A Model for Predicting the Size Distribution of

Product from a Granulating Drum*

By J. D. LITSTER,** A. G. WATERS** and S. K. NICOL**

Synopsis II. Experimental Equipment and Techniques

Carefully controlled granulation experiments were carried out on a hae- All experiments were carried out in a stainless steel
matite ore with a wide size distribution (0-'8 mm) over a range of mois- batch granulating drum 0.285 m long and 0.310 m
ture contents. At any one moisture content, thefinest size fractions could
in diameter. The drum contained six wedge shaped
be classified solely as layering particles whilst the largest size fractions
acted only as nuclei particles. Between these two extremes,a proportion lifter bars, each 4 mm high (after Kapur and Fuerste-
of particles in any one size fraction behaved as nuclei particles, the re- nau7)). No scraper bar was used. The drum rotated
mainder being layering particles. Hence each particle size fraction could on two rollers and the drum speed was controlled by
he characterised by a partition coefficient. The partitioning of an indi- a variable speed gear box. A measured mass of
vidual size range between nuclei and adhering particles was shown to be water was added to the granulation mix via a syringe
dependenton the moisturecontent. For each moisturecontent, thepartition and tube arrangement. The 6 mm diameter stainless
curve could be modelled using a log-normal function. The variation of steel tube, with 1 mm diameter holes drilled at a
adhering layer mass with granule size was investigated and it was found 5 mmnspacing, ran the full length of the drum. The
that adhering layer thickness was approximately proportional to nuclei holes were aligned so that the water sprayed onto the
diameter at lower moisturecontents. At higher moisturecontents there was
material tumbling in the drum and not onto the
an optimum nucleiparticle size for which the ratio of granule mass to total
nuclei mass was a maximum. drum itself. The sample mass (5 kg) and drum speed
A population balance model to predict granule size distribution was de- (20 rpm) used correspond to a space factor of 12 %
velopedand tested against experimental data. The important modelpa- and a Froude number of 0.034. These values cor-
rameterswere thosedescribing theparticle partition curve. Excellent agree- respond to cascade zone movement of the charge in
ment was found between experimental and model predicted granule size the granulating drum which has been recommended
distributionsfor thefull range of moisturecontentscovered. for good granulation.8~
For the experiments reported in this paper, a single
I. Introduction component system consisting of a high grade haematite
Population balance type models have been used ore from Western Australia was used. The experi-
successfully to model the kinetics and the change in mental procedure was as follows. The ore was care-
size distribution of pelletisation processesl'2~ where fully screened and the sized fractions were then dried
the feed is very fine, typically 90 % by mass less than in an oven at 105 °C for 12 h. The required mass
0.075 mm. The pellets that form have a fairly nar- of each of the size fractions was weighed and placed
row size distribution the shape of which is conserved in the granulating drum. The drum was sealed and
as the pellets grow.3~ In contrast, the feed to the rotated for 2 min to mix the feed. While the drum
granulating process prior to iron ore sintering has continued to rotate, the required amount of water
a wide size distribution, typically 0 10 mm, and the was added over a period of 0.5 to 1 min. The drum
size distribution changes significantly during the was allowed to rotate for a further 5 min and then
granulation process. stopped. The granulated material was transferred
It is largely accepted that the principal mechanism to a rotary sample divider where a 1 kg sample and
in sinter feed granulation is the layering of fine two 0.5 kg samples were split out. The 1 kg sample
particles onto larger nuclei particles. Microscopic was weighed before and after drying in an oven for
studies4,5~ have shown clearly that large granules or 12 h at 105 °C to determine the total moisture content.
" quasi -particles " consist of a nucleus surrounded One 0.5 kg sample was used for frozen granule size
by a layer of fines. However, the approach to charac- analysis and the other was kept as a spare.
terising the granulation process has generally been Granule size analysis was performed by freezing
empirical in nature. Only Rankin and Roller6~ the granules in liquid nitrogen then screening. The
have attempted to predict granule size distribution screened samples were dried for 12 h then weighed.
based on the feed size distribution, but their model This technique is fully described elsewhere.6~ In
assumptions have not been fully tested. In this present some experiments a size analysis was performed on
work, carefully controlled laboratory experiments have each granule size fraction obtained from the frozen
been undertaken to study mechanisms involved in the size distribution screening. This was done by first
granulation of iron ore with a wide size range. Based wet screening the individual granule size fractions at
on these experiments, a preliminary model to predict 0.063 mm, then dry sizing the plus 0.063 mm material.
granule size distribution given the feed size distribu- All size analyses were carried out using a full ~/2
tion is proposed and tested. sieve series.
* Manuscript received on December 10, 1985; accepted in the final form on September 12, 1986. © 1986 ISIJ
** Central Research Laboratories, The Broken Hill Proprietary Co., Ltd, Shortland, New South Wales 2307, Australia.

(1036 ) Research Article

 Transactions ISIJ, Vol. 26, 1986 (1037)

III. Results and Discussion grow an adhering layer of fines depends on its momen-
tum. Using tracer studies, Sastry and Fuerstenau9~
1. Dependenceof Layer Thicknesson Nuclei Particle Size
showed that transfer of a small amount material
The dependence of adhering layer thickness on from one pellet to another by abrasion during a col-
nuclei particle size has received little attention in the lision was an important growth mechanism in pelleti-
literature but it is necessary to know the nature of sation. In the pranulation process, a collision between
this dependence to develop a population balance a large, high momentum particle and a small, low
model for the prediction of granule size distribution. momentum particle could result in the transfer of
Rankin and Roller° assumed that layer thickness was some of the adhering layer from the small particle
independent of nuclei particle size although this to the large particle. As a result of this mechanism,
assumption was not fully tested. In the present work, the mass of adhering layer that a particle could hold
three experiments were carried out to test this assump- would be dependent on the particle mass.
tion. Feed to the drum consisted of three size frac-
tions : minus 0.25 mm (layering particles), 1 to 2 mm 2. Granulation of a Wide Size Range Feed
and 4 to 5.6 mm (nuclei particles). The proportion A series of experiments were carried out using a
of fines was held constant (25 %) and the mass ratio feed with a wide size range and varying the moisture
of large to small nuclei particles was varied. The content from 3.5 to 5.4 %. The feed size distribution
mass of granules formed from each nuclei particle used was similar to that of an operating sinter plant
size range was measured and the mass of adhering
(see Table 2). Granule size distributions for the
particles was determined by difference. different moisture contents are shown in Fig. 1.
Results are shown in Table 1. If the layer thick- Clearly, the extent of granulation increases with
ness was independent of particle size, the mass of moisture. For each experiment, the size distribution
adhering layer would be proportional to the total of particles making up each granule size fraction was
surface area of nuclei particles, i.e., measured by the wet screening procedure described
Mass of adhering layer on 4-5.6 mm particles above. From this data, the partitioning of particles
of a particular size into the various granule size frac-
Mass of adhering layer on 1 2 mm particles
tions can also be determined. An example of the
Total surface area of 45.6 6 mm particles
complete results from one experiment is given in
Total surface area of 1 2 mm particles Table 3.
From the particle partitioning data, it is possible
From Table 1, it is clear that this was not the case.
The ratio of the mass of fines adhering to the large to observe which particle size fractions act as nuclei
and which act as adhering particles. Assuming all
particles compared to that adhering to the smaller size enlargement is by layering and the extent of
particles was 3 to 5 times greater than that predicted
on this assumption. Rather, Table 1 shows that the coalescence is minimal, particles which act as nuclei
mass of adhering fines was approximately proportional will appear in granules of the same sieve size fraction
to the nuclei particle mass rather than surface area. or the next larger sieve size fraction due to the adher-
ing layer of fines increasing the particle size by a
On this basis, the proportion of layering fines on each
nuclei size fraction mass was within ± 30 % of the small amount. Particles which form part of the
observed ratio. Given a constant ratio of adhering adhering layer around a larger nucleus particle will
layer mass to nuclei particle mass, it follows that layer appear in all the larger granule size fractions. The
thickness is proportional to nuclei particle size. This partition coefficient ai is therefore defined as the mass
assumption will be used in the model development fraction of particles from size fraction i which act
outlined in Chap. J V. The dependence of adhering
Table 2. Feed size distribution.
layer mass on nuclei particle mass suggests that during
the granulation process, the ability of a particle to

Table 1. Distribution of la yering material on two

distinct nuclei size fractions.

(1038) Transactions Is", Vol. 26, 1986

Fig . 1. Variation in granule size distribution with moisture content.

Table 3. Granule size distributions and particle partition distributions for 4.3 /o moisture,

Research Article
n

Transactions ISIJ, Vol. 26, 1986 (1039)

as nuclei particles, and can be calculated from the where, x0,5: the particle size with a partition coeffi-
experimental data as follows : cient of 0.5
6 : a measure of the spread of size range of
mZZ+m2j+1 intermediate particles.
a1_ .......................(1)
The position of the partition curve moves towards
:m11
1=i the larger particle size fractions with increasing mois-
ture content. Thus, particles that act as nuclei at
where, m2~: the mass of particles of size fraction i
low moisture contents may adhere at higher moisture
which are found in granule size frac-
contents. For example, at 3.5 % moisture, 75 % of
tion j.
the 0.5 mm particles are nuclei while at 5.4 % mois-
Values of ai =1 and cai= 0 correspond to nuclei and
ture 85 % are adhering (see Fig. 2). The variation
adhering size fractions, respectively. Figure 2 shows
of x0,5 and 6 with moisture content is shown in Fig. 4.
the values of a1 as a function of particle size for various
x0,5increases with moisture. This is analogous to the
moisture contents. It can be seen that there are a
increase in cut off size between adhering and nuclei
significant number of intermediate size fractions where
some of the particles act as nuclei and some of the particles with moisture found by Rankin and Roller.e~
The value of 6 decreased with increasing moisture
particles are adhering, i.e., a2 lies somewhere between content. In other words, the range of intermediate
0 and 1. For example, at a moisture content of
sizes became smaller as the moisture content increased.
5.0 %, ai = 0.1 corresponds to a particle size of
The values of these parameters at a given moisture
0.29 mm whereas a2=0.9 corresponds to 1.4 mm.
content will be strongly dependent on the feed particle
Particles between these two sizes behave as nuclei
size distribution. At this stage, no attempt has been
and adhering particles. These observations differ
made to quantify the effect on x0,5and 6 of changing
from the conclusions of some previous workers. Furui
et a1.4~and Vidal et a1.10~while acknowledging the
presence of intermediate particles claimed they took
no part in the granulation process and remained
unchanged. It was claimedl0~ that all granules were
made up of a large nucleus surrounded by a layer of
ultrafines. Intermediate sized particles neither acted
as nuclei or adhered to larger particles. From Fig. 2,
however, it is clear that the intermediate particle
size ranges do take part in the granulation process,
some as nuclei and some as layering particles. These
conclusions are supported by microscopic observation
of green feed granules sampled from an operating
sinter plant (see Photo. 1). The curves shown in
Fig. 2 closely resemble a cumulative log normal
distribution plot and the same data is plotted on log Photo. 1. Section of green feed granules showing intermediate
probability paper in Fig. 3. In all cases a straight size particles (125-250 1em) acting as nuclei (A) and
line gives a good fit to the data. Thus, for a par- adhering particles (B).
ticular moisture content, the relationship between
partition coefficient and particle size can be repre-
sented by the equation :

a(x)
(x) = 6 ~2r
1 - J ln(x)
-o expL_ (t-ln(x0,5))2
262
...........................(2)

Fig .3. Log-probability plot of partition coefficient vas.

Fig. 2. Variation in partition coefficient with particle size. particle size.

Research Article
nn

(1040) Transactions ISIJ, Vol. 26, 1986

V ~vrx3
1,= 6 and Sp = ~Srx2

Nuclei particles of size x to x+dx will form granules

of size x+24 to x+dx+24. From Eqs. (3) and (7)
the mass of granules of size x+24 to x+dx+24 is:
My2(x+24)dx = Myi(x)a(x)dx

3Pt +yi(x)a(x)24(x)Mdx
Ppc5v x
........................... (8)
Rearranging

y2(x+24)
=yi(x)a(x)
l+K'24(x)
] ..........(9
Fig . 4. Variation in x05 and a with moisture content. x where K= 3Pcc5SI PJ)cbv
which is a combination of mate-
rial properties.
Since 4 is proportional to x.
the size distribution.
Iv. Model Description 24(x)=Ax ........................(10)
With information about particle partitioning, a where, A : a constant.
population balance model can be developed to predict Substituting in Eq. (9):
the final granule size distribution from the feed particle
size distribution. At this stage, the model is for y2(x+Ax) = yi(x)a(x)(1 +KA) ............(11)
a single component feed only and is an equilibrium Integrating Eq. (7), the total mass of adhering mate-
model taking no account of granulation kinetics. rial on nuclei of all sizes
The major assumptions of the model are:
(1) Granulation occurs solely by the mechanism = MKA y1(x)a(x)dx ...............(12)
of fine particles layering on larger nuclei particles.
(2) For a particular size fraction, the proportion
of particles acting as nuclei can be represented by 0 The total mass of particles available for layering
a partition coefficient, a(x), which decreases from
1 and approaches 0 as the particle size decreases. = M 1- yi(x)a(x)dx ........:.....(13)
(3) The thickness of the adhering layer of fines
is proportional to the size of the nucleus particle. 0 Equating Eqs. (12) and (13) :
The model is developed as follows. All masses are
on a dry basis. Let y1 and y2 be the size distribution 1-
density functions of the feed particles and product A = -- '
foyi(x)a(x)dx
-- (14)
granules respectively. Consider a particle of size x K yi(x)a(x)dx
with a layer thickness of 4(x).
Mass of particles size x to x+dx available as nuclei
0 a(x) is given by Eq. (2) above. Thus, given x0,5,a
= Myi(x)a(x)dx .......................(3) and K, Eqs. (2), (14) and (11) can be solved to give
the granule size distribution. In a form to use a
No. of particles of size x to x+dx available as nuclei discrete size distribution, Eqs. (11) and (14) are:

- 6My1(x)a(x)dx y2z= yi~,a~(1+AA) ........................(11-a)

P ,jvrx3 ..................... (4)
1- yiia'i(xii-x11_1)
Mass of adhering layer on a particle of size x to A=t =1 ............(14-b)
x+dx K y1zaz(xiz-x11_1)
= pzcpsrx24
(x) ........................(5) i=1

and x21,= xtz+Ax1i .................................(15)

Therefore,
Total mass of adhering layer on all particles of size The model has three parameters-K, x0,5 and a.
x to x+dx K is a lumped parameter which relates layer thickness
to the mass of layering particles. The voidage of
= PL~Srx24(x)6My1(x)a(x)dx (6) the adhering layer will be approximately 0.3 to 0.4.
Therefore, on a dry basis, p~/p, will be in the range
0.6 to 0.7. Allen"" gives values of cS/qv for a wide
range of materials. All values lie in the range 1.5
to 2.7. Therefore, a realistic range for K is 2.7 to
5.6. Figure 5 shows the granule size distribution

Research Article
 Transactions ISIJ, Vol. 26, 1986 (1041)

predicted by the model using the feed size distribution V. Comparison of Model and Experiment
shown in Table 2 and values of x0,5 and of 0.45 mm Comparisons of experimental and model predicted
and 0.737, respectively (4.3 % moisture). Values granule size distributions are shown in Figs. 6 to 10.
of K of 2.7, 4.0 and 5.6 are used. Clearly the granule In each case the values of x0,5 and Q used were those
size distribution is insensitive to changes in K over calculated from the experimental data (see above).
a realistic range of values. A value of K of 4 is It can be seen that for all moisture contents, there is
therefore chosen for use in the model. Thus, the excellent agreement between the experimental and
model becomes effectively a two parameter model. model predicted size distributions. As the partition
It should be noted, however, that the assumption that coefficients were taken from experimental data, these
granule size distribution is insensitive to the value of results confirm the model assumption that the adher-
K is only strictly valid when the shape factor ratio ing layer thickness is proportional to nuclei particle
0s/cbvis independent of particle size. This is a valid size is adequate.
assumption for a one component system for particles The experimental distribution of the total mass of
larger than 0.125 mm.6~ Although the particle shape layering material with granule size can be determined
factors appear explicitly in the model only in the from the particle size distribution of each granule
lumped parameter K, it is not implied that particle size fraction and is shown in Fig. 11 as a function of
shape is unimportant in the granulation process. moisture content. By the same argument used for
Particle shape is very important in determining the the particle partitioning calculations, nuclei particles
efficiency with which particles act as nuclei5~ and are defined as those from the same size fraction as the
hence the partition coefficient curve for a particular granules or one size fraction smaller, i.e.,
material can be expected to depend strongly on cS
and cv. (Mass granules/Mass nuclei) _ m2~1(m~_~~+m~,)
........................(16)

Fig. 5.
Sensitivity of model to the value of K.
x0 5=0.45 mm, U=0.737

Fig. 6.
Comparison of model and experimental
size distributions at a moisture content
of 5.4 °.

Research Article
(1042) Transactions ISIJ, Vol. 26, 1986

Fig. 7.
Comparison of model and experimental
size distributions at a moisture content
of 5.0 %.

Fig. 8.
Comparison of model and experimental
size distributions at a moisture content
of 4.3 %.

Fig. 9.
Comparison of model and experimental
size distributions at a moisture content
of 4.0 %.

The preliminary experiments described in Sec. III. 1 ratio of granule mass to nuclei mass exhibits a maxi-
suggested that this ratio was independent of granule mum value. In other words, at higher moisture
size. Figure 11 confirms that this is approximately contents, there is an optimum nuclei particle size.
true for a wide size range feed at the lower moisture This suggests that there are competing mechanisms
contents. As the moisture content is increased, the which control the relationship between adhering layer

Research Article
 Transactions ISIJ, Vol. 26, 1986 (1043)

Fig. 10.
Comparison of model and experimental
size distributions at a moisture content
of 3.5 0

Fig. 11.
Distribution of layering material as a
function of nucl ei particle size.

thickness and particle size and that the process is nuclei while other particles of the same size are
more complex than the simple mechanism suggested adhering.
in Sec. 111.1. However, it is clear from a comparison (2) The distribution of particles between adhering
of experimental and model predicted granule size and nuclei characteristics can be represented in terms
distributions that the simplifying model assumption of a partition curve. The shape and position of the
is justified. partition curve are functions of the moisture content.
The preliminary model developed in this paper The variation in partition coefficient with particle
shows great potential for use in the study of the size can by adequately represented `by a log-normal
granulation process and for comparing the granu- function.
lating ability of different ores. There are several (3) The thickness of the adhering layer of particles
areas in which the model requires development and is approximately proportional to the nuclei particle
testing. Before the model can be applied to sinter size at lower moisture contents. At higher moisture
feed, further work is necessary to : contents, there is an optimum nuclei size for which
1) Predict changes in model parameters with changes the ratio of granule mass to nuclei mass is a maximum.
in feed size distribution, (4) The preliminary model presented in this
2) Extend the model to cover multicomponent feeds. paper shows excellent agreement with measured
granule size distribution.
VI. Conclusions
From this study the following conclusions are Nomenclature
drawn : A : Proportionality constant between layer thick-
(1) There is a smooth transition from completely ness and particle size
adhering to completely nuclei particles as particle K: Model parameter related to particle density
size increases. In the intermediate size ranges, for and surface area characteristics
one particular size fraction, some particles act as M: Total mass of system (dry basis)

Research Article
(1044) Transactions ISIJ, Vol. 26, 1986

m27: Mass of particles of size fraction i appearing Acknowledgement

in granules of size fraction j (dry basis) The authors wish to thank Mr P. W. Roller for
Number of size fractions useful comment and discussion.
Sp: Particle surface area
Dummy integration variable
REFERENCES
VP: Particle volume
1) K.V.S. Sastry and D. W. Fuerstenau: Trans. Soc. Min.
x: Particle or granule size
Eng. AIME, 262 (1971), 43.
xo.5 Parameter in partition curve equation
2) M. Cross : Ironmaking Steelmaking, (1977), No. 3, 159.
xi,; Top size of size fraction i for feed size distri- 3) P. C. Kapur and D. W. Fuerstenau: Ind. Eng. Chem. Process
bution Des. Dev., 5 (1966), No. 1, 5.
x1,, Mean size of size fraction i for feed size dis- 4) T. Furui, M. Kawazu, K. Sugawara, T. Fujiwara, M.
tribution Kagawa, A. Sawamura and S. Uno : Nippon Steel Tech.
x2i : Top size of size fraction i for granule size Report Overseas, (1977), No. 10, 36.
distribution 5) P. W. Roller: B.H.P. Tech. Bull., 26 (1982), No. 1, 44.
y~(x) Feed size distribution density function (dry 6) W.J. Rankin and P. W. Roller: Proc. Joint Symp. ISIJ
mass basis) and AIMM, ISIJ, Tokyo, (1983), 13.
7) P. C. Kapur and D. W. Fuerstenau: Trans. AIME, 229
y2(x) Granule size distribution density function (dry
mass basis) (1964), 348.
8) Y. Ishikawa, M. Kase, M. Sasaki, K. Satoh and S. Sasaki:
a (x) Partition coefficient AIME Ironmaking Proc., 41 (1982), 80.
ai: Partition coefficient for particles of size frac- 9) K.V.S. Sastry and D. W. Fuerstenau: Powder Tech., 7
tion i (1973), 97.
cS: Surface area shape factor 10) R. Vidal, G. Mennier and E. Poot: Proc. Agglomeration
'85
0" Volume shape factor , Toronto, AIME, Book Crafters Inc., Chelsea, (1985),
4(x): Layer thickness 181.
Pt: Layer density (dry basis) 11) T. Allen: Particle Size Measurement, Chapman and Hill,
Pp: Particle density (dry basis) London, (1974), 84.
a: Parameter in partition curve equation