Вы находитесь на странице: 1из 318

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.

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
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

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
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


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

0255-2701/$ see front matter 2004 Elsevier B.V. All rights reserved.
2 . Sahin et al. / Chemical Engineering and Processing 44 (2005) 16

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

Fig. 1. Schematic diagram of the experimental procedure.

. Sahin et al. / Chemical Engineering and Processing 44 (2005) 16 3

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




85 300 C
400 C
500 C

0 10 20 30 40 50

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
315 3.593 1.88 9.002 1.361 9.009 1.50

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




300 C
400 C
80 500 C


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- Ca(OH)2 to control the transport rate of liberated water
cle size shows a distribution which is attributed to the Ca2+ from inside BPH particle to air. By this method, the
content and temperature. In addition, the particles smaller content of BPH can be decreased up to about 0.68 wt.%
than 315 m contain twice higher Ca2+ than all other parti- Ca2+ .
cle sizes contained. Microscopic photographs of borax pen- (d) Calcination of BPH covered with Ca(OH)2 to BA re-
tahydrate in pure, containing 0.67 wt. and 2.5 wt.% Ca2+ alized in 50, 20 and 5 min intervals at 300, 400 and
are given in Fig. 5, respectively. In pure state, the surface 500 C, respectively. This result shows that at temper-
of BPH particles are smooth whereas the quality of par- atures higher than 400 C, the puffed anhydrous borax
ticles are decreased with increasing Ca2+ quantity on the can be obtained by continuous operation. Also, the high
surface. 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
4. Conclusion reaction, adsorption and high temperature application
of BA.
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- References
ing conclusion can be withdrawn from the results obtained
during the calcination of covered BPH particles in a batch [1] R.P. Douglas, F.G. Donald, Refinement of the structure of tincal-
calcinator: conite, Acta Cryst. C47 (1991) 22792282.
[2] G. Nencetti, A. Pennacchi, Chim. Ind. Milan 46 (5) (1968) 518525.
(a) The calcination of BPH with single stage cannot be [3] Kirk-Othmer, Enc. Chem. Thech., 4 ed., vol. 4, 1992, p. 388.
carried out at the temperature higher than 300 C since [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.
agglomeration is started. The calcination of BPH to BA
Tolun, Ind. Eng. Chem. Res. 35 (1996) 14241428.
also takes longer at around 300 C. [6] O. Sahin, U.G. Beker, A.N. Bulutucu, Int. J. Storing Handl. Process.
(b) BPH can be transformed to BA without agglomeration Powder 7 (1995) 165167.
by mixing CaO with BPH in solid state at temperature [7] O. Sahin, A.N. Bulutucu, Chem. Eng. Process. 41 (2002) 135141.
range 300500 C. In this case, the content of CaO in [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
BA must be higher than 7%.
(2003) 232236.
(c) Covering the BPH particles with Ca(OH)2 aqueous [10] D.F. Snell, C.L. Hilton, Encyclopedia of Industrial Chemical Anal-
solution also prevented the agglomeration in tempera- ysis, vol. 7, Wiley, New York, 1968, pp. 373384.
ture range 300500 C. In this kind of operation, the [11] ISO 3424, Sodium perborate for industrial use-determination of bulk
BPH particle surface was covered with a thin layer of 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


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-

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

0255-2701/$ see front matter 2004 Elsevier B.V. All rights reserved.
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 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-
a S 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
H for strongly swirling flows found for example in cyclones
[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
B ELEVATION PLAN Fredriksson [7] reveal that the RNG k model under pre-
dicts the variation of the axial velocity profile across the
Fig. 1. Tangential cyclone configuration.
radial direction and also over predicts the magnitude of the
tangential velocity and the cyclone pressure drop.
2. Cyclone design The Reynolds stress model requires the solution of trans-
port equations for each of the Reynolds stress components
There are a number of different forms of cyclone but as well as for dissipation transport without the necessity to
the reverse flow cyclone represented in Fig. 1 is the most calculate an isotropic turbulent viscosity field. The Reynolds
common design used in the industry. The cyclone consists of stress turbulence model yield an accurate prediction on swirl
four main parts: the inlet, the separation chamber, the dust flow pattern, axial velocity, tangential velocity and pressure
chamber and the vortex finder. Tangential inlets are preferred drop on cyclone simulation [7,6,13,10].
for the separation of solid particles from gases [1]. In this The finite volume methods have been used to discretised
study, the numerical simulation deals with the standard case the partial differential equations of the model using the Sim-
of reverse flow cyclone with a tangential rectangular inlet. ple method for pressurevelocity coupling and the second
Cyclone dimension used in this simulation are as shown in order upwind scheme to interpolate the variables on the sur-
Table 1. 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
3. Computational fluid dynamics approach the turbulence in cyclones. Standard fluent wall functions
were applied and high order discretisation schemes were
Fluent is a commercially available CFD code which also used.
utilises the finite volume formulation to carry out coupled Under the RSM second order upwind for discretisation
or segregated calculations (with reference to the conserva- there is a difficulty to reach the convergence in simulation.
tion of mass, momentum and energy equations). It is ideally The residuals may exhibit cyclic tendencies which mean
suited for incompressible to mildly compressible flows. that the transient pattern occurs. In this instance, the solver
The conservation of mass, momentum and energy in fluid must be changed to a transient solver and makes the time
flows are expressed in terms of non-linear partial differen- step something in the region of 0.025 s or a tiny fraction of
tial equations which defy solution by analytical means. The the residence time of the cyclone. The simulation is then
solution of these equations has been made possible by the solved with a coupling of unsteady and steady state solver
advent of powerful workstations, opening avenues towards in Fluent. For the simulation using RNG k model the
the calculation of complicated flow fields with relative ease. steady state solver is sufficient to reach the convergence. The
For the turbulent flow in cyclones, the key to the success CFD simulation was performed with a Pentium IV 2.8 GHz
of CFD lies with the accurate description of the turbulent HP workstation XW8000 with 512 cache-memory, 1 GB
behaviour of the flow [8]. To model the swirling turbulent RAM-memory, and 110 GB hard disc memory.

Table 1
Cyclone geometry used in this simulations
Geometry a/D b/D De /D S/D h/D H/D B/D Da

Stairmand high efficiency 0.5 0.2 0.5 0.5 1.5 4 0.375 0.305
Bohnet [2] 0.533 0.133 0.333 0.733 0.693 2.58 0.333 0.15
a Unit in meters.
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
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
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
= 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
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-

= 9.47 (5) Shepherd & Lapple
De2 1000

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


2000 CFD RSM
Pressure Drop (Pa)

1500 Shepherd & Lapple


Casal & Martinez

0 200 400 600 800 1000 1200
Temperature (K)

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

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

2500 CFD RSM
2000 1200
Pressure Drop (Pa)

Pressure Drop (Pa)

Shepherd & Lapple

1500 800 CFD RNGk-
Shepherd & Lapple
1000 Coker

Casal & Martinez
500 Casal & Martinez
0 200 400 600 800 1000 1200
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 11

the CFD numerical calculation when compared with experi- 6. Conclusions

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


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-

siderations into process design, Cabezas et al. [5] introduced
There is currently a great deal of interest in the devel- a potential environmental impact (PEI) balance as an amend-
opment of methods that can be used to prevent or at least ment of the Waste Reduction Algorithm [6]. However, this
reduce the generation of pollution, because it may result in algorithm is simply a tool to be used to aid in evaluating the
lower operating costs due to better utilization of raw mate- environmental friendliness of a process [7].
rials and energy and reduced waste treatment and disposal Making clear the effects of reaction conditions such
costs. Researches have shown that the most opportunities as temperature, pressure and concentration as well as en-
of environmental impact minimization exist in the process gineering factors involving back-mixing, heat and mass
research and development and design stage of chemical pro- transfer and so forth, on the performances of reaction pro-
cesses [1,2]. Waste and pollutant generation in many chemi- cess is one of critical bases for benign reactor design. In
cal manufacturing process can be traced back to the reaction this paper, potential environmental impact balance and PEI
system, where reaction pathways that produce high yields rate-law expression are proposed as a tool to be used to
may also generate toxic byproducts or use environmentally effectively discover these effects on environmental perfor-
unfriendly raw materials, solvents, and catalystswhich mance of a reaction process. With this methodology, the
must be dealt with downstream at a significant cost [3]; PEI of waste and pollutants could be tracked throughout
and the size, configuration, and composition of the reactor reaction process within reactors, and appropriate reactor
are very important factors controlling the greenness of a type and reaction conditions, which could lead to envi-
process [4]. Therefore, the design of new reactor is a very ronmentally benign reaction processes, could be selected
important element in preventing pollution at the source. 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.
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-
 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
(4) can track compositions of chemicals, pressure and tempera-
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:
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 15

Such an expression, which shows how the rate depends on changing of PEI throughout reactors; using Eq. (8) combin-
the concentrations of reactants, is called PEI rate-law expres- ing with other rate equations, the influence of operation pa-
sion. Note, PEI rate-law expression should be consistent with rameters such as temperature, concentration, pressure and
the method for calculating the overall PEI in Eqs. (2) and reaction time, etc. on environmental metrics of a chemical
(3), which ignore the combinational impacts that could be process could be analyzed facilely, and then, reactors gen-
associate with mixtures of chemicals which accounts for the erating a minimum of PEI could be designed easily.
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 4. Case study
form of the PEI rate-law expression should be adjusted ac-
cordingly. In this case study, we use the environmental impact indices
Obviously, rPEI can be expressed in terms of reaction tem- below, which can be derived from Eq. (1), to characterize
perature and concentration or partial pressure of chemicals environmental performance of reaction processes [7,8]:
and so on. Therefore, the PEI rate-law expression can be
(1) Potential environmental impact generation rate by non-
helpful to analyze the effect of concentrations and temper-
products Igen
NP .
ature on the PEI transformation rate, rPEI , and to study the
(2) Specific potential environmental impact due to non-
influence of back-mixing on process environmental perfor-
products PEI .
mance. At least, the PEI rate-law expression can reveal the
factors that control the transformation rate of PEI, so as to Igen

provide guidance for the selection of reactor type and opera- PEI =  (9)
p pp
tion conditions, and the inner structure of the reactor, which
produce desired products while creating minimum undesired where pp is the mass flow rate of product p, and PEI has
potential environmental impact. units of potential environmental impact per mass of prod-
ucts. Obviously, the smaller the values of Igen NP and
3.2. PEI balance the more environmental friendly the process, and all others
factors such as economics being equal, the more desirable.
Since desired products are the substance we desire and the Igen
NP would be useful in comparing different designs on an

objective of this paper is to present a methodology of waste absolute basis, while PEI would be useful in comparing
reduction, that is the primary concern is reducing the impact different designs independently of plant size [8].
and the amount of the non-products, the k of the desired These indices characterize some aspects of the genera-
products is not considered in PEI balance. This insures that tion of PEI within a reaction process. They are very useful
the user or producer is not directly penalized for producing in addressing questions related to the ability of the reactor
a chemical that has a high PEI value. Auxiliary materials to produce desired products while creating a minimum of
(catalyst, solvent and so on) used in reaction processes must undesired potential environmental impact.
leave the production process completely as waste or emis- To illustrate the use of the algorithm, a case of allyl chlo-
sion [10], so the PEI of the auxiliary materials should be ride manufacturing was used. Allyl chloride is manufactured
considered. Raw materials that if possible shall be entirely by means of non-catalytic chlorination of propylene. Con-
processed into the desired product are not as a rule com- sider the following reaction system.
pletely converted into the product. The losses that arise are The principle reaction is:
the cause of generated waste and emission [10], thus their Cl2 + CH2 =CHCH3 CH2 =CHCH2 Cl + HCl (10)
potential environmental impact should be taken into account.
To sum up, the PEI balance of reaction system under steady The main secondary reaction is further chlorination of allyl
state is: chloride to 1,3-dichloropropene:
= Iout
(7) Cl2 + CH2 =CHCH2 Cl ClCH=CHCH2 Cl + HCl (11)
where the superscript NP represents materials except de- Another important side-reaction is the additive chlorination
sired products, that is, non-products. Iout
NP and I NP are input
in of propene to 1,2-dichloropropane:
and output rates of PEI caused by non-products, and Igen NP
Cl2 + CH2 =CHCH3 CH2 ClCHClCH3 (12)
the rate at which impact is generated in a reactor.
For a differential flow reactor, dV, we write a PEI balance This reaction system is a kind of complex van de vusse
equation over the reactor: reaction, a typical reaction process involving consecutive
and parallel reactions. It is, therefore, sufficiently complex
dIgen = rPEI dV (8)
to illustrate the algorithm. Due to the limit to the space of
With PEI balance Eq. (7), combining with other equations, the article, we only select temperature, concentration and
designers can reveal how PEI change with reaction time or back-mixing to be the object of research to illustrate the use
axial and radial direction inside reactors, i.e. can track the 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 ),
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
where r1 , r2 , and r3 represent the rates of the reactions  CB
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)
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-    
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]

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 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- 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
+ 1.5(0.024 0.007) v0
Using the same method as mentioned previously, we can
where AA
is the area of curved trapezoid A
(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
, 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 . 4.4. The effect of temperature on environmental
From Eq. (25), Igen of CSTR is the product of v0 and the performance
area of the corresponding rectangle in Fig. 3. For example,
when reaction temperature is 450 K, Igen of CSTR equals 4.4.1. The effect on Igen
the area of the rectangle in Fig. 3 multiplied by v0 . In such When l = 1, from Eq. (27), it can be seen easily that the
case, the value of Igen is positive number. Comparing Fig. 2 value of rPEI decreases (its absolute value increase) as the
with Fig. 3, it can be found that lower Igen will be obtained if reaction temperature rises, that is, Igen can be lowered by
we adopt PFR and lower reaction temperature. For example, increasing reaction temperature. For example, when reaction
when reaction temperature is 500 K, and v0 is 325 L/s, Igen temperature increases from 500 to 600 K, and the outlet
of PFR is 4.8 101 PEI/h, while for CSTR, Igen = 1.1 concentration CB,out = 0.007 mol/K, and v0 is also 325 L/s,
103 PEI/h, which is 2.3 times that of PFR. Igen of PFR and CSTR would decrease from 3.2104 and
3.5104 to 3.7104 and 4.2104 PEI/h, respectively.
4.3.2. The effect on PEI The effect of reaction temperature on Igen can also be found
From Eqs. (9), (21) and (25), we obtain the formulas for from Figs. 2 and 3.
calculating PEI . Suppose v0 is 325 L/s, then for PFR: When not considering the PEI of Cl2 , from the PEI

Igen 13.0 CB rPEI rate-law expression, expressed in Eq. (29), we can find that
PEI = = dCB (31) in the initial stage of the reaction, the determining part
v0 CC,out MC 103 CC,out CB,0 rB
in the right side of Eq. (29) is 1.1 103 k1 CA CB +
for CSTR: 0.049k3 CA CB , due to the concentration of C is very low
Igen 13.0 rPEI while the concentrations of A and B are relatively high.
PEI = = (CB CB,0 ) Comparing the magnitudes of the two activation energies
v0 CC,out MC 103 CC,out rB
in the two terms, we conclude that the value of this part
decreases as the reaction temperature increases. Thus, in
where CC,out is the outlet concentration of C. the early stage of this reaction process, to lower the value
According to Eqs. (31) and (32), using the curves about of rPEI , the reactor should operate at relatively high tem-
rPEI /rB versus CB and corresponding area, the superior- perature. As the reaction proceeds, the concentration of C
ity in environmental index PEI about PFR and CSTR is increasing while CA and CB is decreasing, so that the
can be discovered. When l = 1, the outlet concentration influences of the third term, 0.33k2 CA CC , is becoming sig-
CC,out = 0.011 mol/L and reaction temperature T = 500 K, nificant. Since the magnitude of activation energy in this
using CSTR will obtain lower PEI than using PFR. In term is the maximum, operating at relatively low temper-
this case, for CSTR, PEI = 31 PEI/kg; for PFR, PEI = ature would result in a relatively low value of rPEI in this
26 PEI/kg. Nevertheless, if not considering A , for the stage. Accordingly, for the reaction system in this case, a
same CC,out and T, using PFR will obtain lower PEI PFR operating at higher temperature first and then lower
than using CSTR, for PFR, PEI = 0.30 PEI/kg, while for temperature must generate lower value of Igen than that sim-
CSTR, PEI = 0.80 PEI/kg, which is 2.6 times that of PFR. ply operating at a corresponding uniform temperature. The
Q. Chen, X. Feng / Chemical Engineering and Processing 44 (2005) 1321 19

curves in different temperature in Fig. 2 intersect each other, the total PEI per mass of product streams leaving the system,
this fact also indicate this result. For example, when the inlet Iout , are another category of indices to characterize some
and outlet concentrations of B are 0.024 and 0.007 mol/L, aspects of emission of PEI from a manufacturing process.
respectively, and v0 is 325 L/s, for PFR operating at reaction According to Eq. (1):
temperature of 450K, Igen = 3.9 101 PEI/h, and for PFR
Iout = Iin + Igen (33)
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 Iout is obtained by dividing Iout by the output of products to
second stage operates at 450 K, the inlet concentration of B give:
of the second stage is 0.018 mol/L, Igen = 3.4 101 PEI/h.
Iout Iin + Igen
Obviously, it is 13% lower than that of the PFR simply Iout =  =  (34)
operating at reaction temperature of 450 K, and 49% lower p p p p
than that of the PFR simply operating at 530 K.
Substituting Eq. (9) into Eq. (34), we obtain:
4.4.2. The effect on PEI Iin
Iout =  + PEI (35)
Known from Eqs. (31) and (32), rPEI /rB can be used to p p
study the effect of the reaction temperature. When l = 1,
it is not difficult to find that increasing reaction temperature For the case study, it is not difficult to obtain that Iin =
would decrease the values of rPEI /rB , accordingly the value 4.6 104 PEI/h, i.e. Iin is a constant. Therefore, when the
of PEI also decreases. For example, for CSTR operating at feed condition of a reaction process is fixed, back-mixing
500 K, if its outlet concentration CC,out = 0.011 mol/L, and and reactiontemperature affect Igen and Iout in the same
v0 is 325 L/s, the value of PEI is 31 PEI/kg, while for way. Since p p = v0 CC,out MC 103 , it is not difficult
CSTR operating at 530 K and with the same CC,out , the value to find that, when the feed condition and CC,out are spec-
of PEI is 36 PEI/kg, that is, this CSTR reduced the value ified, back-mixing and reaction temperature influence Iout
of PEI by 16%. Similar to this example, for the case of not and PEI in the same way, too.
considering the potential environmental impact of Cl2 , one Thus, in Section 4.3, when l = 1 and the reaction pro-
can also conclude the effect of temperature on PEI . cess is operated in the specified conditions, the results that
Igen and PEI of CSTR are less than those of PFR indicate
4.5. The effect of concentration on environmental that using CSTR can obtain lower Iout and Iout than us-
performance ing PFR. That means at the given conditions, back-mixing
make the emission of potential environmental impact de-
We here only study the effect of concentration on PEI . crease. By contrast, when not considering the PEI of A, the
Similar to previous method in Section 4.4.2, one can also use results are converse, i.e. using PFR would gain lower PEI
rPEI /rB to study the effect of concentration. From Eqs. (31) generation rate than using CSTR; and the value of PEI of
and (32), it can be found that for a particular outlet concen- CSTR that operate at reaction temperature 500 K and out-
tration of C the value of PEI would be decreased by in- let concentration CC,out = 0.011 mol/L is 2.6 times that of
creasing the value of rPEI /rB . When l = 1, from Eq. (28), PFR. The results indicate that at the given reaction condi-
the value of rPEI /rB can be increased by augmenting CC and tions, back-mixing will increase the emission of potential
decreasing CB within a reactor. Since C is a product whereas environmental impact of the reaction processes, accordingly,
B is a reactant, using CSTR will lead to higher CC and lower reactor designers should try to avoid it in such case.
CB within a reactor than using PFR. Therefore, in this case, In Section 4.4, when l = 1, the results show that a higher
using CSTR would result in lower value of PEI than using reaction temperature is favorable for the decrease of Igen
PFR operating at the same reaction conditions. When not and PEI in the giving conditions. Therefore, increasing the
considering the potential environmental impact of Cl2 , from reaction temperature will lead to the decrease of Iout and
Eq. (30), decreasing CC and increasing CB within a reactor Iout . When not considering the PEI of Cl2 , PFR operating
will increase the value of rPEI /rB , accordingly, using PFR at a high temperature in the first section and low tempera-
would result in lower PEI than using CSTR. These results ture in the remainder section can result in lower generation
are also concluded in Section 4.3.2. and emission of PEI than that operating at a corresponding
uniform temperature.
To study the effects of concentration on environmental
5. Discussion performance of reaction processes, two analyses are intro-
duced in Section 4.5. The results show that when l = 1,
In the case study, the indexes involved are Igen and PEI . operating at relatively low concentration of B and high
However, these two indexes only express the situation of concentration of C is favorable for decreasing the poten-
PEI generation within processes. The environmental perfor- tial environmental impact created in manufacturing a unit
mance of a process also relates to its emission of PEI. There- mass of the product C. This also is true for total PEI leav-
fore, in the WAR algorithm, the output rate of PEI, Iout , and 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 into this methodology as they become available. It is suggest
contrast, when neglecting the PEI of Cl2 , the results are that either the WAR algorithms assessment method and the
converse. PEI indexes, another similar method and indexes, or even a
In the cases study, we have only studied the effects of more sophisticated and comprehensive method and indexes
back-mixing on the environmental performance. Whereas be used to this pollution reduction methodology.
the phenomena of back-mixing is only one of engineering In the case study, the calculations of the impact indexes
factors, which still involve pre-mixing, heat and mass trans- should be accurate to no more than one significant figure
fer, thermostability and so on. However, the effects of those [12]. But two significant figures are used in order to allow
engineering factors not involved in the case study on envi- readers to reproduce the calculations if necessary. It is also
ronmental performance can also be simulated and therefore worth noting that, even being accurate to one significant
well studied through a combination of reactor mathematical figure, one can still discern significant differences in the
model, which relates to corresponding engineering factors, index between different cases.
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- 6. Conclusions
fect the environmental performance of reaction processes
significantly in some cases, can also be studied with the To perform pollution prevention in the process of reac-
help of corresponding PEI balances and the PEI rate-law tor design, studying the effects of reaction conditions such
expressions, because we can ultimately simulate the change as temperature, pressure and concentration as well as vari-
of the environmental indexes with these reaction conditions ous engineering factors on environmental performance of a
through PEI balance and PEI rate-law expression. reaction process is very important. In this paper, potential
The reaction system involved in the case studied is a kind environmental impact balance and PEI rate-law expression
of relatively complex van de vusse reaction, nevertheless the are proposed as a tool to be used to carry out this work. The
reaction system in real manufacturing process may involve form of PEI rate-law expression should be consistent with
more reaction types and, therefore, is more complex than the method for calculating the overall PEI of mixtures of
that one. One can, however, simulate the change of envi- chemicals inside reactors. The effectiveness of the method
ronmental indexes within a reactor by combining traditional has been demonstrated through the reaction of allyl chlo-
reactor mathematical model with the PEI balance, and may ride production. From the processes of the analyses in the
also discover the effects of reaction conditions and engineer- case study, analyzing with PEI balance and PEI rate-law ex-
ing factors on environmental performance by PEI rate-law pression combining with other corresponding equations and
expression and/or combinations it with other reaction rate some mathematic knowledge can give important insights
equations as well as other related equations in reactor math- into how reaction conditions as well as engineering factors
ematical models. affect the environmental performance of reaction processes.
Note that assuming l = 1 and not considering the PEI of This kind of study is one of the bases for developing chem-
Cl2 are just for illustrating the algorithm, these assumptions ical reactors with environmental benign processes.
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]. Acknowledgements
In the case study, the quantitative environmental impact
assessment and the PEI indexes are cited from the WAR al- Financial support provided by the National Natural Sci-
gorithm, it is worth noting that, however, quantitative impact ence Foundation of China under Grant No. 20176405 and
assessment and environmental indexes are likely to improve the Major State Basic Research Development Program un-
with time, and that these improvements can be incorporate 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)
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)

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


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

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

0255-2701/$ see front matter 2004 Elsevier B.V. All rights reserved.
24 T.A. zbelge et al. / Chemical Engineering and Processing 44 (2005) 2332

the hydroxide ions present, metals precipitated as hydrox- Table 1

ides. Furthermore, increase in metal inhibition with decrease Composition of synthetic wastewater
in pH was attributed to the increase in free metal ions in the Constituents (mg/L)
medium [10,11]. Toxicities of HMs are also affected by cell Proteose-peptone 1221.7
residence time in ASP. Nelson et al. [12] reported that toxic NaCl 407.4
effects of cadmium and zinc decreased considerably by in- Na2 SO4 44.6
creasing cell residence time from 1 to 15 days in ASP with K2 HPO4 44.6
recycle. Uloth and Mavinic [13] explained this decrease with MgCl2 .6H2 O 3.7
FeCl2 .2H2 O 3.7
the increasing production of bacterial extracellular polymer CaCl2 .2H2 O 3.7
substances (EPS) at high mean cell residence times. Some MnSO4 0.057
investigators [9,14,15] reported that an increase in substrate H2 MoO4 0.031
and mixed liquor suspended solids (MLSS) concentrations NaOH 0.008
led to increase in metal uptake and decrease in adverse ef- ZnSO4 0.046
CoSO4 0.049
fects of HMs to ASP; they attributed this decrease to the CuSO4 0.076
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 chemical oxygen demand) of a COD of So = 1300 mg/L
of ASP and the combined uptakes of these metals by acti- was fed, in which Proteose-peptone (1221.7 mg/L) was the
vated sludge have been reported [16], as have the effects of sole carbon and nitrogen source. Other minor nutrients were
copper and zinc on activated sludge bacteria/growth kinet- also added in sufficient quantities to make carbon the limit-
ics [17,18]. However, combined uptakes of only copper and ing nutrient (Table 1). Solutions of CuSO4 and ZnSO4 were
zinc by sludge, and the effect of hydraulic residence time in added to the wastewater to provide Cu2+ and Zn2+ ions at
this process have not been reported, although Vian et al. [16] the desired concentrations. Further details are given else-
discussed the effects of long sludge age (kept almost con- where [17,2123].
stant at 9 days) on the combined uptakes of copper, nickel
and zinc in an ASP with recycle. In another study [19], 2.2. Experimental method
transfer of organic matter from wastewater (WW) to acti-
vated sludge was investigated and observed that the transfer Biomass obtained from the wastewater treatment plant
of soluble fraction always occurred from wastewater to ac- of Middle East Technical University (METU) was used as
tivated sludge, whereas bidirectional transfer occurred for seed in the experiments. In the acclimation of MOs to HMs,
the colloidal fraction. The biosorption capacity of activated maximum concentrations of Cu2+ and Zn2+ reached dur-
sludge was found to be at a ratio of chemical oxygen de- ing acclimation were decided from previous work [17,21].
mand (COD)/TSS of about 10400 mg/L, and biosorbable The literature showed that when these metals were present
fraction of wastewater was almost 45% of non-settleable individually in the system, ASP failed to operate at steady
(colloidal + soluble) fraction [19]. A single metal (Zn2+ ) state values of 15 mg/L for Cu2+ and 90 mg/L for Zn2+ , re-
study was reported [20], where zinc biosorption by activated spectively [5]. Therefore, one-third of threshold concentra-
sludge was more effective than by digested sludge in remov- tions of these metals (that is, 5 mg/L for Cu2+ , and 30 mg/L
ing zinc from metal plating effluent. The objective of this for Zn2+ ) were used to constitute the combined threshold
work is to investigate the effect of hydraulic residence time concentration for mixtures of copper and zinc ions. Accli-
on combined uptake of Cu2+ and Zn2+ by activated sludge mation experiments were performed by gradually increasing
in once-thro ASP (OTASP) without recycle, which is not the concentrations of Cu2+ and Zn2+ to two, three and four
encountered in the available literature. 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
2. Experimental values, was obtained at the concentrations corresponding to
twice their combined threshold concentration [21].
2.1. Experimental set-up Experiments without HMs were conducted using syn-
thetic WW inoculated with 10% (v/v) acclimated seed MOs
A modular continuously stirred glass tank reactor (from to twice their combined threshold concentration (that is,
Cole-Parmer, Inc., USA) with an effective volume of 2 L was 10 mg/L Cu2+ , and 60 mg/L Zn2+ ). The reactor was op-
used in the OTASP. Peristaltic pumps were used at the inlet erated as a batch system for 48 h with the growth of new
and at the exit of the reactor for feed/harvest, respectively. biomass. Then, continuous feeding of synthetic WW to the
Aeration and efficient mixing was provided in the reactor; reactor at a desired volumetric flow rate (or desired resi-
temperature and pH were controlled at 251 C and 70.01, dence time) was started. The system reaches steady state
respectively, by electronic controllers. A synthetic wastew- in about 1244 days depending on the influent metal con-
ater with known inlet substrate concentration (in terms of centrations and the hydraulic residence time (HRT); effluent
T.A. zbelge et al. / Chemical Engineering and Processing 44 (2005) 2332 25

samples from the reactor were analyzed for substrate, S, and Secondly, concentrations of the HMs associated with the
biomass, X, concentrations. MLSS concentration was used solution, and solid phases of the effluent were determined.
as the primary indicator for steady state conditions. At each Two of 50 mL aliquots were collected from the effluent of
residence time, steady state was reached when the effluent the reactor, centrifuged at 1300 rpm for 15 min, by using
S (COD) and X (MLSS) concentrations remained constant a RC5C type Dupont centrifuge, to separate the solution
with respect to time. and solid phases, and the supernatants transferred to sam-
Experiments with HMs were performed at two concentra- ple tubes. Then, supernatants and the residual solid phase
tion levels of each metal (low levels: (Cu2+ ) = 1.5 mg/L were acidified by adding 10 mL concentrated HNO3 to each
and (Zn2+ ) = 9 mg/L; high levels: (Cu2+ ) = 4.5 mg/L sample tube. Samples were then analyzed by AAS after per-
and (Zn2+ ) = 27 mg/L, for different values of HRT in the forming the necessary dilutions to the sensitivity range of
range of 240 h using the same synthetic WW having the the instrument as explained above. The average of the results
composition given in Table 1. In these experiments, 1.8 L from the two samples were recorded as CS (metal concen-
of synthetic WW was inoculated with 0.2 L of acclimated tration in the solution phase), and CA (metal concentration
seed MOs in the activated sludge reactor, and the system in the solid phase) values for each HM.
was operated in batch mode for almost 72 h to ensure the The distribution of different metal forms in the sludge
growth of biomass by observing the increase in measured was determined by using the sequential extraction pro-
turbidities of the samples from the reactor. Then, the con- cedure [25]. Activated sludge mixed liquor of 1.5 L,
tinuous operation was started by feeding the reactor with corresponding to 0.2500.100 g of sludge, was collected
synthetic WW containing copper and zinc ions at the de- from the outlet of the reactor at the steady state. In order
sired inlet concentrations. Influent flow rate was adjusted to separate the sludge and supernatant, centrifugation at
in order to provide a required HRT. Samples were taken 1300 rpm for 15 min was carried out using a RC5C type
from the effluent stream at different times in order to deter- Dupont centrifuge. Then the extraction process was started
mine the effluent S and X concentrations until steady state by adding the reagents of KNO3 , KF, Na4 P2 O7 , EDTA
was reached. Then, a sample of 1.5 L from effluent stream (ethylene-diamine-tetraacetic-acid) and HNO3 on the con-
was collected for the metal analyses, and the operation was centrated sludge (order of addition given in Table 2), which
stopped. are capable of extracting exchangeable, adsorbed, organi-
cally bound, carbonates, and sulfides forms of the metals
2.3. Analytical methods in the sludge sample, respectively. To provide complete
extraction, samples were shaken on a platform shaker, then
Substrate concentrations were determined in centrifuged centrifuged at 1300 rpm for 15 min, and supernatants were
sample aliquots by standard COD analysis, and gravimet- decanted into separate tubes. Between each extraction, the
ric method was used to measure the biomass concentrations sludge residue was washed with 50 mL of de-ionized water
[24]. Analyses were repeated at least twice with the sam- by shaking for 30 min, and the samples centrifuged again
ples under the same experimental conditions and the average under the same conditions. Both supernatants from the ex-
values were obtained. traction and washing were combined and acidified with two
drops of 5 M HNO3 before metal analysis, in order to pre-
2.3.1. Distribution of metals among the constituents of serve the solutions. All metal concentrations were measured
activated sludge mixed liquor as before and conversions were made to obtain the concen-
Since metals were distributed between the two phases of trations of each form of each metal as mg metal/ kg sludge,
mixed liquor in the reactor, namely solution and biomass, the for which the metal concentration as mg/L was divided by
amounts of copper and zinc associated with these phases and the dried MLSS concentration (kg TSS/L), determined at
the forms of these metals in the biomass were determined the steady state. The same procedure was followed for the
at steady state condition for each HRT. other reagents, the only exception being that double extrac-
Firstly, the concentrations of HMs in effluent were de- tion was performed with EDTA and supernatants pooled
termined. At least three samples of 40 mL were acidified before the metal analyses. The remaining solid residue in
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 Table 2
9200X-AAS) (diluted 5 times for Cu2+ analysis, and 25 Details about sequential extraction process parameters [25]
times for Zn2+ analysis) and then analyzed. The average of Reagents Molarity pH Extraction (Solution/solids)
these measurements yielded the total HM concentration in (M) time (h) ratio
the effluent, CT (mg/L), for that specific experimental run. KNO3 1 24 50/1
To convert (mg metal/L) to (mg metal/kg sludge), the mea- KF 0.5 6.5 24 80/1
sured metal concentration (as, mg metal/L) at each HRT Na4 P2 O7 0.1 24 80/1
was divided by the steady state MLSS concentration (as, kg 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 ]
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 ]
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
h (h) CT CS CA [(CS + CA ) 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

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 ),
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
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
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
Organically bound Adsorbed
Carbonates 160 Organically bound
700 Sulfides Carbonates
140 Sulfides
mg metal removed / kg sludge

mg metal removed / kg sludge


400 80

300 60


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-
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
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 29

400 fects, a problem which can be overcome if a recycle is used

in the system. Therefore, further studies are planned to be
Organically bound performed in an ASP with recycle.
Carbonates In the literature, equilibrium biosorption of Zn2+ and
Sulfides Cu2+ by nonviable activated sludge was investigated in a
300 packed column adsorber [26], where metabolically induced
mg metal removed / kg sludge

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-
100 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-
0 tion equilibria. The simultaneous sorption of metal cation
0 10 20 30 40 and sulfate anion was probably occurring at pH of 3.0, while
residence time (h) proton exchange predominated at pH of 3.8. They pointed
out the usefulness of non-viable activated sludge as a biosor-
Fig. 6. Distribution of Cu2+ among its forms in the sludge using the data
of Set-2 experiments.
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.
amount of the metals determined by the extraction proce- It was reported that the presence of Cu2+ ions in wastewater
dure should be equal to the amount of metal bound to the inhibited the net specific growth rate of autotrophic biomass
sludge, the discrepancy between the two being denoted by to a larger extent than Zn2+ ions [27]. Therefore, in an ASP
CH . Thus, the material balance for extraction can be shown treating such a WW, it is expected that the concentration
as: and the composition of EPS (consisting of protein, polysac-
CA = CE + CH (2) charide and DNA) produced by microbial cells will vary
[28], since it is known that the EPS formation rate is pro-
where CA is the concentration of metal bound to the sludge portional to the rate of substrate utilization, thus to the rate
(mg/L, or mg metal/kg sludge), CE is the concentration of biomass production [29]. When EPS are not involved in
of metal extracted in the analysis (mg/L or mg metal/kg the process, MOs may not easily display high flocculation
sludge) and CH is the concentration of metal not extracted and metal uptake capacity, as the floc forming bacteria can
by the used reagents in the extraction (mg/L or mg metal/kg interact with each other [30]. However, the precise role of
sludge). Thus, the extraction procedure used is capable of EPS in biosolidliquid separation is not well understood al-
extracting most of the metals bound to the biomass, varying though EPS are thought to be of considerable importance in
between 93.85 and 98.98% of the total amount of the HMs bioflocculation and settling of activated sludge [28]. For ex-
in the sludge. Also, the total concentrations of the metals ample, the EPS content has been shown to be directly [31],
extracted are mostly consistent with the CA values. The ex- or conversely [32] related to sludge volume index (SVI).
perimental error in the sequential extraction analysis falls Besides, the low protein levels were obtained at high car-
within 10%. bohydrate concentrations depending on microbial consortia
It should be noted that these data were obtained by ex- due to the intracellular storage of excess carbon substrate and
tracting the relatively small amounts of biomass, which can lower sludge retention time, particularly, in the pulppaper
exagerate the experimental error. The reasons for working and textile industries lab-scale activated sludge systems.
with small amounts of biomass are two-fold; firstly, the sam- The low protein content was probably due to the inhibition
ples were collected here at the steady state directly from the of EPS production in the aforementioned industies as re-
effluent of the system at one day intervals. Samples can de- ported by Dignac et al. [33]. It was noted that the wastewater
generate due to the continuing interaction between metals composition rather than operating conditions had a strong
and biomass before the analysis, which can cause changes influence on EPS composition and the physicochemical
in the relative amounts of the metal forms in sludge over properties of flocs in an ASP. For example, if the ASP was
the time. Secondly, biomass concentrations in the effluent of operated at high F/M ratios, both carbohydrate and DNA
the reactor are quite low anyhow due to metal inhibition ef- 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) than the amount of organically removed copper at a HRT of
and less negatively charged floc surface was observed [28]. 40 h, while it was lower than that of organically bound cop-
The present OTASP was operated at high F/M ratios [23]; per at a HRT of 20 h. Therefore, it is shown in Fig. 2 that
although no data were obtained about the concentration and the copper uptake at a HRT of 40 h is less than that at 20 h.
the composition of EPS produced in this work, it might be The copper uptake is probably less at a HRT of 40 h due to
predicted from the available literature that the protein con- the effect of changing metabolic activity of microbial flocs
tent of the produced EPS in the flocs was low; thus the set- and the solubility constant of copper carbonate complex in
tling ability of the resulting sludge, most probably with a the medium at the prevailing conditions. The desorption of
high value of SVI was not good. Actually, low settlability of copper into the solution might have occurred due to the sol-
the produced sludge was observed in the experiments. The ubility constant of this complex.
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 4. Conclusions
residence time (SRT), HRT, EPS and F/M ratio on the set-
tlability of produced sludge and metal uptake. In this study, The effect of hydraulic residence time (in the range of
the change in bacterial profile of the MLSS with metal con- 2.540 h) on the removal of Zn2+ and Cu2+ ions by acti-
centration was not followed, but it was reported in the lit- vated sludge was investigated in a once-thro activated sludge
erature [34] that many different species of bacteria isolated process (OTASP) at a constant pH of 7 and temperature of
from activated sludge were shown to produce EPS. These 25 C. It was found that the percentage removals of both cop-
polymers might be in the form of loose slime or capsules per and zinc by activated sludge increased with the increas-
adhered to the cell wall. In activated sludge, slime polymers ing residence time due to the increasing amounts of sludge
remained in the dissolved and colloidal phases of an efflu- produced with time. Higher percent removal of zinc at the
ent, while capsular polymers remained attached to the flocs low level of influent concentrations of both the copper and
and hence settled with the sludges. Some activated sludge zinc, (Cu2+ ) = 1.5 mg/L and for (Zn2+ ) = 9 mg/L, due
bacteria did not produce capsules or extracellular material. to the higher inlet concentration of zinc over that of copper
In these bacteria, metals might be accumulated either in the was replaced by the higher percent removal of copper at the
cytoplasm or by adsorption on to the cell wall. Some re- high level of influent concentrations of each metal, (Cu2+ )
searchers [35] concluded that only the half of the observed = 4.5 mg/L and for (Zn2+ ) = 27 mg/L, although copper is
metal uptake could be attributed to EPS and that cell itself known to be more toxic to the microorganisms than zinc.
played a role in accumulating metals. In another work [36], This suggested that the lower toxicity of zinc at its low con-
EPS were extracted, cells were broken up using ultrasound centration level is compensated by another factor such as
and cell envelopes were sedimented by centrifugation. Only the removal of both metals by precipitation also contribut-
small quantities of copper were found in the cytoplasm, but ing to the removal process. However, it was not possible
substantial amounts of copper were bound to cell envelopes to differentiate between the percent removals of each metal
and EPS. The metabolism of elements occurred in many by biosorption or by precipitation in OTASP. The previous
species of bacteria. It was also reported that four major types study [16] cannot render this information due to the presence
of metal transformation in organisms occurred [37], these of a third metal ion such as Ni2+ in the system. Therefore, a
being chelate formation by the binding of metals to organic further study is required in an activated sludge system with
ligands, shifts in metal valencies, substitution of one metal recycle which will differentiate the percent removals of each
for another and biomethylation of metals to microorganisms. metal by precipitation and biosorption clearly. In this study,
When metals were chelated to substrate molecules, the ad- at short residence times (a range of 2.55 h), the percent-
sorption of these molecules might represent a means of accu- age of zinc bound to the biomass is higher than for copper
mulating metals within a cell. These metabolic transforma- due to the dominating effect of lower toxicity of zinc; this
tions of metal ions by bacteria were different for each metal is also shown by the sequential extraction results yielding
and for different combinations of metals in the medium. the dominant form of both Zn2+ and Cu2+ as organically
Therefore, in such a dynamic system as is ASP, it is not bound indicating that both types of ions are removed by
plausible to use metal uptake as a measure of EPS content. biosorption rather than by precipitation due to the lower de-
As it is seen in Fig. 2, the copper uptake increases with the gree of inhibition of MOs at small sludge ages or HRTs in
HRT up to 20 h then it shows a slight decrease between 20 OTASP. This point can be clarified by investigation in an
and 40 h. This can be explained with the change in the given activated sludge process with recycle. It is also found that
ranking in Fig. 6. The chemical speciation and distribution the least represented forms of both the copper and zinc ions
of copper in an activated sludge slurry can be quantified in the sludge are their sulfides. A further study is under-
using an equlibrium constant approach including complex- way to investigate the effect of operational conditions on the
ation with inorganic and organic ligands and adsorption by composition of extracellular polymer substances in an ASP
activated sludge solids [38]. It can be observed in Fig. 6 that with recycle to compare the effects of sludge residence time,
the amount of copper carbonate per kg of sludge is higher 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)
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, [33] M.F. Dignac, V. Urbain, D. Rybacki, A. Bruchet, D. Snidaro, P.
The inhibitory effects of heavy metals and organic compounds on Scribe, Chemical description of extracellular polymers: implications
the net maximum specific growth rate of the autotrophic biomass in on activated sludge floc structure, Water Sci. Technol. 38 (1998)
activated sludge, J. Hazard. Mater. B100 (2003) 271283. 4553.
[28] D.T. Sponza, Investigation of extracellular polymer substances (EPS) [34] M.J. Brown, J.N. Lester, Role of bacterial extracellular polymers in
and physicochemical properties of different activated sludge flocs metal uptake in pure bacterial culture and activated sludge, Water
under steady-state conditions, Enzyme Microb. Technol. 32 (2003) Res. 16 (1982) 15491560.
375385. [35] B.A. Friedman, P.R. Dugan, Concentration and accumulation of
[29] C.S. Laspidou, E.R. Bruce, A unified theory for extracellular metallic ions by the bacterium Zoogloea, Dev. Ind. Microbiol. 9
polymeric substances, soluble microbial products, active and inert (1968a) 381388.
biomass, Water Res. 11 (2002) 27112720. [36] W.A. Corpe, Metal-binding properties of surface materials from ma-
[30] J.A. Scatt, J. Karanjkar, Immobilized biofilms on granular activated rine bacteria, Dev. Ind. Microbiol. 16 (1975) 249255.
carbon for removal and accumulation of heavy metals from contam- [37] A. Jernelv, A.L. Martin, Ecological Implications Of Metal
inated streams, Water Sci. Technol. 38 (1998) 97204. Metabolism By Micro-Organisms, Swedish Water and Air Pollution
[31] V. Urbain, J.C. Block, J. Manem, Bioflocculation in activated sludge: Research Laboratory, Stockholm, 1975, p. 32.
an analytical approach, Water Res. 27 (1993) 829838. [38] B.R. Fristoe, P.O. Nelson, Equilibrium chemical modelling of
[32] J.A.S. Goodwin, C.F. Forster, A further examination into the com- heavy metals in activated sludge, Water Res. 17 (1983) 771
position of activated sludge surfaces in relation to their settlement 778.
characteristics, Water Res. 19 (1985) 527533.
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


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

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

0255-2701/$ see front matter 2004 Elsevier B.V. All rights reserved.
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
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
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
10000 rmax 1  yj 1
DeffA = + f(r) dr
rmin 1 yA DAj DAk (r)
12 where is the tortuosity factor of the KD306 catalyst, which
was measured using single pellet string reactor technique.
Kp,methanantion,atm -2


2.3. Thermal conductivity eff [11,14]

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
eff fm
s s
673 723 773 823 873
0 where s is given by
T, K
s = 19.90005 + 0.12333T 2.38169( 100
Fig. 1. Reaction equilibrium constants of methanation reaction and shift
+ 0.151983( 100
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
(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
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
Applying orthogonal collocation method, the above equa- = p (rCO2 )w
tions become Deff,CO2
 2 2
2 T 1 T Rp T


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


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


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

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

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

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

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


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

need to develop customized tuning rules that speed up the
Proportional integral derivative (PID) controllers are ex- controller tuning procedure by minimizing the time needed
tensively used in the process industry. It was recently esti- for retuning or even eliminating retuning.
mated that more than 90% of all control loops involve PID The aim of this paper is to develop mathematical pro-
controllers [1]. The simplicity and transparency of PID con- gramming formulations for the development of tuning rules
trol mechanism, the availability of a large number of highly for PID controllers with arbitrary design objectives posed in
efficient, reliable, and cost-effective commercial PID con- the time domain. The resulting mathematical formulation is
trol modules and their acceptance from the operators are a nonlinear programming problem (NLP). A reformulation
among the reasons for their success [2]. Tuning of the PID of the problem is proposed that transforms the formulation
controllers is achieved in two steps. In the first step a simpli- to a mixed-integer linear problem. If the objective function
fied model of the process is obtained and initial tunings are chosen is linear (convex in general) then the resulting for-
calculated using well-known tuning rules (such as the mod- mulation can be solved to global optimality using currently
ified Ziegler-Nichols [2] or IMC-based [3]). In the second available software. The main advantage of the proposed
step, the performance of the loop is assessed and the initial methodology stems from its generality since it can be ap-
tunings are modified in order to take into account the oper- plied to obtain PID controllers tunings for any linear system.
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 2. PID controller tuning
most cases, cannot be translated into the design objectives of
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.
42 G. Syrcos, I.K. Kookos / Chemical Engineering and Processing 44 (2005) 4149

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


uk = uk1 + uk


min = J[xk , uk ], s.t. uk = kC ek min = J[xk , uk ],

yk yk ykU

xk+1 = Axk + buk + Evk

uLk uk uU

yk = Cxk + duk + Fvk

u uk uU

ek = rk yk


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

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
controller gains.

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

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


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



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

uL u uU

Furthermore, we define q binary variables {0, 1} to
{0, 1} ,
denote the selection of tuning factor ( = 1) or not ( =
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
 if the objective is quadratic or to a mixed integer linear pro-
= 1 (7)
gramming (MILP) problem if the objective is linear. In this
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

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

Ik = S ek (15) m = kC,0
m m
, , m (22)
1 ,k = P Pk , , k (23)
Dk = (ek 2ek1 + ek2 ) (16)
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
K = kC ,
K = I
, K = kC D
(18) following linear form:

P q

I q


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


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


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

,k D

Pk = ek ek1 , , k

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

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

S Dk = ek 2ek1 + ek2

Depending on the form of the objective function, the final


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

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

uL u uU

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

k k

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

u uk uU

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

anteed using currently available software. The same holds


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,
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


3. Case studies -1.5

In order to apply the proposed methodology the control 0 50 100 150
Time (min)
engineer needs to supply the model of the process under
investigation in discrete form (see Eq. (1)), an initial con- Fig. 1. Load response of the closed loop IPDT system (proposed: con-
troller tuning (that can be obtained using any classical tech- tinuous line, Luyben [10]: dotted line).
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 Figs. 1 and 2 compare the closed loop behavior obtained
MINOS solvers was used to solve the proposed formulation. using the tuning proposed in this work and the tunings
All results were obtained on an Intel P4 2.66 GHz/512 Mb proposed by Tyreus and Luyben [10]. Fig. 1 shows that a
computer and the computational time was of the order of significant improvement in the closed loop load response is
few seconds. obtained when the proposed methodology is used. The value
of the objective function (Eq. (31)) is = 59.4 for the pro-
3.1. Control of integrating processes with time delay posed and = 197.7 for the Luybens tunings. Fig. 2 com-
(IPTD) pares the two methodologies for a unit step set point change.
The responses are comparable and the proposed method
Integrating processes with time delay are among the most results in = 28.6 while for Luybens tunings = 29.7.
commonly encountered processes. Liquid levels, for in-
stance, exhibit integrating behavior. A typical plant involves 3.2. Systems with inverse response and time delay
a large number of liquid level loops whose performance
affects strongly the overall performance of the regulatory Luyben [14] has recently investigated a class of mod-
control systems. The transfer function of an integrating els that involve both inverse response and time delay. More
process with time delay is the following: specifically he has considered the following type of pro-
kP ds cesses:
g(s) = e (30)
1 z s ds
PID controller tunings for this model have been given g(s) = kP e (33)
(1 + P s)2
by a number of researchers [913]. Chen and Fruehauf
[9] have given an industrial example of the level control that have a positive zero at s = +1/ z as well as time a
in a distillation column where the open loop dynamics delay d. Using the objective function given by Eq. (31) with
follows the IPTD model with parameters kP = 0.2 and = 1/ S and S = 0.1 we obtained
d = 7.4 min.
The proposed methodology has been applied using the KP = 0.356, KI = 0.154 (34)
following objective function:
 for a hypothetical system with z = d = 1.6, kP = 1 and
min = S (|ek | + | uk |) (31) P = 1. A unit step disturbance at t = 0 min was assumed
KP ,KI followed be a unit step change in the set point at t = 50 min.
Fig. 3 compares the closed loop response obtained using the
where NS = f = 300 min. A unit step disturbance at t =
proposed tunings with the tunings given by Luyben [14].
0 min was assumed followed be a unit step change in the set
Luybens tunings correspond to = 19.5 while the ones
point at t = 150 min. The optimization problem was solved
proposed in this work to = 16.7. It is interesting to note
with = 1/ S and S = 0.2 min giving
that the proposed tunings result in improved load as well as
KP = 0.553, KI = 0.0178 (32) set point response.
G. Syrcos, I.K. Kookos / Chemical Engineering and Processing 44 (2005) 4149 45


y(t) 1


0 50 100 150
Time (min)


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


0 10 20 30 40 50 60 70 80 90 100

u(t) 0



0 10 20 30 40 50 60 70 80 90 100

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

y(t) y(t) 1
-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 0.5
u(t) u(t)
-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 4 6 8 10 12 14 16 18 20


u(t) -1


0 2 4 6 8 10 12 14 16 18 20

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 10 20 30 40 50

0 10 20 30 40 50

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 5 10 15 20 25 30

0 5 10 15 20 25 30

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

q =2 q =3 q =4 q =5
0 5 10 15 20 25 30
-0.8 q =2 q =3 q =4 q =5

0 5 10 15 20 25 30

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

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


0 10 20 30 40 50 60


0 10 20 30 40 50 60

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 49

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


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

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

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

0255-2701/$ see front matter 2004 Elsevier B.V. All rights reserved.
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
Fig. 1. Proposed reaction mechanism between PVA and AN for the
ammonium nitrate (CAN) (Merck) was used after vacuum
synthesis of PVA-g-AN copolymer.
drying. Acetic acid (Merck), dimethylformamide (DMF) and
dimethylsulfoxide (DMSO) (Merck) were used without fur-
ther purification. Other reagents used in the study were also downstream side was kept at 23 Pa by a vacuum pump
all Merck products. (Vacuu Brand RD 15, GMBH Co., Germany). Pervapora-
tion experiments were conducted at constant temperatures
2.2. Synthesis of PVA-g-AN ranging 2550 C using different feed compositions. Upon
reaching steady state flow conditions, the permeate vapor
The reaction was carried out in a three-necked flask was collected in liquid nitrogen traps.
equipped with stirrer, condenser and nitrogen gas inlet. Five The composition of permeates was deduced refractomet-
grams of PVA was dissolved in 95 mL of distilled water at rically by digital differential refractometer (Atago DD-5,
80 C. Fifteen grams of AN was added to the solution. Stir- Atago Co. Ltd., Japan).
ring was continued for 1 h at room temperature under N2
atmosphere. Ten mililiter portion of 1.0 102 M HNO3 2.5. Representation of the results
containing 10 mL, 1.0 102 M of CAN was added to the
reaction vessel and polymerization was carried out for a pe- The permeation and separation characteristics of
riod of 3 h at 30 C. The reaction was quenched with 10 mL PVA-g-AN membranes toward acetic acidwater mixtures
of aqueous ferrous salt solution (FeSO4 ) (5.0 102 M) were expressed as permeation rate (flux) (Q), separation
and pH of graft copolymer latexes were adjusted to 1.5 with factor () and pervaporation separation index (PSI).
1.0 M HCl solution, then dried in air at room temperature. The permeation rate, Q was determined by using the equa-
The product was extracted with water at 80 C then with tion
DMF at 55 C for 2 days to remove the unreacted PVA and Q= (1)
polyacrylonitrile (PAN) homopolymer. After the extraction, At
graft copolymer was dried in vacuum oven at 60 C [5,38]. where m, A and t represent the weight of the permeate (kg),
The proposed grafting mechanism between PVA and AN is
given in Fig. 1.

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
Separation factor was calculated using: Thickness (m) Q (kg/m2 h)

15 Low mechanical strength

= (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)


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)

permeation rate of grafted membranes at different grafting

ratio with respect to the feed composition was given. As it
seen from the graph although permeation rate decreases with
4.0 the amount of AN separation factors were almost the same.
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

Q(CH3COOH) (kg/m h)x10
Q (H2O) (kg/m h)





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






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

0.6 4.0

Q (kg/m h)



0.0 0.0
2 4 6 8
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 57

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


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

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

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

reactor (downflow of gas and liquid over randomly packed resolution of about 510% of the column diameter in the
particles). The apparatus was very similar to that being used measurement plane was claimed [912]. Mewes et al. [9]
in the Chemical Reaction Engineering Laboratory (CREL) used water which was introduced through a large number of
at Washington University in St Louis, albeit, with a higher injection nozzles into a 120 mm diameter reactor containing
number of detectors (32 of 1.27-cm-diameter BGO detectors monolith. With increasing liquid velocity, the liquid distri-
as against 9 of 5-cm-diameter Nal detectors) and increased bution improved. The authors did not demonstrate the effect
radioactive source strength (300 mCi as against 85 mCi). of gas velocity on the liquid distribution.
A comprehensive study was performed to quantify the error Toya et al. [13] used X-ray tomography to scan gas/liquid
associated with the measurements and the authors were able flow in a corrugated structured packing known as Sulzer
to detect solids holdup within 3% accuracy. The authors Mellapak 250 Y. The solids profile of the dry structured
also proposed a correlation to predict liquid holdup based packing was reproduced with fair accuracy. The scan was
on a force balance on liquid film and showed that tomog- performed in a 60-cm-diameter column. Very thin liquid film
raphy can reproduce the liquid holdup as predicted by the flowing past the structure at 0.006 m/s was also detected by
model for liquid flow rates up to 1.7 cm/s (no gas flowing). the tomography unit.
Yin et al. [15] investigated liquid distribution in a col- Iluita and Larachi [4] developed a hydraulic mechanistic
umn of 60 cm in diameter containing pall rings. The authors model for prediction of the irrigated pressure drop, liquid
found that the distributor type had a significant effect on the holdup and the effective interfacial area in the preloading
distribution, which was verified using computational fluid zone of structured packing operated in countercurrent flow
dynamics (CFD). of gas and liquid under partial wetting condition. The model
Examples of the use of tomographic techniques to deter- mimics the two phase flow situation by a network of double
mine gas/liquid distribution in structured packing is limited slits consisting of dry and wet slits. The model was success-
in the open literature. Mewes et al. [9] studied flow distribu- ful in predicting the irrigated pressure drop, liquid holdup
tion in monoliths using capacitance tomography. A spatial 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
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 some errors. In the new setup, the angular distance between
detector. two consecutive detectors is 0.2 , which is the same as
The new scanner introduced the following modifications the angle between two consecutive projections. Therefore,
(Fig. 3). there is no need to interpolate between the detector spaces
(Fig. 3b).
2.2.1. Collimator and detectors rotate simultaneously
In the old CT unit, the detectors are stationary in a partic- 2.2.3. Collimator slit width and radioactive source strength
ular view and the collimators move by a desired angle to ob- In order to achieve the best possible resolution, the thick-
tain projections for various chords. The intensity of radiation ness of the ray being attenuated should have infinitely small
as detected by the detectors is not uniform; it is maximum width. However, this is practically impossible since in that
when the collimator passes the center of the detector and case, we need to have an infinitely small slit width of the
minimum at the two edges. To overcome this non-uniformity detector collimator also. If the slit width is too small, the
in the new setup, the collimator position has been fixed at counts registered by the detectors will be very small, which
the center of the detector and the whole detectorcollimator in turn will affect the resolution. Generally, the maximum
rack is rotated (Fig. 3a). 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
2.2.2. Eliminating the need to interpolate between detectors used instead of 5 mm, which was the case in old CT unit.
In the old CT unit, the angular distance between two de- This is expected to result in a higher resolution of the order
tectors is 0.5 whereas a typical angle between two pro- of 2 mm (Fig. 3c).
jections is approximately 0.2 . Therefore, the source in- Moreover, the new CT along with all the associated elec-
tensity in the space between the detectors has to be in- tronics is mounted on wheels. Being completely made of
terpolated. Hence, two artificial projections are incorpo- aluminum instead of stainless steel, the whole assembly is
rated into the reconstruction process, which may lead to 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 63

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- by determining the solids fraction of the packing using com-
tal setups. puted tomography, which was compared to the one measured
The beam attenuation is measured along a number of by water displacement method.
beam paths through the column, which originate from dif- For the first case of a jar filled with water, the results are
ferent angles. Once a set of attenuation measurements is as follows.
completed, the density distribution image is reconstructed The collimator used in the present CT has an aperture of
by using the EstimationMaximization algorithm [5]. This 2 mm and therefore, theoretically, spatial resolution should
reconstruction algorithm is used, since it has the following be around 2 mm. In this test case, a 70 mm domain con-
advantages: (1) it accounts for statistical variations associ- taining 5.1 cm outer diameter glass jar filled with water was
ated with radiation measurements; (2) it readily incorporates reproduced (as shown in Fig. 4b). The diameter of the ob-
non-uniform beam effects; and (3) it ensures that the final ject obtained from CT was 5.3 cm, which indicates that we
reconstruction will contain positive values. To obtain statis- have a spatial resolution of 2 mm in this case. The attenua-
tically significant results, and to reduce the effect of position, tion in the water region should be uniform. In this case, the
the CT scans were obtained by scanning 360 around the scan gives a slightly non-uniform attenuation as shown in
column using a total scanning time of about five hours and Fig. 4c. The calculated error in the attenuation coefficient
collecting numerous beam path attenuations (approximately was found to be within 6.5%.
2500 projections). The final holdup distribution images are For the second case, a hollow ball was dipped inside a
obtained from the attenuation data using the relationships jar filled with water and scanned. The diameters of the ball
and the procedure developed by Chen et al. (2000) [6] for and jar were 6.98 and 18.95 cm, respectively, as shown in
packed bed. Fig. 5a. A scan area of 27 cm in diameter was reproduced
using the reconstruction algorithm. The dimensions of the
2.3. CT scan evaluation objects as reproduced by the scan were 6.97 and 19.20 cm,
respectively, as shown in Fig. 5b, which give a maximum
Since this was a new CT scanner developed to be used spatial error of about 2.5 mm. This is good enough to re-
for the first time, it was imperative to evaluate the perfor- solve relatively small maldistribution, if it exists, inside the
mance of the unit in terms of spatial and density resolution. 30.48-cm-diameter column used in this study. The figure
Two CT scans were performed to evaluate the system. In the shows that the overall error in the estimated total holdup is
first case, a 5.08 cm jar was filled with water and scanned within 12.8%.
(Fig. 4). In the second case, a hollow ball dipped in water It must be noted that the resolution changes with the size
was scanned and gas holdup was evaluated (Fig. 5). Further, of the object being scanned due to different attenuation of
a third validation, which will be discussed in the following the gamma rays. However, the obtained resolutions should
section of the Results and discussion, was also performed 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 This overall average value is close to the values obtained
by CT scanning. Considering the error associated with the
3.1. Solids fraction of the packing measurement of small amount of the displaced water, this
indicates that CT provides adequate measurements of the
Solids fraction of the packing was first evaluated using CT solids distribution in the bed despite its small values.
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 3.2. Liquid distribution in the bed of corrugated structured
D denote the diameter of the packing (30.48 cm). Fig. 6a packing
shows the solids distribution at those three axial positions.
The azimuthally averaged solids fraction is shown in Fig. 6b Fig. 7ac shows the time-averaged cross-sectional liquid
and is approximately equal to 2.5%. The solid fraction was saturation distribution in the column at superficial gas ve-
also measured experimentally by immersing the packing el- locities of 0, 5 and 10 cm/s, respectively. For each gas veloc-
ement in water-filled bath and measuring the volume of wa- ity, the liquid saturation distribution is shown at three axial
ter displaced, which was found to be approximately 3%. 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, liquid distributor design, which produces seven liquid jets
L , as well as the uniformity factor expressed as percentage at the top of the bed. These jets disperse while moving
f (discussed later). Liquid saturation is the ratio of liquid downward. At high liquid superficial velocity (2.17 cm/s),
volume to the bed void volume. It can be seen that, for the the uniformity factors are smaller than those obtained at
flow conditions used, as superficial liquid velocity increases, lower liquid superficial velocities, since, at this high liquid
the cross-sectional averaged liquid saturation increases. For velocity, the jets require longer bed length to disperse and
all the superficial gas and liquid velocities used, the liquid distribute.
saturation varies between 6 to 15%. Fig. 8 shows the time and azimuthally averaged radial
Fig. 7ac also illustrates the liquid saturation distribution liquid saturation profiles at varying superficial gas and liq-
along the length of the column. In general, it is observed uid velocities at the middle axial position (2.5D). The figure
that as the liquid moves downward, liquid saturation in- shows that liquid saturation is nearly flat, which suggests
creases. At all conditions studied, the uniformity factor a fair uniformity of liquid distribution. Moreover, with in-
shows that the liquid distribution improves at the bottom creasing liquid velocities, liquid saturation increases. Simi-
section of the bed (i.e. 3.5D). This could be due to the 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-
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- tributed domain would give a uniformity factor very close
liuta and Larachi [4] mentioned that gas flow rate had a to 0%.
marginal effect on liquid holdup, a fact reflected in this work The domain consisted of 60 by 60 pixels (3600) out of
as well. which 2936 pixels represent the actual cross-section of the
Experiments were also performed to compare the holdup column, and these are the pixels which represent the liquid
and flow distribution in a bed, randomly packed with 3 mm saturation distribution. Fig. 7ac show the uniformity factor
spherical alumina particle, under the same flow conditions in percentage (f %) for different liquid and gas superficial
as was done for structured packing. However, it was evident velocities for the scans performed in this work. In most of the
that the successful operating conditions for structured pack- cases, the uniformity factor is large (above 70%), suggesting
ing were too severe for random packed bed, due to very high a fairly uniform distribution of liquid achieved in the studied
pressure drop. For very low liquid velocity (1 mm/s) and setup.
no gas flow, when the experiment was possible, the liquid
distribution was poor as indicated by a low uniformity factor 3.4. Pressure drop in corrugated structured packing
(40%). However, this information is insufficient to com-
pare the distribution characteristics of structured and random The pressure drop across the length of the packing was
packings. extremely small and was beyond the detectable limit of the
available pressure gauges. The full scale of the available
3.3. Uniformity factor (f ) transducer (validyne differential transducer) was 200 units,
which correspond to a pressure drop of 1 cm of water. In
The uniformity of the liquid flow distribution across the our study, it was observed that the maximum pressure drop
column cross-section was quantified in this work. This is within the gas and liquid flow rates studied was about 1 mm
presented in Fig. 7ac in a tabular form along with time and of water (20 calibration units). A trend was seen wherein
cross-sectionally averaged liquid saturation values. Various the pressure drop increases with increasing gas and liquid
methods have been used in the literature to quantify the flow velocities.
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- 4. Final remarks
main was divided into. The factor varied between 0 and 1.
This method is suitable for flow domains, which are divided It was established that the newly developed CT unit has
into small number of sectors. For cases where large num- a spatial resolution of about 2 2.5 mm. This is quite ac-
ber of sectors are used (e.g. holdup obtained by tomography ceptable given the size of the column used in this study. The
where large number of pixels are used), the maldistribution solid fraction as determined by CT scan was about 2.5%,
values invariably give small values, thus making it impos- which was close to the value of 3% as determined by water
sible to ascertain the quality of distribution. Jiang proposed displacement method.
the relative standard deviation to quantify the uniformity of The liquid saturation increases with increasing superficial
flow [1]. However, since standard deviation can take values liquid velocity. Moreover, the liquid saturation increases as
up to infinity, this quantity will also not provide a proper the liquid phase moves downward. The liquid distribution
feel for the uniformity. was found to be fairly uniform in general as expressed by the
In the present work, the degree of uniformity is cal- uniformity factor which was relatively large, between 70 to
culated based on the t-test method. The circular domain 95%. Liquid distribution was better at the bottom of the bed,
representing the reactor is divided into several sub-domains, compared to the upper section. The effect of gas velocity,
each having a fixed number of pixels. In this work, 10 pix- was in general, found to be very small at the conditions used
els were arbitrarily taken. This constitutes several groups in this study.
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- Acknowledgements
erage of one sub-domain group. If the group passes the
t-test, this indicates that the null hypothesis is satisfied and The financial support provided by ABB Lummus Global,
the two groups are statistically the same. Then 1 is assigned Inc. is acknowledged which made this work possible.
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 References
similar (i.e. passes the t-test) to the group representing the
whole cross-sectional distribution of liquid saturation and [1] Y. Jiang, Flow distribution and its impact on performance of
its average. In this method, a uniform distribution yields packed-bed reactors, Ph.D thesis, Washington University in St Louis,
a uniformity factor of 100%, and a completely maldis- Missouri, 2000.
S. Roy et al. / Chemical Engineering and Processing 44 (2004) 5969 69

[2] R. Krishna, Reactive Separation: a New Paradigm in an Old Bottle, [8] C. Marcandelli, A.S. Lamine, J.R. Bernard, G. Wild, Liquid distri-
Report from Chemical Engineering department, University of Ams- bution in trickle-bed reactor, Oil Gas Sci. Technol.-Rev IFP 55 (4)
terdam, 1999. (2000) 407415.
[3] G.G. Podrebarac, F.T.T. Ng, G.L. Rempel, The production of diace- [9] D. Mewes, T. Loser, M. Millies, Modeling of two-phase flow in
ton alcohol with catalytic distillation:part II. A rate-based catalytic packings and monoliths, Chem. Eng. Sci. 54 (21) (1999) 47294747.
distillation model for the reaction zone, Chem. Eng. Sci. 53 (3) [10] N. Reinecke, D. Mewes, Flow regimes of two phase flow in mono-
(1998) 1077. lith catalyst, in: Proceedings of 5th World Congress on Chemical
[4] I. Iliuta, F. Larachi, Mechanistic model for structured-packing con- Engineering, San Diego, CA, 1996.
taining columns: Irrigated pressure drop, liquid holdup and pack- [11] N. Reinecke, G. Petritsch, D. Schmitz, D. Mewes, Tomographic
ing fractional wetted area, Ind. Eng. Chem. Res. 40 (2000) 5140 measurement techniquesvisualization of multiphase flows, Chem.
5146. Eng. Technol 21 (1998) 718.
[5] S.B. Kumar, M.P. Dudukovic, Jamal Chaouki, Faical Larachi, Milorad [12] N. Reinecke, D. Mewes, Oscillatory transient two-phase flows in
P. Dudukovic, Computer assisted gamma and X-ray tomography: single channels with reference to monolithic catalyst supports, Int.
applications to multiphase flow systems, in: Non-Invasive Monit. J. Multiphase Flow 25 (67) (1999) 13731393.
Multiphase Flows, Elsevier, 1997, pp. 47103. [13] D. Toye, P. Marchot, M. Crine, M. Pelsser, G. LHomme, Local
[6] J. Chen, R. Novica, M.H. Al-Dahhan, M.P. Dudukovic, Study of measurements of void fraction and liquid holdup in packed columns
particle motion in packed/ebullated beds by computed tomography using X-ray computed tomography, Chem. Eng. Process. 37 (6)
(CT) and computer automated radioactive particle tracking (CARPT), (1998) 511520.
AIChE J. 47 (5) (2000) 9941004. [14] L. Spiegel, W. Meier, Distillation columns with structured packings
[7] C. Boyer, B. Fanget, Measurement of liquid flow distribu- in the next decade, Chem. Eng. Res. Des. 81 (A1) (2003) 3947.
tion in trickle bed reactor of large diameter with a new [15] F. Yin, A. Afacan, K. Nandakumar, K.T. Chuang, Liquid holdup
gamma-ray tomographic system, Chem. Eng. Sci. 57 (2002) 1079 distribution in packed columns: gamma ray tomography and CFD
1089. 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


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

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

0255-2701/$ see front matter 2004 Elsevier B.V. All rights reserved.
72 O. Pupier et al. / Chemical Engineering and Processing 44 (2005) 7179

of a storage vessel during real dynamic cycling operation 0.12

with NG [1618]. The purpose of the present work is to in-
vestigate this last point. An automatic cycling experimental 0.10
device has been developed. The global storage performance
Q(CH4) adsorbed (g/g)

of the vessel is measured as a function of the cycle number. 0.08

For each cycle, the gas composition at the outlet of the ves-
sel during the discharge is determined by gas chromatogra- 0.06
phy. The influence on the performance of the system of the
different gas species that compose NG is described. 0.04

2. Experimental
0 1 2 3 4
2.1. Experimental device and procedure
Pressure (MPa)

Fig. 1 shows the experimental laboratory device. A cylin- Fig. 2. Adsorption isotherm of methane in the activated carbon at 298 K.
drical stainless steel vessel of about 2 l (diameter: 110.3 mm;
length: 215 mm) equipped with a central gas diffuser was a constant volumetric flow rate equal to 25 sl min1 was in-
filled with the adsorbent packing. Four thermocouples were troduced into the vessel. Charge was considered over when
distributed radially throughout the vessel with a spacing of the pressure in the vessel was equal to 3.5 MPa and when the
11 mm and one on the reactor wall to evaluate the thermal difference between the mean temperature of the adsorbent
gradient during charge and discharge phases. Heat exchange minus the heat transfer fluid temperature was lower than
through the vessel wall took place with water at 298 K inlet 0.5 K. During each discharge, an almost constant volumet-
temperature flowing in a double casing around the reservoir. ric flow rate of 1 sl min1 was delivered at the outlet of the
Two mass flow controllers (MFC) recorded and controlled vessel. Every 4 min, the gas composition was measured with
the gas flow at the inlet and the outlet of the vessel. A pres- a micro-gas chromatograph. Discharge was considered over
sure transducer recorded the pressure in the vessel. At the when the pressure in the vessel was lower than 0.15 MPa
outlet of the vessel a micro-gas chromatograph allowed the and when the difference between the mean temperature of
analysis of the composition of the delivered gas during dis- the adsorbent minus the heat transfer fluid temperature was
charge. For safety reasons, a safety valve (SV) (calibrated at lower than 0.5 K. As soon as these conditions were reached
4.0 MPa) was on the top of the storage vessel, and the appa- a charge procedure happened.
ratus was placed in a ventilated safety closed room equipped
with gas detectors. 2.2. Preparation and characterisation of the adsorbent
The experimental procedure is described as follows. Be-
fore the start of the first run, the storage vessel was degassed A commercial activated carbon, manufactured by NORIT
during several hours at a temperature higher than 423 K at a was selected for the cycling operation. It is available in a
pressure lower than 10 Pa. Before the cycling operation, the grain form with particle sizes in the range of few millimetres.
data acquisition system was activated. During each charge, Fig. 2 presents the adsorption isotherm of pure methane at




T Micro-Computer


Fig. 1. Experimental laboratory device for cycling operation.

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

720 90 2.4 0.023
Pressure (MPa)


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


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

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,
= ,
2.5 Q1
where Q1 and Qn are the delivered gas volume per adsor-
bent volume at cycle number 1 and n, respectively. Q1 is




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


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


0.6 CO2
Ethane ( C2H6 ) 0.7
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


Propane (C3H8) 0.7
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 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-
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 This work has been done with the financial support of
storage system. After 700 cycles, corresponding approxi- Gaz De France and ADEME.
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 Appendix A. Experimental profiles of the mean ratios
a steady state working mode, as well as the accurate value of for nitrogen, i-butane, n-butane, i-pentane, n-pentane,
the final performance, probably depends on the composition and the hydrocarbons with molecular weight higher
of natural gas and the dynamic working mode selected for than n-pentane (C6+ )
the cycling operation. Moreover, the microporous structure
of the activated carbon should influence the result. Never- For safety reasons, the cycling operation is interrupted
theless, basically, the final performance is the result of the during the weekend. After an interruption, always happening
multi-component adsorption equilibriums at a temperature at the end of a discharge phase, several cycles are necessary
of 298 K for the two characteristic working pressures, 3.5 to join the continuous curve representative of the evolution of
and 0.1 MPa. Gas chromatography analysis of the stream the behaviour of the tank. This phenomenon tends to demon-
composition at the outlet of the vessel shows that, as ex- strate the importance of the kinetics of the multi-component
pected, the different components contained in the natural adsorption equilibrium in the case of a complex mixture
gas, slowly obstruct the adsorption sites for methane. With 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.
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
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.
[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
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-
age monoliths, Extended Abstract, Gas storage workshop, Kingston,
[20] T.K. Bose, R. Chahine, J.M. St-Arnaud, High density Adsorbent and
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


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.

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

0255-2701/$ see front matter 2004 Elsevier B.V. All rights reserved.
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
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 83

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].
Fig. 4. Radial maldistribution index of the local gas holdup at different
axial positions.

3. Results and discussion

the gas holdup clearly, a radial maldistribution index, , is
3.1. Gas holdup introduced, defined as:

The upper surface of the internal is defined as the zero = (yi y)2 (1)
of the axial position. Axial positions are positive above this n1
and negative below this. The radial profiles of the gas holdup
at five axial positions were measured, as shown in Fig. 3. where y can be gas holdup, bubble velocity, bubble diameter
The radial profile of the gas holdup becomes much flatter or liquid velocity. Fig. 4 shows the change in maldistribu-
after flowing through the internal, with an increase in the tion index with axial position at different superficial gas
gas holdup near the wall and a decrease in the center region velocities. It can be seen that the radial maldistribution has
as compared with the gas holdup below the internal. As a minimum a short distance after passing the internal and
the distance above and away from the internal increases, then it increases gradually. The difference between 144
the profile becomes more and more similar to that before and 209 cm is negligible. Therefore, it can be concluded that
contact with the internal. The local radial profiles of the gas the region of the influence of the internal on the gas holdup
holdup at axial positions 144 and 209 cm show almost no is within 140 cm.
difference. This shows that the influence of the internal on Fig. 5 shows axial changes in the cross-sectional aver-
the radial profile of the gas holdup becomes weaker and aged gas holdup at different superficial gas velocities with
weaker with increasing distance away from the internal and an increase in the distance above the internal. The average
is no longer felt beyond a certain distance. The analysis of gas holdup first increases to a maximum at about 40 mm
the local gas holdup profiles shows that the influence of above the internal and then decreases gradually. Note that
the internal persists to a distance of about 1.4 m. In order at the superficial gas velocity of 0.0067 m/s, the maximum
to show the effect of the internal on the radial profile of gas holdup appears somewhat earlier. At a low superficial

Fig. 3. Radial profile of the gas holdup at different axial positions. 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
Fig. 7. Bubble rise velocity radial profile with axial positions.
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 gasliquid system are affected by the physical properties of
vanished and the local gas holdup is the same as in the state each phase, bubble size, local gas holdup and liquid veloc-
below the internal. ity. Larger bubbles have higher rise velocities and the bubble
Fig. 6 presents the experimental results on the gas holdup swarm velocity is higher than that of a single isolated bubble
profiles with and without the internal at the axial position [26]. Bubble breakup is caused by collisions between bub-
of 74 cm for different superficial gas velocities. The inter- bles and turbulent eddies [24]. Thus, the turbulence inten-
nal causes extra flow resistance, which in turn decreases the sity has an important effect on the bubble rise velocity and
liquid circulation velocity and increases gas holdup [5]. In methods that intensify the turbulence intensity can improve
addition, the experimental results show that the radial pro- the bubble rise velocity and its radial profile.
files of the local gas holdup with an internal are flatter than Fig. 7 shows the radial profiles of the bubble rise velocity
those without the internal. Therefore, a properly designed at different axial positions. The radial profiles of bubble rise
internal can have dual function of increasing the gas holdup velocity become much more uniform after the gasliquid
and improving its radial profile. 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
3.2. Bubble rise velocity
increases, the radial profile of the bubble rise velocity be-
comes more and more similar to that below the internal. At
It is important to study the bubble rise velocity and its
the axial position of 144 cm, the radial profile of the bubble
radial profile in a gasliquid system as these are closely re-
rise velocity is about the same as that below the internal.
lated to the hydrodynamics, and mass and heat transfer [25].
To show the effect of the internal on the average bubble
Bubble rise velocity and its radial profile have also signif-
rise velocity more clearly, the cross-sectional average bub-
icant influences on gas and liquid residence time distribu-
ble rise velocity, ub , and the radial maldistribution index of
tions. A suitable bubble rise velocity and radial profile can
bubble rise velocity, ub are introduced. ub is defined as:
improve production efficiency. Bubble rise velocities in a
2 0 g (r)ub (r)rdr
ub = (2)
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
Fig. 6. Influence of the internal on the local gas holdup for different velocity increases after the bubble had passed through the
superficial gas velocity. internal. The change caused by the internal gets smaller and
T. Zhang et al. / Chemical Engineering and Processing 44 (2005) 8187 85

Fig. 10. Bubble Sauter diameter radial profile at different axial positions.
Fig. 8. Average bubble rise velocity as a function of the axial position.

profile is similar to that below the internal. According to

then vanishes gradually with increasing distance from the the analysis of bubble diameter, it can be concluded that the
internal. effective distance of the internal is about 140 cm.
While flowing through the internal, bubbles rise along the
3.3. Bubble Sauter diameter undersurface of the baffles and collide with the tongue-like
bars, and are broken up into smaller bubbles as shown in
Precise knowledge of the bubble size is important for a Fig. 10. The difference in the direction of adjacent baffles
better understanding of the hydrodynamics, mass transfer, increases the liquid turbulent intensity, which is beneficial
and reactor design. Zun [27] concluded that small bubbles to the breakup of large bubbles. With increasing distance
tend to have uniform radial profiles while large bubbles tend from the internal, the turbulent intensity decreases and coa-
to rise in the central region. Smaller bubbles and a uniform lescence becomes dominant until a new equilibrium between
radial profile lead to a larger interfacial area and a lower rise the breakup and coalescence of bubbles is reached.
velocity, which increases mass transfer.
The Sauter bubble diameter, d32 , is commonly used to
3.4. Bubble size distribution
show the statistic size distribution of the bubbles [28].

ni d 3 The bubble size detected by the probe is subject to a
d32 =  i2 (3) probability distribution [28,29] and only the mean statistical
ni di
value of the bubbles is given by d32 . The analysis of the
Fig. 10 shows the radial profile of the bubble Sauter di- bubble size distribution helps us understand the uniformity
ameter at five axial positions. The bubble size is very much of the bubble sizes and its axial evolution. Fig. 11 shows
decreased after flowing through the internal and then in- the axial change of the bubble size distribution at r/R = 0
creases with an increase of the distance from the internal. and r/R = 0.97, respectively. Below the internal, the peak
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. 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.
Fig. 13. Axial change of the radial maldistribution index of liquid veloc-
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- 4. Conclusions
distributed in the radial direction and the turbulent intensity
is increased. The radial profile of the liquid velocity is al- Mass transfer is essential in EL-ALRs. Smaller bubbles
most uniform after passing the internal. Liquid velocity is and a uniform gas holdup radial distribution increase the in-
lower at the center and higher near the wall as compared terfacial area and improve mass transfer. Intensified turbu-
with that below the internal. With increasing distance from lence increases the surface renewal frequency and decreases
the internal, the turbulence intensity diminishes and the wall bubble size. A novel internal to improve mass transfer and
effect becomes more apparent, that is, the liquid velocity in- the hydrodynamic behavior in a gasliquid system is re-
creases at the center and decreases near the wall. The radial ported. Experiments were carried out to study the effect of
profile obtained at the position of 114 cm from the internal the internal on the bubble behavior and liquid velocity in an
is similar to that obtained below the internal and is the same EL-ALR.
as that at the position of 144 cm.
The radial maldistribution index of the liquid velocity, The radial profiles of the gas holdup and bubble rise veloc-
ul , is defined according to Eq. (1). Fig. 13 shows the ax- ity become more uniform after passing through the inter-
ial changes in ul at different superficial gas velocities. ul nal. With increasing distance from the internal, the radial
is much decreased after passing through the internal and profiles of the gas holdup and bubble rise velocity change
reaches a minimum at 5 cm above the internal, then it in- back to be similar to that below the internal. At 144 cm
creases gradually with increasing axial height. According to above the internal, the radial profiles of the gas holdup
the analysis of the radial profile of the liquid velocities, it and the bubble rise velocity are the same as those below,
can be concluded that the region in which the internal influ- while andub reach their minimum 39 cm above the
ences the liquid velocity is only within 110 cm. 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
Fig. 12. Radial profile of the liquid velocity. the internal should be carried out.
T. Zhang et al. / Chemical Engineering and Processing 44 (2005) 8187 87

Acknowledgements [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.
The authors gratefully acknowledge financial support by [9] S. Becker, A. Sokolichin, G. Eigenberger, Gasliquid flow in bubble
the National Natural Science Foundation of China (No. column and loop reactors. Part II. Comparison of detailed experiments
20276035) and by the Ministry of Education Ph.D. Fellow- and flow simulations, Chem. Eng. Sci. 49 (1994) 57475762.
ship Program (No. 20020003060). [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.
Appendix A. Notations [12] Y.C. Hsu, M.P. Dudukovic, Gas holdup and liquid recirculation in
gas-lift reactors, Chem. Eng. Sci. 35 (1980) 134141.
A cross-section of the reactor (m2 ) [13] M. Gavrilescu, R.Z. Tudose, Study of the liquid velocity in
d32 Sauter bubble diameter (mm) external-loop airlift reactor, Bioprocess Eng. 14 (1995) 183193.
[14] J.H. Hills, The operation of a bubble column at high throughputs. I.
H axial distance from the upper surface of Gas hold-up measurement, Chem. Eng. J. 12 (1976) 8999.
the internal (m) [15] M.H. Siegel, J.C. Merchuk, K. Schugerl, Airlift reactor analysis
r radial position in the reactor (m) interrelationships between riser, downcomer and gasliquid behavior,
R radius of the riser (m) including gas recirculation effects, AIChE J. 32 (1986) 15851595.
u local velocity (m/s) [16] J.B. Joshi, V.V. Ranade, S.D. Gharat, S.S. Lele, Sparged loop reactors,
Can. J. Chem. Eng. 68 (1990) 705741.
U superficial velocity (m/s) [17] M.J. Prince, H.W. Blanch, Bubble coalescence and break-up in
local gas holdup air-sparged bubble columns, AIChE J. 36 (1990) 14851499.
radial maldistribution index [18] C. Maretto, R. Krishna, Modeling of a bubble column slurry reactor
for FischerTropsch synthesis, Catal. Today 52 (1999) 279289.
Subscriptions [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)
b bubble 454466.
g gas phase [20] Y. Jin, T. F. Wang, J. F. Wang, T. W. Zhang, CN Patent 1,403,190A
l liquid phase (2003).
u velocity [21] T.F. Wang, J.F. Wang, W.G. Yang, Y. Jin, Bubble behavior in
gas holdup 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.
References Commun. 187 (2001b) 251263.
[23] J. Lin, M.H. Han, T.F. Wang, T.W. Zhang, J.F. Wang, Y. Jin, Exper-
[1] J.C. Merchuk, S. Yehuda, Local holdup and liquid velocity in air-lift imental study on the local hydrodynamic behavior of a three-phase
reactors, AIChE J. 27 (1981) 377388. external-loop airlift reactor, Ind. Eng. Chem. Res., in press.
[2] A.G. Livingston, S.F. Zhang, Hydrodynamic behavior of three-phase [24] T.F. Wang, J.F. Wang, Y. Jin, A novel theoretical breakup kernel
(gasliquidsolid) airlift reactor, Chem. Eng. Sci. 48 (1993) 1641 function for bubbles/droplets in a turbulent flow, Chem. Eng. Sci.
1654. 58 (2003) 46294637.
[3] S. Guet, G. Ooms, R.V.A. Oliemans, R.F. Mudde, Bubble injector [25] K. Tsuchiya, A. Furumoto, L.S. Fan, J. Zhang, Suspension viscosity
effect on the gas-lift efficiency, AIChE J. 49 (2003) 22422252. and bubble rise velocity in liquidsolid fluidized beds, Chem. Eng.
[4] K.H. Choi, W.K. Lee, Circulation liquid velocity, gas hold-up, vol- Sci. 52 (1997) 30533066.
umetric oxygen transfer coefficient in external-loop airlift reactors, [26] R. Krishna, M.I. Urseanu, J.M. van Baten, J. Ellenberger, Rise
J. Chem. Technol. Biotechnol. 56 (1993) 5158. velocity of a swarm of large gas bubble in liquids, Chem. Eng. Sci.
[5] M.A. Young, R.G. Carbonell, D.F. Ollis, Airlift bioreactors: analysis 54 (1999) 171183.
of local two-phase hydrodynamics, AIChE J. 37 (1991) 403428. [27] I. Zun, The transverse migration of bubble influenced by walls in
[6] A.B. Russell, C.R. Thomas, M.D. Lilly, The influence of vessel vertical bubbly flow, Int. J. Multiphase Flow 6 (1980) 583588.
height and top-section size on the hydrodynamic characteristics of [28] S.A. Patel, J.G. Daly, D.B. Bukur, Holdup and interfacial area mea-
airlift fermentors, Biotechnol. Bioeng. 43 (1994) 6976. surements using dynamic gas disengagement, AIChE J. 35 (1989)
[7] C. Bentifraouine, C. Xuereb, J.P. Riba, An experimental study of 931942.
the hydrodynamic characteristics of external-loop airlift contactors, [29] D. Sabtana, A. Macias-Machin, Local bubble-size distribution in
J. Chem. Technol. Biotechnol. 69 (1997) 345349. 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


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

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

0255-2701/$ see front matter 2004 Elsevier B.V. All rights reserved.
90 C.-L. Chen, P.-S. Hung / Chemical Engineering and Processing 44 (2005) 89100

of the polyhedral region of uncertainty [10]. Under the as- then applied in phase II to simultaneously promote satisfac-
sumption of convexity, only the source-stream temperatures tion levels of all objectives with guaranteed least satisfaction
of the HENs are considered to be the uncertain input pa- value.
rameters. With this formulation, the standard definition of Two numerical examples with flexibility preference in
the HEN synthesis problem with minimal total utility con- source-stream temperatures is presented here to demonstrate
sumption and even minimum number of units is extended that the proposed interactive two-phase fuzzy optimization
to include a flexibility specification for the potential HEN method can provide a feasible and better compensatory so-
structure and can be stated as: given hot/cold streams to be lution for multi-objective HEN synthesis.
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- 2. Model formulation
tion (considering nominal case or average of all vertical
operating points), minimal number of matches if desired, Consider the standard HEN synthesis problem with NH
and maximal flexibility for feasible operation. hot and NC cold process streams along with hot and cold
For handling the multiple and conflict design objectives, utilities. Since it is suitable for formulating the simultaneous
a fuzzy decision-making method is adopted to attain the solution which involving the consideration of both utility
compromised solution among all conflict objectives [11]. consumption and operational flexibility, the HEN superstruc-
Therein each design objective is treated as a fuzzy goal, ture proposed by [3,4] is applied for modeling the structure.
and a specific membership function is used to characterize Therein, the isothermal mixing assumption in the simplified
the transition from the numerical objective value to the de- superstructure eliminates the need for nonlinear/nonconvex
gree of satisfaction for the fuzzy objective. The final deci- energy balance. The minimum number of superstructure
sion, therefore, is interpreted as a fuzzy aggregation of these stages, NT , corresponds to max{NH , NC }, as suggested by
multiple objectives and measured by the overall degree of [3]. Fig. 1 illustrates a 2-hot/2-cold/2-stage superstructure.
satisfaction. And the best compromised solution is finally The mathematical programming formulation for minimizing
reached by maximizing the overall degree of satisfaction for utility consumption with specified uncertain source-stream
the decision. In the course of finding the solution, two pop- temperature ranges can be summarized as follows: [3,4,10]:
ular operators, the minimum and the average, are applied as  (0)
the fuzzy intersection operators, the effects of which are ex- minJ (0) = qcui + qhuj (1)
i HP j CP
amined as well. We also proposed an interactive two-phase
fuzzy decision-making method by combining these two op-
zijk , zcui , zhuj ; tik , tjk ; dtijk , dtcui , dthuj ;
erators to take advantages of the both [11,12]. The minimum x qijk , qcui , qhuj ; (2)
operator is used in phase I to maximize the degree of satis-
i HP, j CP, k ST
faction for the worst objective, and the average operator is

Fig. 1. The two-stage superstructure.

C.-L. Chen, P.-S. Hung / Chemical Engineering and Processing 44 (2005) 89100 91

(T in T out )FCpi =
qijk + qcui



  overall heat balances

(T out T in )FCp =


j j qijk + qhuj


(tik ti,k+1 )FCpi =




stage heat balances

(tjk tj,k+1 )FCpj =



in(0) in(0)

ti,1 = Tiin {Ti

, Ti + }

uncertain inlet temperatures


in(0) in(0)

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

t t



t t



feasibility of temperatures

T out t


i,NT +1


Tj tj,1


i,NT +1 Ti )FCpi = qcui
= 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 )


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



dt T

ijk min

T nominal approach temp. bounds

i min

dthu(0) Tmin



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 ;

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
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


Tiout )FCpi = qijk + qcui

k ST jCP



(Tjout Tj )FCpj = qijk + qhuj

k ST iHP


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



(n) (n)

(tjk tj,k+1 )FCpj = qijk


(n) (n)

= (Ti
+ 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


Tiout ti,NT +1


Tjout tj,1

(ti,NT +1 Tiout )FCpi = qcui

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

(Tj tj,1 )FCpj = qhuj


qijk zijk 0


qcui zcui 0


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 )

(n) (n)

dthuj THU tj,1 + (1 zhuj )

dtijk Tmin

dtcui Tmin


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
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
HEN structures with increasing flexibility targets are prone  
to variation, and the true maximal flexibilities of these HEN + zcui + zhuj
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) 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
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 )];

qcui + qhuj (9)

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


ijk ijk

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

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
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 95

single objective. The minimum operator concentrates on im- The new feasible space guarantees the least degree
proving the worst scenario. However, the minimum operator of satisfaction for each fuzzy objective.
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 4. Numerical example
fuzzy intersection operators to take advantages of both and
propose a two-phase optimization procedure [11,12]. The Two numerical examples adapted from [4,10] are supplied
minimum operator is first used in phase I to find the least to demonstrate the efficacy of proposed HEN synthesis strat-
degree of satisfaction for the worst objective, then the aver- egy. To solve this MO-MILP for the HEN model, GAMS
age operator is applied in phase II to promote satisfaction [16] and CPLEX are used as the modeling environment and
levels of all objectives with guaranteed least membership the MIP solver, respectively. The computing machine is a
value. The merit of this method is that we can not only ob- personal computer with an Intel Pentium IV 2.26 GHz CPU.
tain the unique optimal solution by using average operator
but also guarantee each objective to go after their own max- Example 1. The 2-hot/2-cold streams example studied by
imum on the basis of taking the least degree of satisfaction [10], with problem data presented in Table 1, is illustrated.
as the lower-bound constraint. So, now we can summarize With these parameters, the multi-objective MILP formula-
the procedure of the two-phase fuzzy satisfying approach tion has 408 linear equality constraints, 760 linear inequal-
for the multi-criteria optimization problem. ity constraints, 12 binary variables, and 545 positive con-
tinuous variables. Notably, the restriction of MEUmax = 6
Step 1. For a function to be minimized, determine its ideal
in Eq. (7) will be removed should the minimum number of
solution and anti-ideal solutions by directly mini-
matches be simultaneously taken into account as one of the
mizing and maximizing the objective function.
design objectives.
min Js = J 1s According to the model formulation and the problem data,
xV V
(ideal solution ofJs , totally acceptable we solve the multi-criteria MILP synthesis problem by us-
ing the fuzzy optimization procedure discussed in Section
value) (19) 3. The ideal and anti-ideal solutions, obtained by sequen-
max Js = Js0 tially minimizing and maximizing each objective function,
xV V
are shown in Table 3. Within these lower/upper limiting val-
(anti ideal solution ofJs ,
ues, several preference intervals are subjectively selected to
unacceptable value) (20) establish linear membership functions for the fuzzy objec-
Notably, direct maximization and minimization tives. In implementing the two-phase fuzzy optimization, we
should be taken for a maximizing objective to firstly use the minimum operator to maximize the degree of
obtain the ideal and anti-ideal solutions. satisfaction for the worst target, min . The average operator
Step 2. Based on the importance of different objective func- is then applied to optimize the aggregated objectives with
tions and the acceptable ranges for objective val- guaranteed level of satisfaction.
ues, subjectively select suitable lower/upper bounds, At first, only two conflict objectives are considered: the
J 1s Js1 Js0 Js0 for minimizing objective and minimal utility consumption and the maximal flexibility to
J 0s Js0 Js1 Js1 for maximizing objective. De- all source-stream temperatures. And then the third objective,
membership functions for multiple fuzzy ob-
fine the minimal number of matches, would be appended. Results
jectives as given in Eqs. (12) and (13). of two-phase fuzzy optimization with preference intervals of
Step 3. (Phase I). Use the minimum operator to find the [2550, 12750] or [2550, 8850] for utility, [0, 150], [40, 90]
maximal degree of satisfaction for the worst objec- or [40, 70] for flexibility, and [4, 7] for unit numbers, along
tive, min . with either or not considering restrictions on heat loads at
extreme operating points, are listed in Table 4. The resulting
max D = max
xV V xV V (21) HEN structures are also depicted in Fig. 2. Notably, the re-
min(J1 , . . . , JS ) = min duced range of flexibility, [40, 90], implies that the required
minimum tolerance for temperature deviation is at least 40 K
Step 4. (Phase II). Use the average operator to simultane-
and a tolerance of maximum temperature deviation for 90 K
ously promote satisfaction levels of all objectives
with new constraints of guaranteed minimal degree
Table 3
of satisfaction.
The ideal/anti-ideal solutions of various objectives of example 1
J1 + + JS
max D = max (22) Objective function Ideal solution J 1i Anti-ideal solution Ji0
xV V +
xV V S
J1 utility (kW) 2550 12750
where J2 flexibility (K) 150 0
+ J3 units 4 12
V = V {Js min , s S} (23)
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
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 in Table 5. The number of superstructure stages is set as
that the integrated two-phase method can bring the merits of NT = 5. With these parameters, the MO-MILP formulation
the minimum and average operators together. The minimum has 3510 linear equality constraints, 8026 linear inequality
operator maximize the degree of satisfaction for the worst constraints, 31 binary variables, and 5981 positive continu-
objective, and can result in a harmonious solution with ous variables. The ideal and anti-ideal solutions are shown
satisfaction levels for objective functions that are equal or in Table 6. Various preference intervals are sequentially se-
close to each other. The average operator, on the other hand, lected for defining the membership functions, as shown in
can sometimes promote satisfaction levels for other objec- Table 7.
tives with guaranteed minimum value, such as Cases II, V,
and VI. Owing to these advantages, the two-phase method We directly use the ideal/anti-ideal solutions as the pref-
can thus provide the best compromised HEN configuration. erence intervals in Case I. The acceptable utility range is
reduced from [3780,9860] to [3780,4930] in Case II. With
Example 2. This problem consists of five hot streams and such a smaller allowance interval, the resulting utility con-
one cold stream, (NH = 5, NC = 1), along with steam and sumption will be reduced slightly from 3793 to 3660 at the
cooling water as utilities [3]. The problem data are listed 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.
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- [10] A.E.S. Konukman, M.C. Camurdan, U. Akman, Simultaneous flex-
changer networks, Chem. Eng. Sci. 38 (5) (1983) 745763. ibility targeting and synthesis of minimum-utility heat exchanger
[3] T.F. Yee, I.E. Grossmann, Z. Kravanja, Simultaneous optimization networks with superstructure-based MILP formulation, Chem. Eng.
models for heat integration-I. Area and energy targeting and modeling Processing 41 (2002) 501518.
of multi-stream exchangers, Comp. Chem. Eng. 14 (10) (1990) 1151 [11] C.L. Chen, B.W. Wang, W.C. Lee, Multi-objective Optimization for
1164. Multi-enterprise Supply Chain Networks, Ind. Eng. Chem. Res. 42
[4] T.F. Yee, I.E. Grossmann, Simultaneous optimization models for heat (2003) 18791889.
integration-II. Heat exchanger network synthesis, Comp. Chem. Eng. [12] R.J. Li, E.S. Lee, Fuzzy multiple objective programming and com-
14 (10) (1990) 11651184. promise programming with Pareto optimum, Fuzzy Sets and Systems
[5] C.A. Floudas, Nonlinear and Mixed-Integer Optimization: Funda- 53 (1993) 275288.
mentals and Applications, Oxford University Press, New York, 1995. [13] Sakawa, M., Fuzzy Sets and Interactive Multi-Objective Optimiza-
[6] L.T. Biegler, I.E. Grossmann, A.W. Westerberg, Systematic Methods tion, Plenum Press, New York, 1993.
of Chemical Process Design, Prentice Hall, Englewood Cliffs, NJ, [14] L.A. Zadeh, Fuzzy sets, Information and Control 8 (3) (1965) 338
1997. 353.
[7] K.P. Halemane, I.E. Grossmann, Optimal process design under un- [15] G.L. Klir, B. Yuan, Fuzzy Sets and Fuzzy Logics-Theory and Ap-
certainty, AIChE J. 29 (3) (1983) 425433. plication, Prentice Hall, New York, 1995.
[8] R.E. Swaney, I.E. Grossmann, An index for operational flexibility [16] A. Brooke, D. Kendrick, A. Meeraus, R. Raman, R.E. Rosen-
in chemical process design, AIChE J. 31 (4) (1985) 621630. thal, GMAS: A Users Guide, GAMS Development Corporation,
[9] I.E. Grossmann, C.A. Floudas, Active constraint strategy for flexi- 1988.
bility analysis in chemical processes, Computers Chem. Eng. 11 (6)
(1987) 675693.
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


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-

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

0255-2701/$ see front matter 2004 Elsevier B.V. All rights reserved.
102 A. Arpornwichanop et al. / Chemical Engineering and Processing 44 (2005) 101114

provides challenging and interesting problems. This is mined and then the model is updated, the optimization is
mainly due to the inherent complexity of the batch reac- performed on-line to generate a new optimal input profile.
tors which can be characterized by (i) highly nonlinear With the modified optimal profile, a designed controller is
behavior resulting from the dependence of reaction rates on used to control the system to follow this profile until the
concentrations and temperature, (ii) time-varying system; new one is available.
the process variables (e.g. concentration, temperature) and Apart from specifying the optimal set point profile, a con-
parameters change with time, (iii) no steady state operating trol system used to implement such a profile is another im-
condition, (iv) imperfect model; complex kinetic reactions portant issue to be considered. This is because the deviation
occurring within batch reactors are rarely well understood from the desired profile may cause an off-spec product. How-
that leads to an inaccuracy in developing the system model, ever, since it is well known that the control of batch reactors
and (v) lack of measurement information; the product is difficult due to the inherently nonlinear behavior, the use
qualities or key properties to be controlled (e.g. molecular of a linear control technique may give a poor performance.
weight) cannot be measured until the end of batch run or For this reason, many advanced control techniques have been
even if they can be measured sometime (e.g. concentration), developed and applied to the control of batch reactor. These
there is a significant time delay. Only a few of physical include, for example, nonlinear feedforward-feedback con-
quantities such as temperature and pressure are available trol [7], generic model control [8], adaptive control [1], glob-
for direct on-line measurement. That makes direct con- ally linearizing control [9], dynamic matrix control [10],
trol of product properties difficult [2]. Although, in recent linear model predictive control [11], or inverse model con-
years there have been significant advances in developing trol [12]. Reviews on the progress in control techniques that
new sensors for measuring these product properties, they have been applied to the control of batch reactors as well as
have rarely been used in industrial processes due to high their importance and performance are given by [13]. Among
operating cost and expensive investment on the measure- these advanced control methods, a generic model control
ment devices. Thus, the usual practice is to control other (GMC) technique is one of the most studied control algo-
variables that can be measured rapidly in order to obtain rithm. This is because nonlinear process models can be in-
desired product properties instead. terpreted straightaway in the GMC control algorithm so that
In general, optimal batch reactor operation can be car- they do not need to be linearized. Furthermore, its implemen-
ried out by two-step approach; firstly, determining an opti- tation is relatively easy when compared to other model-based
mal set point profile of key operating process variables such control methods; consequently, the application of this con-
as temperature [3]and secondly, tracking the desired profile trol technique appears in many chemical processes.
by a control system [4]. The optimal profile can usually be In this work, an approach based on an on-line dynamic op-
determined off-line by solving an optimal control problem. timization strategy to modify optimal temperature set point
However, as mentioned above, because of the complexity profile is developed for improving batch operation perfor-
of chemical reaction schemes, modeling errors are always mance. To demonstrate the effectiveness of the developed
present and in addition, process disturbances can occur dur- approach, the batch reactor studied by [8], where two paral-
ing the process operation. Due to the existence of this error lel exothermic reactions occur, is chosen here as a case study.
and disturbance, the final product may significantly differ For solving the on-line optimization problem, it needs the
from the desired value, even though the pre-specified opti- knowledge of the current states of the system. Although most
mal profile is tracked perfectly [5]. To realize this fact, it physical quantities (e.g. temperature, flowrate) can be mea-
is necessary to recalculate the optimal profile as an on-line sured frequently and are available as on line measurement,
optimization strategy whenever new feedback information some other properties (e.g. concentration) are measured in-
is available. This strategy could compensate the modeling frequently with time delay. To overcome this difficulty, the
error leading to process operation improvement. Extended Kalman Filter (EKF) is incorporated into the pro-
In order to perform the on-line optimization strategy, the posed strategy in order to estimate the concentrations from
knowledge of current state variables and/or parameters in their delayed measurements. The resulting optimal control
the process models is required. Due to the fact that some of problem is solved by the sequential solution and optimiza-
these variables cannot be known exactly or sometime can tion method which is often referred to as the control vector
be measured with time delay, it is essential to include an parameterization (CPV) method. Once the optimal tempera-
on-line estimator to estimate these process variables using ture profile is modified, a controller based on generic model
available process measurements as well. The sequence of control algorithm (GMC) is applied to control the batch re-
an estimation and optimization procedure is known as an actor temperature following the desired profile. In the GMC
estimation-optimization task [6]. As in several estimation formulation, the EKF is also used to estimate the heat re-
techniques, an Extended Kalman Filter (EKF) has become leased from the chemical reactions. It should be noted that
increasingly popular because it is relatively easy to imple- the proposed strategy would be one of several strategy stud-
ment. It has been found that the EKF can be applied to a ied to promote the applicability of on-line dynamic opti-
number of chemical process applications with great success. mization with set point tracking for improving a batch pro-
Once the estimate of unknown process variables is deter- 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  
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
Qr = H1 (k1 MA MB ) H2 (k2 MA MC )
A + C
D Qj = UA(Tj Tr )

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

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
set to 20 C. Other process parameter values used in the
= +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)
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
= (6) MWD = 160 kg/kmol
dt Vj j Cpj k11 = 20.9057
k12 = 10000
k21 = 38.9057
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 the more efficient approach to obtain a maximum yield in a
that measurement noise is also included in the temperature minimum time or minimum cost and also to reach the spe-
sensors having Gaussian noise with zero mean and 1 C stan- cific final conditions of the products in terms of quality and
dard deviation. In addition, it was assumed that the concen- quantity.
tration (amount of mol) of reactants in the reactor is mea- In general, an objective function in the optimization prob-
sured infrequently and has a sampling time and measure- lem can be chosen, depending on the nature of the problem.
ment delay of 10 min. 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
3. On-line dynamic optimization strategy amount of desired product, are considered to determine an
optimal reactor temperature profile. The first problem for-
The aim of a dynamic optimization problem is to deter- mulation is applied to a situation where we need to increase
mine a control profile minimizing (or maximizing) a given the amount of desired product while batch operation time is
objective function subject to process constraints. With the fixed. This is due to the limitation of complete production
optimal control policy, the controlled system is driven from line in a sequential processing. However, in some circum-
the initial state to a final desired state in an optimal way. stances, we need to reduce the duration of batch run to allow
However, in the presence of modeling error, the pre-specified the operation of more runs per day. This requirement leads
control profile may lose its optimal character [2]. For this to the minimum time optimization problem. These problems
reason, an on-line optimization strategy through a receding can be described in details as follows.
horizon scheme is employed in this work to compensate
such an error. The basic concept is to compute the optimal 3.1.1. Maximum product concentration problem (P1)
control profile based on current feedback information. How- In this type of problem, the objective is to compute the
ever, only the initial value of the optimal trajectory is sent optimal temperature policy maximizing the amount of a de-
to the system as a set point for the controller. After new sired product concentration for a given fixed batch time sub-
information of states is available from either measurement ject to bounds on the reactor temperature. The problem can
or estimation, the optimization is repeated again to generate be written mathematically as
updated optimal set point profile at the next time interval.
The method proposed for improving the batch operation max J = X(tf )
can be divided into two phases: on-line modification of the
reactor temperature trajectory and on-line tracking of the de- subject to
sired temperature trajectory. The first phase involves deter- x = f(x(t), T, p, t) process model
mining an optimal temperature set point profile by solving
the on-line dynamic optimization problem and will be de- x(t0 ) = x(0) initial conditions
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. tf = tf
Since both the on-line dynamic optimization and the
model-based control strategy rely on process models, the where X is the amount of the desired product at a given final
knowledge of current states and/or model parameters is batch time, x is the vector of state variables, x is the derivative
required. However, in most industrial processes, state vari- of x with respect to time (t), T is the reactor temperature, p
ables are not all measurable and some parameters are not are process parameters, tf is the fixed batch time, and TL and
known exactly. As a consequence, there is a need for es- TU are lower and upper bounds of the reactor temperature.
timating these states and parameters. In this work, two
Extended Kalman Filters (EKF) are implemented. The first 3.1.2. Minimum batch time problem (P2)
one is applied to predict the reactant concentration, which The purpose of this optimization problem is to determine
will be used for on-line dynamic optimization, from its the optimal temperature profiles to achieve the desired fi-
delayed measurement. The other one is applied to esti- nal product concentration in minimum batch time, thus the
mate the unknown heat of reaction, which will be used performance index is the final time whereas the desired pro-
for model-based controller, from the frequently available duction concentration is defined as a terminal constraints.
measurements of temperature. The formulation of the minimum batch time problem can be
shown as
3.1. Problem statements min J = tf
As a batch reactor is utilized for the production of a wide subject to
variety of high value products, an optimization of batch op-
erating conditions, e.g. temperature, operating time, etc. is x = f(x(t), T, p, t) process model
A. Arpornwichanop et al. / Chemical Engineering and Processing 44 (2005) 101114 105

x(t0 ) = x(0) initial conditions sideration, this approach can be divided into two categories:
sequential and simultaneous strategy.
TL T TU In the sequential strategy, a control (manipulated) variable
X(tf ) = X profile is discretized over a time interval. The discretized
control profile can be represented as a piecewise constant,
where X is the desired product concentration at the end of a piecewise linear, or a piecewise polynomial function. The
batch run and tf is final batch time. parameters in such functions and the length of time subin-
terval become decision variables in optimization problem.
3.2. Solution of dynamic optimization problems This strategy is also referred to a control vector parameter-
ization (CVP).
Computational techniques for the solution of a dynamic One advantage in the sequential approach is that only the
optimization problem as formulated above have been an ac- parameters that are used to discretize the control variable
tive area of research. There are a number of different tech- profile are considered as the decision variables. The opti-
niques that have been proposed in the literature to solve this mization formulated by this approach is a small scale NLP
class of problems. In general, they are mainly classified into that makes it attractive to apply for solving the optimal con-
three classes. trol with large dimensional systems that are modeled by a
The first one is based on a classical variation method. large number of differential equations. In addition, this ap-
This approach is also known as an indirect method as it proach can take the advantage of available IVP solvers. How-
focuses on obtaining the solution of the necessary conditions ever, the limitation of the sequential method is a difficulty
rather than solving the optimization directly. Solution of to handle a constraint on state variables (path constraint).
these conditions often results in a two-point boundary value This is because the state variables are not directly included
problem (TPBVP), which is accepted that it is difficult to in NLP.
solve [15]. Although several numerical techniques have been In contrast to the sequential solution method, the simul-
developed to address the solution of TPBVP, e.g. control taneous strategy solves the dynamic process model and the
vector iteration (CVI) and single/multiple shooting method, optimization problem at one step. This avoids solving the
these methods are generally based on an iterative integration model equations at each iteration in the optimization algo-
of the state and adjoint equations and are usually inefficient rithm as in the sequential approach. In this approach, the dy-
[16]. Another difficulty relies on the fact that it requires an namic process model constraints in the optimal control prob-
analytical differentiation to derive the necessary conditions. lem are transformed to a set of algebraic equations which
The second class of solutions is based on dynamic pro- is treated as equality constraints in NLP problem [20]. To
gramming. Unlike the variation method, this approach apply the simultaneous strategy, both state and control vari-
applies the principle of optimality to formulate an op- able profiles are discretized by approximating functions and
timization problem, leading to the development of the treated as the decision variables in optimization algorithms.
HamiltonJacobiBellman partial equations that determine The main advantage of the simultaneous approach is a ca-
the solution of the optimal control problem. However, pability in handing constraints on the state variables. This
this approach is quite limited to a simple control prob- is because these constraints can be dealt with by including
lem [17] because of a difficulty in obtaining the solution them directly in the NLP as additional constraints. However,
of the optimality equations. [18] extended the idea of the due to the discretization on both state and control variables,
optimality principle to develop an alternative technique, this leads the simultaneous approach to a large scale opti-
named as iterative dynamic programming (IDP). Although mization problem consisting of a large set of algebraic con-
the implementation of the IDP for solving many optimal straints and decision variables and needs a special solution
control problems can be found in the literature, it is known strategy.
that the IDP algorithm would be slower than most other As the ability of the sequential approach to handle large
gradient-based algorithms [19]. systems without the need to solve excessively large opti-
The last one is based on discretization techniques, re- mization problem, this approach is utilized in this study to
ceived major attention and considered as an efficient solu- solve the optimal control problem. The formulation of the
tion method. The concept of this approach is to transform the optimal control problem as a nonlinear programming is de-
original optimal control problem into a finite dimensional scribed below.
optimization problem, typically a nonlinear programming
problem (NLP). Then, the optimal control solution is given 3.2.1. Sequential approach
by applying a standard NLP solver to directly solve the op- Consider a dynamic process model as in the form of an
timization problem. For this reason, the method is known as implicit function
a direct method. The transformation of the problem can be f(t, x(t), x(t), u(t), p) = 0 [t0 , tF ] (7)
made by using discretization technique on either only con-
trol variables (partial discretization) or both state and con- where x(t) is the set of all state variables, x denotes the
trol variables (complete discretization). Based on this con- 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. Gears type method is used for solving the process model
The time interval of interest is [t0 , tF ] and the function f is which provides the value of the objective function and con-
assumed to be continuously differentiable with respect to all straints. Gradient information of the objective function and
its arguments. For given initial condition x(t0 ), the optimal constraints with respect to the decision variables is eval-
control of a dynamic process represented by Eq. (7) can be uated in an efficient way using adjoint variable approach.
determined by computing u(t) maximizing (or minimizing) Then, a NLP solver determines a new set of control param-
an objective function J of the form eters and sends it back to the model solver. This procedure
is repeated until the optimal value is found satisfying a
J = F(x(tF ); x(tF ); u(tF ); tF ) (8)
specified accuracy.
subject to bounds on u(t) and terminal constraints. To trans-
form such a dynamic optimization problem into a nonlin- 3.3. State and parameter estimation
ear programming problem via the sequential approach, the
control u(t) is approximated by a finite dimensional repre- The implementation of the on-line optimization strategy
sentation. The time interval [t0 , tF ] is divided into a finite requires the knowledge of current states and/or parameters in
number of subintervals (P). In each subinterval, the control nonlinear process models in order to modify a new optimal
u(t) is represented by a set of basis functions involving a profile defined as a set point for a controller. It is known
finite number of parameters that some measurements i.e. concentration are available at
low sampling rate with significant time delay. To overcome
u(t) = j (t, zj ), t [tj1 , tj ] with j = 1, 2, . . . , J 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-
where tJ = tF . The control profile is defined by the param- tension of the Kalman Filter, is designed to reconstruct the
eters zj and switching times tj . In this study, the piecewise current state variables from the delayed state measurement.
constant control is assumed and used because the form of The advantage of the EKF is that it requires information only
the solution is ideally suited for implementation on a digital from the previous sampling time and allows prior knowledge
computer. Thus the set of decision variables for the nonlin- of a system via process models to be used for the estimation.
ear program can be written as The algorithm of the EKF can be seen in Appendix A.
y = {z1 , z2 , . . . , zJ , t1 , t2 , . . . , tJ } (10)
3.3.1. Application to the batch reactor
The resulting nonlinear programming problem of the form Since the concentrations of reactants (MA , MB , MC and
min(or max ) J(y) (11) MD ) in batch reactor are assumed to be measurable with a
y delay of one sampling time; that is, at time k of the reactor,
subject to equality constraints (process model), inequality only information at time k 1 is available. Thus, the EKF
constraints, bounds on control, etc. is solved using an effec- is applied to estimate the value of reactant concentration at
tive decomposition algorithm named the successive reduced current time k from their delayed measurements at sampling
quadratic programming developed by [21]. Computational time k 1. However, since it is expected to exhibit uncer-
procedure of the sequential approach for the optimization tainty in reaction rate constant (i.e. k21 and k22 ) in real plant,
problem (Eq. (11)) is illustrated in Fig. 2. With the initial the EKF is also used to estimate these uncertain parame-
guess of the decision variables (y), an integrator based on ters. The following equations, therefore, are appended for
parameter estimation
Set initial conditions dk21
Guess initial control parameters =0 (12)
Model Solver =0 (13)
Evaluate Eqs. (1)(4), (12) and (13) correspond to Eq. (A.1) in
Objective function the EKF algorithm. Based on the estimate of the current
information, the dynamic optimization problem is resolved
to generate a new optimal temperature trajectory.

yes 4. Generic model control (GMC)

Optimal control
This section presents the design of a controller to control
Fig. 2. Optimal control solution via the sequential approach. 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 )
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
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 
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
[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
(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

setpoint Tj Tjsp Tr(k) Tj(k)

Mi (K-1)

Qr(k) EKF
with simplified
reactor model


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

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

ature controller (setting as a PI controller) through a heat = bNTr (27)
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)
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
Qr (0) = 0
b(0) = 1.8386 106
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 Fig. 4. Optimal temperature profile: 1 interval (1), 5 intervals (2), 10
due to the reason of safe operation. In this case study, the intervals (3), 20 intervals (4), 40 intervals (5)
specified final batch time (tf ) of 200 min is used and the reac-
tor temperature is bounded according to 20 T ( C) 120.
these results that the amount of the desired product C in-
5.1.1. Temperature set point profile determined off-line creases as a number of interval increases. This is due to that
with perfect tracking as the number of intervals enlarges (more degrees of free-
First, simulation studies have investigated the case where dom in the optimization), the approximated optimal profile
the optimal temperature profile is determined by off-line with piecewise constant policy is closer to the actual optimal
computation and perfect tracking of such a profile is as- profile.
sumed. This results in the maximum amount of product C
(maximum conversion) that can be achieved at the end of 5.1.2. Temperature set point profile determined on-line
batch run and is served as a reference to be compared with with GMC controller
results obtained from the proposed strategy. The optimal Next, the proposed strategy using an on-line dynamic op-
control problems have been solved using time interval with timization to update optimal temperature set point profile is
equal length varied from one to 40 intervals to discretize the implemented. Rather than assuming the reactor temperature
profile. The switching time is fixed and the length of each trajectory can be tracked perfectly as in previous studies, the
interval is specified by dividing the batch operation time (tf ) GMC controller is applied here to drive the system to follow
by a number of time intervals (P). Thus, the problem is to the desired trajectory. Considering the time elapsed in the
seek an optimal temperature value (decision variables) in determination of the optimal control problem, with P = 20,
each subinterval. the temperature set point profile is updated every 10 min.
Simulation results with different time interval (P) are re- This means that, to apply this strategy on-line, the computa-
ported in Table 4. Optimal control policy in reactor tem- tional time for the updated temperature set point profile must
perature for each case is shown in Fig. 4. As shown in be less than 10 min. Here, the computational time based on
Table 4, when one time interval (P = 1) is used, the amount Pentium III/850 mHz is approximately 3 s. As a result, this
of product C obtained at the final time (tf = 200 min) is strategy is applicable for on-line implementation. Regard-
7.0171 kmol and the optimal temperature (isothermal opera- ing to the GMC control performance, it was found that the
tion) set point is 88.01 C whereas using P = 20, the amount GMC controller can drive the system from current set point
of product C achieved is 7.0379 kmol. It was found from value to a new one and maintain it at this set point. There-
fore, the GMC controller can be used for tracking the profile
Table 4
obtained from the proposed strategy.
Summary of the results: off-line optimization and perfect tracking Simulation studies carried out to compare the amount
of desired product C obtained from on-line dynamic opti-
Time Product By-product CPU
interval C (kmol) D (kmol) time (s)
mization strategy with that from off-line strategy, are cases
where the perfect model (all parameters correctly specified)
1 7.0171 1.3464 0.1591
5 7.0281 1.3605 1.2600 is used (nominal case), and where plant/model mismatch
10 7.0339 1.3594 1.9778 is introduced by changing parameters in actual plant i.e.
20 7.0379 1.3585 2.8066 pre-exponential rate constant (k0 ) decreased by 50% and ac-
40 7.0402 1.3579 5.0519 tivation energy (Ea) increased by 20% from their nominal
Run on Pentium III/850 computer. 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 Product C (kmol)

Off-line On-line

1. Nominal case
All parameters specified correctly 6.9478 6.9584
2. Plants/model mismatch case
50% k0 of reaction 2 in plant model (k21 = 38.2125) 7.6751 7.8851
+20% Ea of reaction 2 in plant model (k22 = 20400) 8.5825 10.2131
50% k0 and +20% Ea of reaction 2 in plant model 8.5827 10.2137

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 ison of the actual and estimated heat released by reactions,
strategy (C = 6.9478) is close to that obtained by the on-line respectively. It can be seen that the EKF provides an excel-
strategy (C = 6.9584). Fig. 5(a) and (b) shows the response lent estimation of the heat released. With this heat released,
of GMC controller to track the reactor temperature trajectory the GMC controller gives very good temperature control.
that is pre-specified by off-line calculation, and the compar- 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
Fig. 5. (a) Control response (off-line, nominal case): Trsp (dash), Tr
leads to high product C obtained at the final batch time (C
(solid), Tj (dot). (b) Heat released (off-line, nominal case): actual (solid), = 10.2137) compared to the value of C = 8.5827 obtained
estimated (dot). 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

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
Fig. 7. Control response (on-line, mismatch in k21 ): Trsp (dash), Tr (solid), can be seen from Table 6 that the final batch time needed
Tj (dot). 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 ().

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


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

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

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

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

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
Fig. 1. Diagram of the facility for preparation of nano-sized particles. 1: opalescence observed with the organic phase from the ex-
Bottle with carbon dioxide, 2: rotameter, 3: recurrent reactior with jacket, periments was not very good, the agglomeration was sup-
4: bubbler, 5: magnetic stirrers, 6: drecssel bottle. posed to occur during the electron microscopy analysis it-
self. It may result from the irradiation of the samples by the
The experiments started with preparation of solutions of high-energy electron beam used for the analysis. Photograph
calcium (barium) hydroxide in water and storing them for 1 A was taken from samples obtained from the experiment
day. Their exact concentration was determined by titration carried out at a ratio of inorganic phase Ba(OH)2 to organic
with hydrochloric acid. The microemulsion used was aque- phase n-C6 H14 15/35 ml. Photograph B was taken at phase
ous alkali solution/oil (w/o) where the water drops contained ratio Ba(OH)2 /n-C6 H14 = 40/10 ml.
solved calcium or barium hydroxide and organic phase The results sustained the opinion that the synthesis of
n-hexane. nano-particles in microemulsion (w/o) is to be preferred.
The total volume of the microemulsion was 50 ml for all In common emulsions, both phases spontaneously separate
the experiments. The volumes of the two phases were varied from each other while microemulsions are thermodynami-
from 10 to 40 ml, respectively. The volumes of both organic cally stable, do not segregate and appear to be transparent.
and inorganic phases of the system for each experiment are This can be explained with the size of the water droplets
shown in Table 1. [14,11]. Microemulsions are also characterized by the so
The liquid mixture was stirred with the magnetic stirrer called dynamic exchange process. The emulsion droplets
for 60 min at constant speed. The emulsion formed con- in such emulsions constantly integrate and disintegrate, thus
tained only one of the reagents. The other was fed into the exchanging substance between each other.
reactor as gas (CO2 ) at constant flow rate for 60 min. The The substance dissolved in the emulsion droplets was
second reagent and the cooling water were counterflowing quite small amount, i.e. the reagent concentration in the
(Fig. 1). The experiments were carried out at temperature of droplet was low so it could be expected to obtain small par-
2023 C. ticles with almost equal sizes.
In the case studied, the following physical model was sug-
gested. The microemulsion water solution/organic phase was
3. Results and discussion formed preliminarily. It contained only one of the reagents
(in this case solution of Ca(OH)2 or Ba(OH)2 ). The sec-
Solid nano-sized particles of CaCO3 and BaCO3 were ob- ond reagent was added in the form of a gas (CO2 ). The
tained from the experiments carried out using chemical re- 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-
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
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
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
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 Its value was calculated to be 0 = 68.18 . At 0 < 90
GS should increase with the decrease of particle radius A (as is in this case), the process called limited wetting or
[16,17]. wetting with impregnation. Therefore, the change of the free
The surface tension of 0.02 N solution of Ba(OH)2 was energy per unit area can be found by the expression
determined by the formula [18] G = TB /T T/ (17)
Ba(OH)2 = 0 + 0.049(1 aW ) (12)
The interphase tension T1 /T2 corresponds to the so-called
where : surface tension of Ba(OH)2 (dyne/cm), 0 : sur- boundary coefficient of spreading Sk , after equilibrium was
face tension of water at 25 C. It is considered to be established at the interphase surface between the two liq-
71.97 dyne/cm [18], aw : activity of water in the solution. uids. The higher the coefficient of spreading is, the faster is
The water activity in the solution was determined by the the spreading of the wetting liquid. In the system studied,
well-known equation [18] T1 /T2 = Sk = 50.5 > 0. It means that the surface tension of
 n-C6 H14 remained lower than that of the inorganic solution,
tmBa(OH)2 MBa(OH)2
ln aW = B1 + at + bt2 i.e. n-C6 H14 wetted the polar phase of aqueous solution of
  Me(OH)2 .
1000 + MBa(OH)2 mBa(OH)2
1 4. Conclusions
mBa(OH)2 MBa(OH)2 + (13)
The following conclusions can be formulated on the basis
where R: universal gas constant = 8.314441 J/mol, T: tem- of the results obtained
perature (K), mi : solution molality (g/l), Mi : mole weight
(g/mol), B1 , a, b, B2 , E: coefficients calculated by mathe- 1. A technique and process for preparation of nano-sized
matical treatment of the experimental data for heat capacity. particles from alkali carbonates was developed on the
Value of the surface tension of Ba(OH)2 was calculated to basis of the method of reversed micelles and a probable
be 71.88 dyne/cm. Reference data showed that n-C6 H14 = mechanism was suggested.
32.21 [18]. The value of Ba(OH)2 surface tension was close 2. Nano-sized particles of barium and calcium carbonate
to that of water ( 0 ) since the concentration of Ba(OH)2 was were obtained by a chemical reaction in a microemulsion.
low and the solution was very diluted. The particles were studied by electron microscopy and
Forces of interphase tension T1 /T2 act at the contact sur- were found to possess spherical shape and diameters from
face between the two liquid phases of Ba(OH)2 (or Ca(OH)2 ) 20 to 30 nm.
and n-C6 H14 . The thermodynamic condition for a liquid to 3. The nano-sized particles were obtained at various ratios
spread over the surface of another liquid is [19] of the phases forming the microemulsion. They were
T2 / > T1 / + T1 /T2 (14) about the same size, which confirmed the advantage of
the method selected.
where T1 / , T2 / : surface tensions of the liquids related 4. The particles synthesized were observed to agglomer-
to their environment (dyne/cm), T1 /T2 : surface tension at ate, probably due to effects induced by the electron mi-
the interphase boundary between the liquids (dyne/cm). croscopy analysis.
The interphase tension at the phase boundary between an 5. The values of the surface tensions of the individual phases
inorganic solution and oil was assumed to be 50.5 dyne/cm, were calculated, as well as the value of the interphase
according to literary data [19]. The two phases are quite tension. The results confirmed that the inorganic phase
different by polarity. The basic process taking place at the was dispersed in the organic phase.
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 Appendix A. Nomenclature
of wetting to take place. The coefficient of spreading S can
be calculated from the values of the surface and interphase aw activity of water in the solution
tensions according to [20] mi solution molality (g/l)
S = T2 / T1 / T1 /T2 , S = 10.853 (15) r particles radius (m)
o oil
At S < 0, the wetting angle can be determined by the equa- w water
tion [20] A specific area (m2 /kg)
T22 / T22 / T21 /T2 B1 , a, b, B2 , E coefficients calculated by mathematical
cos 0 = (16) treatment of the experimental data for
2T1 / T2 /
heat capacity
i.e. cos 0 = 0.3717 G change of the free energy per unit area
C. Karagiozov, D. Momchilova / Chemical Engineering and Processing 44 (2005) 115119 119

GS specific free energy [6] R. Valencuela, G. Bueno, A. Solbes, Nanostructured ceria-based

catalysts for oxydehydrogenation of ethane with CO2 , Topics Catal.
Mi mole weight (g/mol)
15 (2-4) (2001) 181187.
N number of particles per unit of [7] Y. Mori, Y. Okastu, Y. Tsujimoto, Titanium dioxide nanoparticles
weight produced in water-in-oil emulsion, J. Nanoparticle Res. 3 (23)
R Universal gas constant (J/mol) (2001) 219225.
T temperature (K) [8] B. Palla, D. Shah, P. Garcia-Casillas, J. Matutes-Aquino, Preparation
of nanoparticles of barium ferrite from precipitation in microemul-
S coefficient of spreading
sions, J. Nanoparticle Res. 1 (2) (1999) 215221.
surface tension of Ba(OH)2 (dyne/cm) [9] K. Song, J. Kim, Synthesis of high surface area tin (Sn) oxide
0 surface tension of water at 25 C (dyne/cm) powders via water-in-oil microemulsions, Powder Technol. 103 (3)
T1 /T2 surface tension at the interphase boundary (2000) 268272.
between the liquids (dyne/cm) [10] W. Wang, X. Fu, J. Tang, L. Jiang, Preparation of submicron spherical
particles of silica by the water-in-oil microemulsion method, Colloids
T1/ , T2 / surface tensions of the liquids related
Surf. A: Physicochem. Eng. Aspects 81 (1993) 177180.
to their environment (dyne/cm) [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.
[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, lung, 1998.
Technical Sciences, Sea Sciences and Ecology, vol. II, 2003, [15] K. Tamapy, Kapha xm, Mockba Mp, 1983.
pp. 179184. [16] A. Akoh, Xmecka tepmohamka, Mockba: Bica
[2] X. Dong, C. Choi, B. Kim, Chemical synthesis of Co nanoparticles koa, 1963.
by chemical vapor condensation, Powder Technol. 113 (2002) 75 [17] A. pock, 1. Cboctba ha matepta. Xmecka
79. tepmohamka, 2. Patbop, Eektpoxm, Xmha
[3] Y. Du, J. Fang, M. Zhang, J. Hong, Z. Yin, Q. Zhang, Phase khetka, 1951
character and structural anomaly of Bi4 Ti3 O12 nanoparticles prepared [18] . a eb, .. Aceeb ko xmecke cboctba haphix
by chemical coprecipitation, Mater. Lett. 57 (2002) 802806. mhookomohehthix pactbopob heopaheckx beectb,
[4] A. Hartridge, A. Bhattacharya, Preparation and analysis of zirconia Mockba "Xm", 1988.
doped ceria nanocrystal dispersions, J. Nanoparticle Res. 1 (2001) [19] H. Backes, J. Ma, E. Bender, G. Maurer, Interfacial tensions in binary
7580. and ternary liquidliquid systems, Chem. Eng. Sci. 45 (1) (1990)
[5] E. Bermejo, T. Becue, C. Lacour, M. Quarton, Synthesis of 275286.
nanoscaled iron particles from freeze-dried precursors, Powder Tech- [20] . Cymm, .B. ophob, ko xmecke ochobi
nol. 94 (Issues 1) (1997) 2934. 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


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 tial ones, affect the mass transfer. In any case, these motions
are much smaller in scale than the gas bubbles. As a result,
Mass transfer from swarm of bubbles into turbulent liq- the size of the gas bubble is not a very critical parameter for
uid controls the rate of many chemical and biochemical pro- the estimation of kL . They deduced [1] the following rela-
cesses. It is assumed that the mechanism of mass transport tion for kL :
in liquid phase is due to a renewal of the liquid at the bub-  0.25  0.5
ev D
ble surface. Models of the process differ in the scale of flow, kL = c1 (1)
which is responsible for the renewal.
The first group of models (eddy models) assumes that Different authors predict different values of the constant c1 :
the liquid renewal is due to small-scale eddies of the turbu- 0.301 [2], 0.4 [1], 0.531 [3], 0.592 [4] and 1.13 [5].
lent field. These models are based on idealized eddy struc- The second group of models (slip velocity models) as-
tures of turbulence in the bubble vicinity. Lamont and Scott sumes a gross mean flow of fluid relative to the bubble (slip
[1] have assumed that the small scales of turbulent motion, velocity) and a bubble surface mobility control of this re-
which extend from the smallest viscous motions to the iner- 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
Corresponding author. Tel. +420 2 2435 3298; size into two categories: the small bubbles (d < 1 mm),
fax: +420 2 3333 7335. which behave always as a rigid sphere, and the large
E-mail address: linekv@vscht.cz (V. Linek). ones (d > 2.5 mm), which always have a completely mobile

0255-2701/$ see front matter 2004 Elsevier B.V. All rights reserved.
122 V. Linek et al. / Chemical Engineering and Processing 44 (2005) 121130

surface. An expression for the calculation of kL for the The time mobile depends on bubble diameter and surfactant
large bubbles has been deduced from dimensionless analy- concentration.
sis of mass transfer from rising particle in gravitational field.
d 1/2 ln (d/ htrans )
    mobile = k (8)
g 1/3 D 1/2 csurf
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 stant related with surfactant properties, csurf is the concen-
kL has been obtained by the equation proposed by Frssling tration of surfactant and htrans is the bubble clean segment
[7] for dissolution of a rigid sphere. height at the transition point from mobile to rigid.
    The main difference between the eddy and the slip
g 1/3 D 2/3 velocity models is in the influence of turbulence inten-
kL = 0.31 (3)
sity on mass transfer coefficient: the eddy models predict
an increase while the slip velocity models a decrease of
The transition between these two bubble categories was kL with increasing turbulence intensity, i.e. with increasing
unclear and strongly dependent on the presence of surfac- power dissipated in the liquid. This is due to the fact that
tants. An expression for calculation of the mass transfer kL from rigid bubbles is cca four times lower than from the
coefficients in this transition region was not given. ones with a mobile surface. The small bubbles, which ei-
Recently, Vasconcelos et al. [8] and Alves et al. [9] pro- ther have lower kL (according to Calderbanks conception)
posed another variant of the slip velocity model. Their or become rigid more quickly (according to Alves concep-
model is based on experimentally observed phenomena of tion, mobile diminishes with diminishing bubble diameter
an abrupt change of the dissolution rate of free-floating bub- d, see Eq. (8)), are generated in large amount just at higher
bles held stationary in a downward water flow. Starting from turbulence intensities. As a result, this leads to the decrease
these experiments, they have interpreted the mass transport of average mass transfer coefficient with increasing power
from bubbles in terms of bubble contamination kinetics, us- dissipated in the liquid. Literature data on kL in stirred tanks
ing a stagnant cap model, according to which bubbles sud- supporting both these models can be found.
denly change from a mobile interface to a rigid condition In dispersions with no mechanical agitation (bubble
when surface tension gradients, caused by surfactant accu- columns, air-lifts), the only source of energy is the expansion
mulation, balance out shear stress. The fresh bubble entering of the entering gas which gives the relation e = gvs . Reith
the dispersion has clean, completely mobile interface. For [10], Dillon and Harris [11], Kawase et al. [5] and Vzquez
the bubbles with completely mobile surface, kL is given by et al. [12] revealed strong dependence of liquid-side mass
the equation, which follows from the well-known penetra- transfer coefficient on gas flow rate in bubble columns:
tion model with exposition time equal to d/vsl : kL vs 0.25/0.5 . Slight dependence or independence of kL

vsl 1/2 on vs was found by Schumpe and Deckwer [13], Bouaifi
kLmobil = 1.13 D (4) et al. [14], Wongsuchoto et al. [15] and Vasconcelos et
al. [8]. The values of kL corresponded to values predicted
After some time ( mobile ), enough contaminant for transition by Calderbanks model for large bubbles. Values which
to rigidity is accumulated on the surface. For the rigid bub- correspond also to small bubbles were found by Vzquez
bles, an expression for the calculation of kL follows from et al. [12] and Vasconcelos et al. [8] only, who worked with
Frssling [7] equation. solutions of surfactants. Vasconcelos et al. [8] interpreted
 their results in terms of bubble contamination kinetics.
rigid vsl 2/3 1/6
kL = 0.6 D v (5) In dispersions with mechanical agitation, following au-
d thors found an increase of kL with increasing energy dissi-
Gasliquid slip velocities vsl are assumed to be close to pation e: Prasher and Wills [4] (kL e0.25 ); Figueiredo and
single-bubble terminal velocities in still water, vt , on which a Calderbank [16] (kL e0.33 ); Bouaifi and Roustan [17] (kL
correction for turbulence is introduced, vsl = 0.65vt [9]. As- e0.22 ); Panja and Rao [18] (kL e0.15 ); and Linek et al.
suming monosized bubble dispersion and considering only [19] (kL e0.14 ). Power dissipated in the liquid by an agi-
two possible values of kL (kLmobil and kL , depending on tator is proportional to the third power of impeller rate (e
surface mobility), they deduced the following relation for f3 ). Thus, we can add the results of authors that presented kL
the average mass transfer coefficient: as a function of impeller rate f: Koetsier and Thoenes [20]
(kL f0.9 e0.3 ) and Yoshida and Miura [21] (kL f0.6
kmobil mobile + kL (R mobile ) e0.2 ). Robinson and Wilke [22] measured kL that did not
kL = L (6)
R depend on power dissipated and Hassan and Robinson [23]
obtained coefficients that decreased with increasing power
where R is the average bubble residence time. dissipated. It was shown [32] that the reported independence
VL on or the decrease of kL with increasing power dissipated
R = (7)
q(1 ) was a result of misinterpretation of kL a data.
V. Linek et al. / Chemical Engineering and Processing 44 (2005) 121130 123

Figueiredo and Calderbank [16] have analysed the suit- efficients in the stirred vessel has been analysed by Linek
ability of Calderbanks model [6] for small and large et al. [3]. Alves et al. [9] measured the overall volumetric
bubbles to the interpretation of mass transfer coefficients in mass transfer coefficients kL a by the peroxide decomposi-
stirred vessels. They analysed an effect of the gas phase mix- tion steady-state technique with air as a gas phase, show-
ing model used for kL a evaluation by the dynamic method ing that kL decreased slightly with power dissipated. Linek
based on the inlet gas exchange N2 /air. In their original et al. [25] showed, however, that the steady-state methods
work [6], the kL a evaluation assumed that the interchange with air absorption/desorption give substantially lower kL a
of the gases in gas hold-up is immediate (steady-state model values at high mixing intensities as compared to pure oxy-
of the gas phase). Corresponding kL values were indepen- gen absorption. If the coefficients kL a were measured by the
dent of power dissipated in the liquid by the agitator and non-problematic pressure dynamic method [26] with pure
matched the constant value of 4.03.104 m s1 as predicted oxygen absorption in the same apparatus as in Alves et al.
by Calderbanks model [6] for large bubbles (Eq. (2)). [9], the relevant kL data were well interpreted in terms of the
When the gas phase dynamics was taken into account (the eddy model (Eq. (1)) with c1 = 0.531 [3]. The corrected
flushing out of an old gas by a new one), the correspond- data have shown that the decisive quantity to correlate kL in
ing values of kL were higher and well fitted by the eddy stirred tanks is power dissipated in the liquid phase rather
model (Eq. (1)) with c1 = 0.5, see Fig. 1. (It was shown than the bubble diameter and their rigidity and that kL in-
by Linek et al. [24] that faulty sampling of liquid used by creases with increasing turbulence intensity.
Figueiredo and Calderbank [16] leads to considerably over- It seems from the literature survey that in mechanically
estimated kL a values for kL a > 0.1 s1 . Therefore, kL values agitated dispersions, the intensity of energy dissipation plays
corresponding to this range are not included in the figure.) a dominant role in the value of the mass transfer coefficient
They concluded [16] (in difference with the original work and that the successful applications of the slip velocity
[6]) that kL is dependent on the turbulent intensity expressed models in stirred tanks are probably only a result of errors
as a function of the power dissipated per unit volume and in evaluation of kL a. On the contrary, in dispersions with no
that the coefficient is considerably higher than would be mechanical agitation, both conceptions (eddy and slip
found for bubbles in free rise under gravitational force as velocity) are equally successful. This may indicate that the
given by the model of small and large bubbles. mechanism of mass transfer without mechanical agitation
Vasconcelos variant of the slip velocity model based differs from that in the presence of agitation, as was sug-
on bubble contamination kinetics, Eqs. (4)(8), was used gested by Bouaifi et al. [14]. The same indicate the results by
by Alves et al. [9] to interpret kL data in a double Rush- Lamont and Scott [1]. They compared rates of mass trans-
ton turbine-stirred tank. The application of Vasconcelos fer from turbulent liquid into bubbles flowing in vertical
model to the interpretation of unreliable mass transfer co- and horizontal tube where the bubbles can and cannot rise

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].
124 V. Linek et al. / Chemical Engineering and Processing 44 (2005) 121130

freely through the turbulent liquid, respectively. If the bub- the model of gradual surface contamination (Eqs. (4)(8))
bles could rise freely through the liquid, the mass transfer using three empirical parameters: htrans , k/csurf and vt /vsl .
coefficient did not depend on intensity of turbulence at low The aim of this paper is to make measurements with liq-
Reynolds number. With increasing Re number, the role of uids of various physical properties in order to define the
the intensity increased and at high Re number, the dependen- effect of the liquid properties and operating conditions on
cies for both arrangements (with and without the free rise of the parameter kL and the limits of validity of the literature
bubbles) coincided. It means that both mechanisms operate models for the interpretation of mass transfer coefficients in
simultaneously but they apply at different extent, depending bubble dispersions. The method, which is used for the mea-
on intensity of turbulence. surements, was verified in Part I to minimize misinterpreta-
Surface-active substances reduce the mass transfer coef- tions.
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- 2. Experimental
sion had been drawn. Two possibilities were offered [27]:
surfactant molecules affect mass transfer via hydrodynam- The experiments were performed in a flat-bottomed
ics effects or via the formation of an interfacial barrier layer. cylindrical vessel with internal diameter 0.19 m, equipped
The barrier effect is due to a monolayer of the surfactants with standard Rushton turbine of 0.075 m in diameter lo-
formed at the interface, which can offer a resistance to be cated 0.08 m above the bottom. Four baffles (1/10 of the
crossed by the gas molecule. The resistance exerted by the vessel diameter width) were symmetrically mounted. A
monolayer is a complex phenomenon, possibly depending on membrane-covered polarographic oxygen probe was sit-
the effects of absorbing gassurfactantwater interactions, uated in the bottom to measure the equilibrium partial
which are related to the molecular structure of the surfac- pressure of oxygen in the bulk of liquid. Gas (pure oxygen
tant: the polarity of the hydrophilic group, the molecular or air) was taken from a pressure flask and introduced 2 cm
weight of the hydrophilic group, the hydrophilic chain length under the impeller through a tube of i.d. 4 mm. The bubble
and the orientation of the molecules at interface. Hydrody- column was realized by keeping zero rotational frequency of
namics effects are induced by the difference in composition the agitator. The experiments were performed with the same
and physical properties of the liquid at the interface and in solutions as those used in the kinetics measurement: 0.8 M
the bulk of liquid. Surfactants (but also electrolytes [28,29]) sodium sulphite solution, pure or with the addition of vis-
tend to create different spatially oriented structures in close cosity or surface tension-changing additives (see Table 1).
proximity of the interface and in bulk liquid and thus, their Physical properties of the solutions are given in Part I of
presence causes an energetic barrier for the surface renewal this work. The vessel was filled with liquid to the height of
due to additional requirements of reordering the liquid en- 226 mm, which amounts to 5.9 l. Superficial velocities of
tering or leaving the interface. For example, if the liquid aeration gas were 1.8, 3.6 or 5.4 mm s1 , which correspond
turbulent eddies penetrate to the interface, they would bring to volumetric flow rates 5.1 105 , 10.2 105 and 15.3
about an increase in the free energy per unit area of inter- 105 m3 s1 , respectively. The impeller rotational speeds
face due to the replacement of the interfacial liquid layer were 0, 5.5, 10, 14.33 and 18.83 l/s. Temperature was kept
by inner liquid of higher surface tension. This dampens the at 30 1 C by a cooling coil.
rejuvenation of the interfacial liquid by fresh bulk liquid, The specific power dissipated by agitators in the liquid,
renders mass transfer more difficult and lowers the value eagit , was measured by a strain gauge mounted on the im-
of mass transfer coefficient. Llorens et al. [27] decided the peller shaft. The total specific power e dissipated in the liq-
hydrodynamic effect to be the principal. They offered the uid by the agitator and the rising bubbles was calculated as
surface pressure (i.e. the difference in surface tension of e = eagit + gvs . The volumetric mass transfer coefficient
the solution without and with s , the surfactant) to rep- and interfacial area were measured by the Danckwerts plot
resent a measure of the additional work needed to replace method described in detail in Part I.
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 3. Results
surfactant. Zieminsky and Whittemore [28] offered the ionic
strength and Machon and Linek [29] the electrical surface 3.1. Volumetric mass transfer coefficient
potential to correlate the reduction of kL in electrolyte solu-
tions as measures of the additional work needed to replace a The overall volumetric mass transfer coefficients mea-
surface element in these solutions. Vzquez et al. [12] used sured by pure oxygen absorption were expressed as a func-
a simple correlation between the mass transfer coefficient tion of the total specific power dissipated in the liquid as
and surface tension of solution under the form kL 1.35 . follows:
Vasconcelos et al. [8] interpreted mass transfer coefficients
in air-lift and bubble columns with solution of surfactants by 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 of electrolyte solutions [24]. Marked reductions of kL a
given for all batches in Table 1. Taking into account gas are observed with ocenol (up to 10 times) and long-fiber
flow rate vs as additional independent parameter has not carboxymethylcellulose (CMC) TS.20 (up to four times).
improved fidelity of the correlation as follows from a com- The ocenol has a strong antifoaming effect that enhances
parison of the mean deviations of the relation (9) with and the coalescence rate of bubbles in dispersion. The resulting
without taking into account vs (see Table 1). Fig. 2 clearly coefficients are even lower than those predicted by a corre-
shows that kL a data for pure sulphite solution agree well lation suggested for coalescent batches like pure water [24].
with a recommended correlation for non-coalescent batches 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- pure oxygen absorption were expressed as a function of the
tion of Sokrat 44 (which increased the viscosity to the same total specific power dissipated in the liquid as follows:
extent as CMC) made no reduction in kL a.
kL = 1 e2 (11)
3.2. Specific interfacial area 1, 2 values together with mean deviations of the data are
given for all batches in Table 1. The results show clearly that
The specific interfacial areas measured by pure oxygen kL increases with increasing power dfissipated in the liq-
absorption were expressed as a function of the total specific uid with exponent 2 equal approximately to 0.25 in agree-
power dissipated in the liquid as follows: ment with the eddy model (1) (see Fig. 4). Exceptions
a = 1 e2 (10) are the solutions with the addition of long-fiber CMC TS.20
and ocenol. In these solutions, the coefficients increase with
1, 2 values together with mean deviations of the data are power but the exponent 2 is lower than 0.25 ( 2 = 0.09
given for all batches in Table 1. Taking into account gas and 0.188, respectively). The correlation (11) also fits well
flow rate vs as an additional independent parameter has not the data measured in the bubble column, which are repre-
improved fidelity of the correlation as follows from a com- sented by the points for e < 100 W m3 in Fig. 4. kL data
parison of the mean deviations of the relation (10) with and for all Newtonian batches without surface-active agents (i.e.
without taking into account vs (see Table 1). This agrees pure sulphite solution and with the addition of Sokrat and
with the finding of Westersterp et al. [30], who stated that short-fiber CMC) are well fitted by the equation
the interfacial area is independent of gas flow rate at im-  0.243  
peller speeds higher than a critical one. The critical im- ev D
kL = 0.915 0.623 (12)
peller speed for Rushton turbine and sulphite solution equals v
9.4 s1 [29]. All but one of the impeller speeds used in this
with a mean deviation of 19%. If the theoretical value 0.25 of
work were higher than this critical value. The results given
the power exponent and 0.5 of Schmidt number exponent
in Fig. 3 show that ocenol and short-fiber CMC TS.5 pro-
are used, the equation changes to
duce a marked decrease of the interfacial area (up to five and
three times, respectively), and that a marked increase is pro-  0.25  0.5
ev D
duced by polyethylenglycol (PEG) (up to three times) and kL = 0.448 (13)
a smaller one by the addition of Sokrat (up to two times).
with a mean deviation of 20%. Since the deviation is prac-
3.3. Mass transfer coefficient tically the same as the deviations of correlation (11), cor-
relations (12) and (13) describe the effects of viscosity and
Mass transfer coefficients were calculated by the ratio diffusivity satisfactorily. Experimental and calculated values
kL a/a. The coefficients evaluated from the data measured by 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 3.4. Effect of surfactants
= 0.25 and c3 = 0.5 fits our and literature data on Newto-
nian batches with a mean deviation of 20%. Experimental Additions of substances for surface tension depression
and calculated values of kL from Eq. (14) are compared in (ocenol, PEG, CMC TS.20, see Table 1 of Part I) induced
Fig. 6. 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

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
Fig. 7. Relative decrease of mass transfer coefficient in the presence of by application of an improper gas phase mixing model of the
surfactant (in comparison to its value in pure sulphite solution) as a
dispersion, leading to an incorrect value of the driving force
function of surface pressure . Comparison with literature data measured
in bubble columns [12,8] and in wetted-wall column [27]. 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
40 up to 80%). The results are given in Fig. 7 together with the same for all bubbles. Thus, kL data used in this work are
literature data measured in bubble columns [12,8] and in measured by pure oxygen. The absorption of air has only
wetted-wall column [27]. In the figure, the relative decrease been used to demonstrate the possible reason of discrepancy
of kL (in comparison to its value in pure sulphite solution, in the literature data. Fig. 9 shows that kL data measured
(kL )s /kL ) is plotted as a function of surface pressure . It is with air can be interpreted in terms of Calderbanks model:
apparent that neither the surface pressure nor surface tension data measured in pure sulphite solution are independent
alone (cf. the values of given in Table 1 of Part I) is of bubble diameter and correspond to Eq. (2) for large

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.
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 be the result of interpretation of physically inconsistent kL
gas hold-up and interfacial area as d = 6/[(1)a].) The data.
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 Acknowledgements
surfactant, which shifts the transition between large and
small bubbles to higher bubble diameters. The decrease This work was supported by the Grant agency of Czech
of kL values with increasing stirring intensity (see Fig. 8 for Republic through the project no. 104/98/1126 and by Czech
ocenol) can be explained by an increasing amount of small Ministry of Education through the scientific project no.
bubbles. Such interpretation is frequently used in literature 223400007.
for the data measured by absorption of diluted gas.
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 [13] A. Schumpe, W.D. Deckwer, Gas holdups, specific interfacial areas,
and mass transfer coefficients of aerated carboxymethyl cellulose
vs superficial velocity, m s1 solutions in a bubble column, I&EC Process Des. Develop. 21 (1982)
vt single bubble terminal velocity in still water, m s1 706711.
vsl slip velocity, m s1 [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.
Greek symbols
Proc. 40 (2001) 97111.
, , constants in correlations (9), (10) and (11) [15] P. Wongsuchoto, T. Chrinpanitkul, P. Pavasant, Bubble size distribu-
gas hold-up tion and gas/liquid mass transfer in airlift contactors, Chem. Eng. J.
kinematics viscosity of the liquid phase, m2 s1 92 (2003) 8190.
=( s ), surface pressure, N m1 [16] M.M.L. Figueiredo, P.H. Calderbank, The scale-up of aerated mixing
vessels for specified oxygen dissolution rates, Chem. Eng. Sci. 34
density of the liquid phase, kg m3 (1979) 13331338.
 =(liquid gas ), kg m3 [17] M. Bouaifi, M. Roustan, Bubble size and mass transfer coefficients
s, surface tension of solution with or without in dual-impeller agitated reactors, Can. J. Chem. Eng. 76 (1998)
surfactant, N m1 390397.
[18] N.C. Panja, D.P. Rao, Measurement of gasliquid parameters in a
mobile time for transition of bubble surface to rigidity, s
mechanically agitated contactor, Chem. Eng. J. 52 (1993) 121129.
R average bubble residence time, s [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.
References [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.
[1] J.C. Lamont, D.S. Scott, An eddy cell model of mass transfer into Eng. J. 5 (1973) 7175.
surface of a turbulent liquid, AIChE J. 16 (1970) 513519. [21] F. Yoshida, Y. Miura, Gas absorption in agitated gasliquid contac-
[2] Y. Kawase, B. Halard, M. Moo-Young, Liquid-phase mass trans- tors, I&EC Process Des. Develop. 2 (1963) 263268.
fer coefficients in bioreactors, Biotech. Bioeng. 39 (1992) 1133 [22] C.W. Robinson, C.R. Wilke, Simultaneous measurement of interfacial
1140. area and mass transfer coefficients for a well-mixed gas dispersion
[3] V. Linek, M. Fujasov, M. Korda, T. Moucha, Gasliquid mass in aqueous electrolyte solutions, AIChE J. 20 (1974) 285294.
transfer coefficient in stirred tank interpreted through models of [23] I.T.M. Hassan, C.W. Robinson, Mass transfer coefficients in mechan-
idealized eddy structure of turbulence in the bubble vicinity, Chem. ically agitated gasaqueous electrolyte dispersions, Can. J. Chem.
Eng. Proc. (in press). Eng. 58 (1980) 198205.
[4] B.D. Prasher, G.B. Wills, Mass transfer in an agitated vessel, I&EC [24] V. Linek, V. Vacek, P. Bene, A critical review and experimental
Process Des. Develop. 12 (1973) 351354. verification of the correct use of the dynamic method for the de-
[5] Y. Kawase, B. Halard, M. Moo-Young, Theoretical prediction of termination of oxygen transfer in aerated agitated vessels to water,
volumetric mass transfer coefficients in bubble columns for Newto- electrolyte solutions and viscous liquids, Chem. Eng. J. 34 (1987)
nian and non-newtonian fluids, Chem. Eng. Sci. 42 (1987) 1609 1134.
1617. [25] V. Linek, P. Bene, J. Sinkule, Critical assessment of the steady-
[6] P.H. Calderbank, M.B. Moo-Young, The continuous phase heat and state Na2 SO3 feeding method for kL a measurement in fermentors,
mass transfer properties of dispersions, Chem. Eng. Sci. 16 (1961) Biotechnol. Bioeng. 35 (1990) 766770.
3961. [26] V. Linek, P. Bene, J. Sinkule, T. Moucha, Non-ideal pressure step
[7] N. Frssling, ber die Verdnstung fallenden Tropfen, Gerlands method for kL a measurement, Chem. Eng. Sci. 48 (1993) 15931599.
Beitage Geophys. 52 (1938) 170216. [27] J. Llorens, C. Mans, J. Costa, Discrimination of the effect of sur-
[8] J.M.T. Vasconcelos, J.M.I. Rodrigues, S.C.P. Orvalho, S.S. Alves, factants in gas absorption, Chem. Eng. Sci. 43 (1988) 443450.
R.L. Mendes, A. Reis, Effect of contaminants on mass transfer [28] S.A. Zieminski, R.C. Whittemore, Behavior of gas bubbles in aqueous
coefficients in bubble column and airlift contactors, Chem. Eng. Sci. electrolyte solutions, Chem. Eng. Sci. 26 (1971) 509520.
58 (2003) 14311440. [29] V. Machon, V. Linek, Effect of salts on the rate of mass transfer
[9] S.S. Alves, C.I. Maia, J.M.T. Vasconcelos, Gasliquid mass transfer across a plane interface between a gas and mechanically agitated
coefficient in stirred tank interpreted through bubble contamination aqueous solutions of inorganic electrolytes, Chem. Eng. J. 8 (1974)
kinetics, Chem. Eng. Proc. 43 (7) (2004) 823830. 5361.
[10] T. Reith, Physical aspects of bubble dispersions in liquids, Ph.D. [30] K.R. Westerterp, L.L. Van Dierendonck, J.A. de Kraa, Interfacial
Thesis, Delft Techn. Univ., 1968. areas in agitated gasliquid contactors, Chem. Eng. Sci. 18 (1963)
[11] G.B. Dillon, I.J. Harris, The determination of mass transfer coeffi- 157176.
cients and interfacial areas in gasliquid contacting systems, Can. J. [31] A.A. Kulkarni, J.B. Joshi, Simultaneous measurement of flow pattern
Chem. Eng. 44 (1966) 307312. and mass transfer coefficient in bubble columns, Chem. Eng. Sci.
[12] G. Vzquez, M.A. Cancela, C. Riverol, E. Alvarez, J.M. Navaza, 59 (2004) 271281.
Application of the Danckwerts method in a bubble column. Effects [32] V. Linek, P. Bene, V. Vacek, F. Hovorka, Analysis of differences
of surfactants on mass transfer coefficient and interfacial area, Chem. in kL a values determined by steady-state and dynamic methods in
Eng. J. 78 (2000) 1319. 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


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

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

0255-2701/$ see front matter 2004 Elsevier B.V. All rights reserved.
132 I. Rodriguez-Donis et al. / Chemical Engineering and Processing 44 (2005) 131137

by homogeneous azeotropic batch distillation [1]. In con- ponents are selected. From the analysis of residue curve map
trast to batch homogeneous distillation processes, studies features, in particular the decanter phases composition and
on heterogeneous batch distillation (HBD) were scarce the distillation region separatrices curvature, the sequence of
[24] until a new method for the synthesis of heterogeneous products to be drawn from the rectifier column can be pre-
distillation in a batch rectifier was defined and a complete dicted. The number of distillation tasks in the whole process
set of rules for the selection of heterogeneous entrainers for is set by the products targeted purity.
the separation of non-ideal mixtures was established [5,6].
Simulation and experimental results have shown key dif-
ferences between homogeneous and heterogeneous batch 2. Selection of a suitable entrainer for the separation
distillation [6]. Indeed, some advantages of heterogeneous of n-hexaneethyl acetate mixtures by heterogeneous
batch processes compared to homogeneous systems are: (i) azeotropic batch distillation
more suitable candidate entrainers and hence, more design
alternatives for the separation of non-ideal mixtures, (ii) Liquidvapour equilibrium data for the binary mixture
simplified distillation sequences thanks to the liquidliquid n-hexaneethyl acetate have been scarcely reported in the
phase split occurring in some parts of the column and in the scientific literature [8]. Those authors reported a mini-
decanter, (iii) the addition of a smaller amount of entrainer to mum boiling temperature azeotrope with an average mole
the original mixture, (iv) the use of more flexible reflux poli- fraction of n-hexane equal to 0.6565 and a temperature
cies through any combination of the entrainer-rich phase and of 65.15 C at 101.3 kPa. A more detailed experimental
the distillate-rich phase, (v) the still path can cross unstable study, at atmospheric pressure, of liquidvapour equilib-
separatrices allowing the separation of components located rium was performed by Acosta et al. [9]. Their estimation
in different basic distillation regions. The overall cost re- of the azeotropic mixture corresponds to a temperature of
duction generated by these advantages makes heterogeneous 64.85 C with a molar fraction of n-hexane of 0.657 with
batch distillation an attractive option for the separation of an average experimental error for the temperature and com-
non-ideal mixture generated by specialty industries. position measurements of 0.02 C and 0.003 mole fraction,
The aim of this paper is to study the performance of het- respectively.
erogeneous batch distillation for the separation of the mix- The candidate homogeneous and heterogeneous en-
ture n-hexaneethyl acetate. This azeotropic mixture is often trainers are investigated applying the rules reported by
used in pharmaceutical industries for separating bioactive Rodriguez-Donis et al. [1,5]. The existence of binary and
substances by liquid chromatographic processes. The sepa- ternary azeotropes is sought by using literature data [8]
ration of this mixture by batch distillation was first studied and by simulating liquidvapour and liquidliquidvapour
by Rodriguez-Donis et al. [7], by using acetone as a homo- equilibrium by using the BibPhyAddIn Macro in an Excel
geneous entrainer. As the separation by using a single batch spreadsheet [10]. Confirming the need for a complex hybrid
distillation process with acetone was not possible, a hybrid process in case a homogeneous entrainer like acetone is
process was proposed involving a homogeneous azeotropic used [7], no homogeneous entrainer matching the reported
distillation column with acetone and a liquidliquid extrac- rules has been found whereas some suitable heteroge-
tion column with water. To circumvent this complex pro- neous entrainers have been identified; the composition and
cess, we propose in this paper a simpler process by using temperature of their azeotropic mixtures with the original
heterogeneous azeotropic batch distillation. Based on en- components are displayed in Table 1 [8].
trainer selection rules established by Rodriguez-Donis et al. As seen in Table 1, the boiling temperature of the pos-
[5], which consider the thermodynamic behaviour of mul- sible entrainers is either lower or higher than that of the
ticomponent mixtures, an initial list of potential entrainers original components. All candidates form a binary hetero-
partially miscible with one or both original azeotropic com- geneous azeotrope with n-hexane. Methanol and acetonitrile

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 Binary azeotrope with Ternary azeotrope
n-hexane (68.7 C) ethyl acetate (77.1 C)
( C) xmolar a T ( C) xmolar a T ( C) xmolar a T ( C)

Methanol 64.7 0.5100b 50.6 0.708 62.05

Acetonitrile 81.5 0.5755b 54.4 0.6730 76.1
Water 100.0 0.7860b 61.8 0.7000b 70.5 0.5764b 60.7
Nitromethane 101.2 0.7271b 62.0 Zeotropic
a The mole fraction indicated is that of the lightest component in the mixture.
b Heteroazeotrope.
I. Rodriguez-Donis et al. / Chemical Engineering and Processing 44 (2005) 131137 133

Table 2 Table 3
Binary interaction parameters for NRTL and UNIQUAC models Purity of n-hexane-rich phase (xI ) and entrainer-rich phase (xII ) at 25 C
Binary parameters Aij Aji aij Entrainer Binary azeotrope with Ternary azeotrope
Binary coefficients for NRTL model (cal/mol) n-hexane
n-Hexaneethyl acetate 720.731 33.844 0.1521 xnhexane
I xnhexane
n-Hexaneacetonitrile 924.616 1212.64 0.2000
Ethyl acetateacetonitrile 580.783 253.247 0.1997 Methanol 0.9152 0.1198 0.5346
Acetonitrile 0.9176 0.0605 0.3991
Binary coefficients for UNIQUAC model (cal/mol) Water 0.7028 0.2481 0.1265
n-Hexaneethyl acetate 371.892 133.73 0.1754 0.0324
n-Hexanemethanol 1451.225 7.831
Ethyl acetatemethanol 770.688 147.777 Nitromethane 0.9356 0.0893 0.2464 -
n-Hexanewater 2292.190 324.669
Ethyl acetatewater 768.058 214.271
n-Hexanenitromethane 1080.270 7.335 mixture n-hexaneethyl acetate-entrainer displayed in Fig. 1
Ethyl acetatenitromethane 181.416 42.973 indicate the temperature and the topological stability of all
singular points. As the heteroazeotropes are unstable nodes,
add a homogeneous minimum boiling azeotrope with ethyl they are recovered in the decanter where the phase rich in
acetate. Nitromethane forms a zeotropic mixture with ethyl one of the original components LW is removed as distillate,
acetate, whereas water is the only component that forms an whereas the other original component is retained into the
additional heterogeneous azeotrope with ethyl acetate and still at the process end. In some cases, the composition of the
also adds a heterogeneous ternary azeotrope to the result- phase rich in original component defined by the liquidliquid
ing mixture. Experimental liquidvapour equilibrium data equilibrium at the decanter temperature may not meet purity
are available for all binary entrainer- ethyl acetate mixtures requirements; then, further distillation tasks are needed.
[11]. However, for the binary system entrainer-n-hexane that One of the key thermodynamic parameter in heteroge-
exhibits a significant miscibility gap, the only experimental neous batch distillation is the decanter split ratio . It is de-
thermodynamic information is given in Table 1, except for fined by the position of the heteroazeotrope composition on
methanol for which liquidvapour equilibrium data were re- its liquidliquid tie line at the decanter temperature or alter-
ported by Gmehling et al. [11]. All previous experimental natively by the mole ratio of the entrainer-rich phase LR to
data are used to regress binary interaction parameters for the the overall liquid phase Lo into the decanter as follows:
UNIQUAC and NRTL models with ChemCAD IV minimiz- LR = Lo (1)
ing the mean square deviation between the experimental and
calculated mole fraction in the liquid-phase (Table 2). Then, Then, the molar quantity of entrainer-weak/original
ternary liquidliquidvapour equilibrium is determined by component-rich phase LW is given by equation (2):
using these binary coefficients with BibPhyAddIn [10]. As LW = (1 )Lo (2)
a result of the use of these binary interaction parameters
regressed on binary systems to predict the behavior of a As stated by Rodriguez-Donis et al. [6], the reflux policy
ternary system, we have detected the existence of the ex- to be used is strongly influenced by the split ratio at the
perimental heterogeneous ternary azeotrope in the mixture decanter. If the liquid reflux needed at the top of the column
n-hexaneethyl acetatewater, which has the lower boiling is lower than LR , then the distillation can be performed by
point in the system (unstable node). using only the reflux of entrainer-rich phase. Otherwise, the
For the synthesis of heterogeneous batch distillation separation of original components requires the reflux of a
the liquidliquid envelope at the decanter temperature is combination of both decanted phases.
considered in addition to the residue curve map. There- Nitromethane shows the simplest residue curve map with
fore, the binary interaction parameters used in predicting one unstable curved separatrix dividing the triangle in two
liquidliquid equilibrium are estimated from binary hetero- basic distillation regions. Methanol and acetonitrile give rise
geneous azeotrope or liquidliquid equilibrium data [8,10]. two binary azeotropic mixtures and three distillation regions
Table 3 shows the calculated purity of original compo- that are bounded by two unstable curved separatrices. Water
nents in each phase split at 25 C for all heterogeneous shows the most complicated residue curve maps, due to the
azeotropes reported in Table 1. The thermodynamic mod- presence of a ternary azeotrope and a miscibility gap with
els and binary coefficients used in the calculation of the both the n-hexane and the ethyl acetate component. In all
liquidliquidvapour equilibrium, liquidliquid equilibrium four cases, the heteroazeotrope (binary or ternary) has the
at 25 C and the separatrices are reported in Table 2. lowest boiling temperature of the system. As it can be seen in
From the thermodynamic information given in Tables 13, Table 3, all entrainers except water provide the n-hexane-rich
the residue curve maps are drawn in Fig. 1 for each entrainer phase LW as distillate product with a purity better than 0.91.
and the batch distillation task sequence required to perform Water is not a desirable entrainer because of the existence
the separation of the original components is deduced using of ternary azeotrope whose n-hexane-rich phase has a water
published rules [5]. The residue curve maps of the ternary 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 (MESH) equations which are solved for the whole column,
purity of both liquid phases at 25 C, methanol and acetoni- decanter included and taking into account the liquidliquid
trile are better entrainers than nitromethane. Methanol is a phase split. Numerical treatment of the Differential Alge-
light entrainer that would result in a ternary system with a braic Equation (DAE) system and discrete events handling is
low boiling temperature that in turn will likely increase the performed with DISCo, a numerical package for hybrid sys-
cost of the condenser cooling system and the make-up of tems with a DAE solver based on Gears method. The col-
fresh entrainer to be added in the next batch. Besides, the umn technical features and operating conditions are shown in
mole ratio of the product-to-be n-hexane in the methanol Table 4. A sequence of two operational batch steps, namely
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 Table 4
by heterogeneous azeotropic batch distillation. 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
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

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
Fig. 2. Simulation and experimental results of the separation of n-hexane n-hexane to be recycled at the top of the column is deter-
ethyl acetate by using batch distillation with acetonitrile as a heavy
heterogeneous entrainer.
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
an infinite reflux step and a distillate removal step described definition of an optimal reflux policy.
in detail below are simulated. Fig. 2 shows the simulation results, the still path and the
The amount of heterogeneous entrainer to be added to column liquid profile, at the end of batch step 1 (infinite
the original binary mixture is established from material bal- reflux) and at the beginning and end of batch step 2 (dis-
ances considering that the entrainer composition in the still tillate removal). Table 5 displays significant process values.
should be less than 0.01 at the end of infinite reflux batch As can be seen in Fig. 2 and Table 5, only 12.2% of acetoni-
step 1. Simulation shows that this composition is reached trile is added to the initial azeotropic binary mixture and the
after two hours, which is the end of batch step 1. It should resulting ternary mixture lies in the distillation region I. it
be noted that the initial ternary composition F is placed in has been observed [6] that heterogeneous batch distillation
distillation region I that is not limited by the vertex of ethyl requires less entrainer than homogeneous batch distillation.
acetate (see Fig. 2). Therefore, in order to reach the vertex The liquid profile in the column at the end of the infinite
of ethyl acetate located in region II at the end of the distil- reflux operation links the still composition FR to the bi-
lation process, the still path must cross the unstable separa- nary heteroazeotrope n-hexaneacetonitrile in the first top
trix connecting the heteroazeotrope with the homoazeotrope tray of the column through the vapour line. Then, applying
n-hexaneethyl acetate. As established by Rodriguez-Donis the reflux policy detailed in Table 4, distillate removal starts.
et al. [6], such a crossing is possible with different reflux Five minutes after the beginning of the distillation step,
policies, either the reflux of entrainer-rich phase or a com- 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 in the composition for all components are found to be 0.1
phases. The reason is that the liquid phase leaving the first and 0.3%, respectively.
tray of the column top has a composition (xT ) inbetween the As shown in Fig. 2, the still path obtained experimentally
heteroazeotrope and the acetonitrile-rich phase composition (circles) is in excellent agreement with the still trajectory
and is located in a distillation region (III) different from re- calculated by simulation. The selected reflux policy permits
gion (I) which contains the still composition at infinite reflux the still path to cross the separatrix into another distillation
FR . Consequently, the simulated still path crosses the un- region than the initial feed region. Hence, the still path is
stable separatrix separating distillation regions I and II (full able to reach the ethyl acetate vertex and this component re-
black line in Fig. 2). At the end of batch step 2, the still mains pure into the still at the end of the distillate removal
path reaches the ethyl acetate vertex when all n-hexane is step. Such a behavior is not possible with a homogeneous
drawn as distillate product. The operating parameters and entrainer that gives rise to a similar residue curve map be-
are set to obtain the binary heteroazeotrope at the column cause the distillation process is restricted to the distillation
top during the whole step 2. Therefore, the split ratio is region where the initial composition of the mixture is lo-
kept constant. The n-hexane-rich phase is drawn as distillate cated. In this case, ethyl acetate could not be obtained as an
product whereas the acetonitrile-rich phase can be recov- isolated product.
ered at the decanter at the end of the step 2. Table 5 reports Table 5 displays the quantity and composition of the main
the simulated purity and recovery yield of all components. streams of the process. n-Hexane is drawn of as distillate
Simulation computation time is typically less than 4 min on product with a molar purity of 96.8%. Another binary batch
a Pentium III-based computer. 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
4. Experimental verification of the separation of the phase obtained from decanter can be reused in the next batch
mixture n-hexaneethyl acetate by heteroazeotropic distillation process. The make-up of fresh entrainer for the
batch distillation in a bench scale rectification column next cycle of n-hexaneethyl acetate separation is approxi-
mately 52%.
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 5. Conclusions
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 Acetonitrile was chosen as a suitable heterogeneous
3 mm 3 mm. The total packed height is considered to be entrainer for the separation of the azeotropic mixture
equivalent to 45 theoretical plates and the total liquid hold n-hexaneethyl acetate by heteroazeotropic batch distil-
goes up of 52.5 ml. The condenser is subcooled at 25 C and lation from the comparison of the residue curve maps of
total liquid reflux is provided at the top of the column by a several candidate entrainers. We have confirmed the feasi-
solenoid valve. The open/close time of the solenoid valve is bility of the heteroazeotropic batch distillation via simula-
defined considering that 70% of the condensed vapour is re- tion and by experimentation in a bench batch distillation
turned to the column. The 30% remaining condensed vapour column. Good agreement was obtained between simulation
is collected into the decanter (70 ml) where the light phase and experimental results. The novel process with a single
is drawn as distillate (n-hexane-rich phase) whereas a por- heterogeneous distillation column shows a great potential
tion of the heavy entrainer-rich phase is recycled to the top over a reported hybrid process with a homogeneous distilla-
of the column with a peristaltic pump. tion column and a liquidliquid extraction column. Indeed,
Experiments taking into account these operating condi- the consumption of entrainer is low and the separation of
tions (Table 4) are carried out in the bench distillation col- all three components can be performed in a single batch
umn and the results are presented in Fig. 2 and Table 5. distillation step using the defined reflux policy. The ethyl
Analysis of the still composition is performed every 0.5 h. acetate purity is high enough to enable its recycling in a
The analysis of purity of the light and heavy phases in the chromatography process whereas the n-hexane-rich phase
decanter and of the distillate product is made at the end of must be purified in a simple non azeotropic batch distillation
the process (4.8 h). The acetonitrile, n-hexane and ethyl ac- process.
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
[1] I. Rodriguez-Donis, V. Gerbaud, X. Joulia, Entrainer selection rules
done using internal standards; the samples were prepared by for the separation of azeotropic and close boiling temperature mix-
weighing on a SARTORIUS BP 211D balance with a pre- tures by homogeneous batch distillation, Ind. Eng. Chem. Res 40
cision of 10 mg. The standard accuracy and reproducibility (2001) 27292741.
I. Rodriguez-Donis et al. / Chemical Engineering and Processing 44 (2005) 131137 137

[2] R. Dssel, J. Stichlmair, Separation of azeotropic mixtures by batch [7] I. Rodriguez-Donis, U. Juregui-Haza, E. Pardillo-Fondevila, Sepa-
distillation using an entrainer, Comp. Chem. Eng. 19 (1995) s113 ration of n-hexane-ethyl acetate mixtures by azeotropic distillation,
s118. Latin Am. Appl. Res. 29 (1999) 119127.
[3] J. Khler, H. Haverkamp, N. Schadler, Zur diskontinuierlichen Rekti- [8] J. Gmehling, J. Menke, J. Krafczyk, K. Fischer, Azeotropic Data,
fikation azeotroper Gemische mit Hilfsstoffeinsatz, Chem. Ing. Tech. Verlag Chemie, Weinheim, 1994.
67 (1995) 967971. [9] J. Acosta, A. Arce, J. Martnez-Ageitos, E. Rodil, A. Soto, Vapour
[4] S. Skouras, S. Skogestad, Separation of ternary heteroazeotropic liquid equilibrium of the ternary system ethyl acetate + hexane
mixtures in a closed multivessel batch distillation-decanter hybrid, + acetona at 101.32 kPa, J. Chem. Eng. Data 47 (2002) 849
Chem. Eng. Proc. 43 (2004) 291304. 854.
[5] I. Rodriguez-Donis, V. Gerbaud, X. Joulia, Heterogeneous entrainer [10] S.A. ProSim, 2000, http://www.prosim.net, Reference Manual.
selection rules for the separation of azeotropic and close boiling [11] J. Gmehling, U. Onken, VapourLiquid Equilibrium Data Collection,
temperature mixtures by heterogeneous batch distillation, Ind. Eng. DECHEMA Chemistry Data Series, vol. 1, 12 parts, DECHEMA,
Chem. Res. 40 (2001) 49354950. Frankfurt am Main, 1982.
[6] I. Rodrguez-Donis, V. Gerbaud, X. Joulia, Feasibility of heteroge-
neous batch distillation, AIChE J. 48 (2002) 11681178.
| |

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
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