Вы находитесь на странице: 1из 9

Chemical Engineering Science 57 (2002) 1685 – 1693

www.elsevier.com/locate/ces

Modelling of granulation by a two-stage auto-layering mechanism


in continuous industrial drums
R. Venkataramanaa , P. C. Kapur a; ∗ , S. S. Guptab
a Tata Research Development and Design Centre, 54-B Hadapsar Industrial Estate, Pune, 411 013, India
b Research & Development Division, Tata Steel, Jamshedpur, 831 001, India

Received 30 September 2001; received in revised form 26 November 2001; accepted 25 January 2002

Abstract

A mechanistic model for the granulation of particulate materials with a wide size distribution in a large-scale continuous drum is
presented. It takes cognizance of the e6ect of relevant process variables: feed size distribution, moisture content, binders such as lime,
residence time distribution, feed rate, etc. The model is based on the auto-layering mechanism of granule growth and incorporates a
piecewise linear model for granulation kinetics. Laboratory scale tests on a batch drum are used to provide kinetic parameters. The
size-dependent residence time distribution of agglomerating mass in the continuous drum is represented by a combination of mixed and
plug ;ow regimes operating in parallel. The model is customized for a continuous drum in an iron ore <nes sintering plant. The predicted
granule size distributions are in good agreement with the plant data under widely varying operating conditions. The modelling framework
provides scope for modifying the individual modules for drum residence time distribution or the granulation mechanism and growth
kinetics. ? 2002 Elsevier Science Ltd. All rights reserved.

Keywords: Granulation; Auto-layering mechanism; Modelling; Growth kinetics; Industrial drums; Residence time distribution

1. Introduction which can be embedded in a population balance paradigm.


Adetayo, Litster, Pratsinis, and Ennis (1995b) used an em-
Size enlargement of <ne moist particulate solids by gran- pirical two-stage coalescence kernel and incorporated it into
ulation or balling is an important unit operation in several the discretized population balance of Hounslow, Ryall, and
process industries, for example, drugs and pharmaceuticals, Marshall (1988) to model the granulation of fertilizers. Ade-
foods, fertilizers, ceramics, iron and steel, and minerals and tayo and Ennis (1997) devised a cuto6 kernel for granule
materials processing (Pietsch, 1991; Ennis, Green, & Davis, coalescence and were able to match the limestone balling
1994). Recently, Ennis and Litster (1997) have broadly data of Kapur and Fuerstenau (1964) and the fertilizer gran-
classi<ed the granule growth regimes as: (1) wetting and ulation data of Adetayo et al. (1993). The predictive capa-
nucleation, (2) consolidation and growth, and (3) breakage bility of population balance models is critically dependent
and attrition. Size enlargement occurs through several mech- on the choice of the coalescence kernel. However, the ker-
anisms such as coalescence, layering or snowballing, and nels reported so far are mostly empirical or semi-empirical
abrasion transfer (Kapur, 1978), acting alone or in combi- (Adetayo, Litster, & Cameron, 1995b), although recently
nation. attempts have been made to understand the phenomena at
Rotary drums and disks are commonly employed for the the microscopic level and utilize this knowledge either for
granulation of fertilizers and iron ore <nes sinter and pellet delineating the growth regimes or for developing the coa-
feeds. The raw feed is normally characterized by a wide size lescence criteria (Ennis, Tardos, & Pfe6er, 1991; Simons,
distribution in the case of fertilizers and sinter feeds while Seville, & Adams, 1993; Thornton & Ning, 1998; Iveson &
the pellet feed is relatively narrower in size distribution. Litster, 1998; Liu, Litster, Iveson, & Ennis, 2000).
Coalescence appears to be the main mechanism of growth Auto-layering is the principal mechanism in the gran-
in the case of fertilizers (Adetayo, Litster, & Desai, 1993) ulation of sinter feeds (Litster, Waters, & Nicol, 1986;
Kapur, Kapur, & Fuerstenau, 1993). Litster and cowork-
∗ Corresponding author. Tel.: +91-20-6871058; fax: +91-20-6810921. ers developed population balance models for the granula-
E-mail address: pck@vsnl.com (P. C. Kapur). tion of single (Litster et al., 1986) and multicomponent

0009-2509/02/$ - see front matter ? 2002 Elsevier Science Ltd. All rights reserved.
PII: S 0 0 0 9 - 2 5 0 9 ( 0 2 ) 0 0 0 7 0 - 2
1686 R. Venkataramana et al. / Chemical Engineering Science 57 (2002) 1685–1693

sinter feeds (Waters, Litster, & Nicol, 1989). A novel of granulation is to eliminate <nes by the simple expedient
size-interval-by-size-interval marching algorithm was de- of layering these onto coarse particles which act as seeds.
veloped by Kapur et al. (1993) to describe the auto-layering Raw feed is introduced at one end of the drum through a
mechanism which incorporates an intrinsic size exclusion belt conveyor. Water is sprayed onto the materials in the
principle that prevents layering of particles larger than initial stages of the drum. As solids move through the drum,
the maximum possible layer thickness. Granule growth granulation occurs primarily due to auto-layering and the
ratio and layer thickness are the important parameters in product granules are discharged from the other end of the
auto-layering models and are dependent on feed character- drum. Millions of tonnes of sinter feed is granulated in this
istics. Kapur and coworkers (Kapur, 1995; Venkataramana, manner worldwide.
Gupta, & Kapur, 1999) developed correlations for these
parameters as functions of measurable process variables. 2.1. Auto-layering mechanism
In spite of their widespread industrial applications, very
few attempts have been reported on modelling continuous Granulation in a continuous operation depends on three
industrial granulation drums. The models mentioned above principal factors: mechanism of growth, growth kinetics and
were tested with the experimental data from laboratory-scale material ;ow characteristics of the drum. Accordingly, the
batch drums. One of the main di6erences between a batch present model has been developed by incorporating individ-
and a continuous process lies in the material transport. In the ual models for these three phenomena. In the auto-layering
case of iron ore <nes sinter feed, due to the heterogeneous mechanism, sinter feed particles are classi<ed as seeds, inter-
nature of moist raw materials, mixing is highly nonuniform mediates and <nes (Litster et al., 1986; Kapur et al., 1993).
and as a consequence, nonideal ;ow patterns develop in the The coarse particles act as nuclei or seeds onto which the
continuous drum. Adetayo et al. (1995a) developed a model <nes are layered. Depending primarily upon the moisture
for a continuous drum but assumed that the materials have content, intermediate size particles act as either nuclei or
a uniform plug ;ow. The only exception appears to be the layering <nes, or take no part in granulation (Furui, Kawazy,
model developed by Kapur, Sastry, and Fuerstenau (1981). Sugawara, Fujiwara, Kagawa, Sawamura, & Uno, 1977;
They assumed a tanks-in-series model to represent the ma- Nagano, Noda, Yanaka, Shiobara, & Yamaoka, 1985; Lit-
terial transport and tested the predictions with data from an ster et al., 1986; Peters, Beer, Kropla, & Rinne, 1989). The
industrial pelletization drum at steady state. It is important layering mechanisms depend on the feed characteristics and
to note that because of the coupling between granule growth moisture content. It is convenient to describe these mecha-
kinetics and residence time distribution (RTD), it is neces- nisms in terms of two limiting postulates (Kapur, 1995). In
sary to have an independent veri<cation of the growth kinet- the t-postulate, a layer of <nes of <xed thickness t is formed
ics, say in a batch drum. Otherwise, the modelling exercise on the coarse particles, irrespective of the seed size x (Peters
could reduce to mere parameter <tting rather than represent- et al., 1989). The granule size y becomes
ing the actual process.
We have previously developed a model based on the y = x + 2t: (1)
two-stage auto-layering mechanism and tested it for the This mechanism implies that the rate of pick-up of sticky,
granulation of sinter feeds in a laboratory-scale batch drum slime-like mass of <nes is proportional to the surface area
(Venkataramana et al., 1999). In the present paper, the batch of the rolling granule. The layer thickness can be related
model is extended for a continuous granulation process. The to the distribution of <nes and moisture content by taking
model is customized for a large-scale continuous drum in a a mass balance on <nes that are coated (Venkataramana et
sintering plant and validated under di6erent operating con- al., 1999). In k-postulate (Litster et al., 1986), granule size
ditions. The modelling framework presented here could be is proportional to the seed size:
applicable to a variety of granulation devices operating un-
y = kx: (2)
der di6erent situations and for di6erent materials.
This mechanism arises when the rate of layering is propor-
tional to the volume of the granule. The constant of propor-
2. Model of the continuous granulation drum tionality k is strongly dependent on the moisture content, and
it can be derived, at least for idealized systems, by taking a
The iron ore <nes granulation process is carried out in water balance (Kapur, 1995). It is readily shown that the pro-
a long cylindrical drum. The feed consists of a variety of portionate growth of granules should lead to self-similar or
materials with widely di6erent chemical compositions and self-preserving granule size distributions. This conclusion is
size distributions. Typical sinter feed consists of iron ore in reasonable agreement with the experimental results shown
<nes, coke <nes, ;uxes (limestone and dolomite or dunite), in what follows. Earlier, the proportionate growth hypothe-
recycled sinter, and other iron-bearing wastes generated in sis was invoked to model the crushing and layering mech-
the steel plant. Water and calcined lime are used as the anism that prevails in granulation of relatively coarse, nar-
binding media. Raw feed is characterized by a broad size row size dispersed sands (Kapur, 1971). Litster and Waters
distribution, ranging from 10 to ¡ 0:063 mm. The objective (1988) suggested that layering occurs in two stages: uniform
R. Venkataramana et al. / Chemical Engineering Science 57 (2002) 1685–1693 1687

coating of <nes in the <rst stage followed by proportionate


growth of the intermediate granules. This mechanism can be
designated as a mixed postulate. Detailed analysis of gran-
ules produced in the laboratory-scale batch drum indicated
that neither k nor t-postulate alone could describe the gran-
ule size distribution adequately (Venkataramana, Gupta, Ka-
pur, & Ramachandran, 1997). Hence, the general two-stage
growth mechanism was modelled and incorporated into the
batch granulation model (Venkataramana et al., 1999).
The two-stage growth model is essentially a population
balance model for granulation by auto-layering. However,
rather than assuming some form of a collision kernel, our
model simulates the layering process by making mass
balances on individual size fractions and uses the two
growth factors, t and k to compute the granule growth.
In the <rst stage, all particles <ner than a particular size
are assumed to form a uniform layer of constant thickness
over the coarse particles. The maximum size of <nes that
can adhere is a function of moisture content but is inde-
pendent of nucleus size. As a result, the size of nuclei
increases by 2t. Because of its highly nonlinear structure, Fig. 1. Size distributions of iron ore <nes feed and of granules produced
the proportionate growth second stage is simulated using a in a batch drum and a continuous industrial drum (moisture 6.57%; <nes
3% −0:147 mm; lime 0.86%).
size-interval-by-size-interval marching algorithm (Kapur et
al., 1993). The granules, formed at the end of the <rst stage,
are used as the intermediate feed to the second stage. The distribution is consistently <ner than the laboratory granule
<ne particles in the intermediate feed are coated onto the size distribution. On the whole, the former distribution is
granules and the <nal size of granules is obtained by scal- closer to the feed size distribution than the latter distribu-
ing the size of intermediate granules by the proportionality tion. The di6erence in the performance of batch and contin-
constant, k. The computational details are given in the cited uous drums is primarily due to the material transport in the
references (Kapur et al., 1993; Venkataramana et al., 1999). latter device. Apparently, a fraction of the coarse feed by-
This steady state or equilibrium granule size distribution is passes the granulating charge and the residence time in the
used for computing the granule growth time, discussed later. drum becomes size dependent. One can perhaps draw sim-
ilar conclusions from the sampling trials carried out on a
large-scale granulation drum (Waters, Litster, Nightingale,
2.2. Residence time distribution & Williams, 1987). The drum space factor, de<ned as the
per cent volume occupied by the agglomerating charge, was
Material ;ow and mixing of solids and moisture play a found to decrease from about 14% at the inlet to nearly
signi<cant role in large-scale continuous drums. Intimate 7% at the exit. This suggests that the larger feed particles
mixing of coarse particles, <nes and the binding media are and granules roll out faster than the smaller granules and
important to achieve the desired granule size distribution and <ne particles, which leads to the inference that the residence
quality. However, industrial drums operate far from the ideal time of discrete entities is size dependent. It is assumed that
operating conditions. The situation becomes more complex the larger particles and=or granules follow a plug ;ow-like
as the materials are also heterogeneous in nature, di6ering regime while the smaller particles have a near mixed ;ow
as they do in their size distribution, chemical composition, pattern. Accordingly, a ;ow model is proposed which com-
particle shape and density. The nonuniform mixing pattern prises a fully mixed ;ow in parallel with an ideal plug ;ow
results in a size-dependent distribution of residence time. It is (Fig. 2). The RTD in the fully mixed regime is given by
 
then necessary to identify a reasonably realistic material ;ow
Em ( ) = exp − ; (3)
model for the drum and embed it into the overall granulation f f
model.
where is the mixed ;ow parameter, the time and f the
Granules made from an identical feed in the laboratory-
time duration of the initial near-linear growth region, after
scale batch drum and plant-scale continuous drum turn out
which granule growth virtually ceases, as discussed in the
to be signi<cantly di6erent in appearance and size distri-
next section. If (x) is the fraction of x size particles in feed
bution (Fig. 1). A comparison of the size distributions of
f(x) reporting to the plug ;ow regime, the size distribution
feed and plant granules reveals that the particles above
of feed to the fully mixed ;ow becomes
6 –8 mm grow but little in the plant drum. Moreover, in
the less than about 6 mm size range, the plant granule size fm (x) = [1 − (x)]f(x): (4)
1688 R. Venkataramana et al. / Chemical Engineering Science 57 (2002) 1685–1693

Fig. 2. Residence time model of the continuous drum.

This feed results in a granule size distribution, Gm (y; ) in a


residence time interval, , when calculated by the two-stage
granule growth model. The plug ;ow of duration, p , can
be represented by the delta function
Ep ( ) = ( − p ): (5)
The feed to the plug ;ow stream is
fp (x) = (x)f(x): (6)
This feed results in a granule distribution designated as
Gp (y; p ). The drum output is the sum of fully mixed and
plug ;ow streams. Hence, at steady state the granule size
distribution in the drum discharge is given by the following
expression:
 f
G(y) = Em ( )Gm (y; ) d + Gm (y; f )
0
 ∞ Fig. 3. Kinetics of granulation in a batch drum for feeds di6ering in
× Em ( ) d + Gp (y; p ): (7) moisture and <nes content.
f

Note that Gm (y; ) is a function of fm (x) and Gp (y; p ) is a observed somewhat shorter time durations (Venkataramana
function of fp (x). It is now necessary to have a quantitative et al., 1997). The di6erences are most likely due to the
knowledge of the variation of granule size distribution with nature of the feed, its moisture content and the granulation
time and the time required for the granulation process to conditions.
achieve equilibrium. This information can be obtained from In order to estimate the parameters for the growth kinet-
the kinetics of granulation. ics, several experiments were conducted in the batch drum at
granulation times of 1, 2, 3 and 6 min under di6erent oper-
2.3. Kinetics of granulation ating conditions of moisture content, lime content and <nes
level in raw feed. The <nes level is de<ned here as the cu-
Under normal operating conditions, granules grow rapidly mulative mass per cent of particles smaller than 0:147 mm.
at <rst to a near limiting size and thereafter growth vir- In these experiments, moisture content varied from 5.05%
tually ceases. Growth periods of 12 min (Nagano et al., to 7.18%, lime content from 1.66% to 2.54% and the <nes
1985), 10 min (Mizukami, Murata, Shibuta, Itoh, & Toku- varied from 12.68% to 20.17%. The granule size distribu-
take, 1986) and 30 min (Litster & Waters, 1990) have been tions were measured by screening at the end of each run.
reported. Sampling trials inside a large-scale continuous The calculated granule mean sizes (surface-volume mean
drum showed that growth is steep immediately after the ad- diameter) from a few representative runs are plotted against
dition of water and the mean granule size remained more or time in Fig. 3. It shows that initially the mean granule size
less constant thereafter, indicating that granule growth oc- increases steeply with time and then growth virtually stops,
curs only during the <rst few minutes of granulation (Waters or undergoes a slow rate of growth=decay or even ;uctuates
et al., 1987). In our laboratory batch experiments, we have somewhat. In all likelihood, these phenomena are driven by
R. Venkataramana et al. / Chemical Engineering Science 57 (2002) 1685–1693 1689

Fig. 5. Size distributions of feed, +1 mm fraction of feed, and of granules


Fig. 4. Comparison of measured and predicted granule growth rates. formed after 1, 2, 3 and 6 min of granulation time in a batch drum
(moisture 6.47%; <nes 16.42% −0:147 mm; lime 1.69%).

abrasion transfer mechanism, <rst described by Kapur and


Fuerstenau (1964). Because of inertia and friction within
the sticky charge, a signi<cant lag occurs when the drum is It is now possible to predict f , after rearranging Eq. (9),
started or stopped. These end e6ects introduce considerable the time duration of the steep near-linear growth region, re-
noise in the measurement, more so in experiments of short quired in Eq. (7). The equilibrium mean granule size,y( P f)
durations. In view of these uncertainties, it is assumed for can be computed from the granule size distributions simu-
modelling purposes that the growth kinetics can be divided lated by the two-stage growth model discussed earlier for
into the following two idealized stages: the given conditions of moisture and feed characteristics.
y( )
P = xP + b ; 0 ¡ 6 f ; (8a)
2.4. Self-similarity of granule size distributions
y( ) P f );
P = y( ¿ f ; (8b)
where y( )
P is the mean granule size at time and f is the It is known that the feed size distribution has a
time to reach the end of the initial fast growth stage. The template-like e6ect on the granule size distribution (Kapur,
parameter xP is the mean feed size and b is the granule growth 1995). Fig. 5 shows the granule size distributions produced
rate, de<ned as the rate of change of mean granule size with at di6erent time intervals along with the size distributions
time: of feed and that portion of feed (¿ 1 mm) which acts as
y(
P f ) − xP nuclei. It will be noticed that the granule size distributions
b= : (9) resemble that of the nuclei fraction in the feed more than
f
the total feed, suggesting the auto-layering mechanism and
Both the moisture content and the <nes in the feed the template e6ect in the proportionate growth postulate.
(−0:147 mm) a6ect the rate of growth. The current under- Fig. 6 demonstrates that the distributions in Fig. 5 are ap-
standing of the interaction between moisture and <nes and proximately self-preserving when plotted as a function of
their combined e6ect on growth rate is still not clear enough a dimensionless normalized size, y= y. P This master curve,
to derive a theoretical model. Therefore, the experimental which is the <ngerprint of the agglomeration system, can
data for the computed growth rates were utilized to develop be <tted to an appropriate mathematical function, which in
the following least-square model: turn permits us to predict the size distribution of granules
b = −31:69 + 5:77 ∗ MST + 2:11 ∗ FNS for any mean granule size by simply rescaling the size. As
a consequence, the problem of kinetics of granulation is
−0:33 ∗ MST ∗ FNS (mm=min); (10) reduced to determining the variation of mean granule size
where MST and FNS are the measured moisture content and with granulation time.
the <nes level in the feed, respectively. The model and the P y)is
If Z(y= P the functional form of the master curve in
experimental granule growth rates are compared in Fig. 4. Fig. 6, then the evolution of granule size distribution as a
1690 R. Venkataramana et al. / Chemical Engineering Science 57 (2002) 1685–1693

where p1 and p2 are the <tting parameters. Fig. 7 compares


the granule size distributions back calculated from the simi-
larity function with the experimental data for a di6erent feed
with a lower amount of <nes. The granule size distributions
at di6erent times were computed using Eq. (12) after rescal-
ing the size with experimentally determined mean granule
sizes at the corresponding times.

3. Plant sampling campaigns and data collection

As mentioned earlier, the goal of this work was to develop


a mathematical model for an industrial continuous granu-
lation drum. Data collected from the granulation drum in
Sinter Plant # 2 at Tata Steel, Jamshedpur, India was used
for model testing and validation. Granulation is carried out
in a 15 m long cylindrical drum with a diameter of 3:8 m.
The drum rotates at a speed of 5:2 rpm and the throughput
is about 400 tph.
The raw sinter feed to the drum consists of Base Mix,
recycled sinter <nes, coke and ;ux trimmings and calcined
Fig. 6. Self-similar size distributions of nuclei and granules showing the
lime. Base Mix is prepared in advance in the Bedding and
template e6ect of nuclei size distribution (moisture 6.47%; <nes 16.42%
−0:147 mm; lime 1.69%). Blending Plant and is essentially a mixture of blended iron
ore <nes, coke breeze, limestone, dunite, calcined lime and
iron bearing steel plant wastes. A number of sampling cam-
paigns were conducted in Sinter Plant # 2 to collect samples
of Base Mix, recycled sinter <nes, calcined lime and Green
Mix (the product of the granulation drum). The amount of
each sample collected was about 60 kg and the entire sam-
ple was used for determining the size analysis using stan-
dard sieve series in order to minimize the possible errors
in sampling. Screen analysis of granules did not pose any
problems as they were reasonably strong and did not break
during sieving. This is possibly due to the cementing e6ect
of the calcined lime. The Base Mix and Green Mix sam-
ples were analysed for their moisture content. Flow rates of
Base Mix, recycled sinter <nes, coke and ;ux trimmings,
calcined lime and water were recorded for determining the
<nal composition of the raw sinter feed and moisture avail-
able for granulation. Several sampling campaigns were con-
ducted to collect data over a wide period of time and range
Fig. 7. Comparison of granule size distributions back calculated from the of variables.
similarity function, Eq. (12), with the experimental data (moisture 6.54%;
<nes 12.68% −0:147 mm; lime 1.70%).

4. Model customization and validation


function of time is given by
P Model parameters were <tted and tuned by using the plant
G(y; ) = Z(y=[xP + b ]); (11)
data and the laboratory batch data. Data from the experi-
which can be incorporated into the continuous granulation ments on batch kinetics were employed to <t the parameters
P y)
model. Note that the function Z(y= P can be obtained from in the self-similar size distribution function in Eq. (12).
the nuclei portion (+1:0 mm) of the feed, as evident from The granule growth rate, b, was calculated using Eq. (10)
Fig. 6. The following self-similar size distribution function, and the time duration of the near-linear growth region, f
also shown in Fig. 6, was <tted to the experimental data: was computed using Eq. (9). The equilibrium mean granule
     p2  P f ), was predicted using the two-stage auto-layering
size, y(
y y growth model (Venkataramana et al., 1999). The plant
G = 100 1 − exp −p1 ; (12)
y( )
P y( )
P drum data were used to tune the RTD model parameters in
R. Venkataramana et al. / Chemical Engineering Science 57 (2002) 1685–1693 1691

Fig. 8. Split function; fraction of particles reporting to the plug ;ow Fig. 9. Comparison of plant and model-predicted granule size distributions
regime. under normal conditions (moisture 6.33%; <nes 14.42% −0:147 mm; lime
2.27%).

Eqs. (3) and (5), and derive the split function. The mixed
;ow parameter, and the duration of plug ;ow, p were
obtained as 0.4145 and 1:8 min, respectively. Utilizing a
few sets of data from the drum, the following equation
for the split function (fraction of particles in the feed ex-
hibiting plug ;ow) was developed and its parameters were
estimated by <tting the model in Eq. (7) to plant data by
standard search techniques:
(x) = 0 {1 − exp(−1 x2 )}: (13)
The split function, (x), was assumed to be invariant of op-
erating conditions other than drum speed and feed rate which
were kept constant. Fig. 8 shows a plot of the split function
in Eq. (13) along with estimated values of the parameters,
’s. It will be seen that a majority of particles greater than
about 6 mm in the feed bypass the granulation mix and do
not undergo any signi<cant growth. This result is in confor-
mity with the experience of plant operators. Fig. 10. Comparison of plant and model-predicted granule size dis-
tributions under high feed moisture (moisture 7.66%; <nes 11.54%
The model was subsequently used in a predictive mode
−0:147 mm; lime 2.24%).
without adjusting any parameters. Only a few representa-
tive results can be shown here. Fig. 9 shows the model pre-
dictions obtained under normal plant operating conditions. raw feed and in this case too the model predictions are in
However, industrial processes operate under ;uctuating con- good agreement with the actual granule size distributions.
ditions due to the variations in the raw material characteris- Not infrequently, new materials are granulated or operat-
tics. For example, the size distribution of feed depends on ing conditions are altered in the plant. One would normally
the performance of the grinding units and the nature and resort to laboratory- or pilot-scale tests to assess the “gran-
hardness of ores, coke and limestone. Similarly, the mois- ulability” of the new feed and establish the appropriate op-
ture content of materials varies with season and each compo- erating conditions. In such a situation the model could be
nent of feed can have a di6erent moisture content depending highly useful as a predictive tool for translating the results
on its moisture absorbing capacity and particle porosity. It obtained in the laboratory studies to the industrial drum. In
is therefore important that the process model performs well an alternate but simpler approach, a few sets of experiments
even when plant operating conditions are di6erent from the on kinetics of granulation can be conducted on the labora-
normal. Fig. 10 shows a condition where moisture avail- tory batch drum to determine the maximum stable granule
able for granulation was much higher than what is optimum. size and the time duration of linear growth. This informa-
Here too, the model predictions are quite close to the plant tion along with the self-similar distribution of the granules
data. Fig. 11 represents conditions of low level of <nes in can then be input to the continuous granulation model to
1692 R. Venkataramana et al. / Chemical Engineering Science 57 (2002) 1685–1693

5. Concluding remarks

Granulation of heterogeneous raw materials with broad


size distributions by a two-stage auto-layering mechanism is
a complex process. The granules from a large-scale continu-
ous drum exhibit lesser growth and a smaller mean size than
those from a laboratory-scale batch drum. The main di6er-
ence between batch and continuous processes is the material
transport and the residence time distribution of particles in-
side the continuous drum that appears to be size dependent.
It is necessary to incorporate the underlying granulation
mechanism, growth kinetics and material ;ow characteris-
tics for modelling the industrial continuous drums. In the
proposed model, growth mechanism and kinetics are based
on extensive previous and present investigations carried out
on a laboratory-scale batch drum (Venkataramana et al.,
Fig. 11. Comparison of plant and model-predicted granule size distribu- 1997, 1999). The material transport, including the split func-
tions under low <nes content (moisture 6.57%; <nes 2.99% −0:147 mm; tion, is admittedly empirical but seemingly plausible. It is
lime 0.86%). inferred from the di6erence in batch and continuous drum
granulated products (Fig. 1), drop in hold-up along the drum
length (Waters et al., 1987), plant operators observations
and back <tting to the plant data.
Even though the model is quite successful in translat-
ing results from the laboratory-scale batch drum to the
large-scale continuous drum and predicts with reasonable
accuracy the granule size distributions under widely vary-
ing operating conditions, the modular nature of the model
allows one to modify any of the model components. For
instance, the material ;ow or RTD representation can be
readily modi<ed if a more de<nitive transport model is
made available.

Notation

b granule growth rate, mm/min


Em residence time distribution in the mixed
Fig. 12. Comparison of plant and model-predicted granule size distribu- ;ow regime, min−1
tions using an alternate simpler approach (moisture 7.79%; <nes 9.45% Ep residence time distribution in the plug
−0:147 mm; lime 2.27%). ;ow regime, min−1
f(x) feed size distribution, mm−1
fm (x) size distribution of feed to the mixed
predict the granule size distribution from the plant drum, ;ow regime, mm−1
without using the two-stage growth model and the model for fp (x) size distribution of feed to the plug ;ow
granule growth rate. Results obtained by this approach are regime, mm−1
presented in Fig. 12 for one of the representative situations. FNS <nes (−0:147 mm) content of feed, %
In order to verify the simpler approach, experiments were G overall granule size distribution, mm−1
conducted on the laboratory drum under operating condi- Gm granule size distribution resulting from
tions similar to those practiced in the plant to determine the mixed ;ow, mm−1
equilibrium mean granule size, y( P f ) and the time duration Gp granule size distribution resulting from
of the near-linear growth region, f . These data were used plug ;ow, mm−1
as input to the continuous granulation model for translating k granule growth ratio, dimensionless
the laboratory results to the plant scale. It will be noticed MST moisture content of feed, %
from Fig. 12 that the model was reasonably successful in t layer thickness, mm
predicting the size distribution of granules from the plant x seed size, mm
drum using the laboratory granulation kinetics data. xP mean feed size, mm
R. Venkataramana et al. / Chemical Engineering Science 57 (2002) 1685–1693 1693

y granule size, mm Kapur, P. C. (1978). Balling and granulation. Advances in Chemical


yP mean granule size, mm Engineering, 10, 55–123.
ZP self-similar size distribution function, Kapur, P. C. (1995). Particle population balance in granulation of iron
ores by an auto-layering mode. In: S. P. Mehrotra, & R. Shekhar
mm−1
(Eds.), Mineral processing: Recent advances and future trends (pp.
703–717). New Delhi: Allied Publishers.
Greek letters Kapur, P. C., & Fuerstenau, D. W. (1964). Kinetics of green pelletisation.
Transactions of AIME, 229, 348–355.
(x) split function, dimensionless Kapur, P. C., Kapur, P., & Fuerstenau, D. W. (1993). An auto-layering
model for the granulation of iron ore <nes. International Journal of
mixed ;ow parameter, dimensionless
Mineral Processing, 39, 239–250.
time, min Kapur, P. C., Sastry, K. V. S., & Fuerstenau, D. W. (1981). Mathematical
p residence time of particles exhibiting models of open-circuit balling or granulating devices. Industrial
plug ;ow, min & Engineering Chemistry Process Design & Development, 20(3),
f time duration of linear growth, min 519–524.
Litster, J. D., & Waters, A. G. (1988). In;uence of the material
properties of iron ore sinter feed on granulation e6ectiveness. Powder
Technology, 55, 141–151.
Acknowledgements Litster, J. D., & Waters, A. G. (1990). Kinetics of iron ore sinter feed
granulation. Powder Technology, 62, 125–134.
The authors thank the managements of Tata Steel, Litster, J. D., Waters, A. G., & Nicol, S. K. (1986). A model for predicting
Jamshedpur and Tata Research Development and Design the size distribution of product from a granulating drum. Transactions
Centre, Pune for the permission to publish this paper. Con- of Iron & Steel Institute of Japan, 26, 1036–1044.
Liu, L. X., Litster, J. D., Iveson, S. M., & Ennis, B. J. (2000).
tinuous encouragement from Prof. Mathai Joseph and Prof. Coalescence of deformable granules in wet granulation processes.
E.C. Subbarao, TRDDC and Dr. T. Mukherjee, Tata Steel American Institute of Chemical Engineers Journal, 46(3), 529–539.
is gratefully acknowledged. Mizukami, S., Murata, H., Shibuta, K., Itoh, R., & Tokutake, K. (1986).
Agglomeration rate of sinter mix in drum mixer. Transactions of Iron
& Steel Institute of Japan, 26, B1.
References Nagano, S., Noda, H., Yanaka, H., Shiobara, K., & Yamaoka, Y. (1985).
Characteristics of sinter feed and its in;uence on sintering. In: C. E.
Adetayo, A. A., & Ennis, B. J. (1997). Unifying approach to modelling Capes (Ed.), Proceedings of the Fourth international symposium on
granule coalescence mechanisms. American Institute of Chemical agglomeration, Toronto, Canada (pp. 191–202).
Engineers Journal, 43(4), 927–934. Peters, K. H., Beer, H., Kropla, H. W., & Rinne, K. (1989). Granulation
Adetayo, A. A., Litster, J. D., & Cameron, I. T. (1995a). Steady state behaviour of sinter mix components, with particular respect to coke
modelling and simulation of a fertilizer granulation circuit. Computers breeze particle size and its in;uence on the iron ore sintering process.
& Chemical Engineering, 19(4), 383–393. In M. Cross, & R. Oliver (Eds.), Fifth international symposium
Adetayo, A. A., Litster, J. D., & Desai, M. (1993). The e6ect of on agglomeration (pp. (51–70)). Rugby, UK: Institute of Chemical
process parameters on drum granulation of fertilizers with broad size Engineering.
distributions. Chemical Engineering Science, 48(23), 3951–3961. Pietsch, W. (1991). Size enlargement by agglomeration. New York:
Adetayo, A. A., Litster, J. D., Pratsinis, S. E., & Ennis, B. J. (1995b). Wiley.
Population balance modelling of drum granulation of materials with Simons, S. J. R., Seville, J. P. K., & Adams, M. J. (1993). Mechanisms
wide size distribution. Powder Technology, 82, 37–49. of agglomeration. Proceedings of the Sixth international symposium
Ennis, B. J., & Litster, J. D. (1997). Particle size enlargement. In: R. on agglomeration, Nagoya, Japan (pp. 117–122).
Perry, & D. Green (Eds.), Perry’s chemical engineers’ handbook (7th Thornton, C., & Ning, Z. (1998). A theoretical model for the stick/bounce
ed.). New York: McGraw-Hill. behaviour of adhesive, elastic-plastic spheres. Powder Technology, 99,
Ennis, B. J., Green, J., & Davies, R. (1994). Particle technology: The 154–162.
legacy of neglect in the U.S Chemical Engineering Progress, 90, Venkataramana, R., Gupta, S. S., & Kapur, P. C. (1999). A combined
32–43. model for granule size distribution and cold bed permeability in the wet
Ennis, B. J., Tardos, G., & Pfe6er, R. (1991). A microlevel-based stage of iron ore sintering process. International Journal of Mineral
characterization of granulation phenomena. Powder Technology, 65, Processing, 57, 43–58.
257–272. Venkataramana, R., Gupta, S. S., Kapur, P. C., & Ramachandran, N.
Furui, T., Kawazy, M., Sugawara, K., Fujiwara, T., Kagawa, M., (1997). Modelling of iron ore sinter feed granulation. Tata Search,
Sawamura, A., & Uno, S. (1977). Technology of preparation of raw 66 –72.
materials to be sintered. Nippon Steel Technical Report Overseas, 10, Waters, A. G., Litster, J. D., & Nicol, S. K. (1989). A mathematical model
36–46. for the prediction of granule size distribution for multicomponent sinter
Hounslow, M. J., Ryall, R. L., & Marshall, V. R. (1988). A discretized feed. Iron & Steel Institute of Japan International, 29(4), 274–283.
population balance for nucleation, growth and aggregation. American Waters, A. G., Litster, J. D., Nightingale, R., & Williams, R. (1987).
Institute of Chemical Engineers Journal, 34(11), 1821–1832. A study of the factors a6ecting the granulation of iron ores. In
Iveson, S. M., & Litster, J. D. (1998). Growth regime map for liquid-bound Proceedings of 15th Australian Chemical Engineering Conference
granules. American Institute of Chemical Engineers Journal, 44(7), (CHEMECA 87), Melbourne, Australia (pp. 104 –111).
1510–1518.
Kapur, P. C. (1971). The crushing and layering mechanism of granule
growth. Chemical Engineering Science, 26, 1093–1099.

Вам также может понравиться