Mk = (di m1 )k wi (4) Table 3
i=1 Feed composition for monoporosis and terracotta in Paradyz [11]
The zero moment of the first order m1 is identical to the Monoporosis Terracotta
mean grain size of the entire set.
Solid components (kg) 21300 22000
A common feature of the classical absolute measures of Including feldspar (kg) 3834 11660
skewness and flatness is that they do not have a finite numer- Clay (%) 11502 7700
ical interval which theoretically would include all possible Dolomite (%) 1704 1980
values. In order to interpret the coefficients of flatness and Calcium carbonate (%) 1704
skewness more easily, original modified forms presented by Raw crushed minerals (%) 660
Eqs. (5) and (6) were proposed. Liquid components (kg) 2100 2600
Including water (kg) 2051 2550
M4 (M2 )2 Trisodium polyphosphate (kg) 49 50
K1m = (5)
M4
M3 Table 4
K2m = (6) Ball size and weight in the mills in Ceramika Paradyz [11] and Opoczno
(M2 ) +M
3/2
3 S.A. [10]
A characteristic feature of these modified coefficients is that Ball diameter (m) Weight (kg)
their values are included always in the strictly determined
Ceramika Paradyz 0.025 4000
intervals: the value of modified skewness coefficient K2m is
0.030 8000
always in the interval 1, 1, while the value of the modified 0.040 8000
flatness coefficient K1m is always in the interval 0, 1. 0.045 6500
Ball diameter (inch) Weight (kg)

3. Experimental Opoczno S.A. 3/4 1700

7/8 2750
Industrial mills in which the experimental grinding was 1 4150
performed were in two factories. The first one was in Ce- 1(1/4) 8000
1(1/2) 1800
ramika Paradyz Ltd. at Tomaszw Mazowiecki, and the other 1(3/4) 3050
one in Opoczno S.A., at Opoczno. Selected technical data 2 4550
are given for each mill in Table 1. Depending on future use,
further in the production, the feed differing in the percent-
age of feldspar, clay and additives was applied. In both fac- The mills were filled with corundum grinding media. The
tories, the ground raw materials were a starting material for nominal packing were the same for both materials ground
wall and floor tile production. at Opoczno and for the mills operating at Paradyz (Table 4).
The composition of raw material mixtures used in the pro-
duction of wall tiles (stoneware) and floor tiles (terracotta)
in Opoczno is given in Table 2. Data characterising the feed 4. Results and discussion
in the mill at Paradyz are given in Table 3.
For particle size analysis, around 400 g samples were
Table 1 taken from the mills at determined time intervals.
Selected technical data of the mills The time in which a sample was taken was counted since
Mill Ceramika Paradyz Opoczno S.A. the beginning of the grinding cycle for each raw material
Inner diameter (m) 2.5 3.1
and it was, respectively
Water volume (m3 ) 38 38 grinding at Ceramika Paradyz: monoporosis 80, 140,
Frequency of rotations (min1 ) 12.63 13
200, 260, 320, 380 and 460 min, terracotta 300,
 A. Heim et al. / Chemical Engineering and Processing 44 (2005) 263266 265

Table 5
Comparison of calculated mean particle size m1 , flatness coefficient K1m and skewness coefficient K2m . Terracotta grinding in the mill at Opoczno S.A.
Grinding time (min)

60 120 180 240 300 360 420 480

m1 41.07 22.72 23.62 14.59 13.18 11.94 11.30 10.87

K1m 0.9454 0.9357 0.9612 0.9865 0.9791 0.9854 0.9637 0.9661
K2m 0.7774 0.731 0.7823 0.7957 0.8056 0.8012 0.8006 0.8105

70
m 1 [ m]
60 Opoczno_terracotta
Parady _terracotta
50 Opoczno_stoneware
Parady _monoporosis
40

30

20

10
t [min]
0
0 200 400 600 800 1000

Fig. 1. Changes in the mean particle size m1 for raw materials ground at Opoczno and Paradyz.

360, 420, 480, 540, 600, 660, 720, 780, 840 and 900
min;
grinding at Opoczno S.A.: stoneware 15, 45, 75, 105,
135 and 175 min, terracotta 60, 120, 180, 240, 300, 360,
420 and 480 min.
The samples were subjected to particle size analysis using
a laser analyser. Results obtained enabled the calculation
of a mean particle size (zero moment m1 ) and modified
coefficients of skewness K2m and flatness K1m . Examples of
results obtained for one sample are given in Table 5.
Fig. 1 shows changes in the mean particle size m1 as a
function of time for raw materials ground in the mills at
Opoczno and Paradyz. In all cases, in the initial grinding
period, an intensive change of the mean particle size m1
takes place. For terracotta and silica ground at Opoczno,

1,2
K1m, K2m

0,8

0,6
K1m (Opoczno - stoneware)
.
K1m (Paradyz - monoporosis)
0,4 K1m (Opoczno - terracotta)
.
K1m (Paradyz - terracotta)
K2m (Opoczno - stoneware)
.
K2m (Paradyz - monoporosis)
0,2 K2m (Opoczno - terracotta)
.
K2m (Paradyz - terracotta) m1 [m]

0
0 10 20 30 40 50 60 70

Fig. 2. Dependence of the modified flatness coefficient K2m and skewness coefficient K1m on mean particle size m1 for raw materials being ground.
266 A. Heim et al. / Chemical Engineering and Processing 44 (2005) 263266
the technologically required particle size is obtained in a
shorter time (480 and 175 min, respectively) as compared to
the grinding of terracotta and monoporosis at Paradyz (900
and 460 min, respectively). The shorter time of grinding at
Opoczno is a result of a different set of grinding balls and
different composition of the feed mixture.
When analysing the dependence of skewness coefficient
K2m and flatness coefficient K1m on the mean particle size,
it is found that coefficients K1m and K2m are stable in the
tested range of variability of m1 (Fig. 2).
In the range of mean particle size m1 (from 10 to 40 m),
almost constant values of the K1m (0.98) and K2m (0.8)
were observed. The nature of changes, K1m and K2m , for
the tested grinding process reveals a far reaching similarity
to the changes that are observed in a laboratory-scale grind-
ing. However, in the laboratory-scale experiments, tests were
performed for a bigger range of changes in the mean parti-
cle size m1 [2,3,1216]. The coefficients K1m and K2m were
also constant in the range 20 m1 100 m. For higher
values of m1 , the coefficient K1m decreased approaching a
boundary value 1.
For very small values of m1 , the coefficient K1m reached
values close to 0.9. The flatness coefficient was growing up
to the boundary value 0.9, for big mean particle sizes.
To find analogous tendencies in the change of coeffi-
cients K1m and K2m for industrial grinding, it is necessary
to carry out the grinding process, for which the mean par-
ticle size would be in the range above 100 m and below
10 m.
Results obtained in Opoczno and Paradyz mills confirmed
applicability of the theory of statistical moments in the de-
scription of particles of the material ground in industrial ball
mills. However, it is necessary to extend the range of vari-
ability of the mean particle size.
Acknowledgements
The work was carried out under research Project no.
3T09C 005 23 financed by the State Committee for Scien-
tific Research for the years 20022005.
Appendix A. Nomenclature
di mean grain size from size grade i (m)
K1 flatness coefficient
K1m modified flatness coefficient
K2 skewness coefficient
K2m modified skewness coefficient
m1 mean grain size (m)
M2 central moment of the 2nd order
M3 central moment of the 3rd order
M4 central moment of the 4th order
t grinding time (s)
wi percentage part of mass of grains from size
grade i
References
[1] A. Heim, Determination of particle size distribution of ground ma-
terial, Rudy i Metale 38 (3) (1993) 74.
[2] A. Heim, T.P. Olejnik, Simulation of milling in ball mills, Proceed-
ings of the 2nd Israel Conference for Conveying and Handling of
Particulate Solids, Israel, 1997, p. 813.
[3] A. Heim, T.P. Olejnik., Description of grinding in a ball mill using
statistical moments, Proceedings of the 3rd Israeli Conference for
Conveying and Handling of Particle Solids, Israel, vol. 1, 2000, p.
632.
[4] L.G. Austin, A review introduction to the description of grinding as
a rate process, Power Tech. 5 (1971) 123135.
[5] B. Beke, The Process of Fine grinding, Akademiai Kiado, Budapest,
1981, pp. 2123.
[6] R.P. Gardner, L.G. Austin, J. Inst. Fuel. London 35 (1962) 174.
[7] K. Sedlatschek, L. Bass, Powder Met. Bull. 6 (1953) 148.
[8] E. Verdes, Ph.D. thesis, SZIKKTI, Budapest, 1984.
[9] A. Heim, M. Pietrzykowski, Grinding in a ball mill, Powder Hand.
Process. 5 (4) (1993).
[10] Data from Technological Department, Opoczno S.A.
[11] Data from Technological Department, Ceramika Paradyz Sp. z o.o.
[12] A. Heim, T.P. Olejnik, Opis kinetyki procesu rozdrabniania niek-
trych materiaw skalnych za pomoca teorii momentw, Inzynieria
Chemiczna i Procesowa, Zeszyty Naukowe P., vol. 21, 1997,
pp. 101108.
[13] T.P. Olejnik, Zastosowanie teorii momentw do opisu procesu roz-
drabniania w myniekulowym, Ph.D. thesis, Faculty of Process and
Environmental Engineering, Technical University of Lodz, 1999.
[14] A. Heim, T.P. Olejnik., Wpyw czestosci obrotowej myna kulowego
na skad granulometryczny produktw rozdrabniania, Proceedings
XXXVI Seminar Fizykochemiczne Problemy Mineralurgii, 1999,
pp. 6373.
[15] A. Heim, T.P. Olejnik, Wpyw masy nadawy oraz kul na kinetyke
procesu mielenia opisana teoria momentw, Zeszyty Nauk, Politech-
niki, dz Inz. Chemiczna i Procesowa, vol. 838, no. 27, 2000,
pp. 105114.
[16] A. Heim, T.P. Olejnik, A. Pawlak, Z. Korzen, Dobr parametrw
procesu mielenia w mynie kulowym w skali pprzemysowej w
oparciu o teorie momentw statystycznych, Proceedings of the Con-
ference Kruszywa mineralne surowce rynek technologie
jakosc, Polanica Zdrj, 2001, pp. 113122.
 Chemical Engineering and Processing 44 (2005) 267272

On possible instability of throughputs in complex milling circuits

V. Mizonov a , V. Zhukov a , A. Korovkin a , H. Berthiaux b,
a Department of Applied Mathematics, Ivanovo State Power Engineering University Rabfakovakaya 34, 153003 Ivanovo, Russia
b Laboratoire de Gnie des Procds des Solides Diviss, Campus Jarlard, Ecole des Mines dAlbi-Carmaux, 81000 Albi, France

Received 15 July 2003; received in revised form 5 January 2004; accepted 11 February 2004
Available online 25 June 2004

Abstract

The objective of the study is to develop an effective and universal approach, based on the theory of Markov chains, to calculate the basic
parameters of the steady state regime of complex technological systems for processing granular materials. In particular, we focus on the
response of a complex milling system to fluctuations of the feed rate to the system. It is shown that depending on the characteristics of the
classifiers the fluctuations can grow unlimitedly, so that the stability of throughputs in the circuit can be lost.
2004 Elsevier B.V. All rights reserved.

Keywords: Throughputs; Milling circuits; Classifiers

1. Introduction
Application of complex technological systems in produc-
tion of fine powders is a well known tendency in improve-
ment of the production efficiency. In particular, complex
milling circuits allow reducing the specific energy, which is
necessary for grinding. This is due to the correspondence
of the size of grinding bodies and that of current particles,
and also to the evacuation of particles, which are already
ground enough, from the process. Application of networks
of classifiers allows obtaining very high sharpness of cut,
even though separate classifiers of the network could be
fairly ineffective. A common feature of such circuits is the
use of recycling classifier-formed particle flows, formed by
a classifier, and directed to further grinding, or to repeated
classification. For instance, in particle size controlled pro-
duction of flour, such networks can currently include more
than 50 mills and classifiers, while three stages containing
1520 hydrocyclones are needed in the paper pulp process-
ing. However, this feature predetermines some difficulties in
the mathematical modelling of the circuits, as well as some
problems in their industrial operation.
Corresponding author. Present address: Centre Poudres et Procedes,
Campus Jarland, Ecole des Mines dAlbi Carmaux, Albi Ct Cedex, Albi
81013, France. Fax: +33 5 63 49 30 25.
E-mail address: berthiaux@enstimac.fr (H. Berthiaux).
During the last years, certain attention has been devoted
to the application of the theory of Markov chains to model
complex technological systems as well as separate processes
of them [17]. This theory appears to be an effective and
universal mathematical tool also allowing the efficient com-
putational analysis with modern computers. An algorithm
for modelling and calculation of steady-state parameters of
circuits of arbitrary complexity was proposed in [8,9]. It is
implicitly based on the Markov chain approach and uses its
basic ideas. The basic operator of the approach is the block
matrix of a circuit, which is to be constructed (according
to developed rules) of matrices of grinding and classifica-
tion for mills and classifiers the circuit consists of. However,
there is no clear correlation between the algorithm and the
general strategy of building Markov chain models that does
not allow developing it. Besides that, the approach deals
with state (throughput and fraction size distribution) inde-
pendent matrices, i.e., represents a linear model of the cir-
cuit. At the same time, the flow rate of the recycling particle
flow is not known in advance, as it will depend on the char-
acteristics of the classification which, in turn, depend on the
mass flow rate to the classifier. This may cause considerable
deviation of real throughputs from calculated ones even at a
small fluctuation of the feed rate to the circuit for example.
The objective of the study is an attempt to estimate con-
ditions, under which a circuit has stable steady-state char-
acteristics, as well as conditions, under which this stability

0255-2701/$ see front matter 2004 Elsevier B.V. All rights reserved.
doi:10.1016/j.cep.2004.02.021
268 V. Mizonov et al. / Chemical Engineering and Processing 44 (2005) 267272

can be lost. For this, simulations based on the Markov chain (a)
formulation for calculating the transient behaviour a circuit,
B22
are performed. B23
1 2 3
B0 B11 B12=B21
2. Application of the theory of Markov chains to model (b)
steady-state parameters of complex circuits
B22
B23
The simplest technological systems with recycling flows 1 2 3
are shown schematically in Fig. 1. In this, the scheme (a) B0 B11 B13=B21
corresponds to a closed milling circuit, and the scheme (b) B12
to a network of classifiers, in which, the fine product of the
first classifier has the repeated classification in the second Fig. 1. Technological systems with recycling flows: (a) closed milling
circuit; (b) network of classifiers. (3-absorber of particles).
apparatus. These basic elements can be a part of more com-
plex circuits. First let us examine the closed circuit milling
system. ones standing for transitions among fractions, i.e., to grind-
The chain model corresponding to the closed circuit ing, horizontalfor transitions among elements of the cir-
milling system with localised models of all its elements is cuit. The matrix of transition probabilities has the following
presented in Fig. 2. It can be constructed in different ways, form:

but first let us examine the model shown in Fig. 2a. Every P 11 P 12 0

column of the set of cells corresponds to an element of P = P 21 P 22 0 (1)
the circuit: a mill; a classifier; or an absorber. The cells 0 P 31 I
within columns correspond to fraction numbers with the
total number of fractions equal to r. The fraction size de- where P is the block matrix consisting of smaller matrices of
creases with increasing fraction number. The state of the size r r. The matrices on the main diagonal describe transi-
system is characterised by the set of probabilities fij to tions among cells (rows) within the corresponding columns,
occupy the cells, every of which can be interpreted as the the matrices on the diagonals neighbouring to the main one
relative mass content of particles in the cell ij. In partic- describe forward (upper diagonal) and backward (lower di-
ular, the set fij , I = 1,2,. . . r, corresponds to the particle agonal) transitions between elements of the circuit, i.e., be-
size distribution in the hold-up of the jth element of the tween columns of cells. I is the identity matrix of the same
circuit. size as P ij , O is the zero matrix of the same size.
After a transition time t the state of the system changes, An example of the matrix P expressed in numbers is
and these changes occur due to transitions of particles among shown below. The total number of fractions, which are un-
the cells. The arrows show possible transitions, the vertical der observation, is taken equal to 3 (r = 3). The transition

(a) (b)

Element of the circuit Fraction number (size)

feed
11 12 13
Fraction number (size)

11 12 13 1r
21 22 23
Element of the circuit
feed

31 32 33 21 22 23 2r

r1 r2 r3
31 32 33 3r
Mill Classifier Absorber

fine

Fig. 2. Two variants of constructing the cell model of the closed circuit milling system.
 V. Mizonov et al. / Chemical Engineering and Processing 44 (2005) 267272 269

q
probabilities are denoted as pp where p is the index of the with the probability to stay in it equal to zero, placed just
cell where the material transits to from the cell q. before the real absorber which is not included in the chain.
11 12
p11 0 0 p11 0 0 0 0 0
p11 p21
0 0 p22

0 0 0 0

21 21 21

p11 p21 p31 0 0 p32 0 0 0
31 31 31 31

p11 p12
12
0 0 0
12 0 0 0 0

P = 0 p21 0

0 p 22 0

0 0 0 (1a)
22 22
0 p31 0 p32 0 0 0
0 0

32

32

p13
12
1 0 0
0 0 0 0 0

0 0 0 0 p22 23 0 0 1 0
0 0 0 0 0 p33 32 0 0 1
The mass content of fractions in the cells is described by the
state vector F , which is the column block vector consisting In this case, the identity matrix I in (1) is to be replaced by
of vectors f j related to the jth column of the set of cells the zero matrix of the same size, and the matrix P becomes:

placed one after another: P 11 P 12 0

f1 P = P 21 P 22 0 , (6)
P 31
F = f2 (2) 0 0
f3 and the steady-state (asymptotic) distribution can be calcu-
lated by the formula:
where every vector f j has the size r 1, and consists of
elements fij , f2j , . . . , frj . F = (I P)1 F 0 , (7)
The feed vector F 0 is similar to F , and contains mass
which follows from the formula for the sum of the infinite
fractions, which are added to the circuit after every transi-
geometrical progression written in matrix notation.
tion. In the case shown in Fig. 2a, it contains only the first
After separating the vectors f i from F it becomes pos-
non-zero vector of the three, which is the fraction size dis-
sible to define fraction size distribution in every element
tribution f 0 in the feed to the circuit. The rest two vectors
of the circuit, and then, using (4) and (5)-throughputs and
are the zero vectors of the same size.
hold-ups in them.
If the transition time t is chosen, and matrices P ij are
The algorithm of the circuit matrix construction can be
defined, the matrix P can be constructed and successive
strongly simplified, if we suppose that, the particle flow in
evolution of the state vector F can be calculated as:
the mill can be presented as the perfect plug flow, and take
F n+1 = P(F n + F n0 ). (3) the transition time equal to the particle residence time in
the mill. In this case, the probability to stay within the cell
The hold-up in the jth element of the circuit and the total
presenting the mill is equal to zero. Since the residence time
flow rate from it to the pth apparatus of the circuit are:
for a mill is normally much longer than it is for a classifier,
r
 the same holds true for the probability to stay in the cell
Mj = fij (4) presenting the classifier.
i=1 In this case, the constructing of the circuit matrix is more
r
 convenient if the axes are directed as it is shown in Fig. 2b.
Bpj =
ij
fij pip (5) Under the above assumptions this matrix takes the form:
i=1
0 I C 0
j
where Bp is the mass flow rate from the jth to the pth element
P = G 0 0, (8)
ij
of the circuit, pip is the probability of transition of the ith 0 C 0
size fraction between these elements.
Under a continuous constant feed to the circuit there exists where G is the matrix of grinding in the mill related to the
a steady-state (asymptotic) regime of the circuit operation. transition time t, which is a lower triangle matrix of size r
However, this solution cannot be obtained with the matrix r (r is the number of fractions under observation) C, the ma-
(1) because it contains the absorbing state, the probability trix of classification, which is a diagonal matrix of the same
of which grows unlimitedly. In order to get round the size consisting of discrete values of a grade efficiency curve
difficulty, the absorbing cell can be presented not like the [8,9,11]. Substitution of the matrix (8) into Eq. (7) gives:
end of the chain, but as an intermediate element of the chain F = K1 F 0 (9)
270 V. Mizonov et al. / Chemical Engineering and Processing 44 (2005) 267272
where,
I I C 0
K= G I 0 (10)
0 C I 3. Stability of throughputs in closed circuit systems
is the final form of the matrix of the circuit, which can be
constructed on the basis of the following rule:
The matrix K of the circuit is the block matrix of size
m x m blocks, i.e., smaller matrices presenting processes
in elements of the circuit, every of which has the size
r r, where m is the number of elements of the circuit
(including the absorber), r is the number of fractions under
observation.
The main diagonal of K contains the identity matrices of
size r r with the sign minus.
Every column of K belongs to an element of the circuit.
The matrix of the element is to be placed in the row of this
column, whose number is equal to the number of element,
where the particle flow is directed.
This rule is universal and can be used for an arbitrary
circuit. The vectors f i of the block vector F is the fraction
size distribution in the particle flow before the ith element.
The matrix K for the network of classifiers shown in Fig. 1b
has the following form:
I I C2 0
K = C1 I 0 (11)
0 C2 I where, x and xc is the current and cut size respectively, s
where, the absorber for the coarse product from the first
classifier is not included into the chain because information
on the state is irrelevant to the objective of this study.
It is necessary to note that the matrix K does not deal with
the hold-ups at all. The same rule for its construction was
suggested in [8,9] on the basis of particle fractions flows bal-
ance only, without referring to Markov chains models. Here
it was obtained as a particular case of the developed model,
which is presented by the matrix P and allows describing
also the evolution of hold-ups in general case provided the
smaller matrices of P are known.
Once the types of equipment a circuit consists of have
been decided upon, we have the preliminary information
on the corresponding matrices, and can calculate the circuit
using the algorithm described above. After calculating the
throughputs, we can select sizes of the equipment chosen.
Thus, we get the complete information on the circuit param-
eters, which are supposed to be reached in the real operation.
If the matrices of the equipment are throughput independent,
small errors in describing them give also a small error in pre-
dicting the circuit parameters. However, if they are through-
put dependent, first, the error can become much larger, and
second, small variation of operating parameters of the circuit
may cause large variation of throughputs, up to conditions
when the circuit fails to work. Further on, we will try to
express mathematically these ideas and illustrate them with
some simulations.
3.1. Closed circuit milling system
Let us turn back to the closed circuit milling system shown
in Fig. 1a. The throughput dependent matrix of grinding was
introduced as follows. For the fixed throughput Bs the ma-
trix P was constructed on the basis of the linear selection
and homogeneous breakage functions [8,10]. It was sup-
posed that the only result of the throughput change was the
variation of the mean residence time. Under the assumption
of a constant hold-up, it gives:
P(B) = P(Bs )tm /tms = P(Bs )(M/B)/(M/Bs ) = P(Bs )Bs/B
(12)
where, tm is the mean grinding time, M, the constant is the
hold-up in the mill, and the subscript s indicates the reference
regime of grinding.
The matrix of classification C was introduced on the basis
of the formula by Molerus and Hoffmann [11] for the grade
efficiency curve:
1
qx = (13)
1 + x/xc exp((s/2)(1 x/xc ))
is the parameter related to the efficiency of classification.
With s Eq. (13) gives the grade efficiency curve of
the perfect classifier with the cut size xc . The elements of
C were calculated as Cjj = qx (xj /xc ) where, xj is the mean
size fraction.
When the throughput in a classifier is changed, both the
cut size xc and the parameter s can be changed. An increase
of the throughput practically always leads to the decrease of
s (the efficiency decreases) but the cut size can both increase
and decrease. Both parameters are changing the outflows
of fine and coarse material, but it can be shown that the
influence of the cut size is much stronger. That is why the
assumption that xc = xc (B), but s = s(B) was introduced.
Despite, the dependence of the cut size on the throughput
can be very complex, as the first approximation it can be
taken as a linear one around the reference throughput:
xc (B) = xc (Bs ) k(B Bs ) (14)
where, xc (Bs ) is the cut size at the reference throughput,
which was used to choose the classifier, k = (xc /B)B=Bs ,
and its positive value is assigned to the decreasing function
xc (B), which is being met more often for industrial classi-
fiers.
Investigation of transient processes after deviation of the
feed rate was done as a numerical experiment according to
the following procedure. The basic value of the feed rate B0
 V. Mizonov et al. / Chemical Engineering and Processing 44 (2005) 267272 271

Fig. 5. Transient process in the network of classifiers at B0

= 0, 1: (a)
Fig. 3. Transient process in the closed milling circuit: (a) k > 0 (1,2: k = l(1), k = 2(2); (b) k = 1(1), k = 2(2).
B0
= 0.15, 3,4: B0
= 0.15, 1,3: k = 0.15, 2,4: k = 0.3) and (b) k < 0
(B0
= 0.15, (1) k = 0.6, (2) k = 0.9).
sults of numerical examples and shows areas in the graph
k0 = f(B0
) providing the stable and unstable operation of
got a small perturbation B0
of the impulse kind. Then, the the circuit.
evolution of perturbations of throughputs and fraction size
distributions in the circuit caused by the impulse was ob- 3.2. Network of classifiers
served. If the perturbations were getting down during itera-
tions, the steady-state regime was considered stable, if they The general matrix of the network of classifiers shown
were growing unlimitedly, or the system was removed to an- in Fig. 1b is given by Eq. (11) where, C1 and C2 are the
other steady-state regime, the basic steady-state regime was matrices of classification in the first and the second ap-
considered unstable. Fig. 3 shows the evolution of through- paratus, respectively. Investigation of throughputs stability
put in the closed milling circuit as the function of itera- was carried out according to the procedure described above.
tion number n after introducing the impulse perturbation As far as a network normally contains identical classifiers,
to the feed with homogeneous fraction size distribution at or at least classifiers of the same type, their characteris-
s = 5. tics were taken identical (xc1 (Bs ) = xc2 (Bs ) = 0.5; k1 = k2
For positive values of k (Fig. 3a), providing the stable op- = k).
eration of the circuit, the perturbation is going down (curves The transient process that follows after introducing the im-
1,3). If k is larger than the critical value k0 , and B0
> 0, pulse perturbation to the feed is shown in Fig. 5. At the stable
the perturbation is growing unlimitedly and leads to the conditions stabilization of throughputs can be monotonous
mill blockage (curve 2). If B0
< 0, the throughputs tran- (curve 1 in Fig. 5a) and oscillating (curve 1 in Fig. 5b).
sit to other steady-state values, which are completely differ- The loss of stability for identical classifiers is always of
ent in comparison to preliminary calculated ones (curve 4). fast stabilizing oscillation character but the oscillations oc-
It means that the character of stability loss depends on the curs around the throughput values, which are different in
sign of the feed perturbation. The regimes 2 and 4 are ob- comparison to the nominal ones. The reason for the appear-
viously unstable and their basic predicted parameters can- ance of these auto-oscillations is namely the similarity of
not be realized in practice. At k < 0 the loss of stability is
of oscillating character (Fig. 3b). Fig. 4 generalises the re-

Fig. 4. Areas of stable and unstable operation of the closed milling circuit Fig. 6. Areas of stable and unstable operation of the network of classifiers
(os = 5; s = 15). (os = 5, s = 15).
272 V. Mizonov et al. / Chemical Engineering and Processing 44 (2005) 267272
the characteristics of the classifiers. If the value of k is out
of the stable area, an increase of the feed to the first clas-
sifier leads to the decrease of it to the second one, and, re-
spectively, the decrease of the recycling particle flow. The
classifiers work like in opposite phases, and this supports
the auto-oscillations. A generalization of the results of nu-
merical experiments is shown in Fig. 6.
4. Conclusions
In this work, a convenient and effective algorithm for cal-
culation of the steady-state parameters of complex techno-
logical systems for processing granular materials, based on
the theory of Markov chains, has been proposed. It allows
a fast construction of a model representing a circuit of ar-
bitrary complexity and its computational analysis. It is also
shown that, the throughputs in a complex milling circuit
with recycling particle flows calculated for a steady-state
regime can lose the stability and grow unbound, even after
a small perturbation of the feed rate if the characteristics
of the classifiers do not satisfy the conditions of stability.
It may cause the blockage of the mill in a closed milling
circuit, and a high deviation of operating parameters from
calculated ones in a network of classifiers. This suggests
that a reliable modelling of complex milling systems can-
not be done only on the basis of steady-state parameters of
the equipment employed: it is necessary to analyse the re-
sponse of the system parameters to variation of operating
conditions.
References
[1] R. Nassar, S.T. Chou, L.T. Fan, Modelling and simulation of deep-bed
filtration: a stochastic compartmental model, Chem. Eng. Sci., 1986,
pp. 412017412027.
[2] A. Tamir, Applications of Markov Chains in Chemical Engineering,
Elsevier Publishers, Amsterdam, 1998, p. 604.
[3] H. Berthiaux, J.A. Dodds, Modeling classifier network by Markov
chains, Powder Technol. 105 (1999) 266273.
[4] V.E. Mizonov, E.A. Barantseva, K. Marikh, H. Berthiaux, Application
of the theory of Markov chains to model mechanical processes in
chemical engineering. in: the Proceedings of the Fourth International
Conference on Theoretical and Experimental Foundations of Design
of High Effective Processes and Equipment for Chemical Technology,
Ivanovo, 2628 June, 2001, pp. 9294 (Russian).
[5] C. Dechsiri, J.C. Abanades, H.G. Dehling, A.C. Hoffman, P.C.
Wright, A stochastic model for mixing an segregation in slugging
fluidised beds, Rcents Progrs en Genie des Procds 14 (2000)
631638.
[6] A.T. Harris, R.B. Thorpe, J.F. Davidson, Stochastic modelling of the
particle residence time distribution in circulating fluidised bed risers,
Chem. Eng. Sci. 57 (2002) 47794796.
[7] K.K. Yin, H. Yang, P. Daoutidis, G.G. Yin, Simulation of population
dynamics using continuous-time finite state Markov chains, Compt.
Chem. Eng. 27 (2003) 235249.
[8] V. Mizonov, V. Zhukov, S. Bernotat, Simulation of Grinding: New
Approaches, ISPEU Press, 1997, p. 108.
[9] V.P. Zhukov, V.E. Mizonov, et al., Mathematical modelling and struc-
tural optimisation of complex milling circuits, Theor. Found. Chem.
Eng. 32 (1998) 288293 (Russian).
[10] V.E. Mizonov, H. Berthiaux, et al., On one approach to describe
the kinetics of grinding, Chem. Chem. Technol. 42 (1999) 124126
(Russian).
[11] O. Molerus, H. Hoffmann, Darstellung von Windsichtrtrenn-kurven
durch ein Stochstisches Modell, Chem. Ing. Tech. 41 (1969) 340
344.
 Chemical Engineering and Processing 44 (2005) 273277

Optimising design of continuous grinding mill-classifier systems

P.B. Kis a, , Cs. Mihlyk b , B.G. Lakatos c
a Institute of Natural Sciences, College of Dunajvros, H-2401 Dunajvros, P.O. Box 152, Hungary
b Department of Mathematics and Computing, University of Veszprm, H-8201 Veszprm, P.O. Box 158, Hungary
c Department of Process Engineering, University of Veszprm, H-8201 Veszprm, P.O. Box 158, Hungary

Received 2 September 2003; received in revised form 24 March 2004; accepted 15 April 2004
Available online 26 June 2004

Abstract

An optimising design method is presented for designing closed grinding circuits to meet the requirements for the particulate product
expressed as a combination of the mean value and standard deviation of the particle size. The grinding circuit is modelled by a discrete
mathematical model describing the axial mixing and breakage of particles in the mill and a possible time delay in the recycle line. The design
problem is formulated as a nonlinear programming task in which the convective flow velocity and the axial dispersion coefficient of the ground
material are selected as design variables.
The method proved to be a useful tool for optimising design of grinding systems, and can be applied as a decision aid for process engineers
to improve the efficiency of processes and maintaining the quality required for the particular products.
2004 Elsevier B.V. All rights reserved.

Keywords: Continuous grinding; Grinding mill-classifier circuit; Optimising design; Nonlinear programming; Simulation

1. Introduction rial to be ground may affect the efficiency and productivity

Grinding is an important operation within the field of
processing technology. It is aimed to produce particulate
material of a given particle size distribution prepared for
subsequent processing or as an end product. Further, since
grinding is an energy consuming operation, appropriate se-
lection of configuration and design of grinding systems is
of primary importance.
A number of methods have been used for optimisation of
grinding circuits. Espig and Reinsch [1] presented the refer-
ence state method, Gupta et al. [2] applied an enumeration
method based on their expert system, Lestage et al. [3] used
linear programming for real-time optimisation, and Mitra
and Gopinath [4] used a genetic algorithm for multiobjec-
tive optimisation of grinding circuits. Morrell [5] developed
new autogenous and semi-autogenous models for design and
optimisation of grinding.
In most of the previous works dealing with optimisation,
the mills were described as perfectly mixed units. In a num-
ber of grinding systems, however, mixing and breakage of
particles occur simultaneously, so that mixing of the mate-
Corresponding author. Tel.: +36 25 551244; fax: +36 25 551240.
E-mail address: piros@mail.duf.hu (P.B. Kis).
of systems significantly. Besides, in industrial operations,
grinding mills practically always are coupled with classi-
fiers, thus it is advisable to consider the design and optimisa-
tion of grinding circuits taking into account both mixing and
product classification with recycling the oversize particles.
The paper addresses the problem of optimising design of
a grinding mill-classifier system in order to satisfy the prod-
uct requirements expressed as a combination of the mean
and variance of particle size. A discrete model is used for
describing the material transport in the mill, and the optimal
conditions are chosen by numerical experiments and non-
linear optimisation.
2. Mathematical model
The continuous grinding mill-classifier system under
investigation is shown schematically in Fig. 1. This is a
network of operational modules for which we have the
following assumptions:
1. The mixer is a static ideal mixing unit;
2. The mill is described by a discrete mathematical model,
developed on the basis of the axial dispersion model

0255-2701/$ see front matter 2004 Elsevier B.V. All rights reserved.
doi:10.1016/j.cep.2004.04.009
274 P.B. Kis et al. / Chemical Engineering and Processing 44 (2005) 273277

equations describing processes in the first feed section, in

internal sections, and in the last outlet section of the mill,
respectively.
Introducing now the notation xi(1/2) = ihx hx /2, the
discrete mathematical model of the grinding mill-classifier
circuit, shown schematically in Figs. 1 and 2, is formed by
the following set of recurrence algebraic equations.
Fig. 1. Grinding mill with product classification and recycled oversize
fraction.
First section of the mill:

(y1 , xi , tn+1 ) = (1 VF S(xi(1/2) )) (y1 , xi , tn )

describing transport and the fundamental grinding equa- I
tion describing the kinetics of breakage of particles [6];

+ VB (y2 , xi , tn ) + pk,i (y1 , xi , tn )
3. The classifier is a static device and is described by a
k=i
classification function [7];
4. The recycle line is described by plug flow so that it rep- + ai (tn ) + (VF VB ) (xi(1/2) )
resents some delay in recycling the oversize fraction of (yJ , xi , tnd ) i = 1, 2, . . . , I. (1)
particles.
Internal sections of the mill:
With these assumptions, the discrete mathematical model
of the grinding mill-classifier system is as follows. Let the (yj , xi , tn+1 ) = VF (yj1 , xi , tn ) + (1 VF
interval [0,Y] represent the length of the mill. We subdi-
vide [0,Y] into J equal subintervals of length hy , denoted VB S(xi(1/2) ))(yj , xi , tn )
by yj = jhy , j = 1, 2, . . . , J, the discrete axial coordinate I


of the mill. Similarly, if x stands for the particle size, the + VB (yj+1 , xi , tn ) + pk,i (yj , xi , tn )
interval [xmin , xmax ], representing the total size interval of k=i
particles is subdivided into I equal subintervals, and we de- j = 2, . . . , j 1, i = 1, 2, . . . , I. (2)
note the ith size of particles by xi = xmin + ihx , I = 1, 2,
. . . , I. Then, the rectangular domain [yj 1 ,yj ] [xi1 ,xi ] is Last section of the mill:
called the (j,i)th cell of the model, and as a consequence, the
2D-discrete computational scheme of the mill consists of J (yJ , xi , tn+1 ) = VF (yJ1 , xi , tn )
columns termed sections and J I cells as it is illustrated + (1 VF S(xi(1/2) ))(yJ , xi , tn )
in Fig. 2. I


Finally, since the process is modelled as a discrete time + pk,i (yJ , xi , tn ) i = 1, 2, . . . , I.
system with as the time step, the simulation time is denoted k=i
by tn = n where the integer n is taken from 0 to 8. (3)
Let (yj ,xi ,tn ) denote the total amount of particles in the
(j,i)th cell at time tn . Then, the governing equations of the Here, parameter VF denotes the velocity of forward flow
system [6,7] because of the boundary conditions at the inlet of particles in the mill in axial direction, while VB is the ve-
and outlet of the mill depending on the feed and classifica- locity of flow backwards. VF and VB are intrinsic parameters
tion conditions can be divided into three groups. Namely, of the discrete model [6], while parameters Pe, u and D are
originated in the axial dispersion model. Namely, Pe is the
Peclet number related to the convective flow velocity u and
the axial dispersion coefficient D as Pe = uY/D where Y de-
notes the length of the mill. The Peclet number characterises
the axial mixing of particles in the mill in dimensionless
form, analogously to the dimensional dispersion coefficient
D. It is proved [6] that relations VF = (1+Pehy )/(Peh2y ) and
VB = (1/Peh2y ) are hold between these parameters where
and hy are, in essence, parameters of the numerical scheme.
In Eqs. (1)(3), S() denotes the breakage selection func-
tion by means of which S(xk )
S(xk ) = Ks xk (4)

provides the rate of breakage of particles of size xk . The

Fig. 2. The jth column (a) and the (j,i)th cell (b) of the discrete model second function describing the kinetics of breakage is the
of the grinding mill. breakage distribution function B(,) which usually is used
 P.B. Kis et al. / Chemical Engineering and Processing 44 (2005) 273277
in the form [8]:
   
xi xi
B(xk , xi ) = + (1 ) ,
xk xk
xmin xi < xk xmax , 0 1. (5)
In expressions (4) and (5), , Ks , , and are consti-
tutive parameters. Based on these relations, we can precise
also functions pk,i in Eqs. (1)(3): pk,I = S(xk )[B(xk ,xi )
B(xk ,xi1 )], I = 1, 2, . . . , I; k = i, . . . , I.
Let us note that in order to have stable numerical scheme
in simulation the condition (Peh2y )/(2 + Pehy +
Peh2y S(xi(1/2) )) is required to be satisfied.
The first term on the right hand side of Eq. (1) gives
the amount of particles of ith size at the moment of time
tn which neither moved forwards nor broke during one
time step of simulation. The second term represents those
particles of the same size that moved backwards from the
second section. The third term expresses the amount of
particles that remained in the first section and have bro-
ken from some size higher than xi to that subinterval of
size. Term ai (tn ) denotes the amount of particles of the ith
size that freshly enters the system. Finally, the last term
(VF VB )(xi(1/2) )(yJ , xi , tnd ) is the fraction of par-
ticles of the ith size that left the mill at the moment of time
tnd and was classified by the classifier to recycle into the
mill for further grinding. Here, function i , 0 i = (xi )
1, describes the operation of the classifier, whilst parame-
ter d denotes the discrete time delay in the recirculation line.
The variation of the amounts of particles in the internal
sections of the mill is described by Eq. (2). The first term on
the right hand side of Eq. (2) represents the fraction of par-
ticles of the ith size that moved forwards from the previous
column, while the third term gives that part of particles of the
same size that moved backwards from the next column. The
second term expresses the amount of particles that have not
changed at all. Finally, the last term describes the amount of
those particles that remained in the section and has broken
from some size larger than xi to the ith subinterval.
Finally, the size composition of particles in the last section
of the mill is described via Eq. (3). The first term on the
right hand side expresses the amount of those particles that
moved forwards from the previous column. The second term
represents particles that have not changed at all, while the
third term gives the amount of particles that remained in the
last column and have broken from some size larger than xi
to the interval in question.
3. Formulation of the optimisation problem
The particle size distribution of the product is one of the
most important product characteristics of grinding circuits.
Instead of the complete distribution, however, we are limited
mostly to determining a few characteristic parameters of the
product size such as the mean particle size and the variance,
275
containing information about the average deviation of the
particle size from its mean. Following the model formed by
Eqs (1)(3), the mean particle size and the standard deviation
of size at the outlet are computed as
I

(yJ , xi , tn )
x(tn ) = xi(1/2) I (6)
i=1 i=1 (yJ , xi , tn )
and

 I

(yJ , xi , tn )
s(tn ) =  xi(1/2) I
2 x(tn )2 (7)
i=1 i=1 (y J , xi , tn )
where x denotes the mean particle size and s stands for
the standard deviation, being the positive square root of the
variance.
Assuming now that the requirements for the quality of
the product are formulated in terms of prescribed mean size
0 and standard deviation 0 , the objective function for
choosing the design variables is formulated as
Q(d) = w1 (xs (d) 0 )2 + w2 (ss (d) 0 )2 (8)
where w1 and w2 are weighing factors, expressing the
significance of the mean value and standard deviation terms,
according to the requirements, xs (.) and ss (.), denote the
steady state values of the mean size and standard deviation,
respectively, computed by solving the set of Eqs. (1)(3),
and d is the vector of the design variables.
Based on these considerations, the purpose of optimisa-
tion is to determine the values of the design variables d of
the grinding circuit to produce such particulate product that
the objective function be minimal.
In this formulation, precising the initial conditions for
Eqs. (1)(3), this set of equations is solved recurrently in
time to generate the steady state values xs (.) and ss (.), pro-
vided the system parameters and the feed conditions are
given. This means that the set of Eqs. (1)(3) is solved in
each step of searching the optimal solution, monitoring at
the same time if the steady state was reached.
4. Results and discussion
In the simulation runs, the design parameters were chosen
as d = (u,D), i.e. the kinetic parameters of breakage and
the classification function were fixed. The size distribution
of particles in the feed stream was constant, selecting 0 for
x = xmax /2 and 0.05 for x > xmax /2, while the initial size
distribution of the material in the mill was the same in each
run. The largest particle size was chosen xmax = 103 m,
and the values of the kinetic parameters were Ks = 104 ,
= 1.00, = 3.00, = 0.60, = 0.48. The classification
function was a jump function of the form
0, i = 1, . . . , [I/2]
i =
1, i = [I/2] + 1, . . . , I
276 P.B. Kis et al. / Chemical Engineering and Processing 44 (2005) 273277

Table 1
Values of the objective function in the environment of local minimum
Convective flow Axial dispersion coefficient, D (m2 /s)
velocity, u (m/s)
0.004 0.005 0.006 0.007 0.008 0.009

0.018 40.0625 33.1646 26.9248 21.5284 16.7450 12.6413

0.019 19.6625 14.8512 10.7871 7.4337 4.7141 2.6914
0.020 6.2962 3.7217 1.8271 0.9630 0.0904 0.1579
0.021 0.3416 0.0025 0.3538 1.4282 2.9864 5.1978
0.022 1.8501 3.8073 6.4657 9.7005 13.5153 17.9604
0.023 10.9857 15.3290 20.2159 25.7795 31.7940 38.4365
0.024 27.8646 34.5650 41.8590 49.7104 58.0859 66.9463

The model allows applying more general classification func-

tions, describing non-ideal classification of particles. Since,
however, the parameters of classification were not taken on
as design variables, this ideal classification appeared to be
satisfactory for simulation purposes.
Selecting parameters u and D as design variables for fixed
length of the mill, also the mean residence time t = Y/u
and the Peclet number Pe = uY/D become determined.
The intervals of feasible solutions u = [0.018, 0.024] m/s
and D = [0.004, 0.009] m2 /s, and the value Y = 6.0 m were
selected on the basis of measurement data by Van Nierop
and Moys [9,10]. Finally, the parameters of the numerical
scheme were chosen as d = 10, I = 20, J = 20 and = 0.1.
Based on these considerations, the task of optimising
design was formulated as

min Q(d) = (xs (u, D) 56)2 + 0.01 (ss (u, D) 55)2
d respect to the design variables u and D.
(9)
{subject to the constraints given by the steady state solutions
of Eqs. (1)(3)}.
The computer program for solving the nonlinear program-
ming problem (9) was developed using Matlab, applying the
simplex method for searching the optimal solution.
The optimal values of the design variables, obtained
as a solution of problem (9), are uopt = 0.021 m/s and
Dopt = 0.005 m2 /s, and the objective function takes value
Q(0.021,0.005) = 0.0025 here, as it is shown in Table 1.
A number of numerical experiments were carried out to
investigate the sensitivity of the objective function around
the optimal solution. The values of the objective function in
the environment of optimal point are also shown in Table 1,
and these very values are shown graphically in Fig. 3.
Both the data in Table 1 and the surface Q(u,D) in Fig. 3
illustrate well that the sensitivity of the objective function
with respect to the convective flow velocity is significant.
At the same time, variation of Q as a function of the ax-
ial dispersion coefficient seems to be smaller, which means
that the mean residence time has stronger influence on the
properties of the particulate product than the axial mixing
of particles.
Fig. 3. Surface of the objective function Q illustrating the sensitivity with
5. Conclusions
An optimising design method was presented for designing
continuous grinding circuits to meet the requirements for the
particulate product expressed as a combination of the mean
value and standard deviation of the particle size. A discrete
mathematical model, describing the axial mixing and break-
age of particles in the mill and some time delay in the recy-
cle line, was applied to describe the behaviour of the closed
circuit. The task was formulated as a nonlinear program-
ming problem in which the convective flow velocity and
the axial dispersion coefficient of the ground material were
selected as design variables. The sensitivity study showed
that the mean residence time has stronger influence on the
properties of the particulate product than the axial mixing of
particles.
The method proved to be a useful tool for optimising
design of grinding systems, and can be applied as a de-
cision aid for process engineers to improve the efficiency
of processes and maintaining the quality required for the
particular products.
 P.B. Kis et al. / Chemical Engineering and Processing 44 (2005) 273277 277

Acknowledgements Greek letters

parameter of the selection function S
The work was supported partially by the Hungarian parameter of the breakage function B
Research Foundation under Grant T034406. The financial parameter of the breakage function B
support is gratefully acknowledged. Csaba Mihlyk is parameter of the breakage function B
grateful for the financial support of the Jnos Bolyai Re- amount of ground material
search Fellowship of the Hungarian Academy of Science. classification function
discrete time step

Appendix A. Nomenclature

ai feed rate of particles of the ith size References

B breakage distribution function
d delay in the recycle expressed in time steps [1] D. Espig, V. Reinsch, Computer aided grinding circuit optimization
d design vector utilizing a new mill efficiency curve, Int. J. Miner. Process. 4445
D axial dispersion coefficient (m2 /s) (1996) 249259.
[2] R. Gupta, K.S. Shishodia, G.S. Sekhon, Optimization of grinding
hx length of subinterval of particle size
process parameters using enumeration method, J. Mater. Process.
hy length of subinterval of the axial coordinate Technol. 112 (2001) 6367.
Ks parameter of selection function S [3] R. Lestage, A. Pomerleau, D. Hodouin, Constrained real-time opti-
pk,i coefficients in Eqs. (1)(3) mization of a grinding circuit using steady-state linear programming
Pe Peclet number supervisory control, Powder Technol. 124 (2002) 254263.
[4] K. Mitra, R. Gopinath, Multiobjective optimization of an industrial
Q objective function
grinding operation using elitist nondominated sorting genetic algo-
s computed standard deviation rithm, Chem. Eng. Sci. 59 (2004) 385396.
ss steady state value of s [5] S. Morrell, A new autogenous and semi-autogenous mill model for
S selection function scale-up, design and optimisation, Miner. Eng. 17 (2004) 437445.
t time (s) [6] Cs. Mihlyk, T. Blickle, B.G. Lakatos, A simulation model for
analysis and design of continuous grinding mills, Powder Technol.
tn discrete time
97 (1998) 5158.
u convective flow velocity (m/s) [7] P.B. Kis, Cs. Mihlyk, B.G. Lakatos, Mathematical models for
VB flow velocity forwards in axial direction simulation of continuous grinding processes with recirculation, Acta
VF flow velocity backwards in axial direction Cybernetica 15 (2002) 529545.
x particle size (m) [8] L.G. Austin, A discussion of equations for the analysis of batch
grinding data, Powder Technol. 106 (1999) 7176.
xi discrete particle size (m)
[9] M.A. Van Nierop, M.H. Moys, Premature centrifuging, oscillation
x computed mean particle size (m) and axial mixing in industrial grinding mill load, Minerals Eng. 11
xs steady state value of x (1998) 437445.
y axial coordinate (m) [10] M.A. Van Nierop, M.H. Moys, Axial mixing of slurry in an auto-
yi discrete axial coordinate (m) genous grinding mill, Int. J. Mineral Process. 65 (2002) 151164.
Y length of the mill (m)
w weighing factor of the objective function
 Chemical Engineering and Processing 44 (2005) 279285

Air classification of solid particles: a review

M. Shapiro , V. Galperin
Laboratory of Transport Processes in Porous Materials, Faculty of Mechanical Engineering,
Technion-Israel Institute of Technology, Haifa 3200, Israel

Received 22 June 2003; received in revised form 14 January 2004; accepted 11 February 2004
Available online 1 July 2004

Abstract

An overview is given of modern air classification devices, their operation principles, features and parameters, as cut size, cleanness
and recovery. We outline the separation zones common to existing classifiers. In practice, various gravitational and centrifugal classifiers
are employed, including those with vertical and horizontal air streams, cascade classifiers, fluidized bed separators, inertial, vortex, rotor
classifiers, etc. Choice of the classifier type and designs is dictated by the technological requirements (throughput, cleanness, etc.) and the
properties of materials to be classified. Advantages and limitations of the classifiers and basic trends in their development are considered.
2004 Elsevier B.V. All rights reserved.

Keywords: Air classification; Gravitational; Centrifugal; Particles

1. Introduction coarse fraction content in the coarse product Cc . Conversely,

Air classification is a technological method of separating
particulates according to their size, as represented by particle
equivalent diameter [1]. The latter may be either geometric
or hydrodynamic/aerodynamic equivalent diameter, depend-
ing on the size-measurement method used. The initial or
original (o) material is divided into two products: coarse (c)
and fine (f). Separation is characterized by a cut size d50 ,
which is a border between these fractions (see below defini-
tion of this quantity in terms of fraction cleanness and recov-
ery). Particles with sizes above d50 constitute coarse fraction
and those with sizes below d50 are fines. Due to various
stochastic factors (air turbulence, particle collisions), some
fines get into the coarse product and vice versa. Quality of
particle separation is characterized by mass contents of each
fraction in the suitable product. Denote the mass contents
of the coarse fraction (in per cent) in the feed, coarse and
fine products by Co , Cc and Cf , respectively. Similarly, Fo ,
Fc and Ff are the mass contents of the fine fraction in the
corresponding materials. Fractional cleanness is the mass
content of any useful fraction in the corresponding product,
namely, fine fraction content in the fine product Ff , and
Corresponding author. Fax: +972 48 294923.
E-mail address: mersm01@tx.technion.ac.il (M. Shapiro).
the content of fines in the coarse product Fc , or coarse in the
fines product Cf is fractional dirtiness. The mass balance
may be expressed as mo = mc + mf , where mo , mc and mf
are masses of initial, coarse and fine products, respectively.
Product yield is the mass of a suitable product relative to
the feed, i.e. coarse product yield is c = mc /mo , fine product
yield is f = mf /mo . It is clear that c + f = 1 or 100%.
Recovery of a certain fraction is the ratio between the
masses of any fraction in the suitable product and in the
original material. The recovery shows which part of useful
particles contained in the feed gets into a suitable product.
The coarse and fine fraction recoveries are expressed by
c = mc Cc /mo Co , f = mf Ff /mo Fo , respectively. Cleanness
and dirtiness characterize quality of the separated products.
In contrast, fraction recovery characterizes the separation
efficiency. In particular if the original product is composed
only of particles having cut size, its fraction recovery is 50%.
In continuous technological processes direct measurement
of products masses is difficult. The recoveries are usually
estimated by determination of the fraction content in sepa-
rated products. This may be done by analyzing their probes,
e.g. by sieving, sedimentation method etc. Using the frac-
tion contents, one can estimate the recoveries of coarse and
fine fractions by the following respective expressions: c
= Cc (Co Cf )/[Co (Cc Cf )], f = Ff (Fo Fc )/[Fo (Ff
Fc )]. Their average gives total separation efficiency: av

0255-2701/$ see front matter 2004 Elsevier B.V. All rights reserved.
doi:10.1016/j.cep.2004.02.022
280 M. Shapiro, V. Galperin / Chemical Engineering and Processing 44 (2005) 279285
= 0.5(c + f ). The above definitions of c , f are suitable
when the cut size is given. Otherwise, c , f may be esti-
mated by grade efficiency curves [1].
2. Particle separation zones
An interior space of a classifying device, where solid
particles interact with air stream, is the separation zone.
Four basic separation zones are known, as described
by Rumpf [2], which include gravitational-counterflow,
gravitational-crossflow, centrifugal-counterflow and centri-
fugal-crossflow zone.
2.1. Gravitational-counterflow zone
The gravitational-counterflow zone exists in air stream ris-
ing up with velocity u inside a vertical chamber with parallel
walls (Fig. 1a). Particles experience gravity G and drag Fd
forces acting in opposite directions. Coarse particles, hav-
ing terminal settling velocity VTS larger than air flow ve-
locity, move downwards, against the air stream, and fines
rise with the stream. The cut size d50 is calculated from the
equilibrium G = Fd combined with zero particles absolute
velocity, i.e. VTS = u. Theoretically, the cut size particles
are indefinitely suspended in the separation zone. Practi-
Fig. 1. Separation zones: (a) gravitational-counterflow zone, (b) gravitational-crossflow zone, (c) centrifugal-counterflow zone, (d) centrifugal-crossflow
zone.
cally all particles are separated due to unavoidable stochastic
factors.
2.2. Gravitational-crossflow zone
A horizontal air stream entering this zone expands within a
chamber and converges towards the outlet (Fig. 1b). Particles
fed in the chamber with zero velocity accelerate horizontally
by the drag force Fd and also fall down. At the end of the
acceleration interval the horizontal velocity component vx of
all particles almost reaches the are flow velocity u and they
do not separate in this direction. Separation occurs due do
particles vertical motion across the air stream. Each particle
falls at its own terminal settling velocity. Accordingly, due
to the horizontal motion they land on the bottom at different
chamber locations: coarse-close to the inlet, smaller-close
to the outlet. Coarse fraction is gathered on the bottom of
the chamber, fines do not reach the bottom and are carried
away by the air flow through the outlet.
Thus, in the gravitational-crossflow zone particle separa-
tion is a two-dimensional process, where their trajectories
are in fact ballistic tracks. Unlike the counterflow zone, the
cut size depends not only on the particles terminal velocity,
but mainly on the chamber length and height. These param-
eters are chosen in such a way that particles with the cut
size land at the farthest point A of the bottom (Fig. 1b).
 M. Shapiro, V. Galperin / Chemical Engineering and Processing 44 (2005) 279285
2.3. Centrifugal-counterflow zone
This separation zone is characterized by a flat air vortex
prevailing in a cylindrical chamber with tangential inlet and
central outlet (Fig. 1c). In this vortex air rotates and flows
radially towards the chamber centre. The radial air move-
ment (radial sink flow type) serves as the particles separa-
tion track.
Particles fed tangentially with air stream are involved in
rotation. Due to the centrifugal (inertial) force Fc their tra-
jectories gradually deviate from the air streamlines. Sepa-
ration is governed by the balance between Fc and the drag
force component Fdr induced by the radial air movement.
Coarse particles drift towards the chamber walls, where they
fall down to the outlet. Fines move towards the chamber
axis where they are eventually carried out by the air stream.
Particles with cut size theoretically move with zero radial
velocity, but practically go to one of the products. As the
centrifugal force depends on the radial position, so does the
cut size d50 . This distinguishes the separation process in the
centrifugal-counterflow zone from the separation occurring
in the gravitational-counterflow zone. To keep the cut size
constant along the radial direction, the classifier should be
designed to provide constant tangential air velocity. This
feature is characteristic of the Archimedean spiral or free
vortex flow [2].
2.4. Centrifugal-crossflow zone
This zone may be created in an air vortex for example
in a cylindrical chamber with inlet and outlet placed on the
opposite sides of the chamber axis (Fig. 1d). The fed par-
ticles enter with the air stream through whirl blades, which
create a screw-type flow. The blade profiles should be de-
signed such that the radial component of the air velocity
everywhere vanishes. Particles rotate while drifting radially
towards the chamber walls with size-dependent velocities.
As they simultaneously proceed along the chamber axis they
reach the wall in different locations. Larger particles arrive
Fig. 2. Cascade classifiers: (a) zigzag, (b) shelf classifier, (c) with horizontal scavenging.
281
to the wall faster; smaller ones make it farther along the ax-
ial direction. The chamber is designed in such a way that
the cut size particles reach the periphery at its edge (point
A). The separation process in the centrifugal-crossflow zone
is similar to that in cyclones; so these devices have similar
design principles.
3. Air classifiers types and designs
Numerous air classifier designs are described in the
literature [3,21]. No universal design method can be recom-
mended for separation of different materials, as each device
type has its specific technical requirements and suitable
for specific product. The separation zones described above
serve as a basis for systematizing air classifiers, which
may have several zones used in combination in the same
device.
Air classifiers may have external or internal aerodynamic
cycles. The former devices are constructed independently
of dust collectors and fans and connected to them via
pipelines. The internal-cycle (recirculating-air) devices
contain these elements in a single unit, and hence need
less production area and normally require lower invest-
ments. They have several technical shortcomings outlined in
subsection 3.5.
Air classifiers may also be distinguished by method of
powder feed. Particles may be fed either within air stream, or
separately by freely pouring them within the device. Other
systematizing may be done with respect to the air stream
form, classifier design features and operation mode (contin-
uous or batch).
Below we review principal schemes of most commonly
used air classifiers to summarize the most important features
of their design and operation. Modern trends of classifier
design account for stochastic nature of separation and aero-
dynamic parameters of equipment, creation of precise air
velocity field, repeated cleaning of both fractions, arrange-
ment of several classifying stages, etc.
282 M. Shapiro, V. Galperin / Chemical Engineering and Processing 44 (2005) 279285
3.1. Gravitational air classifiers
3.1.1. Counterflow classifiers with vertical (rising) flows
Early classifiers of such type [3] were constructed as verti-
cal chambers having within a rising air stream (Fig. 1a) with
particles fed from the top or from the side. Being inexpensive
and simple in construction and maintenance, these devices
have one important shortcoming. Namely, the fed particles
break air flow uniformity and create significant turbulence.
As such, coarse particles prevent fines from rising with the
air stream. This in combination with the single-stage oper-
ation mode underlies low separation efficiency of counter-
flow classifiers with rising flows. These classifiers are used
only for separation of coarse bulk materials, containing two
types of particles of well differing sizes, e.g. coarse sand or
pebbles of 110 mm mixed with dust 0.10.2 mm.
3.1.2. Crossflow classifiers with horizontal air stream
Separation in these devices known as winnowing ma-
chines [3], is achieved due to the difference between tra-
jectories of coarse and fine particles in the separation zone
(Fig. 1b). Their operation and efficiency are strongly af-
fected by the stochastic factors of the process, in particular
by uncertainties in feeding and particles aerodynamic inter-
actions. In most cases coarse particles prevent proper clas-
sification of fines. Separation efficiency of these devices is
usually low. They are normally used for separation of solid
particles according to densities (e.g. grain from peel), rather
than by size. Sometimes crossflow separation in horizontal
streams is used in combination with other separation prin-
ciples.
3.2. Cascade air classifiers
These devices, known as Zigzag classifiers [4], are cas-
cades of several inclined branch pipes of a rectangular cross
section arranged to form a vertical zigzag channel (Fig. 2a).
Air flowing from below scavenges the particles fed from
above in cross-current direction. The flow field within the
chamber has vortexes in the turn points, where the fines are
dragged by the stream, and the coarse particles cross it to-
wards the opposite wall and pour down. Separation occurs in
a cross-flow process. Sequential pipes arrangement enables
multiple scavenging to improve the separation efficiency.
Several types of Zigzag classifier are produced by Alpine
[5] for separation of ferruginous ore, fertilizers, plastic,
wooden dust, etc. at cut sizes ranging from 0.1 to 10 mm.
The number of parallel chambers varies from one to 16
with output ranging from 1 to 10 t/h.
Although multiple scavenging improves separation,
coarse particles crossing the air stream on each cascade step
collide and mix with fines. This precludes sharp separation
and decreases the efficiency. Another disadvantage of the
device is its high aerodynamic resistance (pressure drop of
about 2.53.5 kPa), resulting from the vortexes formed in
cascades turn points.
These shortcomings prevail also in Shelf cascade classi-
fier [6], where instead of pipes, inclined shelves are mounted
on the inner walls of a vertical vessel (Fig. 2b). Strong turbu-
lence prevailing under the shelves cause high pressure drops
(of up to 45 kPa) and high energy consumption of these
devices.
Cascade principle was also combined with cross-count
horizontal scavenging [7] in a vertical vessel divided into
separation and precipitation chambers, separated by a per-
meable grid with concave shelves (Fig. 2c). Particles fed
from the top pour onto the upper shelf, glide down and fall
of its edge where they are scavenged by the cross-current
stream. Part of fines is carried into the precipitation chamber
and the rest of the mixture falls onto the next shelf for re-
peated separation. After multiple separations, coarse fraction
exits from the separation chamber and fines leave the pre-
cipitation chamber. The finest fraction exits with the stream
and is collected in cyclones or filters.
This design practically or to a considerable extent elim-
inates particles collisions in the separation zone, which
sharpens separation. As the flow field in the chamber has no
inherent vortexes the device has lower air resistance and en-
ergy consumption. Several prototypes of this classifier were
used in phosphate industry for separation of crushed phos-
phate ore at 1 mm cut size. One such plant with output of 25 t
per hour and pressure drop of 800900 Pa yielded 9698%
cleanness of fines and 7885% of coarse fraction [8].
3.3. Fluidized bed classifiers
In these devices classification occurs in the space over
a fluidized bed from which fine particles are carried away
with the air stream. The air velocity must essentially ex-
ceed the maximal fluidization velocity of fine fraction. These
classifiers operate on the gravitational counterflow separa-
tion principle with fluidized bed used to increase particles
residence time in the separator. Coarse particles can move
horizontally across the bed to the exit, which is another ad-
vantage of this device.
Fluidized bed classifiers may have straight or widening
forms. The latter (Fig. 3a) consist of a conical chamber with
a grid permeable for particles fed from below, inlet and
outlet pipes and a branch for coarse product. Air velocity
in the holes is high so that all the particles move far up-
wards into the cone, where the stream expands and the par-
ticles slow down. The coarse part returns to the grid and the
fines are carried away. The cones upper diameter is cho-
sen in such a way that the air velocity does not exceed the
coarse particles terminal velocity. These particles exit from
a pseudo-boiling layer formed above the grid through the
outlet branch. Pass-through grids allow arranging several
cones one over another i.e. to design multi-stage classifiers
(Fig. 3b). Each stage has its (coarse) product with size de-
pendent on the chamber diameter.
Classifiers with pass-through grids did not find wide use
in practice mainly because of their big size. For example, a
 M. Shapiro, V. Galperin / Chemical Engineering and Processing 44 (2005) 279285
Fig. 3. Fluidized bed classifiers: (a) single-stage, (b) two-stage, (c) and
(d) continuous operation.
three-stage air classifier of 600 kg/h is about 4.5 m tall. Ad-
ditionally, particles crushing resulting from impacts against
the grid, may affect the product quality.
In a later generation of fluidized bed classifiers the
feed is poured onto the grid from above (Fig. 3c, d), so
crushing is eliminated [9]. Several prototypes of such flu-
idized bed classifiers with cylindrical or conical chambers
of diameter 3501700 mm and throughput of 0.510 t/h,
were constructed for separation of sand, potassium chlo-
ride and ground phosphate particles with sizes from 50
to 1.5 mm [10]. The recovery of coarse fraction was
7592%.
Fig. 4. (a) Franken/Buell gravitational-inertial classifier, (b) Vortex classifier, (c) gravitationalcentrifugal classifier.
283
One disadvantage of the fluidized bed classifiers with
cylindrical chambers of large diameters (above 1 m) is for-
mation of dead zones within the fluidized bed far from the
axis, where the particles do not move. These zones are elim-
inated in FB classifiers with chambers of oblong (oval) form
(as in Fig. 3d) [11]. Such a device with 40 t/h throughput
and 160 m cut size was built in the Karatau phosphate
works milling plant. It yielded the fine product coarse frac-
tion cleanness of 86% and separation recovery of 78% [10].
Fluidized bed classification plant constructed in the
Technion (Fig. 3d) has 30200 kg/h feeder output, air flow
rate of up to 400 m3 /h allowing to change the cut from 50
to1000 m [12]. The efficiency is about 90% and above,
depending on the material. The grid of changeable slope
allows controlling the rate of discharge of coarse particles
out of the separator. Investigations showed good perfor-
mance stability of the classifier in the wide range (up to
four times) of feed rate. These devices have both products
recoveries higher at least by 1520% with respect to other
gravitational classifiers.
3.4. Inertial air classifiers
These devices are employed for fine powders (40400 m)
and also coarser particles. In the Franken or Vandenhoek in-
ertial separators [13,14] (Fig. 4a) particle-air mixture enters
from the top and falls down in the inlet pipe. The air turns
abruptly to the outlet pipe inclined by about 4550 and
carries the fine particles. Coarse fraction proceeds straight
down because of high inertia towards the discharge pipe.
On its way this material undergoes repeated separation in a
secondary air flow carrying fines into a curvilinear cham-
ber. Then they exit via the outlet pipe with blades mounted
for better air distribution. These classifiers, manufactured
by Buell Co., are used for separation of pulverized lime-
stone at cut size of 150 m with feeds from a few pounds
to 600 t/h [15]. In the Buells variant of this device [14]
blades have different angles to increase the efficiency. Such
devices were used for separation of 151000 m particles.
In the Buell separators the secondary air stream interferes
with the primary stream causing the fine particles moving
284 M. Shapiro, V. Galperin / Chemical Engineering and Processing 44 (2005) 279285
in the curvilinear chamber to mix with fed material. This
was eliminated by adding a branch pipe in the centre of the
curvilinear chamber to provide a separate exit of fines [16].
One such 20 t/h classifier was tested in phosphate works to
separate 50400 m fraction from ground phosphorite. It
had coarse product cleanness of 91% and recovery of 77%.
3.5. Centrifugal air classifiers
Two main types of these classifiers are known, distin-
guished by method of creating air rotation. In vortex clas-
sifiers the air is rotated by stationary vanes (or tangential
pipes); in rotor classifiers this is done by rotating blades. The
latter devices may have open or closed aerodynamic cycle.
3.5.1. Vortex air classifiers
Vortex air classifiers have centrifugal-counterflow separa-
tion zone. Alpine A-G manufactures several such classifiers
on the basis of the Rumpfs developments. One device of
this type (Fig. 4b) is designed as a flat horizontal cylinder
containing a vortex chamber. Particle-laden air passes via in-
clined vanes into the chamber, through which the air moves
along a spiral trajectory, carrying the fines out. Coarse parti-
cles are drawn to the wall region and exit via a screw outlet.
These classifiers are used for separation of fine powders, e.g.
limestone, quartz, pigments, etc. at cut sizes of 5100 m
[1,17].
3.5.2. Gravitationalcentrifugal air classifiers
Gravitationalcentrifugal air classifiers are two-stage de-
vices having gravitational and centrifugal separation zones,
with air circulation in most cases induced by orientable
blades. The feed enters the classifier in the air stream into
a widening cone (Fig. 4c), where the coarsest fraction is
removed by gravitational separation. The stream with re-
maining material moves up into the centrifugal zone via the
blades having changeable angle to adjust the cut size. Fines
are carried away with the stream, whereas the coarse parti-
cles move to the wall and descend to the exit. These devices
Fig. 5. (a) Rotor classifier, (b) classifier with profiled blades, (c) circulating air classifier.
with throughputs of up to about 20 t/h are widely used in
cement and coal milling plants [18].
3.5.3. Rotor classifiers
Rotor classifiers utilize rotating blades for air circu-
lation. In one such classifier [19] (see Fig. 5a) particle
laden air enters from below and rises upward into a con-
ical vessel, containing a drive-activated rotor with a cone
and rotating blades. These blades whirl the air to create a
centrifugal-counterflow separation zone in the upper part
of the vessel. The cut size is controlled by adjusting the
drives rotational velocity. These devices with throughputs
of up to 1.5 t/h are used for separation of fine quartz powder,
cement, ground limestone etc. at cut sizes of 20100 m
and coarse product recovery of 7080%. Similar classifiers
were developed by Hosokawa Co. [17] for separation of
ground coal, talc, aluminum oxide, etc. A variant of rotor
classifiers (Blade Classifier) [20] has profiled blades similar
to axial compressors (Fig. 5b). The space above the blades
is arranged to form a centrifugal-crossflow zone, (rather
than the counterflow zone, characteristic of the Hosokawa
classifiers). In this zone particle separation occurs as de-
scribed in Section 2.4. Among the various blade profiles
tested, the highest efficiency was registered for that pro-
viding the solid body air rotation. The blade classifiers
allow separation 2060 m powders, e.g. quartz, pigments,
alumina with 7080% both products recovery.
3.5.4. Circulating air classifiers
Circulating air classifiers are widely used, mainly in ce-
ment industry. One type of such devices produced by Sturte-
vant [22] (Fig. 5c) consists of two coaxial exterior and inte-
rior vessels. Particles entering the device from the top onto
the plate are scattered within the interior vessel by a rotat-
ing plate. There they rotate and move to periphery. The fan
creates a circulating air flow that rises within the interior
vessel and descends within the intermediate space between
the two. The air stream returns into the inner vessel through
the vanes. Coarse particles fall down in the inner vessel be-
 M. Shapiro, V. Galperin / Chemical Engineering and Processing 44 (2005) 279285
ing separated as in the centrifugal-crossflow separation zone
(see Fig. 1d). When the air abruptly turns to the vanes, fines
proceed down to their exit. Air-particle motion within the
intermediate space is similar to that in cyclones.
These classifiers may process up to 800 t/h of bulk solids
[22]. They have rather poor dust/air separation in the inter-
mediate zone, resulting in particle accumulation, which re-
duces their efficiency. Cut size control in these devices is
difficult.
4. Conclusions
Choice of the classifier type is dictated by product re-
quirements, technological and economic factors. Rational
design and improving separation quality is done by creating
a well-defined and stable air flow, reducing turbulence, elim-
inating particles collisions, controlling feed and repeated
cleaning of both fractions. Among gravitational type clas-
sifiers, fluidized-bed based devices provide stable operation
and sharp separation in the cut size range of 501000 m.
Smaller-size particles may be separated by rotor classifiers,
enabling easy cut-size control and providing products re-
coveries up to 80%.
References
[1] H. Rumpf, Particle Technology, Chapman and Hall, London, 1990.
[2] H. Rumpf, Fortschritte und Probleme auf dem Gebiete der Wind-
sichtung, Staub, vol. 47, 1956.
[3] A. Taggart, Handbook of Mineral Dressing, John Willey & Sons,
USA, 1950.
285
[4] F. Kaiser, Der Zickzack-Sichter ein Windsichter nach neuen Prinzip,
Chemie-Ing.-Techn. 35, 1963.
[5] Zerkleinerungsmaschinen, Windsichter, Extruder, Alpine Prospekt
017/2R, 1968.
[6] M. Barsky, Powder Fractionation, Nedra, Moscow, 1980.
[7] V. Galperin, et al., Device for Separating Bulk Materials, USSR
Patent No. 1619535, 1984.
[8] V. Galperin, et al, Dry Classification of Phosphate Ore in Chilisai
Phosphate Field, Chem Industry, 12, Moscow, 1988.
[9] A. Vasilenchuk, et al., Gravitational Air Classifier with Fluidized
Bed, USSR Patent 228641, 1968.
[10] V. Galperin, et al., Industrial tests of the Air Classifiers with Flu-
idized Bed, Proceedings of Mines and Chemistry Institute for Raw
Materials, issue 57, Moscow, 1985.
[11] V. Galperin, Gravitational Air Classifier with Fluidized Bed, USSR
Patent 1619535, 1984.
[12] V. Galperin, M. Shapiro, Separation of Solid Particles in a Fluidized
Bed Air Classifier, Powder Handling and Processing, 11, No. 2,
AprilJune 1999.
[13] J.F. Franken, Apparatus for Classifying Particulate Material, USA
Patent 3006470, 1961.
[14] J.P. Vandenhoeck, Classifier with gas flow distributor, USA Patent
3006470, 1961.
[15] Buell, A Division of Fisher-Klosterman, Inc., http://www.fjinc.
com/Buell/cat1.html 2003.
[16] V. Galperin, et al., Air Classifier, USSR Patent 608568, 1978.
[17] Hosokawa, http://www.hosokawa.com.web/powderequip/
CLASSIFIERS/HTM, Powder and Particle Processing, 2003.
[18] G.K. Smyshliaev, Air Classification in Mineral Processing, Nedra,
Moscow, 1969.
[19] V. Rybin, S. Bockshtein, Investigation of Impingement-Vortex Clas-
sifier, Proceedings of the Institute of Building Materials, Gosstrojiz-
dat, 1960.
[20] M. Dubinsky, V. Galperin, Blade Air Classifier, USSR Patent 177766,
1965.
[21] I. Klumpar, et al., Air Classifiers, Chemical Engineering, 3 March,
1986.
[22] Sturtevant Inc., Whirlwind Air Classifiers, http://www.sturtevant.com,
2003.
 Chemical Engineering and Processing 44 (2005) 287291

Application of a vertical venturi separator for improved recycling

of automotive tyres
W. McBride , S. Keys
School of Engineering, Mechanical Engineering Discipline, University of Newcastle, Newcastle, NSW 2308, Australia

Received 1 September 2003; received in revised form 6 January 2004; accepted 11 February 2004
Available online 19 June 2004

Abstract

This paper reports on the successful application of a large vertical venturi separator for the removal of tangled steel and rayon fibres from
crumbed rubber product as part of a tyre recycling process. This represents an advance over the other separation units encountered during this
project in cleaning efficiency and scalability. The venturi separator was able to unwind the synthetic fibres from the steel fibre by keeping the
agglomerate in a suspended state within the throat of the venturi. A chaotic particle motion was observed at the throat which aided in teasing
the synthetic away from the steel fibre. Overall the venturi separator offered far superior cleaning ability, suitable throughput rates and simple
construction/implementation. The concept could easily be applied to other industries where an aerodynamic classifier offering potentially
unlimited particle retention times may be beneficial.
2004 Elsevier B.V. All rights reserved.

Keywords: Vertical venturi separator; Tyre recycling; Automotive; Classifier

1. Introduction duction. In this transportation stage a magnetic separation

The work reported on in this paper was in response to an
industry problem on the removal of agglomerated steel and
synthetic fibres from the saleable rubber crumb as part of
an automotive tyre recycling plant.
One of the difficulties in the conversion of disused tyres
into saleable rubber crumb is the removal of the various
refuse materialsSteel, Rayon and Nylon. The recycling
process for the production of the rubber crumb commences
with primary pulping of the tyres. The pulping can be done
at either ambient or cryogenic temperatures depending on
the process chosen. The cryogenic method provides a lower
level of mechanical work to shred the tyres but this is off-
set by the problems inherent in cryogenic temperatures. In
both ambient and cryogenic processing the primary pulping
is generally a mechanical tearing process working on either
complete tyres or pre-cut sections. In the ambient temper-
ature installation under review material from the primary
mill was transferred to a secondary mill for further size re-
Corresponding author. Tel.: +61 2 4921 5208; fax: +61 2 4921 6946.
E-mail address: william.mcbride@newcastle.edu.au (W. McBride).
unit was in use to collect the liberated steel fibre. This pro-
cess consisted of a standard metal tramp belt unit operat-
ing transverse to the product stream from the primary to the
secondary mill. One difficulty with the magnetic separation
unit was the need to spread the primary product as a thin
layer onto the transfer belt to assist the magnetic separation.
If the steel fibre was buried under any depth of rubber there
was insufficient force generated from the magnet and the
steel fibre would remain in the product. After the primary
steel separation phase the product entered the second mill
for further size reduction and more fibre liberation. Typical
particle size from the primary mill was +20 mm with the
secondary mill producing 6 mm product.
The primary separation of synthetic fibre was achieved
using a number of cyclones assuming that the rubber par-
ticles would dis-engage from the air stream but the fibre
would be carried on. Although the cyclones did remove a
large amount of the synthetic fibre there was a percentage
of free fibre and steel/synthetic fibre agglomerate contami-
nating the saleable product.
The formation of the steel/synthetic fibre/rubber agglom-
erates is a by-product of the milling process and in part
the tyre manufacturing process. The steel belts used in auto

0255-2701/$ see front matter 2004 Elsevier B.V. All rights reserved.
doi:10.1016/j.cep.2004.02.023
288 W. McBride, S. Keys / Chemical Engineering and Processing 44 (2005) 287291

A significant problem with the style of separator shown

Nomenclature as Fig. 1 is that any feed bias to one location will result
ap particle acceleration in a reduced local airflow rate and subsequent reduction in
A duct area the cleaning efficiency; an inherent negative feedback loop
Cd aerodynamic drag factor (non-dimensional) exists in the airflow path.
Fd drag force on a particle due to passing air The steel/synthetic fibre agglomerates formed during the
g acceleration due to gravity milling process are extremely difficult to separate using con-
M mass ventional equipment due to the twisted nature of the steel
r, rp particle radius fibres. If sufficient steel is contained in the agglomerate then
S displacement from origin the magnetic separation process may remove the entire lump
U initial velocity at the first stage of product cleaning. However, this then
V velocity contaminates the steel refuse with both rubber and fibre, re-
Z vertical distance into convergent section ducing both the sale price of the steel refuse and the net
of venturi rubber crumb yield available. If the agglomerate passes be-
a density of air yond the magnetic separator it is unlikely to be removed by
s density of solid the air separators as a large lump presented to the separator
unit shown as Fig. 1 is essentially passed through with lit-
tle cleaning. This results in the agglomerate arriving in the
tyres are a twisted strand wire rope and in the shredding pro- final products bin, reducing the value of the rubber crumb.
cess these formed short twisted lengths of steel fibre. The Although a number of air separation units, other than as
rotary nature of the milling process allowed the synthetic shown by Fig. 1, were reported by our client to be available
fibre to become tangled with the steel fibre often trapping and trialled, none were reported as being able to adequately
crumb rubber as part of the agglomerate. These agglomer- remove the entwined synthetic fibre from the steel fibre to
ated lumps tended to remain tangled throughout the cleaning release the rubber product, and separate the synthetic fibre
processes and often contained sufficient mass and density from the steel fibre.
to pass through the separators as product, reducing the fi-
nal crumb quality.The client trialled a number of secondary
fibre removers with little improvement in final product, a 2. Venturi separator
schematic of a typical secondary fabric separator used at the
plant is shown as Fig. 1. The difficulty with the processing of the agglomerate lies
with the twisted nature of the synthetic and steel fibres. The
existing methods of separation were failing as the agglom-
erate only had a short residence time in the air stream. The
concept of the venturi separator was to make use of the dif-
fering velocity profiles available in the venturi to prolong
the residence time for the agglomerates. Fig. 2 illustrates the
test facility used in this project.
The testing program investigated different total air flow
rates through the venturi and the effects on particle sizes
trapped by the filters compared to the removal efficiency of
the agglomerates.
Central feed tube
The results from the testing program were impressive. In
operation the crumb composite (crumbs, synthetic fibre and
steel fibre) gained momentum from the feed point down to
FEED

To Filter
the mouth of the venturi. With the throat velocity set to
prevent the largest particles from exiting the venturi, an ex-
tremely fast and efficient cleaning was achieved (typically
less than 10 s of residence time). As the material fell down
Cone for material dispersion
into the throat it was progressively reacted on by the op-
posing air flow and accelerated back into the venturi dif-
fuser. A large degree of interparticle collision was evident
in the continuous falling/elevating process providing a ran-
dom or chaotic motion. This randomness tended to tease
the loose end of the synthetic fibre from the steel fibre.
Air flow
As the particle circulated the synthetic fibre was progres-
Fig. 1. Typical style fabric separator. sively pulled from the steel fibre. Once the steel fibre was
 W. McBride, S. Keys / Chemical Engineering and Processing 44 (2005) 287291
Storage
Hopper
Screw Feeder
Central Feed Tube To Filters
Fig. 2. Venturi separator equipment.
cleaned, now having a low aerodynamic drag/density ratio, it
exited the venturi under gravity. This left a batch of cleaned
rubber crumb in the venturi, free of steel and synthetic
fibres.
In order to minimise air separation from the venturi walls
the unit was designed with a 3.3:1 area expansion ratio with
a 6 included angle on the divergent side. The implication
of this limitation is that a second stage of processing is
likely to be necessary to reclaim the smaller rubber frac-
tion lost to the filters in the main venturi separator. The
primary rubber crumb product has a particle size range of
approximately 6 mm. The testing with the venturi separa-
tor effectively produced a main product of 6, +1 mm. The
fraction below 1 mm was lost to the filters with the lib-
erated fabric. However, with the fabric teased away from
the steel inclusions it is expected that a simple cyclone
could be tuned to recover the majority of the 1 mm rubber
crumb.
289
Fd
V
Mg
Fig. 3. Force balance on a particle: aerodynamic due to air flow (up) and;
gravity (down).
2.1. Analysis of particle forces in the venturi
Force balance on a particle can be determined from the
following equations and with reference to Fig. 3.
Drag force on a particle is given by
Fd = 21 a V 2 CdA = 21 a V 2 Cdr 2 (1)
Body force = Mg = 43 r 3 s g (2)
where V is the local air velocity, r the particle radius, g the
acceleration due to gravity, a and s are the air and particle
density, respectively.
Equating 1 and 2 gives
2 a Vdz Cdr = 43 rp3 s g
1 2 2
(3)
2 a Vdz Cd = 43 rp s g
1 2
If Cd, a and s are considered as constants then (3) reduces
to
rp Vdz
2
With a venturi having an area ratio of 3.3:1, the velocity ra-
tio and thus, particle size range trapped in the venturi is in
theory, 11:1. This means if the velocity at the throat is set
to suspend an 11 mm particle then particles just less than
1 mm would be lost to the filters. In reality this bandwidth
was diminished by the layout of the test facility and vari-
able particle Cd. In the layout shown as Fig. 2 the material
entered from a drop tube at the top of the venturi, this al-
lowed the particles to accelerate for nearly 1.2 m prior to the
venturi constriction. If the throat velocity were set to sus-
pend an 11 mm particle then the gravity force is balanced by
the particle drag at the throat, however, there will be no net
acceleration. This means that if an 11 mm particle entered
the throat with any vertical velocity there would be no de-
celeration of that particle and it would pass through. Thus,
the bandwidth of the unit is limited by the method in which
the particles are introduced into the separator. In the testing
work undertaken, a nominal bandwidth of 6:1 was observed.
This could have been increased by altering the feed from a
vertical drop to a lower horizontal entry.
If a particle enters the converging section of the venturi
with an initial velocity of U then it is possible to predict the
290 W. McBride, S. Keys / Chemical Engineering and Processing 44 (2005) 287291

Vdz

To Filters
Z

Butterfly valve to
dump product

Fig. 4. Elemental slice through a convergent section of the venturi.

Multiple units in series

for continious processing
maximum particle size that can be captured by the venturi
Storage
using Hopper

V 2 = U 2 2aS. (4)
Screw Feeder

From the drag Eq. (1) and noting the F = MA then it can be
shown that the acceleration on a particle at some position
Z (Fig. 4) from the start of the converging section can be
determined by
3a Cd 2
ap = V (5)
8rs dz
Where ap is the particle acceleration of a particle with radius
r.
If the venturi is segmented into slices Z thick it can be
shown that the velocity exiting the slice is given by
seperator

Vexit = Uentry + 2x(g ap )Z
(6)
Application of (6) allows evaluation of the lowest point in
the venturi a particle will fall through and (5) will allow for
the calculation of the static suspension position of a given
particle.
Application of these equations for the test data obtained
does not provide a highly accurate match as the equations
are based on particles with a constant Cd. The raw rub-
ber crumb contained a range of particle shapes and corre-
sponding Cds which introduced significant variances into
the computations. However, the equations clearly indicate
the operational background of the unit.
3. Discussion
The use of a vertical venturi separator was successfully
used to clean steel and synthetic fibres from rubber crumb
produced from an automotive tyre recycling plant. The major
advantage of the venturi separator was an inherent positive
feedback loop that exists at the throat. If more material was
introduced to the throat then a higher velocity was generated
at the throat to help retain particles (dependant upon air
mover characteristics). The more conventional cleaning units
observed had a negative feedback inherent in their design,
with an increased local load of product the air velocity would
Magnetic
Crumb Product Belt
Fig. 5. Expected batch system layout.
bypass the area resulting in unprocessed material exiting the
system.
From the testing work undertaken it was unclear if the
venturi unit could be operated in a continuous mode with
clean rubber exiting the throat with a continuing input of
product, or if the unit needed to be operated in a batched
mode with a relatively low loading. Given that the cycle
time for the most heavily tangled agglomerate was in the
order of 10 s the batch operation should not present a limi-
tation to throughput with typical crumb plants operating at
around 6 t/h max. From the information gained in the re-
search program it was determined that a single batch mode
venturi with a 250 mm throat would be able to process the
full plants throughput.
Fig. 5 illustrates the proposed concept for the plant im-
plementation of the process. Unfortunately this system was
never trialled as the company in question was liquidated for
other reasons.
 W. McBride, S. Keys / Chemical Engineering and Processing 44 (2005) 287291
4. Conclusions
Through a program of experiments it was found that the
use of a classically designed venturi could be used to un-
tangle synthetic fibre from steel fibres formed during the
milling process of automotive tyres. With appropriate air
flow settings the venturi was able to hold the largest particle
sizes in the throat area, allow the clean steel to pass straight
through, and remove only the finer fraction of the rubber
with the synthetic fibres. It is believed that the fine rubber
could be easily reclaimed from the fibres using a traditional
cyclone separator.
291
This application of a classic venturi as essentially a
notch pass filter (aerodynamically large sections pass
straight through whilst aerodynamically small particles are
removed through the top) appears a significant improve-
ment over the alternate separation units witnessed during
this project. The alternate units were quite capable of re-
moving free fibre but large particles caused a bias in the
airflow pattern away from the lump. This resulted in a
lowered cleaning efficiency. The venturi unit displayed
an increase in throat velocity with the introduction of
more material causing an increase in the local cleaning
efficiency.
 Chemical Engineering and Processing 44 (2005) 293296

Particle movement during granular intermingling

in a pulsated bottom mixer
Mikls Nemnyi , Attila J. Kovcs
Faculty of Agricultural and Food Sciences, Institute of Agricultural, Food and Environmental Engineering,
University of West Hungary, Vr 2, Mosonmagyarvr H-9200, Hungary

Received 14 August 2003; received in revised form 2 February 2004; accepted 11 February 2004
Available online 28 July 2004

Abstract

Many bulk commodities, such as cereal grains loaded into a silo require active mixing during storage. A special pulsated bottom mixer was
patented at our university for gentle mixing of granular solids stored in large containers or silos. This work intended to reveal the fundamentals
in displacement of single particles mixed by different pulsation orders. A glass-walled simulator (filled with wooden cylinders) was build in
order to be able to follow particles inside the simulated mixer. Three pulsating pistons were attached to the bottom of the simulator. Their
movements provide the mixing effect. Selected cylinders (some with different densities and friction coefficients) were followed during mixing.
Their positions were recorded after certain pulsations. From the measurements the mixing velocity of cylinders was calculated. Examinations
were focused on to give restrictions of the mixing: density and friction of the particles in connection with the pulsation. No correlations were
found between the mixing velocity and density or friction of the particles.
2004 Elsevier B.V. All rights reserved.

Keywords: Particle mixing; Density; Friction; Simulation

1. Introduction
Research on storage, mixing, and preservation techniques
of bulky materials has been initiated 15 years ago at the Uni-
versity of West Hungary, Faculty of Agricultural and Food
Sciences, Institute of Agricultural, Food and Environmental
Engineering. A pilot size storage container was built with in-
flatable air tubes at the bottom of the bulk. Inflating and de-
flating them with compressed air enable to move and mix the
grain layers being placed above them [1]. Using this patented
method the formation of rotting parts can be prevented and
the existing centers can be eliminated. On the other hand,
the mixing system with ventilation can increase the safety of
storage [2,3]. The investigations proved that this method is
also suitable for other granular materials than cereal grains.
Based on the pilot scale measurements mathematical mod-
eling was started in order to describe the processes. Discrete
element method (DEM) was used to model the mixing. This
was originated from a simple mechanical model to formulate
Corresponding author. Tel.: +36 96 566 635; fax: +36 96 566 641.
E-mail address: nemenyim@mtk.nyme.hu (M. Nemenyi).
the contact forces between two spheres and a particle and
wall, respectively [4,5]. The model considers among others,
the stiffness, damping, and friction coefficients of the par-
ticles. However, the results of modeling of pulsated bottom
mixer were inaccurate and did not follow exactly our mea-
surements. Therefore, examinations were set up to clarify
the principles.
A laboratory glass-walled mixing simulator (dimensions:
0.7 m 0.5 m 0.02 m) was built based on the pilot-scale
apparatus (Fig. 1). Three pulsating pistons were attached at
the bottom of the simulator. The circulation of the particles
is due to the sequence of the pulsation of the pistons. The
pattern of motion using the pistons is similar to the motion
achieved by vibrating processes [6]. The difference between
the two processes is that any kind of motion pattern can be
acquired by using the pulsating mixer independently from
the shape of vessel.
Consequently, the direction of movement of particles with
different densities is independent in our system.
The basic aim of this examination was to find out the
effects of densities and friction coefficients of particles on
the velocities of movement and mixing.

0255-2701/$ see front matter 2004 Elsevier B.V. All rights reserved.
doi:10.1016/j.cep.2004.02.024
294 M. Nemenyi, A.J. Kovacs / Chemical Engineering and Processing 44 (2005) 293296

70.00 cm
particles

50.00 cm
42.50 cm vessel
4.00 cm

2 1 3
(a) 17.00 cm (b) pistons 7.00 cm

Fig. 1. Schematic diagram of the laboratory glass-walled mixing simulator of (a) cylinder in the middle move down; (b) cylinders in the middle move
upwards.

Fig. 2. Photo of the mixing simulator during measurement (1, 2, 3: selected cylinders; sequence: 2). (a) Before movement, (b) after 50 sequences, (c)
after 100 sequences, (d) after 150 sequences, (e) after 200 sequences, and (f) after 250 sequences.
 M. Nemenyi, A.J. Kovacs / Chemical Engineering and Processing 44 (2005) 293296
Table 1
The sequences of the pistons (starting point: all three cylinders are down)
Steps Sequence 1 (Fig. 1a)
1 Pistons #2 and #3
2 Piston #1
3 Pistons #2 and #3
4 Piston #1
2. Material and methods
The simulator (Fig. 1) was filled with wooden cylin-
ders with the diameter of 20 mm and the length of 18 mm
( = 700 kg/m3 , = 0.25). The examined parameters
were: (i) mixing of particles inside the bulk; and (ii)
moving (mixing) characteristics of selected cylinders with
different parameters. In each measurement three cylin-
ders with the same properties were followed. Different
properties of cylinders were studied, such as: wooden
cylinders with coarse surfaces (coated with sandpaper):
fine = 0.63; middle = 0.73; rough = 0.91; aluminum
cylinders ( = 0.24, = 2500 kg/m3 ); wooden cylinder
with meat insert ( = 1600 kg/m3 ); hollow metal cylin-
der ( = 0.25, = 6180 kg/m3 ). Figs. 1 and 2 show
the diagram and picture of pulsated bottom mixer. The
lifting of a piston was taken by 0.5 s, and the lowering
was done by 0.1 s. Two moving sequences were utilized
(Table 1).
After 10 sequences the positions of the three selected
cylinders were recorded and a photo was taken from the
Fig. 3. MATLAB-based evaluation of the mixing shown in Fig. 2.
295
Sequence 2 (Fig. 1b)
Up () Piston #1 Up ()
Up () Pistons #2 and #3 Up ()
Down () Piston #1 Down ()
Down () Pistons #2 and #3 Down ()
mixer (Fig. 2). Each measurement was taken until 300500
sequences reached that is each selected cylinder rotated at
least two times.
3. Results and discussions
The evaluation of measurements was carried out using the
MATLAB 6.5 software (Fig. 3). The path of the three se-
lected cylinders visualized and the velocity of each cylinder
was calculated. The velocity of a cylinder was calculated
by dividing the sequences by the distance that the cylinder
drove. The velocities of three cylinders were averaged.
Based on the measurements the following was established.
The velocity of cylinder is independent of the friction coef-
ficient (Figs. 4 and 5). This means that the smooth surfaced
wooden or aluminum cylinders move with the same speed
as the rough sand-paper covered one. Higher average veloc-
ities with 0.2 cm per movement was found in sequence 2
(where the middle examined cylinder moves upward), than
in sequence 1 (where a cylinder in the middle moves down).
296 M. Nemenyi, A.J. Kovacs / Chemical Engineering and Processing 44 (2005) 293296

0. 8 be concluded from this that the wooden cylinder with a

0. 7 density of 700 kg/m3 moves with the same speed as the
hollow metal cylinder having the density of 6180 kg/m3 .
0. 6
Velocity, v [distance:

These measurements also proved that exact mixing effect

cm/movement]

0. 5
could be reached even under a significant density difference
0. 4 (1:10). It was also proved that the denser particles could be
0. 3 moved in any directions into the bulk.
0. 2 The results of the above measurements can be used for
Average: 0.59 improving the mathematical descriptions. However, in this
0. 1
kind of mixing a major revision and maybe simplification
0. 0
of the models seems to be suitable.
0. 0 0.2 0. 4 0.6 0. 8 1.0
Friction coefficient,

Fig. 4. Velocities of the selected cylinders with different friction coeffi-

cients; middle section moves downwards.
4. Conclusions

Measurements were carried out in order to correct the

0. 6
existing discrete element methods that describe mixing ef-
0. 5 fect of granules in a bulk. Different densities and friction
coefficients of particles were studied to find correlations
Velocity, v [distance:

0. 4 between the parameters and the mixing characteristics.

cm/movement]

There was found no effect of these parameters on mixing.

0. 3
Therefore, the DEM modeling work in this case has to be
0. 2 reevaluated.

0. 1 Average: 0. 38

0. 0 Acknowledgements
0. 0 0.2 0. 4 0.6 0. 8 1.0
Friction coefficient, The authors express their thanks to Ms. Mrta Varga (Hun-
Fig. 5. Velocities of the selected cylinders with different friction coeffi- garian Academy of Sciences, Research Group of Process
cients; middle section moves upwards. Engineering of Agricultural Products) and Mr. Roland Pap,
Ph.D. students for their help; and the founder of Hungarian
0.7
Research Funds (OTKA) # T032666 and # F035247.

0.6 Middle pa rt do wn
[distance: cm/movement]

0.5 Midd le p art up References

Velocity, v

0.4
[1] M. Nemnyi, Z. Srkny, K. Kacz, Moving the cereal bulks by pneu-
0.3 matic system, Hung. Agric. Eng. 5 (1992) 2425.
[2] K. Kacz, Z. Srkny, Intensive mixing the cereal bulks by pneumatic
0.2 method during storage, Hung. Agric. Eng. 6 (1996) 5758.
aluminum
metal
wood+

[3] M. Nemnyi, J. Gyenis, A.J. Kovcs, Intermingling of cohesionless

wood

met al

0.1
0 Proceedings of Third Israeli Conference for Conveying and Han-
0 0 4 6
De ns ity ratio [ /
Fig. 6. Velocities of the selected cylinders with different densities.
One of the reasons of this is that the distance of penetration
of pistons into the bulk was not the same on the sides than
in the middle (Fig. 1).
The densities of cylinders also do not play a role in the
velocities according to our measurements (Fig. 6). It can
particles with special regard to grains in pulsated bottom mixer, in:
8 10 dling of Particulate Solids, Dead Sea, Israel, 29 May1 June 2000,
wo o d ] pp. 8.248.28.
[4] Zs. Ulbert, J. Szpvlgyi, J. Gyenis, Y. Tsuji, Modelling and simulation
of particle mixing in gravity and pneumatic mixing tubes containing
helicoidal mixer element, Recents Progres en Genie des Procedes 11
(1997) 299306.
[5] P.A. Cundall, O.D.L. Strack, A discrete numerical model for granular
assemblies, Geotechnique 29 (1979) 4765.
[6] K. Erdsz, Vibrcis technika az lelmiszeriparban, Mezogazdasgi
Kiad, Budapest, 1984 (Vibration technique in food industry, Agri-
cultural Press, Budapest).
 Chemical Engineering and Processing 44 (2005) 297302

Estimating the homogenization efficiency of mammoth silos

by process dynamics and simulations
Comparing the results of process dynamics with the simulations
D.L. Schott a, , L.A. van Wijk b , W.J. Vlasblom a
a Department of Marine and Transport Technology, Delft University of Technology, Mekelweg 2, 2628 CD Delft, The Netherlands
b ESI Eurosilo BV, P.O. Box 1047, 1440 BA Purmerend, The Netherlands

Received 26 August 2003; received in revised form 10 March 2004; accepted 19 March 2004
Available online 25 June 2004

Abstract

Previous publications of the authors introduced the homogenization theory of mammoth silos and process dynamics as an indicator of
the homogenization efficiency of mammoth silos. Here, it is shown that process dynamics can be used for estimating the homogenization
efficiency mainly for silos with a large H/D, given a negative exponential auto correlation function for modeling of the input properties. For
smaller silos with equal volume process dynamics underestimates the homogenization results, due to the existence of top and bottom cones.
2004 Elsevier B.V. All rights reserved.

Keywords: Homogenization efficiency; Mammoth silos; Silos; Process dynamics

1. Introduction The aim of this paper is to examine the usefulness of pro-

cess dynamics (PD) for obtaining reliable estimations of the
In previous articles on Homogenization in mammoth homogenization efficiency. Therefore, we will first briefly
silos the homogenization theory of these silos was pre- discuss the homogenization theory and PD in, respectively,
sented by Schott et al. [1,2], as well as process dynamics Sections 2 and 3. Then the results of simulations of the ho-
(PD) as a tool to predict the homogenization efficiency by mogenization theory and PD will be compared. This com-
Gerstel and Schott [3,4]. One method to achieve homoge- parison in Section 4 is mainly based on the reduction of the
nization in mammoth silos is presented in Fig. 1. From the standard deviation (y /x ) of the material properties. This
homogenization theory follows that no analytical solution paper concludes with a discussion regarding estimating the
can be obtained due to silo geometry, stacking and reclaim- homogenization efficiency using PD.
ing methods. PD does give an analytical solution if the input
auto correlation is known and can be analytically described,
but this solution is based on an infinite number of layers
and is valid only for the silo part where the reclaiming and 2. Modeling using homogenization theory
stacking angle are constant. Or, in other words, PD does not
allow for the start and bottom cones of the homogenization For the derivation of the homogenization theory is referred
method. Oosterhuis et al. [5] performed experiments with to previous publications of the authors [1,2]. The main result
an inclined screw conveyor on a free surface. It was inves- is the reduction of the standard deviation of the material
tigated whether this screw can be used to obtain large-scale property from the output compared to the input as shown in
homogenization in mammoth silos (Fig. 1). These experi- Eq. (1):
ments were carried out at Delft University of Technology.


 m  bs bs 
 (n 1) v v (i, j)
y 1  s=1 s i=as j=as i,s j,s x
Corresponding author. Tel.: +31 15 278 3306; fax: +31 15 278 2492. = m
E-mail addresses: dingenaschott@hotmail.com (D.L. Schott),
x vslice n
s=1 s 1
avanwijk@eurosilo.com (L.A. van Wijk). (1)

0255-2701/$ see front matter 2004 Elsevier B.V. All rights reserved.
doi:10.1016/j.cep.2004.03.015
298 D.L. Schott et al. / Chemical Engineering and Processing 44 (2005) 297302
cd < 0 >0
Fig. 1. Inclined stacking by controlled dump (left) and reclaiming using
an inclined conveyor (right). The symbols and indicate the stacking
and reclaiming angles.
This equation is valid under the condition that the volumes
of the slices (vslice ) are mutually equal. As can be seen from
Eq. (1) the reduction of the standard deviation is a function
of:

y 4Vci
= (2Vs+r
x 4
3Vs+r
volume distribution, i.e. the number of layers intersected
by the conveyor during reclamation, and the volume of
each layer in this slice (vi,s and vj,s ). This is fixed by the
chosen silo geometry, stacking and reclaiming method,
and number of layers (ns ) and slices (m);
input properties, i.e. the fluctuations of certain physical
or chemical material properties in the input flow modeled
by the auto correlation function x .
This equation cannot be solved analytically due to
the volume distribution. Therefore we have built a sim-
ulation program to calculate the homogenization effi-
ciency (y /x and the improvement of the autocorrelation
function).
3. Modeling using process dynamics (PD)
The homogenization process, existing out of a stacking
and reclaiming process, is linear and due to constant stacking
and reclaiming angles the system is time invariant as well.
Applying s requires a continuous and linear time invariant
system, this first requirement is met by taking infinite thin
slices. How the auto covariance function of the output is
obtained is in detail explained in previous publications of
Gerstel and Schott [3,4], here only the result based on a
first order negative exponential input correlation function
(Eq. (2)) is presented in Eqs. (3)(5).
x () = e||/Vci
where Vci is the characteristic volume of the input properties
(m3 ).The auto covariance function for 0 z Vs+r and
for z Vs+r is, respectively, Eqs. (3) and (4):
4x2 Vci 3
yy (z ) = 4
2Vs+r + z3 + 6z Vci2 3Vs+r
2
z
3Vs+r
+(6Vci3 3Vs+r
2
Vci )ez /Vci
(3Vci3 + 3Vs+r Vci2 )e(Vs+r +z )/Vci
(3Vci3 + 3Vs+r Vci2 )eVs+r +z /Vci (3)
4x2 Vci2
yy (z ) = (2Vci2 Vs+r
2
ez /Vci
Vs+r
4
(Vci Vs+r + Vci2 )e(Vs+r +z )/Vci
+ (Vci Vs+r Vci2 )eVs+r z /Vci ) (4)
The reduction of the standard deviation is obtained by using
z = 0 in Eq. (3):
V /V
3 + 6V 3 3V 2 V 6V 2 (V + V
s+r ci ci ci s+r )e s+r ci ) (5)
ci
4. Results
In order to compare the results from PD with the simula-
tions, the input properties, and silo geometry have to be the
same. As the results of PD in the previous section are based
on a first order negative exponential correlation function,
this function was used as well for the simulations. Three
different input properties were used with different charac-
teristic volumes (Vci ): 40 and 400 m3 . The silo volume was
approximately 40,000 m3 , with a height of 50 m and a diam-
eter of 32 m. This relatively high silo was chosen because
constant angles over the silo height are required (as indi-
cated in Fig. 1) for a thorough comparison with PD. As the
input properties are realizations of a stochastic process 400
repetitions were done per simulation.
4.1. Process dynamics (PD)
A characteristic volume of the input property (Vci ) equal
to 40 m3 means that the signal is very incoherent. This indi-
cates that homogenizing will improve the coherence of the
signal and reduce the amount of fluctuations very well. Fig. 2
shows the improvement of the auto correlation function de-
pending on the stacking and reclaiming angles. The y-axis
represents the auto correlation function as a function of
the lag V on the x-axis. The plot on the left-hand side shows
the results of the homogenization process with a horizontal
stacking method, whereas the other plot shows the results
on the basis of inclined stacking ( = 30 ). The plots repre-
(2) sent the correlation function of the input (solid line) and the
 D.L. Schott et al. / Chemical Engineering and Processing 44 (2005) 297302 299

in = 10 in = 0 = 10
[-] = 0 = 20 = 30 [-] =30 = 20 = 30
1 1

0.8 0.8

0.6 0.6

0.4 0.4

0.2 0.2

0 0
0 1 2 3 4 0 1 2 3 4
v x103 [m3] v x103 [m3]

Fig. 2. PD results: input and output auto correlation functions with Vci = 40 m3 .

in = 10 in = 0 = 10
[-] = 0 = 20 = 30 [-] =30 = 20 = 30
1 1

0.8 0.8

0.6 0.6

0.4 0.4

0.2 0.2

0 0
0 1 2 3 4 0 1 2 3 4
v x103 [m3] v x103 [m3]

Fig. 3. PD results: input and output auto correlation functions with Vci = 400 m3 .

correlation functions of the output (dashed lines). The cor-
relation functions of the output differ since the reclaiming
angle is different as indicated in the legend of the plot.
The improvement of the characteristic volume can be de-
rived from the figure. For example: the characteristic volume
of the output property (Vco ) is approximately 2000 m3 when
the stacking is horizontally ( = 0 ) and the reclaiming an-
gle = 20 . This improves to Vco 4000 m3 when the
stacking angle = 30 and the reclaiming angle = 20 .
A better correlated input signal will leave less space for
improvement compared to the previous. Fig. 3 shows the
improvement of the auto correlation functions for Vci =
400 m3 . It follows that the characteristic volume of the output
property (Vco ) is just slightly higher compared to Vci =
40 m3 : Vco 3400 m3 versus Vco 3200 m3 for = 0
and = 30 .
Results of the reduction of the standard deviation obtained
from PD follow in the next subsection.
The requirement for applying PD is that the system should
be a linear time invariant system. This is the case in the area
where both the reclaiming and the stacking angles are con-
stant. This part is indicated by const in Fig. 5. The top and
bottom cones are indicated by top and bottom respec-
tively. This notation is also used in the following tables.
The stacking method can be implemented in two ways;
using an inclined conveyor (ic) and controlled dump (cd).
The difference is shown in Fig. 4.
These tables show that the PD results match quite
well with the results of the constant part and the total

4.2. Comparison of PD with simulations

Tables 1 and 2 both show the reduction of the standard

deviations relative to the mean value of the input calculated
by the simulations and PD for different stacking () and <0 cd < 0
reclaiming () angles. Vci in the upper left of the tables
indicates the characteristic volume of the input property used Fig. 4. Inclined stacking using an inclined conveyor (left) and inclined
for the calculations and simulations. stacking using controlled dump (right).
300 D.L. Schott et al. / Chemical Engineering and Processing 44 (2005) 297302

Table 1
Results of the reduction of the standard deviation from process dynamics and simulations for Vci = 40 m3 , H = 50 m and D = 32 m.
Vci = 40 m3 PD (y /x ) Average y /x of 400 simulations

Total Const Top Bottom

0 30 0.119 0.166 0.122 0.246 0.299

10 cd 20 0.123 0.138 0.126 0.140 0.175
10 ic 20 0.123 0.144 0.126 0.170 0.190
20 cd 10 0.123 0.133 0.125 0.138 0.143
20 ic 10 0.123 0.144 0.125 0.178 0.166
30 cd 30 0.085 0.089 0.074 0.087 0.097
30 ic 30 0.085 0.105 0.074 0.113 0.114

Table 2
Results of the reduction of the standard deviation from process dynamics and simulations for Vci = 400 m3 , H = 50 m and D = 32 m
Vci = 400 m3 PD (y /x ) Average y /x of 400 simulations

Total Const Top Bottom

0 30 0.364 0.385 0.323 0.506 0.580

10 cd 20 0.375 0.361 0.333 0.358 0.432
10 ic 20 0.375 0.364 0.333 0.385 0.438
20 cd 10 0.375 0.350 0.332 0.346 0.371
20 ic 10 0.375 0.361 0.331 0.402 0.385
30 cd 30 0.263 0.240 0.200 0.228 0.260
30 ic 30 0.263 0.258 0.200 0.268 0.274

homogenization efficiency of the silo. As we have to deal Table 3

with top and bottom cones (Fig. 5) where the homoge- Volume of the different silo parts of a silo with H = 50 m and D = 32 m
nization efficiency will be less (as shown by the last two H = 50 m; D = 32 m Volume in parts of silo (m3 )
columns in Tables 1 and 2), the question is whether PD Total Const Top Bottom
is a good approximation of the whole silo content and not
only for the constant part. Even though the homogenization 0 30 40212 32703 2503 5006
10 20 40212 30523 4118 5572
of the top and bottom part is less, the approximation of 20 10 40212 28100 5733 6379
PD is within 3% of the simulation results. Comparing the 30 30 40212 13566 12112 14535
rows with controlled dump and inclined conveyor shows
preferable results for the controlled dump stacking method
over the inclined conveyor method (Tables 1, 2 and 5). where cd is the method to prefer in terms of homogenization
The slightly underestimation of the homogenization effi- efficiency.
ciency in Table 1 and the slightly overestimation in Table 2 For a smaller silo, with equal storage volume, the volume
can be a result of the realization of the stochastic input prop- in the parts of the silo change, the constant part becomes
erties, although these are based on the analytical input auto less and the top and bottom parts have an increasing influ-
correlation function used for PD. Another possible cause is ence as can be seen in Table 4 and Fig. 5. The bottom row
the number of layers; in practice the layer thickness is 10 cm, shows that due to the stacking and reclaiming angles no part
this value was used in the simulations. But, PD is based on with stationary homogenization efficiency is left. As a re-
infinite thin layers so using layers of 10 cm can make the
difference with the volume distribution of different layers in Table 4
a slice (Section 2). Volume of the different silo parts of a silo with H = 25 m and D = 45 m
Table 3 gives an overview of the volume in the different H = 25 m; D = 45 m Volume in parts of silo (m3 )
parts of the silo. The influence of the top and bottom cones
of the small stacking and reclaiming angles seems to be Total Const Top Bottom
rather small, and there is no difference between controlled 0 30 39761 19162 6866 13733
dump (cd) and inclined conveyor (ic) stacking. So the 10 20 39761 2156 16687 20918
difference between cd and ic can be explained only by the 20 10 39761 1357 18204 20200
30 30 39761 0 16527 23234
different volume distributions in the top and bottom cones,
 D.L. Schott et al. / Chemical Engineering and Processing 44 (2005) 297302 301

CL

top

CL
constant

bottom

Fig. 5. Distinction of areas in the mammoth silo for different H/D.

Table 5
Results of the reduction of the standard deviation from process dynamics and simulations for Vci = 400 m3 , H = 25 m, and D = 45 m
H= 25 m; D = 45 m; Vci = 400 m3 PD (y /x ) y /x simulations
Total Const Top Bottom

0 30 0.364 0.313 0.157 0.353 0.420

10 cd 20 0.375 0.223 0.150 0.195 0.240
10 ic 20 0.375 0.237 0.150 0.219 0.248
20 cd 10 0.375 0.201 0.155 0.192 0.203
20 ic 10 0.375 0.237 0.154 0.242 0.227
30 cd 30 0.263 0.136 0.117 0.145
30 ic 30 0.263 0.182 0.177 0.179
Homogenization of the same amount of material with equal stacking and reclaiming methods in a smaller silo will be better compared to a tall silo
(Fig. 5) because the amount of material between the first and last intersected layer is larger. So, the auto correlation of the output material will be
improved at a higher extent compared to the silo with a large H/D.

sult the estimation by PD is getting worse, but is at least vi,s volume of layer i in slice s (m3 )
an underestimation of the total homogenization results (see vj,s volume of layer j in slice s (m3 )
Table 5). vslice volume of a slice (m3 )
Vci characteristic volume of the input properties
(m3 )
5. Conclusions Vco characteristic volume of the output
properties (m3 )
With material properties modeled by a negative exponen- Vs+r volume of stacking + reclaiming cone (m3 )
tial auto correlation function we have seen that the estima-
tion by process dynamics is quite well for H = 50 m and D Greek letters
= 32 m. For silos with a smaller H/D the estimation is get- yy auto covariance function of the output
ting worse, because the volume where the process dynamics auto correlation function
can be applied is getting smaller and top and bottom cones x auto correlation function of the input x
dominate the results. Compared to tall silos, the homoge- x standard deviation of the input signal x
nization result in small silos with equal storage volume will
, z lag (m3 )
in general be better because the amount of material between
the first and last intersected layers in a slice is larger. Of
course this is valid for equal stacking and reclaiming meth-
ods. References

[1] D.L. Schott, A.W. Gerstel, L.A. van Wijk, Homogenization theory of
mammoth silos, in: Proceedings of the Seventh ICBMH, University
Appendix A. Nomenclature
of Newcastle, NSW, Australia, 2001, pp. 199208.
[2] D.L. Schott, L.A. van Wijk, C. Kraaikamp, Large scale homogeniza-
m number of slices tion in mammoth silos: importance of modeling input properties, Int.
ns number of layers intersected by slice s J. Miner. Process. 71 (14) (2003) 179199.
302 D.L. Schott et al. / Chemical Engineering and Processing 44 (2005) 297302

[3] A.W. Gerstel, D.L. Schott, Homogenisation in mammoth silos, in:
Proceedings of the Seventh ICBMH, University of Newcastle, NSW,
Australia, 2001, pp. 189198.
[4] A.W. Gerstel, D.L. Schott, Upgrading of bulk material in mammoth
silos, Port Technol. Int. 17 (2002) 109116.
[5] E.J. Oosterhuis, D.L. Schott, L.A. van Wijk, Reclaim screw in mam-
moth silos operating on a free surface: comparison between horizontal
and inclined operation on free flowing bulk solids, in: Proceedings of
the Fourth CHoPS, Budapest, Hungary, 2003.
 Chemical Engineering and Processing 44 (2005) 303313

Assessing the homogeneity of powder mixtures

by on-line electrical capacitance
N. Ehrhardt a , M. Montagne a , H. Berthiaux a, , B. Dalloz-Dubrujeaud b , C. Gatumel a
a Laboratoire de Gnie des Procds des Solides Diviss, UMR CNRS 2392, Ecole des Mines dAlbi-Carmaux,
Campus Jarlard, route de Teillet, 81 000 ALBI, France
b Dpartement de Mcanique Energtique, UMR CNRS 6595, Ecole Polytechnique Universitaire de Marseille,

Technopole de Chateau-Gombert, 5, rue Enrico Fermi, 13 453 Marseille, France

Received 27 August 2003; received in revised form 10 March 2004; accepted 19 March 2004
Available online 25 June 2004

Abstract

In this paper, we try to improve our comprehension of one of the most important challenge in the particulate solids industry: the assessment
of the homogeneity of powder mixtures. A two-electrode electrical capacitance method is used to measure the permittivity of mixtures while
flowing in a 1-D set up. An emphasis is placed on the possibility of following homogeneity evolution through permittivity measurement,
without the necessity to calculate the volumetric proportions of each component from an effective medium formulation. The methodology is
applied to examine segregation in a funnel, as well as the efficiency of a laboratory drum mixer.
2004 Elsevier B.V. All rights reserved.

Keywords: Homogeneity; Capacitance; Powder mixing

1. Introduction important mismatch between the refinement attained at the

With the increased variety of formulated products of the
particle form that appears every day on the market and
that are issued for a wide range from industry (pharma-
ceutical, agro-food, cement, plastics, . . . ), particulate pro-
cesses in general, and mixing processes in particular, are
pushed to a never-ending battle against their own perfor-
mances. This is due to the fact that the end-used properties of
the products that are manufactured nowadays are more and
more complex, especially because these must cover more
and more functions. Because product engineering had be-
come a major engineering science producing its own tools
both at the research and the educational levels, a prod-
uct ought to become an intelligent delivery system, a
small process able to optimise its own action when it is
used.
However, it is of common knowledge that mixers, as much
other equipment dealing with particles, are all designed and
operated from the purest empiricism. Indeed, there exists an
Corresponding author. Fax: +33 5 63 49 30 25.
E-mail address: berthiau@enstimac.fr (H. Berthiaux).
formulation step and the development of processes able to
manufacture the designed and desired product. If we take
biodisponibility of a pharmaceutical tablet as an example,
the trust that we have in formulation must be contrasted with
the following traditional chemical engineering interroga-
tion: How to make sure that the process will deliver pills,
all having the same biodisponibility?. In practice, this dif-
ficulty in controlling the product has led the companies and
some regulatory institutions, such as the FDA, to emphasise
on freezing the process by avoiding any change. In many
process industries, the main keyword concerning powder or
granular material mixers is undoubtedly reproducibility,
with the meaning that once an equipment is qualified and
validated to perform a certain function, operating conditions
are fixed and the cost of any variation in the process pa-
rameters may become prohibitive [1]. For example, in the
pharmaceutical field, even a slight change in the mixing pro-
cess would lead to clinical trials for ensuring that the drugs
therapeutic effect has not been affected. In addition, current
sampling procedures are widely contested [24], especially
because of the limited number of samples typically thirty
that can be assumed in the context of production with

0255-2701/$ see front matter 2004 Elsevier B.V. All rights reserved.
doi:10.1016/j.cep.2004.03.016
304 N. Ehrhardt et al. / Chemical Engineering and Processing 44 (2005) 303313
respect to the overall number of possible samples about
one million.
In other words, the technological improvement of the
products does not guarantee the same progress at the man-
ufacturing level at all, mostly because of a lack of instru-
mentation and control of the processes, more deeply linked
to the lack of understanding and reliability of particle char-
acterisation in general. If we except the mineral industry,
the industrial applications of particle control systems are
still scarce, the only well-known example being particle
size distribution measurement through LASER diffraction.
However, the situation is not so dramatic if we look at the
progress made in the field of particulate instrumentation
research in the last decade, with a constant special insight
provided by the international conference series Control of
Particulate Processes [5].
Optical techniques were developed both in situ and on
line over a broad range of applications and various types of
light sources. Main examples are solids concentration mea-
surements in particle fluidisation [6,7], as well as gassolid
flow [8], or sample composition in dry blending operations,
this latter beginning from the pioneering work of Harwood
et al. [9] up to more recent advances in this scientific field
[1012]. The main interest of these techniques lays in the
fact that optical probes are easily available on the market,
so that qualitative monitoring of mixers is possible at an in-
dustrial scale. However, these probes only provide local in-
formation of the mixture (typically 1/40 of a tablet), so that
in essence, they still sample the powder flow in a way that
may be intrusive and not always representative of the overall
stream.
Electrical capacitance tomography (ECT), as well as elec-
trical resistance tomography (ERT) for liquidsolid systems,
experienced a flourishing development these recent years
for a wide variety of applications [1318]. Their major
advantage is that they provide, in a non-intrusive way, a
cross-sectional view of a stream that may contain at the
same time a liquid, a solid and a gas. In that sense, they
are able to give a radial image, which in turn, and through
the help of other multiple electrode sensors in the axial di-
rection, can be reconstructed in three dimensions. However,
this huge amount of information must be contrasted with the
low resolution at which it is given. This has probably limited
the application of this technique to powder mixtures that re-
quire an analysis at a very fine scale. X-ray, -ray, as well
as magnetic resonance imaging (MRI), must also be cited
in the category of the tomographic techniques. They were
mostly used to investigate fluidisation processes, but still
seem difficult to employ at the level of a factory [1921].
LASER-induced fluorescence (LIF) technique must finally
be cited, especially because some pharmaceutical ingredi-
ents can naturally provide a signal when excited by the ad-
equate beam [22].
In this work, we develop an electrical capacitance sensor
that consists of only two electrodes placed around a glass
tube. This set up is considered not in the view of imaging
the media (such as in classical ECT methods), but with the
objective of obtaining a global dielectric information and
follow its variation while powder is flowing into the tube,
with the aim to measure the composition of binary or ternary
mixtures in the axial direction of the powder flow stream.
It is waited that this method will provide, through specific
iterative algorithms, a quantitative and structural information
about a binary mixture flowing out of a mixer (either batch
or continuous) in relation with its own processing.
2. Theoretical and conceptual developments
2.1. Powder blend homogeneity: macro and micro
considerations
Homogeneity is a key concept for many chemical engi-
neering applications, especially those in which a transforma-
tion occur (precipitation, chemical reaction, grinding, . . . ),
because of the coupling between kinetics aspects and con-
tact time of the reagents due to the flow conditions. In
the field of powder mixing, in most cases, no transforma-
tion takes place and the quality of the product is the fact
of the particulate flow dynamics itself. Also, the main dif-
ference with fluid mixing lays in the fact that a perfect mix
cannot be performed at a molecular scale, and is still diffi-
cult to assess at the scale of a particle. This is due to a wide
variety of particleparticle interactions, resulting in a con-
flict between mobility of single particles, which is essential
to achieve a good mix, and resistance to segregation, which
is a fact of the mixtures structure. In that sense, cohesive
particulate media (typically below 50 m) can be viewed as
macrofluids, while free-flowing systems (typically supe-
rior to 500 m) can be viewed as microfluids. Between
these two extreme behaviours that sometimes may occur
in a single product (case of aerable powders) particle and
particle packets trajectories and motion are both influenced
by gravity and surface forces.
As felt above, the concept of powder blend homogene-
ity cannot be dissociated from the scale at which a mixture
is observed. For continuous mixers, homogeneity must be
defined at the level of the blend flowing out of the vessel,
and falling down to a conveyor or any other transportation
equipment. In the case of a batch mixer, the mixture struc-
ture can be particularly complex to define, but in practice
and when the mixer is emptied, the content of the vessel
is discharged through a pipe to a container or a tableting
machine. Hence, in both cases and if we take into account
the entire cross-section of the pipe, the mixture flow can be
assimilated as a 1-D particle stream.
Thus, let us consider a powder mixture flow as a mono-
dimensional layer divided into N consecutive elementary
samples. In a first approach, the size of these can be ar-
bitrary fixed, not too small to prevent from statistical er-
rors (inversely proportional to the square root of the par-
ticle number) and not too large for ensuring the validity
 N. Ehrhardt et al. / Chemical Engineering and Processing 44 (2005) 303313
Fig. 1. Two mixtures (a and b) of identical macromixing state can exhibit different micromixing structures that may result in products of different quality
and effects.
of the following analysis. From a global point of view,
the distribution of the compositions is a first indicator of
the mixtures homogeneity, a perfect blend (at the scale of
scrutiny of the samples size) corresponding to a peak cen-
tred on the mean value. Hence, the following standard de-
viation (Eq. (1)) can be considered as a good macromixing
index (the coefficient of variation can also be employed).


N
i=1 (xi )
2
= (1)
N 
This latter formula, which is commonly used in the in-
dustrial practice, also holds for the case of an estimation of
the mixture quality using random sampling. However, the
main problem in this global analysis lays in the well-known
(since the famous Danckwerts example in chemical reac-
tors [23]) fact that two different structures can correspond to
the same intensity of segregation criteria. Fig. 1 illustrates
this idea, which has a very practical sense in the context of
pharmaceutical blending.
In effect, let us assume that the two processes depicted
here concern pharmaceutical tablets produced in packages
containing six consecutive unit doses. These are of identi-
cal macromixing state, as standard deviations are equal. But
in one case (namely Fig. 1a), a patient will take a slightly
under-dosed treatment, and another one a slightly overdosed
treatment, while in the second possibility (Fig. 1b), an equi-
librium of the dose will be reached after the second take.
Indeed, structure defaults can provoke negative therapeutic
effects that cannot be diagnosed with a macroscopic index
of segregation intensity.
As introduced by Danckwerts in the mid-fifties [24,25],
and more recently reviewed by Gyenis [26], the mixing
structure can be accounted for by the use of the autocorrela-
tion function R(r) (Eq. (2)). The distance r0 , corresponding
to the equation R(r) = 0, at which two sample compositions
can be considered to be uncorrelated, is a main character-
istic of the structure of the mixture and has an important
305
practical consequence: tablets should probably be produced
at this segregation scale to statistically ensure that these
have the right (desired) composition. However, the figure
given here is still global, as we do not know anything about
the location and the distribution of the segregated zones in
the mixer, and also on zones of acceptable defaults. As a
consequence, the size of the tablets that corresponds to the
segregation scale may be not so adequate as it is said to be.

 Nr
(xi )(xi+r )
R(r) =  i=1N (2)
i=1 (xi )
2
An illustrating example for this idea is provided in Fig. 2,
in which a series of consecutive elementary samples con-
taining an active ingredient has been artificially generated.
One may easily find that the mean content is = 5%, while
the global standard deviation is = 1.34%. If we refer to
a pharmaceutical standard, this is not a good mixing, being
the corresponding coefficient of variation equal to 26%. As
it can also be seen from the data, an important deviation
from the mean occurs for sample nos. 11 and 12, which
have the particularity of being corrected at a small scale
(sample nos. 13 and 14). From the point of view of mon-
itoring of the mixing process, it is certainly interesting to
search for this local default that makes the overall mixture
probably inadequate, despite of the fact that other regions
seem to be much better mixed.
If we look at the variations of the corresponding auto-
correlation function (Fig. 3), the characteristic distance r0
covers between 1 and 2 elementary samples, which means
that these must be grouped two by two or three by three (for
more security) to avoid any correlation. Doing so, six or ten
tablets can be defined depending on the way the elementary
ones were grouped together (see also Fig. 2). However, this
procedure did not really change the situation, as it gave rise
to an overdosed (no. 4) and an under-dosed tablet (no. 5) and
four perfectly dosed tablets (nos. l3 and 6). This is due to
306 N. Ehrhardt et al. / Chemical Engineering and Processing 44 (2005) 303313

Fig. 2. Numerical example showing that a non-randomised structural default (heterogeneous zone between sample nos. 1114) cannot be cleared by
increasing sample size either by a factor 2 (top case) or 3 (bottom case).

the fact that this structural default induces segregation at a
relatively short scale, and cannot be adequately detected by
the criteria r0 . The reason for this failure is fundamentally
linked to the global (but structural) character of this crite-
rion that characterises homogeneity at the scale of all the
N samples. In other words, this analysis would give a bet-
ter result if the problematic elementary samples were not
placed coast to coast in the mixture, but dispatched along
the whole data set.
2.2. Methodology for detecting structural defaults
Because of the above-mentioned type of problem, it is ad-
visable to have a better insight on the local structure of the
mixture, with the aim of disregarding samples of inadequate
composition rather than grouping them. For this reason, we
derived another type of analysis, which we claim to be more
sensible to structural defaults and also easier to adapt to
on-line measurement of continuous processes in which the
series of data is, in principle, infinite. Let us first define a
window that consists of n consecutive elementary sam-
ples (of course n < N) placed at the beginning of the data
series. For this packet of samples, a mean 1,n and a stan-
dard deviation 1,n can be defined by the following Eqs. (3)
and (4).
n
1
1,n = xi (3)
n
i=1

Fig. 3. Autocorrelogram corresponding to the present example, and indicating that a characteristic interval of two samples should be considered for
ensuring homogeneity.
 N. Ehrhardt et al. / Chemical Engineering and Processing 44 (2005) 303313 307


n curve is the consequence of a good mixture achieved at a
i=1 (xi 1,n )2
1,n = (4) small scale, while a rough one corresponds to the existence
n of heterogeneous regions in the blend.
Let us further displace this window from one sample to From the graph, it is also possible to calculate the number
the right (of, say, Fig. 2), defining thus another mean 2,n of samples n0 that form part of the default, and should prob-
and another standard deviation 2,n . If the composition of ably be taken away from the production line. In effect, it is
the newly considered sample (number n + 1) is close to x1 , easy to demonstrate Eq. (7). In the present example, as np
then both 2,n and 2,n will be close to 1,n and 1,n , re- = 8 and n = 5, one will verify that the heterogeneous zone
spectively. In the contrary, if a small structural default ex- effectively covers four consecutive elementary samples.
ists, it will be detected by this analysis because the variation n0 = np n + 1 (7)
of the corresponding mean and standard deviation will be
important, even if it concerns segregation at a small scale.
Generalisation of Eqs. (3) and (4) for packets beginning with 3. Experimental set up development and methodologies
sample number k is obviously given by Eqs. (5) and (6).
n+k1 Like in any micromixing state consideration, the experi-
1 
k,n = xi (5) mental illustration of the concept needs the development of
n a sensor. Because even slight structural defaults must be de-
i=k

tected, it can be felt that this technique may cover the fol-
n+k1 lowing properties:
i=k (xi k,n )2
k,n = (6)
n (1) Be adapted on-line: The three dimensional knowledge
of the repartition of the components in a vessel (as in
For the above example, one can observe that the seventh
a batch mixer) is generally of a few interest for the
packet first considers the small segregated region, while it is
practitioner, especially because an important disruption
not touched after the fourteenth if the size of the window is
of the mixture occurs when emptying the mixer.
equal to five samples. This results in a clearly marked peak
(2) Be non-intrusive: This is the better way to avoid any
on a graph representing the variation of these local variances
cutting error due to the insertion of a probe in the
with time, while the means are affected at a much lower
powder flow.
level (see Fig. 4).
(3) Be exhaustive: As all defaults are to be screened, no
The reason for this lays in the fact that even a small
partial sampling procedures that would induce a statis-
deviation in composition has a consequence on the standard
tical bias can be developed. For instance, the derivation
deviation value, so that the local standard deviations are as
of a certain part of the powder flow for analysis cannot
great as many samples in the packet demonstrate a deviation
be considered here.
from the mean. In our case, it can be seen that the number
(4) Have a good spatial and temporal resolution.
np of packets that are affected by the perturbation is equal to
8 out of 16, which also helps in magnifying the default. For On the other hand, as only the global content in the sam-
instance, it is felt that a smoothed local standard deviation ple is to be derived, imaging and tomographic techniques

Fig. 4. Evolution of the local means and standard deviations with the number of packet considered (packet width equal to five elementary samples),
magnifying the existence and the location of the heterogeneous zone.
308 N. Ehrhardt et al. / Chemical Engineering and Processing 44 (2005) 303313

Fig. 5. Capacitance sensor developed.

that also provide spatial distributions are perhaps too pow-
erful tools to be employed with respect to the information
effectively needed. From the above-mentioned techniques, it
can be understood that the best method profile should com-
bine the advantages of fibre-optics (flexibility, global infor-
mation) and ECT (non-intrusive, no sampling) sensors.
3.1. Electrical capacitance set up
Classical ECT set ups are composed by a ring of elec-
trodes (typically 12, see [27,28]), which is generally placed
around the external part of the pipe itself. The correlation
between the dielectric measurements provided by the elec-
trodes allows for a tomographic mapping of the pipe in
the radial direction. In the present case and as commented
above, the radial distribution of the components is of a few
interest (at least in a first approximation) as we are only
concerned with the overall composition in the measuring
volume. For this reason, we developed a capacitance sensor
with the aim of measuring the variations of the global ca-
pacity of a mixture during its own flow. It thus consists of
only two copper electrodes covering practically the whole
section of a glass tube of external diameter 12 mm (see Fig.
5). A PVC screen allows isolating the electrodes from the
aluminium full cylinder (of dimensions 40 mm 30 mm)
that covers the overall measuring cell. This latter also pre-
vents from external disturbances, as well as the coaxial
cables, making the liaison with the capacitance meter.
The sensor is connected to a capacitance meter consisting
of the electronic circuit depicted in Fig. 6, and further to a
computer through an analogic/numeric converter. Measure-
ments are taken at a frequency of 5000 scans per second
and the sensibility of the circuit is high enough (close to
0.1 pF) to detect the capacity variations of some picofarads
provoked by changes in the composition of the measured
volume. Preliminary results also show that capacity and di-
electric permittivity could be linked to each other through a
linear relation over the whole measuring range in which we
work. It must be stated that such a statement is only true if the
measured space between electrodes is fully filled with mate-
rial. Otherwise, the measurements would show very strong
non-linear character.
The calibration of the sensors is performed with a liquid
(namely n-butyl acetate) of known dielectric permittivity,

Fig. 6. Sketch of the electronic circuit for capacitance measurements.

 N. Ehrhardt et al. / Chemical Engineering and Processing 44 (2005) 303313 309

and is repeated every week. Three sensors of exactly the Table 1

same characteristics were constructed and can be linked to
the capacitance meter at the same time. As capacitance is
strongly influenced by temperature and humidity, a cool-
ing air equipment is also placed near the sensors, ensuring
then a temperature comprised between 15 and 16 C. In
the same time, humidity of the laboratory is controlled to
be sure it is rigorously the same for each experimentation.
Each sensor proceeds to measurements on samples whose
volume is approximately 2.3 cm3 and the electrode area is
approximately 0.38 cm3 .
3.2. The effective medium theory
The effective medium theory consists in considering
the real medium, which is quite complex, as a fictitious
model medium (the effective medium) of identical prop-
erties. Bruggeman [29] had proposed a relation linking
the dielectric permittivity of the medium to the volumetric
proportions of each component of the medium, including
the air through the porosity of the powder mixture. This
formula has been rearranged under a symmetrical form by
Landauer (see Eq. (8), where i is the permittivity of pow-
der i at a dense state, m is the permittivity of the mixture
and Pi the volumetric proportion of powder i) and cited by
Guillot [30] as one of the most powerful model.
Nc
 i m The mixture to be analysed is placed inside vertical walls
Pi = 0
i + 2m
i=1
Some characteristics of the products used
Products True density Dense state relative d50 (m)
permittivity
Salt 2.21 5.43 565
Silicon carbide 3.15 44.44 565
permittivity, weighting the powder in order to isolate the
porosity and, thanks to the effective medium equation (see
Eq. (8)), to deduce the dense permittivity of the powder.
This method is less precise than the previous one, but we
proceeded to several measurements and we averaged them,
thus obtaining better results.
In this study, commercial sugar, silicon carbide and pow-
der that enter in the composition of cement (AC) were fi-
nally chosen for their high permittivity difference. Sugar and
SiC were both sieved between 500 m and 630 m in order
to avoid segregation by particle size. Another advantage is
that silicon carbide particles can be separated from sugar in
the blends by water dissolution of the latter, filtration of SiC
and finally drying. Table 1 gives the physical properties of
these powders. Note that true densities were measured by a
classical pycnometric method.
3.4. Experimental rig and mixers
(8)
in a vibrating channel operating at low amplitude to avoid

3.3. Particulate system considered and characterisation

Because the objective of the study is to calculate the pro-

portion of a definite component in a binary or in a ternary
mixture from the data, the particulate products to be cho-
sen may first demonstrate a certain permittivity contrast at
a dense state. It must be stated that in the case of a par-
ticulate material, this does not correspond to the dielectric
constant of a massive block of the material for which elec-
tric contacts between grains do not exist. For instance, the
dense state corresponds to a hypothetical block of material
having the same quality of dielectric contact as a granular
material for which porosity is occupied by vacuum.
The method used in determining the dense dielectric per-
mittivity of a definite material consists in defining a fluid in
which the material cannot dissolve and further to measure
the effective permittivity of various particulate media/fluid
mixtures corresponding to various porosity. The intercep-
tion point of the curves representing the variation of the per-
mittivity of the material with porosity (through the effective
medium equation, see section 3.2) gives the dense state
permittivity.
It is clear that this method does not work in the case of a
particulate medium/fluid mixture that saturates the sensor. Fig. 7. Sketch of the experimental rig showing the intial mixture (1), the
In this way, we developed a second method that consists vibrating channel (2), the upper funnel (3), the sensors (4), the static mixer
in filling the sensor with powder, measuring the blend (5), the lower funnel (6), the belt conveyor (7) and the capacimeter (8).
310 N. Ehrhardt et al. / Chemical Engineering and Processing 44 (2005) 303313

Fig. 8. View of the discharge profiles through a funnel.

particle segregation during transport, as well as ensuring (sugar at the bottom and SiC at the top). This preparation
mass flow of the blend. The particles are then falling at the flowed through the glass tube equipped by the first sensor
centre of a funnel (without accumulation) that is connected and then through a lower funnel equipped by the two other
to the glass tube surrounded by a first capacitive sensor (de- sensors at its outlet. The discharge profiles are represented
noted sensor H). At the end of the tube, a Sulzer static in Fig. 8.
mixer consisting of six mixing elements is placed. These The profile in the upper funnel clearly shows the step of
mainly allow for a radial motion of the particles and few permittivity applied to the system, whereas the profile after
axial mixing occurs in normal flow regime (see [31]). The the lower funnel is clearly spread. This shows us that the
mixture further flows by gravity to another funnel placed in
the upper part of another glass tube that supports the last two
sensors separated by a known distance (see Fig. 7). Assum-
ing mass flow in this last section, the correlation between
the responses of these sensors (denoted M and B) allows for
the calculation of the flow rate and the frequency to consider
avoiding any sample crossing.
For instance, the glass tube must always be full of solids
with a constant flow. To ensure this, the end of each tube
was squeezed and the bottom of these tubes were placed at
a short distance from a conveyor belt operating at constant
speed (5 m min1 ), thus gently deposing the mixture as a
layer onto the belt. This set up also allows a validation of the
method by comparing the sample compositions in the belt
to the compositions given by the closest sensor. With this
special set up, all particle flow rates become quickly equal
from the vibrating channel to the conveyor belt (in fact, the
amplitude of the vibrations can be adjusted for this task).

4. Results and discussion

4.1. Following the discharge profile through a funnel

In this experiment, we studied the effect of a funnel on

the discharge profile of two different and successive layers Fig. 9. Photograph of the Erweka Rotocube mixer.
 N. Ehrhardt et al. / Chemical Engineering and Processing 44 (2005) 303313 311

Fig. 10. View of the initial discharge profiles before and after the static mixer.

SiC is mixed with sugar at the beginning of the experiment
because it flows preferentially in the middle of the funnel,
whereas components situated near the walls are flowing at
the end.
We can also conclude that this experimental device gen-
erates a high segregation by density difference that give rise
to a preferential flow in the centre of the funnel. It thus
greatly perturbs the measures and the mixing process. In or-
der to prevent from segregation, it is therefore particularly
important not to fill the funnels, so that a solution could be
to convey powders directly to the centre of it.
4.2. Kinetics of mixing in a laboratory drum mixer: the
Rotocube
In this section, we apply the above methodology for the
case of the Rotocube mixer (see Fig. 9), and in particular,
we examine the influence of the number of revolutions at a
fixed rotational speed (30 rpm) on the mixture quality.
The first experiment was realised without any action of the
Rotocube and we just divided the initial preparation into
three successive layers placed in a vibrating conveyor: the
lower one was constituted by half of sugar, the middle one

Fig. 11. Standard deviation of dielectric permittivity before and after the static mixer versus the number of revolutions in the Rotocube .
312 N. Ehrhardt et al. / Chemical Engineering and Processing 44 (2005) 303313
by all the SiC and the upper one by the rest of sugar. All the
preparations constituted of 20% by mass of SiC and 80% by
mass of sugar. Fig. 10 clearly shows that this experimental
device prevents from segregation inference, even if there is
still left some, probably due to the conveyors vibrations. It
must be commented that the nominal value in SiC content is
not reached because of the small thickness of the SiC layer
and of a slight phenomenon of mixing between SiC and
sugar when passing from the first layer to the second one.
Experiments were then realised with several number of
revolutions of the mixer. The entire mixture flowed through
the device and the standard deviation of the whole mixture
(3600 samples) could be derived for each experiment. Fig. 11
shows the evolution of the standard deviation of the permit-
tivity versus the number of revolutions. We can immediately
see that the two plots are quite close, which shows that the
static mixer mixes in a radial way, an effect that cannot be
detected by the sensors we used. Concerning the shape of the
plots, it is clearly shown that the mixing quality is improved
by the Rotocube since the first revolution. Then, quality is
slowly bettered up to four revolutions and begins slowly to
decrease after four revolutions. This can be due to a change
in the mixing mechanism in the Rotocube (passage from a
shearing mechanism to a diffusion one), which may change
the mixing structure. Note that the presence of oscillations,
which are softened by the static mixer, may be due to seg-
regation in the Rotocube , but also in the vibrating system.
5. Conclusions
Hopefully, dielectric constants of current materials are
covering a wide range of values, which also means that such
a method can be applied in various industrial applications.
This study has shown the possibility to measure segrega-
tion problems when discharging mixtures through a funnel.
It has also shown the effect of the number of drum revolu-
tions at a fixed speed on the quality of a mixture, as well as
the effect of a static mixer. The axial structure of the mix-
tures through autocorrelation functions could also be stud-
ied from these data, but this has not been reported here for
clarity of the paper. In addition, it must be remembered that
we just studied the evolution of the dielectric permittivity,
and that in most cases, we will have to follow the volumet-
ric compositions of each component in order to characterise
the homogeneity of the medium. The capacitive method is
indeed full of promise for particulate systems, and it would
be interesting to explore it much into details, particularly
for determining the proportions of each component of the
mixture.
Appendix A. Nomenclature
d50 median diameter at 50%
n consecutive elementary samples defining a window
(of course n < N)
n0 number of samples that form part of the default
and should probably be taken away from the
production line
np number of packets that are affected by the
perturbation
N number of considered consecutive elementary
samples into which the powder mixture flow
is divided
Nc number of components in the powder mixture,
including the air trough the porosity of the medium
Pi volumetric proportion of the component i in the
powder mixture
r spatial (or temporal) variable corresponding to
the distance (or time elapsed) between two samples
r0 distance (or time elapsed) between two samples
corresponding to the equation R(r) = 0
R(r) autocorrelation function
xi weight composition in a key component in the
elementary sample i
Symbols
mean observed for the N consecutive samples
standard deviation for the N consecutive samples
k,n mean for a packet of n samples beginning at
sample number k
k,n standard deviation for a packet of n samples
beginning at sample k
i dense dielectric permittivity of the component
i in the powder mixture
m dielectric permittivity of the powder mixture
References
[1] F.J. Muzzio, T. Shinbrot, B.J. Glasser, Powder technology in the
pharmaceutical industry: the need to catch up fast, Powder Technol.
124 (2002) 17.
[2] J. Berman, A. Schoeneman, J.T. Shelton, Unit dose sampling: a tale
of two thieves, Drug Dev. Ind. Pharm. 22 (11) (1996) 11211132.
[3] F.J. Muzzio, P. Robinson, C. Wightman, D. Brone, Sampling practices
in powder blending, Int. J. Pharm. 155 (1997).
[4] S. Massol-Chaudeur, Caractrisation de ltat de mlange de poudres
cas de mlanges faiblement doss, Ph.D. Thesis, INPL Nancy,
2000.
[5] J. Lister, Forward to special edition of powder technology from
Control of Particulate Processes VI, Powder Technol. 124 (2002)
175.
[6] T. Wang, Z.J. Lin, C.M. Zhu, D.C. Liu, S.X. Saxena, Particle velocity
measurements in a circulating fluidized bed, AIChE J. 39 (8) (1993)
14061410.
[7] J. Aguillon, K. Shakourzadeh, P. Guigon, A new method for local
solid concentration measurements in circulating fluidized bed, Pow-
der Technol. 86 (1996) 251255.
[8] J.J. Nieuwland, R. Meijer, J.A.M. Kuipers, W.P.M. van Swaaij,
Measurements of solids concentration and axial solids velocity in
gassolid two phase flows, Powder Technol. 87 (1996) 127139.
[9] C.F. Harwood, R. Davies, M. Jackson, E. Freeman, An optic probe
for measuring the mixture composition of powders, Powder Technol.
5 (1971/1972) 7780.
 N. Ehrhardt et al. / Chemical Engineering and Processing 44 (2005) 303313
[10] R. Weinektter, L. Reh, Characterization of particulate mixtures by
in-line measurements, Part. Part. Syst. Charact. 11 (1994) 284290.
[11] D. Steinmetz, B. Hocine, M. Poux, C. Lagurie, Analyse du mlange
de poudres en fluidisation gazeuse laide de la mthode de dflu-
idisation et par capteur fibre optique, Powder Technol. 89 (1996)
139147.
[12] O. Berntsson, L.G. Danielsson, B. Lagerholm, S. Folestad, Quantita-
tive in-line monitoring of powder blending by near infrared reflection
spectroscopy, Powder Technol. 123 (2002) 185193.
[13] T. Dyakowski, L.F.C. Jeanmeure, A.J. Jaworski, Applications of
electrical tomography for gassolids and liquidssolids flows a
review, Powder Technol. 112 (2000) 174192.
[14] K.L. Ostrowski, S.P. Luke, M.A. Benett, R.A. Williams, Application
of capacitance electrical tomography for on-line and off-line analysis
of flow pattern in horizontal pipeline of pneumaic conveyer, Chem.
Eng. J. 77 (2000) 4350.
[15] E. Brunazzi, A. Paglianti, S. Pintus, A capacitance probe and a new
model to identify and predict the capacity of columns equipped with
structured packings, Ind. Eng. Chem. Res. 40 (2001) 12051212.
[16] W. Warsito, L.S. Fan, Measurement of real-time flow structures in
gasliquid and gasliquidsolid flow systems using electrical capac-
itance tomography (ECT), Chem. Eng. Sci. 56 (2001) 64556462.
[17] T. Loser, M. Geweke, D. Mewes, Measuring the Phase Distribution
in a Transient GasSolid Flow Using Capacitance Tomography. 3rd
Israeli Conference for Conveying and Handling of Particulate Solids,
Dead Sea, Israel, 2000, pp. 15.1115.15.
[18] T. Dyakowski, A.J. Jaworski, Application of tomographic tech-
niques for imaging fluidized beds, 3rd Israeli Conference for Con-
veying and Handling of Particulate Solids, Dead Sea, Israel, 2000,
pp. 15.115.10.
[19] A. Kantzas, Computation of holdups in fluidized and trickle beds by
computer-assisted tomography, AIChE J. 40 (7) (1994) 12541261.
313
[20] D. Newton, M. Fiorentino, G.B. Smith, The application of X-ray
imaging to the developments of fluidized bed processes, Powder
Technol. 120 (2001) 7075.
[21] N. Sommier, P. Porion, P. Evesque, B. Leclerc, P. Tchoreloff,
G. Couarraze, Magnetic resonance imaging investigation of the
mixing-segregation process in a pharmaceutical blender, Int. J. Pharm.
222 (2001) 243258.
[22] C.K. Lai, D. Holt, J.C. Leung, C.L. Cooney, G.K. Raju, P. Hansen,
Real-time noninvasive monitoring of dry powder blend homogeneity,
AIChE J. 47 (11) (2001) 26182622.
[23] J. Villermaux, Gnie de la raction chimique Conception
et fonctionnement des racteurs, Tec&Doc, Lavoisier ed, 1984,
p. 204.
[24] P.V. Danckwerts, Continuous flow systems: distribution of residence
times, Chem. Eng. Sci. 2 (1953) 111.
[25] P.V. Danckwerts, Theory of mixtures and mixing, Res. (London) 6
(1953) 355361.
[26] J. Gyenis, Assessment of mixing mechanisms on the basis of con-
centration patterns, Chem. Eng. Process. 38 (1999) 655674.
[27] R.C. Waterfall, W.W. Loh, Arko, J. Covy, G.P. Lucas, Electrical
tomography for multiphases flow applications, 3rd Israli Conference
for Conveying and Handling of Particulate Solids, Dead Sea, Israel,
2000, pp. 1725.
[28] UMIST, Industrial Process Tomography at UMIST, http://www.
tomography.umist.ac.uk.
[29] D.A.G. Bruggeman, Ann. Phys. (Leipzig) 24 (1935) 636.
[30] T. Guillot, Contribution la modlisation des proprits lectromag-
ntiques des mlanges alatoires du type dilectrique/conducteur,
ONERA, Note technique 3, 1992.
[31] S. Muerza, H. Berthiaux, S. Massol-Chaudeur, G. Thomas, A dy-
namic study of static mixing using on-line image analysis, Powder
Technol. 128 (2002) 195204.
 Chemical Engineering and Processing 44 (2005) 315321

Solid transport in a pyrolysis pilot-scale rotary kiln:

preliminary resultsstationary and dynamic results
N. Descoins a , J.-L. Dirion a, , T. Howes b
a UMR CNRS 2392 Laboratoire de Gnie des Procds des Solides Diviss, Centre nergtiqueEnvironnement,
Ecole des Mines dAlbi-Carmaux, Campus Jarlard, 81013 Albi CT Cedex 09, France
b School of Engineering, The University of Queensland, St. Lucia 4072, Australia

Received 14 July 2003; received in revised form 18 December 2003; accepted 11 February 2004
Available online 26 June 2004

Abstract

Experiments for the investigation of the flow of granular solids in a pyrolysis pilot-scale rotary kiln are presented. These experiments
consisted first in measuring the volumetric filling ratio (steady-state experiences) for several operating conditions and second in recording the
exit flow rates after a positive or negative step in one of the operating parameters (dynamic experiences). A dynamical model computing the
evolution of the flow rate of granular solids through the kiln has been developed based on Saeman model [Chem. Eng. Prog. 47 (1951) 508].
The simulations are compared with experimental results; the model gives good results for the rolling mode, but for the slipping mode too.
2004 Elsevier B.V. All rights reserved.

Keywords: Rotary kiln; Dynamic; Solid flow-rate; Pyrolysis

1. Introduction a thermal part,

The processes of pyrolysis provide a solution for the
thermal treatment of solid wastes and are a complemen-
tary and often advantageous technology to incineration and
gasification. For example, in comparison with conventional
waste incineration, pyrolysis allows waste treatment with
extremely low dioxin and furan emissions, substitution of
fossil fuels by process gas and pyrolysis coke, and a wide
range of waste materials can be processed. Rotary kilns are
well suited as pyrolysis reactors because of their good mix-
ing capability, ability to use different materials and efficient
heat transfer. In order to develop design tools for industrial
scale pyrolysis rotary kilns adapted to various wastes or
mixed waste, it is necessary to correctly understand and de-
scribe the physicalchemical phenomena occurring inside
the pyrolysis vessel.
Our goal is to develop a global dynamic model of pyrol-
ysis rotary kilns. This global model can be viewed as three
parts with coupling terms:
a solid phase motion part,
Corresponding author. Tel.: +33 5 63 49 31 52;
fax: +33 5 63 49 32 43.
E-mail address: dirion@enstimac.fr (J.-L. Dirion).
a chemistry part.
The thermal part concerns the modelling of the different
thermal fluxes between the solid bed, the gas phase and the
kiln wall in order to determine the temperature profiles along
the kiln. The three thermal transfer modes (conductive trans-
fer, convective transfer, radiative transfer) generally need to
be taken into account because of the experimental operating
conditions, i.e. high temperatures inside the kiln, long con-
tact times between the solid bed and the wall, gas flow in
the kiln, etc. The chemistry part must be able to represent
the kinetics and the behaviour of the solid bed. For pyrol-
ysis processes, the kinetics are required for solid degrada-
tion reactions and for secondary gasgas reactions too. The
behaviour of the solid bed and the transfer of the gaseous
species inside the porous media must be well described dur-
ing the thermal degradation. Finally, the solid phase motion
part must be able to describe the solid bed profile and the
solid flow rate across the length of the kiln and at the exit
of the kiln. This paper is only concerned with the last point.
Many theoretical and experimental studies carried out in
the past deal with the study of the solid motion inside a
rotary cylinder, drum or kiln [18]. Saemans model, de-
veloped in 1951, is undoubtedly the first attempt for mod-
elling such motion [1]: this mechanistic model is widely

0255-2701/$ see front matter 2004 Elsevier B.V. All rights reserved.
doi:10.1016/j.cep.2004.02.025
316 N. Descoins et al. / Chemical Engineering and Processing 44 (2005) 315321
used nowadays because of its good predictive ability and its
simplicity. It is based on a geometrical analysis of the move-
ment of one solid particle in the bed; the model assumes the
bed is in rolling motion. Some authors have developed para-
metric models based on the determination of experimental
residence time distributions [24]. More recent attempts to
improve the description of the solid phase motion exist but
the developed models tend to be rather complex to use [5,6].
The majority of the studies conducted are concerned with
steady-state studies [16] and there is little published work
dealing with the time dependent axial flow of a granular
solid through an inclined rotating cylinder [7,8].
This paper presents the first experimental results for the
solid motion inside a pilot-scale rotary kiln. Such data are
useful to enable efficient pyrolysis reactions inside a rotary
kiln to be carried out, through the prediction of residence
time and material hold-up, and the evaluation of different
surfaces and temperature profiles according to the operating
conditions. In the first part, the pilot-scale rotary kiln and the
principle of the experiments will be described. An original
dynamic solid motion model will be presented in the second
part, this dynamic model is derived from the original static
model of Seaman [1]. The static and dynamic experimental
results are finally compared with the simulated results.
2. Experimental
2.1. Pilot-scale rotary kiln
The pilot-scale pyrolysis reactor is an externally heated
rotary kiln. The kiln is an nickelchrome hollow cylinder 4 m
long with 0.21 m internal diameter. The cylinder is heated
by five independent electric elements allowing a temperature
profile along the kiln to be prescribed. The pilot operates
under a positive working pressure of nitrogen (30 mbar) and
with a maximum temperature of 1000 C. Nitrogen is used as
a carrier gas for the vapours produced by pyrolysis reactions.
Nitrogen and solids can flow in counter-current or co-current
way. Rotational speed is adjustable between 0.5 and 8 rpm
(8.3 103 and 0.13 tr s1 )and the cylinder can be tilted at
angles of up to 7 . The continuous feeding system consists
of a 30 dm3 hopper with a vibrating cylindrical conveyor;
solid feed rates are adjusted by a potentiometer between 10
and 80 kg h1 (2.8 103 and 2.2 102 kg s1 ).
The inside wall of the rotating cylinder is smooth, so a
metal grid (netting with 2 cm mesh) has been rolled inside
the cylinder in order to artificially increase the roughness of
the wall for several experiments presented later.
2.2. Materials
The experimental results concern trials at room tempera-
ture with crushed rice. The physical properties of the crushed
rice are given in Fig. 1. These values will be necessary for
the comparison between experimental measurements and the
Bulk density Diameter Dynamic angle Wall friction
(kg/m3) (mm) of repose (deg.) angle (deg.)
(with the grid) (without the grid)
873 1% [1 5] 33 2 21 2
Fig. 1. Physical properties of crushed rice.
model presented in the next section. The angle of repose is
directly measured inside the kiln and reflects the bulk char-
acteristics of solid bed in rolling mode; the rolling mode
has been obtained only for the experiments with the metal
grid rolled inside the rotating kiln. The wall friction angle
will be used for the slipping regime (i.e. cases without the
grid). The wall friction angle is measured according to this
procedure: the rice is laid on a metal plate (with the same
roughness of the wall kiln) and the wall friction angle is
determined as the solids begin to fall along the tilted plate.
2.3. Experimental methods
Measurements of volumetric filling ratio were carried out
for several operating conditions in stationary regime accord-
ing to the procedure explained in the following. The station-
ary regime was identified when the exit flow rate became
constant. An electronic balance allowed to measure the flow
rate; the acquisition period was fixed in order to have an ab-
solute incertitude for the solid flow rate less than 2 kg h1
(5.6 104 kg s1 ).
The experimental method was the following one:
the rotation of kiln, the vibrating conveyor and the elec-
tronic balance were switched on,
as soon as the exit solid flow rate was constant, the rotation
of kiln and the vibrating conveyor were switched off,
the balance was replaced by an empty container and the
rotation was restarted,
when the kiln was empty then the total mass of solid in
the container was measured,
the volumetric filling ratio was deducted from this mass
(with an incertitude about 1%).
Experiments were performed to measure the transient re-
sponse of the exit flow rate after a positive or negative step
change in the operating parameters (only experiments with
a step change in the rotation speed are presented here). The
step change was applied after the stationary regime was iden-
tified.
3. Modelling and experimental results
3.1. Dynamic model
The dynamic model of solid transport through the ro-
tary kiln is based on the geometrical model derived by
 N. Descoins et al. / Chemical Engineering and Processing 44 (2005) 315321 317

Fig. 2. Schemes of the rotary kiln with the main variables used in the equations.

Saeman [1]: Ss corresponds to the cross section of solid and can be ex-

 pressed in function of the local bed height. The Eq. (1) can be
4 tan() dH(z) inserted in the above equation and finally a partial differen-
Qv (z) = nR3 + cot()
3 sin() dz tial equation can be derived for the computation of the transi-

 3/2 tory profiles of bed (more details are given in Appendix A):
2H(z) H(z)2
(1)
R R2 1/2 H(z, t) UT tan() 1/2 H(z, t)
FH F (1 FH )1/2
t sin() H z
This equation gives the local volumetric solid flow rate (Qv )

UT 3/2 H(z, t)
according to the kiln radius (R), the different operating vari- = R cot FH with FH
ables (, n), the angle of repose of materials () and the local z 3 z
depth of the solid bed (H). The spatial variable (z) is defined 2H(z, t) H(z, t)2
from the kiln end; for example, H(z = 0) corresponds to the = and UT = 2nR (4)
R R2
depth of the solid bed at the exit of the kiln. Two views of Two boundary conditions are necessary. The first condition is
the kiln are presented in Fig. 2 with the main variables used always for the constant exit height (at z = 0) and the second
in the equations. one defines the continuity of the flow between the vibrating
The equation is obtained by a geometrical analysis of feeder and the kiln at the inlet of the kiln (at z = L):
the path that one particle follows in travelling through the
rotary kiln. For this analysis, the rolling regime is assumed, H(z = 0, t) = H S

i.e. particles roll down continually on the bed surface (the H(t) 
active layer), then they enter the passive region where they z z=L
are carried upward by the rotating wall until they enter the
3/2
active layer again. The angle of repose is assumed to be 3QEv tan() 2H(z = L, t) H(z = L, t)2
=
constant all along the kiln. 2UT R2 R R2
In the stationary regime and for a non-reactive solid, the tan()
volumetric solid flow rate is constant and equal to the inlet
cos()
flow rate given by the vibrating feeder (QEv ). From Eq. (1),
the following ordinary differential equation is directly ob- (5)
tained and can be solved with a boundary condition (exit The initial profile is computed by the Saeman model
height fixed) in order to compute the bed profile along the (Eq. (1)). The inlet flow rate and/or the operating parame-
kiln: ters can be modified during the resolution of the dynamic

3/2 equation. Spurling et al. [7] give a similar equation, but the
dH(z) 3QEv tan() 2H(z) H(z)2 tan() computed variable is not the local height of the bed, but the
=
dz 4nR 3 R R 2 cos() maximum half angle subtended by the bed at the cylinder
(2) axis.
The dynamic model can simulate the filling phase of
The mass conservation equation is written for the transitory the kiln, provided there are some physical considerations.
regime and for a non reactive solid: In fact, the volumetric flow rate cannot be set to 0 in the
Ss (z, t) Qv (z, t) Saemans equation because a mathematical singularity oc-
= (3) curs. So the flow rate is set to a very low value, close to 0,
t z
318 N. Descoins et al. / Chemical Engineering and Processing 44 (2005) 315321
Fig. 3. Dynamic simulation: transient response for the filling phase of the
kiln.
and anyway the mass flow rate will be less than one particle
per second, which is the physical limit for granular matter.
The Fig. 3 shows an example of a filling phase simulation.
The blue line represents the initial steady state profile, i.e.
the kiln is empty. The dynamic simulation starts when
the feeder volumetric flow rate is increased: the bed depth
profile evolution can be seen for different times (200, 500,
1000, 1500 , 2500, 4000 and 8000 s). The bed depth pro-
file converges on a new steady-state profile at 8000 s. This
last profile can be also simulated with the Saemans equa-
tion, the final value of flow rate being used. The dynamic
model can also simulate one or several transitions occurred
by a change in an operating parameter. For example, Fig. 4
presents the dynamic evolution of the bed depth profile after
a change in the rotational speed. The initial bed depth pro-
file is represented by the blue line. The dynamic simulation
starts when the rotational speed is increased by a factor of
two: the bed depth profile evolution can be seen for differ-
Fig. 4. Dynamic simulation: transient response of the exit flow rate to a
step change in the rotational speed.
ent times (1600, 3200, 48,000, 6400 and 8000 s). The bed
depth profile computed by the dynamic model converges on
the profile computed by the steady state model with the final
value of rotational speed.
3.2. Steady state results
The volumetric filling ratio, X, can be evaluated from
Eq. (2) in the steady state: the angle subtended by the bed
at the cylinder axis, (z), is computed from the local height
H(z), and the volumetric filling ratio is obtained by numer-
ical integration:


1 H(z)
(z) = 2 cos 1
R
 z=L
1 Vsolid (z)
X= dz
L z=0 Vcyl (z)
 (6)
1 z=L R2 ((z) sin((z)))/2
= dz
L z=0 R2
 z=L
1
= ((z) sin((z)))dz
2L z=0
The experimental and theoretical volumetric filling ratios are
given in Fig. 5 for 12 experiments.
The bed was in slipping mode for all experiments with-
out the grid inside the kiln because of the smoothness of
the inside wall. So, the angle of repose of the bed mate-
rial have been replaced by the wall friction angle in the
Eq. (1). However in spite of this modification, the predic-
tive ability of the Seaman equation is good: the predicted
volumetric filling ratios are very close to the experimental
ones (results for cases 18). When the grid is rolled in-
side the kiln, the rolling regime was observed and the com-
parison between experimental and simulation results stays
good.
It has been noted that the filling ratios obtained with the
grid are generally larger than without; for example, the vol-
umetric filling ratio values can be compared between the
cases 2 and 10 because the experimental conditions are
close.
3.3. Dynamic results: filling phase of the kiln
In course of the twelve previous experiments, the exit
flow rate has been recorded during the filling of the kiln
until the steady state regime was achieved. The transient
volumetric flow rate along the kiln, and more particularly
the exit flow rate, can be evaluated from the bed height
profiles obtained by the integration of Eq. (4). In Figs. 6
and 7, the experimental exit flow rates without (Fig. 6) and
with the grid (Fig. 7) for several different rotation speeds
are compared with the theoretical ones. The inclination of
kiln is fixed at 1 .
The values of the delay, i.e. the necessary time for the
first solid particles flow out of the kiln are always well eval-
 N. Descoins et al. / Chemical Engineering and Processing 44 (2005) 315321 319

Case Inlet flow Rotation kiln Grid X exp X theo angle

rate speed inclination % % (used in Eq. 1)

-1 -1
kg.h Rev.min deg. deg.

1 15 1.7 1 No 11 13 21

2 23 2.9 1 No 11 12 21

3 26 4.0 1 No 10 10 21

4 35 5.2 1 No 11 10 21

5 21 1.7 2 No 11 12 21

6 34 2.9 2 No 11 11 21

7 47 4.0 2 No 10 10 21

8 62 5.2 2 No 10 11 21

9 9 1.7 1 Yes 12 12 33

10 20 2.9 1 Yes 15 16 33

11 41 4.0 1 Yes 24 23 33

12 51 5.2 1 Yes 24 24 33

Fig. 5. Experimental and theoretical volumetric filling ratio.

uated. The model gives rather good results for the simula-
tion of the subsequent transient phase; however, a slightly
tendency to overestimates the length of the transition phase
for the cases without grid can be observed. These overesti-
mations can be perhaps explained from some visual obser-
vations: for the slipping regime (cases without grid), some
small oscillations of the angle of repose have been observed
in course of the experiments. Mellmann called this type of
motion surging, which is characterised by periodic alter-
nations between adhesive and kinetic friction of the bed on
the wall [9]. An assumption of the model is to have a con-
stant angle of repose.
3.4. Dynamic results: transient response to a step in
operating parameters
In order to evaluate the quality of the developed model,
new experiments have been carried out. These experiments
consisted in fixing operating parameters (inclination angle,
rotation speed, inlet flow rate), waiting for the steady state

Fig. 6. Experimental (markers) and simulated (lines) exit flow rates for a Fig. 7. Experimental (markers) and simulated (lines) exit flow rates for a
filling phase. Experiments without grid inside the kiln. filling phase. Experiments with a grid inside the kiln.
320 N. Descoins et al. / Chemical Engineering and Processing 44 (2005) 315321

oretical curves can be observed at the start of the positive

steps.

4. Conclusion

The developed model for the simulation of the dynamic

flow rate in a rotary kiln has been validated for varied exper-
imental conditions: filling phases, transient responses after
a step in the rotation speed or in the feed rate (not presented
here). It is shown that this dynamic model, based on the Sae-
man equation developed for rolling mode, works with the
slipping mode too: the replacement of the angle of repose
by the wall friction angle is sufficient.
From an industrial point of view, this model is able to
predict and study the influence of the operating parameters
Fig. 8. Transient responses of the exit flow rate for a positive and negative on the filling ratios or the residence times for example. The
step in the rotation speed: experimental (markers) and simulated (lines)
flow rates. Experiments without grid.
model can be used for the design (steady state) and the
control (transient phase) of rotating kilns.
The future steps of the project will be to adapt the model
regime and applying a step (positive or negative) in one of to the flow of reacting granular solids and to incorporate
the operating parameters. The response of the exit flow rate this model in a global model with thermal and chemical
have been recorded until the new steady state was achieved. parts.
In this paper, the experimental and theoretical flow rates
are compared only in the case of a positive and negative
step in the rotation speed of the kiln (Figs. 8 and 9). These Appendix A. Details for the development of the
figures present respectively results without and with the dynamic equation
grid.
In all presented cases, a large step is initially observed. For one reason of simplicity, the indexes relative to the
The exit flow rate is instantaneously modified from its ini- time (t) or to the axis of the kiln (z) have been omitted.
tial value (the initial value of the flow rate is equal to the The mass conservation equation is written for the transi-
final value) as soon as the step is applied. Next, the flow tory regime and for a non reactive solid:
rate returns slowly to its nominal value, i.e. value of the
Ss Qv (Ss Us ) Ss Us
inlet flow rate. The magnitude and length of the transient = = = Ss + Us (A.1)
phase are very well evaluated by the model in the two fig- t z z z z
ures. However, a difference between experimental and the- where Ss and Us are, respectively, the cross-sectional area
of bed and the mean axial velocity:
R2
Ss = ( sin()) (A.2)
2
 
Qv 8 tan() R  
Us = = nR + cot() sin
Ss 3 sin() 2 2 z
sin3 (/2)
(A.3)
sin()
The mean axial velocity is directly given by the Saemans
equation.
If all the terms of Eq. (A.1) are replaced by the Eqs. (A.2)
and (A.3) and by their derivatives respective to the time and
to the space, the new equation is:

  
4
(1 cos()) = nR2 cot() sin4
t z 3 2 z
Fig. 9. Transient responses of the exit flow rate for a positive and negative
2nR tan()  
step in the rotation speed: experimental (markers) and simulated (lines) + sin sin() (A.4)
flow rates. Experiments with a grid. sin() 2 z
 N. Descoins et al. / Chemical Engineering and Processing 44 (2005) 315321 321

The relations between (angle subtended by the bed at Appendix B. Nomenclature

the cylinder axis) and h (height of bed) are:
   H bed height (m)
H = R 1 cos (A.5) L length of kiln (m)
2
n rotational speed, rpm (tr s-1 )
 
2H 1/2
H2 Qv volumetric flow rate (m3 s1 )
sin = 2 (A.6)
2 R R R kiln radius, m
 
 Ss cross-sectional area of bed (m2 )
2 2H H2 t time (s)
cos() = 1 2 sin =12 2 (A.7)
2 R R Us mean axial velocity (m s1 )
  Vcyl volume of kiln (m3 )
2H H2
cos2 = 1 sin2 =1 + 2 (A.8) Vsolid volume of solid bed (m3 )
2 2 R R Xexp experimental volumetric filling ratio
Xtheo theoretical volumetric filling ratio
 
1/2
2H H2 z axial distance from the cylinder discharge
sin() = 2 sin cos =2 2 end (m)
2 2 R R

1/2
2H H 2
Greek letters
1 + 2 (A.9)
R R angle subtended by the bed at the cylinder axis,
2 H angle of repose of the granular solid,
= (A.10) angle between the kiln and the horizontal,
z R sin(/2) z
2 H
= (A.11) References
t R sin(/2) t
With the Eqs. (A.5)(A.11), the partial derivative equation [1] W. C Saeman, Passage of solids trough rotary kilns: factors affecting
with Eq. (A.4) gives partial derivative equation with the time of passage, Chem. Eng. Prog. 47 (1951) 508514.
variable H: [2] J. Mu, D.D. Perlmutter, The mixing of granular solids in a rotary
cylinder, A.I.Ch.E. J. 26 (1980) 928934.

1/2
2H H2 H [3] P.S.T. Sai, G.D. Surender, A.D. Damodaran, V. Suresh, Z.G. Philip,
2 K. Sankaran, Residence time distribution and material flow studies in
R R t

3/2
a rotary kiln, Metall. Trans. B 21B (1990) 10051011.
2 2H H2 H [4] S. Das gupta, S.K. Bhatia, D.V. Khakhar, Axial transport of granular
= nR cot()
2
2 solids in horizontal rotating cylinders. Part 1. Theory, Powder Tech.
z 3 R R z 67 (1991) 145151.

1/2 [5] J. Perron, R.T. Bui, Rotary cylinders: solid transport prediction by
2nR tan () 2H H2 dimensional and rheological analysis, Can. J. Chem. Eng. 68 (1990)
+ 2
sin () R R 6168.
1/2

[6] D.V. Khakhar, J.J. Mc Carthy, T. Shinbrot, J. Ottino, Transverse flow

2H H2 H and mixing of granular materials in a rotating cylinder, Phys. Fluids
1 + 2 (A.12) 9 (1997) 3143.
R R z [7] R.J. Spurling, J.F. Davidson, D.M. Scott, The transient response of
granular flows in an inclined rotating cylinder, Trans. I. Chem. E. 79
The final Eq. (A.13) can be written: (2001) 5161.
[8] P.S.T. Sai, A.D. Damodaran, G.D. Surender, Prediction of axial ve-
1/2 H UT tan() 1/2 H
FH F (1 FH )1/2 locity profiles and solids hold-up in a rotary kiln, Can. J. Chem. Eng.
t sin() H z 70 (1992) 438443.

 [9] J. Mellmann, The transverse motion of solids in rotating
UT 3/2 H
= R cot()FH with FH cylindersforms of motion and transition behavior, Powder Technol.
z 3 z 118 (2001) 251270.
2H H2
= 2 and UT = 2nR (A.13)
R R
 Chemical Engineering and Processing 44 (2005) 323326

Controlling dust emissions and explosion hazards in

powder handling plants
Peter Wypych, Dave Cook , Paul Cooper
Centre for Bulk Solids and Particulate Technologies, Faculty of Engineering, University of Wollongong, Wollongong NSW 2522, Australia

Received 27 August 2003; received in revised form 2 January 2004; accepted 11 February 2004
Available online 17 July 2004

Abstract

This paper presents results from investigations into quantifying and modelling the dust generation and air entrainment mechanisms that
occur during free-falling streams of material. Emphasis is placed on the effects of drop height and product temperature. This information is
important for the efficient design and operation of dust control systems, as well as minimising the risk and severity of dust explosions.
2004 Published by Elsevier B.V.

Keywords: Dust generation; Drop height; Product temperature

1. Introduction predict such information are inaccurate and unreliable. This

The efficient control of dust or particulate emissions in
the workplace and general environment is becoming increas-
ingly important in industry for many reasons:
more stringent OH&S legislation;
decreasing allowable emission levels;
loss of material (e.g. spillage, dust emissions);
risk of dust explosions when handling combustible solids;
maintenance and clean-up costs;
damage to process equipment;
premature failure of dust control systems, especially for
hot processes;
aesthetics;
increasing pressure from the government and/or commu-
nity;
society becoming more environmentally conscious.
The design of a suitable dust control system for a par-
ticular application requires knowledge of dust generation
and air entrainment. The interaction of these two parameters
with the process (e.g. solids throughput, powder properties,
temperature, drop height, velocity) also must be considered
for this purpose. Unfortunately, the existing techniques to
Corresponding author. Tel.: +61 2 4221 3488; fax: +61 2 4221 4577.
E-mail address: peter wypych@uow.edu.au (D. Cook).
paper aims to:
describe and examine dust generation and air entrainment
mechanisms;
present some of the latest research work and results in
trying to quantify and model air entrainment and dust
generation/control;
examine process effects on dust generation;
describe some techniques to minimise/eliminate dust gen-
eration and emissions (e.g. velocity control);
explore the implications of the above findings for dust
explosion hazards.
Some case studies are included to provide examples and
also demonstrate some of the more important issues.
2. Dust generation and air entrainment
One of the most difficult processes in which to control dust
generation is that where a bulk material is dropped under
free-fall conditions to impact on a stockpile or other surface.
Examples of this type of situation are shown in Fig. 1.
The approach generally taken to control the dust in such
situations is to partially or fully enclose the process. A very
common example of this is the case of conveyor transfer
stations, where bulk material is dropped from the end of
one conveyor onto the start of another. Even when only a

0255-2701/$ see front matter 2004 Published by Elsevier B.V.

doi:10.1016/j.cep.2004.02.026
324 P. Wypych et al. / Chemical Engineering and Processing 44 (2005) 323326

Fig. 1. Examples of dust generation processes involving free-fall of bulk material.

very small fraction of the parent bulk material is liberated

as fugitive dust the total quantity of dust generated may be
enormous due to the very large throughput of the parent
material.
For the designer of such enclosures there are two major
questions that need to be answered:
(a) what is the volume of dusty air that should be extracted
from the enclosure?
(b) what is the concentration of dust in the air?
As yet, there is no reliable method of answering either of
these questions. This paper considers the most fundamental
situation of a product free falling under steady-state condi-
tions from a hopper at a given height above a stockpile. It
also examines briefly the effect of product temperature on
dust generation.

3. Previous research
The first serious attempt to predict the quantity of air
entrained by a falling stream theoretically was that of
Hemeon [1]. Hemeon modelled the air entrainment process
as one whereby the drag force exerted by each solid parti-
cle was essentially equal to that of a single particle falling
through a quiescent fluid. He deduced that the induced
(entrained) air flow, Qind , is given by (converted here to SI
units):

1/3
0.66g m(hA)2
Qind = m3 /s (1)
(dp p )
where g = gravitational acceleration, m = mass flow rate of
solids, h = drop height, A = cross-sectional area of material
stream, dp = particle diameter and p = particle density.
Tooker [2] then used this approach to formulate a means
of calculating the extract air requirements for enclosures,
such as conveyor transfer stations. The equations of Hemeon
[1] and Tooker [2], and variations thereon, are used widely
in industry. However, they generally grossly over-predict the
rate of air entrainment unless empirical correction factors
Fig. 2. Dust generation and air entrainment for falling stream of material.
are applied. This is a result of the inherent assumption that
the flow in the stream is made up of particles that act in-
dependently, whereas the reality is that: the stream of parti-
cles form a relatively coherent stream, Fig. 2; and particles
within the stream interact with the quiescent ambient to a
much lesser extent than if they were falling independently
through the fluid.
4. Experimental facility and results
Two experimental rigs were used for this study. The first
rig comprised a double-feed hopper arrangement to ensure
a virtually constant mass flow rate of powder during each
test. The hopper arrangement was mounted on a frame that
could be raised or lowered to change drop height and was
suspended on cables connected to three load cells to en-
able the mass flow rate to be recorded. The bulk material
dropped through a hole in a shelf above the stockpile and
air was extracted from underneath the shelf at a rate such
that the air pressure under the shelf was kept exactly equal
 P. Wypych et al. / Chemical Engineering and Processing 44 (2005) 323326 325

Fig. 3. Experimental rig used to measure entrained air and dust generation.
Fig. 4. Specific air entrainment results of Smithers (x symbols) and Glutz
(
symbols) using alumina with drop heights of 625650 mm.
to that of the external ambient air, as shown in Fig. 3. A
pressure observation port between the space under the shelf
and outside allowed the pressure to be equalised between
the surrounding air and the underside of the shelf by smoke
visualisation. Under these conditions, the volume of air ex-
tracted from beneath the shelf was equal to the total vol-
ume of air and bulk material falling through the aperture.
A relatively inert and robust powder, alumina (median size
= 96 m, particle density = 2465 kg/m3 ; loose-poured bulk
density = 1010 kg/m3 ), was selected for testing.
The second rig involved the release of approximately 2 kg
of hot alumina from an electrically heated hopper in a sealed
box. The alumina was dropped onto a plate at the bottom
of the box and the dust liberated during the fall and subse-
quent impact was collected in petri dishes and weighed to
determine the specific dust generation rate G = (dust liber- Fig. 5. Dust generation from a falling stream of alumina as a function of
ated/mass of parent material). The motivation for this work temperature difference (drop height = 500 mm, mass = 2.0 kg).
came from the need to understand the fundamentals of dust
generation during hot solids handling processes, such as The results of the experiments to determine how the rate
those involved in cement and coke production. of dust generation is influenced by the temperature differ-
ence between the bulk material and the surrounding ambient
air are shown in Fig. 5 [6]. Clearly, temperature difference
has a very strong influence on dust generation rates for
5. Results and discussion
materials with a relatively high proportion of fines. For the
alumina used in these experiments a temperature difference
The volume of entrained air was found to be proportional
of 140 C results in an increase in dust generation rate
to the drop height raised to the power of approximately 5/3.
by approximately an order of magnitude. This could ex-
This is in contrast to Eq. (1) where the expected index is
plain the premature failure of dust filtration systems on hot
2/3 [3]. This is an important result and suggests that a better
processes, where existing design and selection techniques
approach to modelling of the air entrainment process might
could be grossly under-estimating actual dust loadings.
come from the use of a modified plume theory.
The specific air entrainment (i.e. air entrainment per unit
mass of parent bulk material) was found to decrease with
6. Dust explosion hazards
increasing mass flow rate of solids, Fig. 4. The solid line
correlates the results of Smithers [4] and Glutz [5] and shows
In many instances, it may be possible to reduce the risk
that specific air entrainment is proportional to m0.67 . If the
and severity of dust explosions by minimising the generation
particles in the falling stream behaved independently of one
and turbulence of dust. In a falling stream of material:
another, each interacting in the same way with the ambient
air, then the specific air entrainment would be constant with the main core of falling powder is surrounded by a tur-
respect to mass flow rate. The experimental results show that bulent layer of fine dust, which escapes easily into the
this is not the case. surrounding air, Fig. 2;
326 P. Wypych et al. / Chemical Engineering and Processing 44 (2005) 323326
air is entrained in the falling stream of dilating powder;
the impact zone is a highly turbulent region where the en-
trained air is ejected into the atmosphere at relatively high
velocity and carries with it a large amount of fine dust;
considerable reductions in dust concentration and turbu-
lence, as well as dust explosion risk and severity, can
be achieved by minimising drop height and hence, the
amount of entrained air, as demonstrated previously.
The situation depicted in Fig. 2 is exacerbated for pneu-
matically filled bins. Here, the velocity at impact can be
quite high, especially when the material reaches high level.
Employing a large drop-out box or tee-bend can reduce
the extent and severity of dust generation. Further reductions
in dust loading, explosion risk and severity may be possible
by using a tangential entry. For example, Hauert et al. [7]
found that the reduced explosion pressures for the tangential
pneumatic filling of wheat flour and corn starch were 35
times smaller than those obtained with central filling.
Significant increases in dust also can occur during
the dilute-phase transport (suspension flow) of granular
materials, such as sugar and wheat. The relatively high
transport velocities can cause particle attrition and a sig-
nificant increase in fines, as well as an increased level of
electrostatic charge generation and risk of spark ignition.
Significant reductions in velocity are achieved by pursu-
ing the dense-phase (non-suspension) mode of flow. Here
the particles are conveyed either in fluidised dense-phase
(e.g. for air-retentive powders, such as flour and starch) or
low-velocity slug-flow (e.g. for granular products, such as
wheat and sugar). The following typical operating condi-
tions for wheat conveyed through a 105 mm 96 m long
pipeline demonstrate the possible (velocity) differences
between dilute-phase and dense-phase.
Dilute-phase (suspension flow): solids mass flow
= 6 t h1 , air mass flow = 0.260 kg s1 , solids loading
= 6, pipeline pressure drop = 30 kPa, transport velocity
= 19.325.0 m s1 .
Dense-phase (low-velocity slug-flow): solids mass flow
= 6 t h1 , air mass flow = 0.065 kg s1 , solids loading
= 26, pipeline pressure drop = 120 kPa, air velocity = 2.9
to 6.3 m s1 .
In addition to the obvious advantages of reduced air flow,
velocity and hence, power, dense-phase offers the following
additional benefits.
Particle velocity is controlled easily in the range
0.252 m s1 .
Particle damage and dust generation are extremely low
and often unable to be detected (e.g. sugar crystals can be
conveyed without even being scratched).
Electrostatic charge generation and risk of spark ignition
are very low.
The concentration of particles inside the pipeline is ex-
tremely high and beyond the upper explosible limit (UEL).
This forms a natural buffer zone or barrier for any
explosion that may occur in a vessel at either end of the
pipeline (i.e. flame propagation is unlikely). This could
avoid the need for explosion suppression and/or isolation,
which would be required normally for a dilute-phase or
dust extraction system.
The frequency and turbulence of particles striking the filter
elements are extremely low.
Filtration requirements (e.g. filter surface area) also are
quite low.
7. Conclusions
Free-falling bulk solids can produce large quantities of
fugitive dust and comprise core and boundary layer regions.
Specific air entrainment increases with drop height, but de-
creases with increasing solids mass flow rate. Higher temper-
atures or transport velocities increase dust generation rates
dramatically. Significant reductions in velocity and dust gen-
eration are achieved by selecting the dense-phase mode of
flow. Such findings should be considered for the efficient
design and operation of dust control and explosion control
systems.
References
[1] W.D.L. Hemeon, Plant and Process Ventilation, Industrial Press, New
York, USA, 1963.
[2] G.E. Tooker, Bulk Solids Handling 12 (1992) 227232.
[3] P. Cooper, P.C. Arnold, Air entrainment and dust generation from
a falling stream of bulk material, Kona Powder Particle 13 (1995)
125134.
[4] T. Smithers, Generation of Dust from a Falling Stream of Material, BE
Honours Thesis, Mechanical Engineering, University of Wollongong,
1994.
[5] A. Glutz, Air Entrainment in Falling Streams of Material, BE Honours
Thesis, Mechanical Engineering, University of Wollongong, 1995.
[6] B. Perrin, Dust Generation from a Heated Bulk Solid, BE Honours
Thesis, Mechanical Engineering, University of Wollongong, 1999.
[7] F. Hauert, A. Vogl, S. Radandt, Dust cloud characterization and its
influence on the pressure-time-history in silos, Process Safety Progress
15 (1996) 178184.
 Chemical Engineering and Processing 44 (2005) 327334

Photocatalytic degradation of trichloroethylene (TCE) over

TiO2 /silica gel in a circulating fluidized bed
(CFB) photoreactor
Tak Hyoung Lim, Sang Done Kim
Department of Chemical and Biomolecular Engineering and Energy and Environment Research Center,
Korea Advanced Institute of Science and Technology, DaeJeon 305-701, South Korea

Received 31 July 2003; received in revised form 12 December 2003; accepted 11 February 2004
Available online 25 June 2004

Abstract

The effects of the ratio of TiO2 /silica gel, superficial gas velocity (Ug ), solid circulation rate (Gs ), initial trichloroethylene (TCE) concen-
tration, wavelength of UV light, and concentrations of H2 O and O2 on photodegradation of TCE over TiO2 /silica gel have been determined
in a circulating fluidized bed (CFB) photoreactor. The TCE conversion decreases with wavelength of the UV light, the initial TCE and H2 O
concentrations in the CFB reactor. The optimum concentration of O2 for TCE degradation is found to be approximately 9 vol.% in the CFB
photoreactor. The CFB photoreactor is an effective tool for high TCE degradation with efficient utilization of photon energy.
2004 Elsevier B.V. All rights reserved.

Keywords: TiO2 /silica gel; Solid circulation rate and holdup; Circulation fluidized bed (CFB) photoreactor

1. Introduction cases whereas the efficiency of gas treatment is commonly

The degradation of trace contaminants from polluted air
and water by using photocatalytic reactor system has been in-
vestigated [1]. The photocatalytic degradation technologies
have several attractive advantages such as complete and effi-
cient remediation of a broad range of pollutants under certain
operating conditions at ambient temperature and pressure
without any chemical additives. Also the most commonly
employed TiO2 photocatalyst is inexpensive, safe, and very
stable with high photocatalytic activity [1]. Therefore, pho-
tocatalytic degradation systems are being actively used as an
economical remediation method for various purification ap-
plications. Studies of heterogeneous photocatalysis involv-
ing gas-solid interfaces are relatively sparse compared to
those of water-solid interfaces. Photocatalytic gas treatment
systems are under developing since they exhibit much higher
photocatalytic activity than the water treatment system [2].
Typical phocatalytic degradation efficiencies in water
treatment are a few percent at best or less than 1% in most
Corresponding author. Tel.: +82 42 869 3913; fax: +82 42 869 3910.
E-mail address: kimsd@kaist.ac.kr (S.D. Kim).
above 10%. Such higher photocatalytic efficiencies in the
gas-solid interface provide a motivation to develop efficient
photocatalytic reactors. The photocatalytic water remedi-
ation systems have been studied in suspensions whereas
gas remediation systems usually employed immobilized
photocatalysts on solid substrates except for reactors using
packed powder layer. The most common immobilized pho-
tocatalytic reactors for gas treatment are annular plug flow
reactors and honeycomb monolith reactors [1].
The idea of using fluidized bed as both uniform light
distribution and an immobilizing support for photocatalysts
has been originally proposed and theoretically evaluated by
Yue and Khan [3]. Experimental application of this idea has
been demonstrated by Dibble and Raupp [4] who designed
a bench scale flat plate fluidized bed photoreactor for pho-
tocatalytic oxidation of trichloroethylene (TCE). Recently,
Lim et al. [5,6] have developed a modified two-dimensional
fluidized bed photocatalytic reactor system and determined
the effects of various operating variables on decomposi-
tion of NOx . Fluidized bed photocatalytic reactor systems
have several advantages over conventional immobilized or
slurry-type photocatalytic reactors [7,8]. The unique reac-

0255-2701/$ see front matter 2004 Elsevier B.V. All rights reserved.
doi:10.1016/j.cep.2004.02.027
328 T.H. Lim, S.D. Kim / Chemical Engineering and Processing 44 (2005) 327334
tor configuration provides both exposure of photocatalyst to
UV light and good penetration of the light into the photo-
catalyst bed that allows more contact of photocatalyst and
reactant gas [8]. Also, UV light can be more evenly dis-
tributed within the given catalyst bed. Furthermore, a newly
developed circulating fluidized bed (CFB) photoreactor sys-
tem enables remote delivery of UV light to the photocata-
lyst, which is ideally suited to the gas remediation since it
provides good contact between UV light and photocatalyst
with reaction gas, higher reactant throughputs, reduces mass
transport limitations [9] and lower pressure drops compared
to conventional bubbling fluidized beds [10]. Although the
two-dimensional fluidized bed photocatalytic reactor system
can be applied to water and gas treatments, a CFB photore-
actor system for gas phase treatment has not been reported
yet.
In the present study, the effects of the initial TCE con-
centration, wavelength of the UV light, H2 O concentration,
O2 concentration, the ratio of TiO2 /silica gel, superficial gas
velocity (Ug ), solid circulation rate (Gs ) on photodegrada-
tion of TCE have been determined in a CFB photoreactor
system.
2. Experimental
2.1. Photocatalyst
In general, TiO2 powder is classified into Geldart C group
having a poor fluidization behavior [5,6]. Therefore, silica
gel (Aldrich, USA) with size of 3060 mesh and surface area
of 490 m2 g1 was used as a support to enhance fluidization
quality. This support is transparent to near UV light. Precur-
sor solutions for TiO2 coating on silica gel were prepared by
using titanium ethoxide, ethanol, HCl, hexylene glycol and
milli-Q water. Titanium ethoxide (5 g) and hexylene glycol
(2.8 cc) were dissolved in ethanol (50 g). After mixing vig-
orously at room temperature for 90 min, water (0.4 cc) and
37 vol.% HCl (0.1 cc) were added to the solution. The molar
ratios of water and HCl to the titanium ethoxide were 1 and
0.11, respectively. After mixing vigorously at room tempera-
ture for 90 min, silica gel was added to the aqueous colloidal
suspensions of TiO2 . Then, it was dried at 80 C for 24 h
and calcined at 450 C for 1 h. The composition of a crystal
structure was verified by Rigaku RINT2000 HRXRD. The
amount of TiO2 coating on silica gel is found to be 0.3 g
of TiO2 per gram silica gel. The BET surface area of the
silica/TiO2 catalyst is somewhat reduced to 404 m2 g1 due
to pore plugging by the impregnated TiO2 .
2.2. Circulating fluidized bed (CFB) photoreactor
A circulating fluidized bed (CFB) photoreactor and
installation of the UV light lamps is shown in Fig. 1.
The CFB photoreactor composed of a riser (13 mm-i.d.
1200 mm-high, Quartz glass), cyclone, downcomer
(23 mm-i.d., Pyrex glass) and a solids feeding system. A
sintered metal plate was used as a gas distributor to provide
uniform fluidization of TiO2 /silica gel photocatalyst.
To improve sealing and the mechanical problems,
non-mechanical valves (loop-seal, L-, J-, and V-valve) are
commonly employed [10]. The present CFB photoreactor
had a loop-seal for photocatalyst feeding and its flow rate
was varied by aeration rate at the bottom of the loop-seal.
The amount of the photocatalyst was 80 g for one batch op-
eration. During the operation, photocatalyst particles were
transported by up-flowing gas at the bottom of the riser.
The transported particles from the riser were collected and
separated in a cyclone. The separated photocatalyst parti-
cles flow down into the downcomer. The powders in the
downcomer flow into the riser again through a loop-seal
by adjusting aeration gas. The flow rate of reactant gases
(548 ppmv TCE/N2 balance, 50 vol.% O2 /N2 balance, N2 )
were regulated by a mass flow controller (MKS, USA) to
match suitable concentrations of TCE, O2 , and H2 O and in-
troduced into the bottom of the CFB photoreactor. Concen-
tration of TCE was measured by online gas chromatography
equipped with ECD (HP 6890 II) and an auto-sampling
valve system (Valco instrument Inc., A60). The minimum
fluidization velocity (Umf ) of TiO2 /silica gel photocatalyst
was determined by the pressure drop and the expanded bed
height in the bed with variation of Ug [5,7].
2.3. The installation of the UV light lamp
Photoillumination was provided with 15 germicidal
white and 15 fluorescent black light lamps (8 W, Sankyo
Denki Japan, G8T8, F8T8). Wavelength of the white light
lamp ranged from 200 to 300 nm with the maximum light
intensity at 254 nm, wavelength of the black light lamp
ranged 300400 nm with the maximum light intensity at
356 nm. The light intensity was measured by UV radiome-
ter (Cole-Parmer, VLX-3 W, USA). The UV light lamps
was affixed at the outside of the CFB riser for effective UV
light irradiation and aluminum foil surrounded the UV light
lamps to minimize loss of light irradiation and to improve
utilization of reflected and deflected lights.
3. Results and discussion
To determine the optimum content of TiO2 in silica gel
for maximum TCE conversion, the TiO2 /silica gel ratio was
varied from 0.10 to 0.42 with variation of Ug (1.3, 1.5,
1.7 m s1 ) as shown in Fig. 2. As can be seen, the conver-
sion increases gradually up to 80% at Ug = 1.5 m s1 with
increasing the TiO2 /silica gel ratio up to around 0.3. When
TiO2 /silica gel ratio is below 0.3, TiO2 content is not suffi-
cient to decompose TCE by photocatalysis in the photore-
actor. Thereafter, TCE conversion remains constant with a
further increase in TiO2 /silica gel ratio due to the arrival of
sufficient participating reaction of TCE by the photodegra-
 T.H. Lim, S.D. Kim / Chemical Engineering and Processing 44 (2005) 327334 329

Fig. 1. Schematic diagram of CFB photoreactor.

dation. Therefore, it may be clamed that the optimum ratio
of TiO2 /silica gel in the present CFB photoreactor is 0.3 at
the given specific geometry of UV light irradiation. There-
fore, the TiO2 /silica gel ratio of 0.3 was employed in all the
experimental conditions.
The effect of Ug on photocatalytic activity of TiO2 /silica
gel as a bed material in the CFB photoreactor is shown in
Fig. 3. As can be seen, TCE conversion over TiO2 /silica gel
photocatalyst exhibits a maximum value at Ug = 1.5 m s1
where UV light transmission and mixing of TiO2 /silica gel
and TCE gas might be at the optimum condition. Thereafter,
TCE conversion decreases with a further increase in Ug due
to bypassing of unreacted TCE gas and reduction of resi-
dence time of TCE gas in the photoreactor [5,7]. Therefore,
it can be concluded that photo degradation of TCE requires
sufficient mixing of reactant and catalyst at the optimum su-
perficial gas velocity to provide the maximum contact be-
tween TiO2 /silica gel and TCE gas-UV light in the CFB
photoreactor.
The solid circulation rate (Gs , kg m2 s1 ) can be con-
trolled by aeration gas in the loop-seal to adjust pressure
balance in the CFB loop. Solid circulation rate is a unique
operating variable in the CFB system as a function of pres-
sure drop across loop-seal of a CFB photoractor [10] and
directly related to solid holdup in the riser where the pho-
todegradation of TCE occurs.
Solid holdup (s ), the volume fraction of solids, and
Priser, pressure drop in the riser are defined as
s = 1 g (1)
330 T.H. Lim, S.D. Kim / Chemical Engineering and Processing 44 (2005) 327334
Fig. 2. Effect of TiO2 /silica gel ratio on TCE conversion with a variation
of Ug .
Priser
= (1 g )(s g )g
L
where, Priser , L, g , s , g are pressure drop across the riser
(PRB PRT ), length of the measuring height, gas holdup,
solid density and gas density, respectively.
The TCE conversion with a variation of wavelength of UV
light as a function of Gs and s is shown in the Fig. 4. As can
be seen, TCE conversion increases linearly with increasing
Fig. 3. Effect of Ug on TCE conversion with a variation of Gs .
Fig. 4. Effect of solid circulation rate on TCE conversion with a variation
of wavelength of UV light Ug .
(2)
Gs at the given wavelengths (254 and 365 nm). With the ger-
micidal white lamp at Gs = 9.2 kg m2 s1 (s = 0.018),
more than 80% of TCE is degraded, and the TCE conversion
increases up to 90% with increasing Gs (15.4 kg m2 s1 )
at s = 0.031. Therefore, it can be seen that solid holdup
(s ) affects the performance of CFB photoreactor since the
optimum s provides better contact between the catalyst
(TiO2 /silica gel) and the reactant gas (TCE) at the given
UV light irradiation as reported in the surface modifica-
tion of HDPE powders in a circulating fluidized bed reactor
[10].
The effect of initial TCE concentration on TCE conver-
sion in the CFB photoreactor is shown in Fig. 5 where the
conversion decreases linearly with increasing the initial TCE
concentration at the given Ug (1.3 and 1.5 m s1 ). It may in-
dicate that the limited and fixed amounts of active sites on
TiO2 /silica gel are present before degradation of TCE as ob-
served for decomposition of toluene and formaldehyde by
UV/TiO2 at the saturated adsorption with the fixed active
sites [1,7,11].
The effect of wavelength of the UV light (germicidal
white and fluorescent black lamps) on TCE conversion in
the CFB photoreactor is shown in Fig. 6. As can be seen,
the germicidal white lamp (254 nm) exhibits higher TCE
conversion than the fluorescent black lamp (365 nm). It has
been reported that variation of photon energy in excess of
band gap of TiO2 does not affect the product mixture from a
photocatalyst reactor [12]. The increase of TCE conversion
is caused by higher photon flux, rather than by increases
 T.H. Lim, S.D. Kim / Chemical Engineering and Processing 44 (2005) 327334 331

Fig. 5. Effect of initial TCE concentration on TCE conversion with a variation of Ug .

in the energy of individual photons. Germicidal white wavelength light is adsorbed more strongly by TiO2 /silica
lamps are about 25% more efficient than fluorescent black gel than the longer one. Therefore, penetration distance
lights to producing photons with wavelengths less than of photons into TiO2 /silica gel is shorter and electrons
385 nm [11]. In addition, it has been reported that shorter and holes are formed closer to the surface of TiO2 /silica

Fig. 6. Effect of wavelength of UV light on TCE conversion with a variation of Ug .

332 T.H. Lim, S.D. Kim / Chemical Engineering and Processing 44 (2005) 327334
Fig. 7. Effect of O2 concentration on TCE conversion with a variation of Ug .
gel. Therefore, photons take less time to migrate onto
TiO2 /silica gel surface and hence have less time to par-
ticipate in energy wasting recombination reactions before
useful surface (or near-surface) chemical reaction take place
[5]. Therefore, the shorter wavelength (245 nm) irradiation
is more effective to enhance photodegradation of TCE than
the longer wavelength (365 nm) irradiation in the CFB
photoreactor.
The effect of O2 concentration on TCE conversion with
variation of Ug is shown in Fig. 7 where the TCE conversion
increases with increasing O2 concentration up to 9 vol.%
and remains constant with a further increase in O2 con-
centration. The role of molecular O2 in the photocatalytic
gas-treatment is three-fold: (1) a conduction band for elec-
tron scavenger to reduce fast electronhole recombination,
(2) a precursor of active oxygen species (e.g. HO2 , HO, O)
that initiate decomposition reactions, and (3) a replenishing
source of the lattice oxygen that could be depleted during
the photocatalytic degradation [9]. Therefore, increasing
oxygen content up to 9 vol.% enhances TCE conversion.
Above 9 vol.% of O2 , TCE conversion remains constant
that may indicate a sign of O2 saturation. In the absence
of O2 at Ug = 1.5 m s1 , it can be noticed that the conver-
sion reaches 0.75 that may imply the lattice oxygen took
part in the photocatalytic degradation of TCE [9]. From
this study, the optimum O2 concentration is approximately
9 vol.% for the maximum TCE conversion in the CFB
photoreactor.
The effect of H2 O concentration on the TCE conversion
in the CFB photoreactor is shown in Fig. 8. As can be seen,
TCE conversion decreases with H2 O concentration at the
given Ug (1.3, 1.5 m s1 ). The presence of H2 O could en-
hance or reduce the photocatalytic reaction rates that depend
on the reactant and H2 O concentrations [4]. The water vapor
in this study negatively affects the photocatalytic conversion
of TCE as reported previously [13]. In general, the presence
of water vapor in photocatalytic reactions is considered to be
favorable because water molecules react with valence band
holes to produce reactive hydroxyl radicals. Indeed, a small
amount of preadsorbed water is found to be essential in pho-
tocatalytic degradation of TCE [9]. Although no water was
added in this study except when the effect of water vapor
on the TCE conversion was determined, water molecules
generated in situ from various reactions byproduct of TCE
photo-degradation seem to be a sufficient source of OH rad-
ical in the photoreaction of TCE. Also, TCE molecules are
oxidized through direct hole transfer pathway that does not
involve OH radical mediators [4]. It can be reasonably accept
that both the hydroxyl radical and the direct hole pathways
are operating in the photocatalytic conversion of TCE. In ei-
ther case, excess water molecules compete for active surface
sites on TiO2 /silica gel photocatalyst with TCE molecules
 T.H. Lim, S.D. Kim / Chemical Engineering and Processing 44 (2005) 327334 333

Fig. 8. Effect of H2 O concentration on TCE conversion with a variation of Ug .

to reduce the photodegradation conversion. Therefore, it can Appendix A. Nomenclature

be claimed that TCE conversion decreases with increasing
H2 O concentration with a variation of Ug in the CFB pho- CO2 concentration of O2 in gas phase (ppmv)
toreactor. CH2 O concentration of H2 O in gas phase (ppmv)
CTCE concentration of TCE in gas phase (ppmv)
g acceleration due to gravity (m s2 )
4. Conclusion Gs solid circulation rate (kg m2 s1 )
L length of the measuring height (m)
The photocatalytic degradation characteristics of trichlo- P pressure (kPa)
roethylene have been determined in a CFB (circulating flu- Umf minimum fluidization gas velocity (cm s1 )
idized bed) photoreactor over TiO2 /silica gel photocatalyst. Ug superficial gas velocity (cm s1 )
The performance of the CFB photoreactor is governed by X fractional conversion of TCE ()
solid holdup as a function of superficial gas velocity and
solid circulation rate to provide the optimum contact be- Greek letters
tween TiO2 /silica gel and TCE gas at the given UV light. The g gas holdup ()
TCE conversion decreases with wavelength of the UV light s solid holdup ()
and H2 O concentration in the CFB reactor. The optimum g gas density (g cm3 )
concentration of O2 for the maximum TCE degradation is s solid density (g cm3 )
found to be approximately 9 vol.% in the photoreactor.

References
Acknowledgements
[1] S. Hager, R. Bauer, G. Kudielka, Photocatalytic oxidation of gaseous
chlorinated organics over titanium dioxide, Chemosphere 41 (2000)
The authors would like to thank for a grant in aid of re-
12191225.
search to Dr. S.D. Kim from the LG-Caltex Co. in the frame [2] J. Peral, X. Domenech, D. Ollis, Heterogeneous photocatalysis for
of the Brain Korea 21 Project. Also, this work was supported purification, decontamination and deodorization of Air, J. Chem.
by Korea Research Foundation Grant (KRF-2000-E00442). Technol. Biotechnol. 70 (1997) 117140.
334 T.H. Lim, S.D. Kim / Chemical Engineering and Processing 44 (2005) 327334
[3] P.L. Yue, F. Khan, Photocatalytic ammonia synthesis in a fluidized
bed reactor, Chem. Sci. Eng. 38 (1983) 18931900.
[4] L.A. Dibble, G.B. Raupp, Fluidized-bed photocatalytic oxidation of
trichloroethylene in contaminated air streams, Environ. Sci. Technol.
26 (1992) 492495.
[5] T.H. Lim, S.M. Jeong, S.D. Kim, J. Gyenis, Photocatalytic decom-
position of NO with TiO2 particles, J. Photochem. Photobiol. A:
Chem. 134 (2000a) 209217.
[6] T.H. Lim, S.M. Jeong, S.D. Kim, J. Gyenis, Degradation charac-
teristics of NO by photocatalysis with TiO2 and CuO/TiO2 , React.
Kinet. Catal. Lett. 71 (2000) 223229.
[7] T.H. Lim, S.D. Kim, Photocatalytic degradation of trichloroethylene
over TiO2 /SiO2 in an annulus fluidized bed reactor, Kor. J. Chem.
Eng. 19 (2002) 10721077.
[8] T.H. Lim, S.D. Kim, Trichloroethylene (TCE) degradation by pho-
tocatalysis in annular flow and annulus fluidized bed photoreactors,
Chemosphere 54 (2004) 305311.
[9] W.Y. Choi, J.Y. Ko, H.W. Park, J.S. Chung, Investigation on
TiO2 -coated optical fibers for gas-phase photocatalytic oxidation of
acetone, Appl. Catal. B: Environ. 31 (2001) 209220.
[10] S.H. Jung, S.H. Park, D.H. Lee, S.D. Kim, Surface modification
of HDPE powders by oxygen plasma in a circulating fluidized bed
reactor, Poly. Bull. 47 (2001) 199205.
[11] Y. Ku, C.M. Ma, Y.S. Shen, Decomposition of gaseous trichloroethy-
lene in a photoreactor with TiO2 -coated nonwoven fiber textile, Appl.
Catal. B: Environ. 34 (2001) 181190.
[12] W.A. Jacoby, D.M. Blake, R.D. Noble, C.A. Koval, Kinetics of the
oxidation of trichloroethylene in air via heterogeneous photocatalysis,
J. Catal. 157 (1995) 8796.
[13] S.B. Kim, S.C. Hong, Kinetic study for photocatalytic degradation of
volatile organic compounds in air using thin film TiO2 photocatalyst,
Appl. Catal. B: Environ. 35 (2002) 305315.
 | |

Publication information: Chemical Engineering and Processing (ISSN 0255- Central and South America), including air speed delivery. Periodical postage paid
CHEMICAL ENGINEERING AND PROCESSING 2701). For 2005, volume 44 is scheduled for publication. Subscription prices are
available upon request from the Publisher or from the Regional Sales Office
at Jamaica, NY 11431.
USA POSTMASTER Send address changes to Chemical Engineering
nearest you or from this journals website (http://www.elsevier.com/ locate/cep). and Processing, Publications Expediting Inc., 200 Meacham Ave, Elmont, NY
Further information is available on this journal and other Elsevier products 11003.
Editors: through Elseviers website: (http://www.elsevier.com). Subscriptions are AIRFREIGHT AND MAILING in the USA by Publications Expediting Inc., 200
Prof. Dr.-lng. A. Gorak accepted on a prepaid basis only and are entered on a calendar year basis. Meacham Avenue, Elmont 11003, NY 11003.
Universitat Dortmund Issues are sent by standard mail (surface within Europe, air delivery outside
Lehrstuhl fur Thermische Verfahrenstechnik Europe). Priority rates are available upon request. Claims for missing issues Orders, claims, and journal enquiries: please contact the Customer Service
Fachbereich Chemietechnik should be made within six months of the date of dispatch. Department at the Regional Sales Office nearest you:
Emil-Figge-Strasse 70
D-44221 Dortmund Author enquiries
Phone: 49 721 608 2391 Orlando: Elsevier, Customer Service Department, 6277 Sea Harbor Drive,
Fax: 49 721 608 3490 Orlando, FL 32887-4800, USA; phone: (+1) (877) 8397126 [toll free number
Submissions for US customers], or (+1) (407) 3454020 [customers outside US]; fax: (+1) (407)
e-mail: gorak.CEP@ct.uni-dortmund.de For information regarding the submission of manuscripts to Chemical 3631354; e-mail: usjcs@elsevier.com
Engineering and Processing, please refer to the Instructions to Authors in the
Prof. Dr.-Ing. A. Wild back of this issue.
Laboratoire des Science du Genie Chimique Amsterdam: Elsevier, Customer Service Department, PO Box 211, 1000 AE
Groupe ENSIC Amsterdam, The Netherlands; phone: (+31) (20) 4853757; fax: (+31) (20)
These instructions are also available on the World Wide Web: http://
1 rue Grandville 4853432; e-mail: nlinfo-f@elsevier.com
www.elsevier.com/locate/cep
BP 451
F-54001 Nancy For detailed instructions on the preparation of electronic artwork, consult the Tokyo: Elsevier, Customer Service Department, 4F Higashi-Azabu, 1-Chome
France Elseviers author gateway at http://authors.elsevier.com Bldg, 1-9-15 Higashi-Azabu, Minato-ku, Tokyo 106-0044, Japan; phone: (+81) (3)
Phone: 49 231 755 2323 5561 5037; fax: (+81) (3) 5561 5047; e-mail: jp.info@elsevier.com
Fax: 49 231 755 3035 Other enquiries
e-mail: Gabriel.Wild@ensic.inpl-nancy.fr Visit Elseviers Author Gateway (http://authors.elsevier.com) for the facility Singapore: Elsevier, Customer Service Department, 3 Killiney Road, #08-01
to track accepted articles and set up e-mail alerts to inform you of when an Winsland House I, Singapore 239519; phone: (+65) 63490222; fax: (+65)
Founding Editor: articles status has changed. The Author Gateway also provides detailed artwork 67331510; e-mail: asiainfo@elsevier.com
Prof. Em. Dr.-lng. Dr. h.c. mult. E.-U. Schlunder, Universitat Karlsruhe, Germany guidelines, copyright information, frequently asked questions and more.
Advertising Information: Advertising orders and enquiries can be sent to:
Honorary Editor: Contact details for questions arising after acceptance of an article, especially Elsevier, 84 Theobalds Rd, London, WC1X 8RR, tel: +44 (0) 20 7611 4117;
Dr. lng. V. Gnielinski, Universitat Karlsruhe, Germany those relating to proofs, are provided after registration of an article for publication. fax: +44 (0) 20 7611 4463; e-mail: k.barton@elsevier.com

USA mailing notice: Chemical Engineering and Processing (ISSN 0225- 2701) Reprints
is published monthly by Elsevier B.V. (P.O. Box 211, 1000 AE Amsterdam, The Reprint orders can be sent to currentreprints@elsevier.com, tel: +44 (0) 20 7611
Editorial Board: Netherlands). Annual subscription price in the USA US$1,965 (valid in North, 4494; fax: +44 (0) 20 7611 4463
P. Aptel, Universite Paul Sabatier, France
P.I. Barton, Massachusetts Institute of Technology, Cambridge, Massachusetts, USA
P. Bekiaroglou, University of Thessaloniki, Thessaloniki, Greece
P. Burghardt, Instytut Inzynierii Chemicznei l, Gliwice, Poland
K.J. Carpenter, Syngenta Global Specialist Technology, Huddersfield, UK
E. Drioli, Universita Della Calabria, Italy
J. Gyenis, Hungarian Academy of Sciences, Veszprem, Hungary
K. Jelemensky, Slovak Technical University, Bratislava, Slovak Republic
J.T.F. Keurentjes, Eindhoven University of Technology, The Netherlands
B. Kraushaar-Czarnetzki, University of Karlsruhe, Karlsruhe, Germany
F. Larachi, Universite Laval, Canada
R. Marr, Technische Universitat Graz, Graz, Austria
D. Mewes, University of Hanover, Hanover, Germany
H. Muller-Steinhagen, German Aerospace Centre, Stuttgart, Germany
K.D.P. Nigam, Indian Institute of Technology Delhi, New Delhi, India
A. Olcay, Ankara University, Ankara, Turkey
L. Puigjaner, Universitat Politencnica de Catulunya, Barcelona, Spain
H. Schoenmakers, BASF AG, ZAT-L540, Ludwigshafen, Germany
S. Skogestad, Norwegian University of Science and Technology, Trondheim, Norway
A. Stankiewicz, DSM Research, The Netherlands
K. Sundmacher, Universitaet Magdeburg, Germany
J. Werther, Technische Universitat of Hamburg-Harburg, Hamburg, Germany
K.R. Westerterp, Twente University of Technology, Enschede, The Netherlands

Aims and Scope

Chemical Engineering and Processing is a journal intended for practising engineers in industry and academia. The aim of the published articles is to demonstrate
how novel discoveries, developments and theories may be used for improved analysis and design of chemical engineering equipment and processes. The Journal
offers space for articles on any branch of chemical engineering and is particularly concerned with mechanical, thermal, chemical and bioengineering processes and
also includes analyses relating to environmental chemical engineering and sustainable industrial development. Advanced knowledge on engineering fundamentals
and processes is presented in such a form that it can be readily implemented into practical application. From time to time, review papers will be published for
individual fields of activity to illustrate and summarise scientific and/or technological progress.

*
1 The paper used in this publication meets the requirements of ANSI/NISO Z39.48-1992 (Permanence of Paper).

Printed in The Netherlands

| |