CFD Simulation and Comparison of Industrial Crystallizers

Chinmay V. Rane,1 Arijit A. Ganguli,1,2 Ekambara Kalekudithi,1,2 Raosaheb N. Patil,2 Jyeshtharaj B. Joshi1,3*
and Doraiswami Ramkrishna4*
1. Department of Chemical Engineering, Institute of Chemical Technology, Matunga, Mumbai 400 019, India
2. Techno Force Solution (I) Pvt Ltd., D‐33 MIDC, Nashik 422 010, India
3. Homi Bhabha National Institute, Anushaktinagar, Mumbai 400 094, India
4. School of Chemical Engineering, Purdue University, West Lafayette, IN 47907

Simulation of flow patterns in eleven industrial crystallizers (standard Messo, Cerny direct contact, Swenson draft tube baffled, Swenson walker,
Swenson evaporative, Oslo cooling, Oslo Krystal, APV Kestner, Batch Vacuum, stirred tank with disc turbine, and stirred tank with pitched blade
impeller) have been carried out using computational fluid dynamics (CFD). Population balance modelling (PBM) is coupled with CFD to obtain better
results. In all the cases, the crystallizer volume was 100 liter and the power consumption per unit volume was 1 kW/m3. The simulation results have
been presented in terms of mean velocity components, turbulent kinetic energy and dissipation rate. On the basis of the flow patterns, these eleven
crystallizers have been compared for crystal size distribution (CSD). It was found that Swenson Evaporative, Krystal and Batch Vacuum provide
relatively low CSD.

Keywords: crystallization, population balance, crystal size distribution, industrial crystallizers, multiphase CFD

INTRODUCTION rate, retain mother liquor and hence poor quality. Therefore, the
flow field needs to be linked with population balance to obtain

C
rystallization as a separation process has become increas- predictions of the size distribution.
ingly popular in the chemical industry due to its ability to
purify while producing a solid with the desired physical
properties. However, increasing requirements for higher product INDUSTRIAL CRYSTALLIZERS
purity and consistency have placed new demands on this Crystallizers can be classified into cooling crystallizers, evaporative
process.[1] In crystallization, molecules are transferred from a crystallizers and vacuum or adiabatic crystallizers[5] (Figure 1).
solute dissolved in a liquid phase to a solid phase through two main Over the years, a large number of crystallizer designs have been
steps: nucleation and crystal growth. Theoretically, nucleation developed for the commercial scale. The following are the widely
should initiate when solute concentration passes the saturation used industrial crystallizers. These have been schematically shown
point and the solution becomes supersaturated.[2] However, it does in Figure 2 and a very brief description is given below.
not occur instantly, and excess of solute remains in solution until a
sufficiently high level of supersaturation is generated to induce APV Kestner Evaporative Crystallizer (Figure 2A)[5]
nucleation. The extent of this supersaturation is referred to as This crystallizer falls in the category of evaporative crystallizers. It
metastable zone width. The growth rate of crystals depends on the consists of feed inlet (1). The feed is pumped (2) through a heat
level of supersaturation which is the driving force. The crystal exchanger (3) to the crystallizer (5). The flow the crystallizer
growth process takes place within the metastable zone and the tangentially below the liquor level (6). The liquid level in the
growth rate is controlled by the fluid dynamic conditions around crystallizer is kept above the top of the heat exchanger to prevent
the growing crystals.[3] Further, crystallization gives product of boiling in the heat exchanger tubes. Vapor‐liquid dispersion (7) is
high purity at low energy costs.[4] There are three main character- present above the entrance level (4). A baffle (8) in the crystallizer
istics of crystal product: mean size, CSD, and morphology. creates a relatively quiescent growth zone where the crystals grow.
Generally, there is a wide CSD in a crystallizer. The behaviour of
the product in succeeding operations like filtration, drying,
transport and storage is governed by CSD. Therefore, there has
existed for long a need for a reliably predictable method for design It is a pleasure for Professor D. Ramkrishna to felicitate Professor J. B. Joshi,
of industrial crystallizers. For a rational design, it is desirable to a coauthor of this paper, with a contribution that could never have been
understand the kinetics of nucleation and growth, the rates of conceived without his commitment to the use of advanced engineering
breakage and agglomeration. All these processes strongly depend methodology to analyze a variety of process equipment which are in
successful commercial operation.
on the flow field in the crystallizer which determines the *Author to whom correspondences may be addressed.
concentration field and hence the local rates of nucleation and E‐mail address: jb.joshi@ictmumbai.edu.in, jbjoshi@gmail.com,
growth. Poor CSD causes problems in manufacturing process, ramkrish@ecn.purdue.edu
particularly in separation of product from mother liquor. Too large Can. J. Chem. Eng. 92:2138–2156, 2014
a crystal size may result in crystals settling to bottom of the tank © 2014 Canadian Society for Chemical Engineering
DOI 10.1002/cjce.22078
and agglomeration. The large crystals, in some applications may Published online in Wiley Online Library
even need size reduction. While fine crystals give low filtration (wileyonlinelibrary.com).

2138 THE CANADIAN JOURNAL OF CHEMICAL ENGINEERING VOLUME 92, DECEMBER 2014
Figure 1. Classification of crystallizers.

The crystal slurry is removed through the outlet (9) and the mother concentric pipes consisting of guide tube (3) and an ejector tube (4)
liquor is recirculated (10).The vapor leaves through the outlet (11). and a baffle (5). The primary circuit is created by fast upward flow
The heating medium to the heat exchanger enters and leaves at of the slurry in the guide tube and a downward flow in the annulus.
(12) and (13), respectively. The slurry is drawn through the slot (8) between the ejector tube
and the baffle and a secondary flow circuit is formed in the lower
Swenson Evaporative Crystallizer (Figure 2B)[5]
region of the vessel. The feed entering the guide tube passes
The slurry leaving the crystallizer (at 8) is divided into two parts. One upwards into the vaporizer section (9) where flash evaporation
is taken as product (9) and the remaining slurry (10) along with fresh takes place. Nucleation therefore occurs in this region and the
feed (1) is pumped (2) through a heat exchanger (3) to the crystallizer nuclei are swept into the primary circuit. The crystal slurry is
(5). The flow enters tangentially (4) below the liquid level (6). The removed from the outlet (6), vacuum is applied from top (7) to
crystal slurry is removed from bottom (9) and vapor from top (11). facilitate the evaporation of the solvent.
Oslo‐Krystal Evaporative Crystallizer (Figure 2C)[5]
Swenson Draft Tube Baffled (DTB) Crystallizer (Figure 2F)[5]
The operation is similar to that of APV Kestner or Swenson
The feed (1) enters beneath the impeller (2) placed in an inner
evaporative crystallizers, where the feed (1) passes through a
draft tube (3). An outer draft tube (4) forms an annular space
pump (2) and a heat exchanger (3). The flow enters in an upward
in which agitation effects are virtually absent, and hence provides
direction at position (4) of the crystallizer (5) below the liquid level
a settling zone (5). The mother liquor is removed and fed back
(6). The solution then reverses the direction and flows through a
using a pump (6) through an elutriation leg (7), where
conical down flow (8). The region above the conical down flow can
classification of crystals takes place to facilitate removal of desired
be termed as nucleation zone as nucleation predominantly occurs
size crystals (8). Vacuum is applied at the top (9) for removal of
above the down flow while the region below can be termed as
vapor.
growth zone owing to the growth of crystals. The rate of liquor
circulation is such that the crystals are maintained in fluidized state Cerny Direct Contact Crystallizer (CDCC) (Figure 2G)[5]
in the vessel and the classification occurs by the hindered settling
It is a continuous direct contact cooling (DCC) crystallizer, where
process. Crystals that have grown to the desired size settle towards
cooling is achieved by direct contact heat transfer between the hot
bottom of the vessel and are discharged from the outlet (9) and the
crystallizing solution and an immiscible coolant drops. The feed
mother liquor (10) is mixed with fresh feed (1) and recirculated.
enters from inlet (1) and the coolant droplets are introduced
Oslo Cooling Crystallizer (Figure 2D)[5] through a distributor (2) which is located below the draft tube (3).
A feed solution enters the crystallizer vessel at point (1). The feed A gentle agitating action is provided by an impeller (4) and the
along with the saturated solution in the crystallizer is pumped (2) crystals are collected at the bottom (5). The low density coolant is
through heat exchanger (3), where it is cooled to form a collected as a separate layer at the top (6). It is then passed through
supersaturated solution. Care is taken to maintain the supersaturated a cyclone separator (7) where the aqueous solution droplets are
solution in the metastable zone. It is then fed through inlet pipe (4) to separated (8) and recycled. The coolant is recirculated through a
the bottom of crystallizer (5), the velocity is controlled to ensure the pump (9)and heat exchanger (10).
crystals remain suspended in the solution. The classification occurs by
Swenson Walker Crystallizer (Figure 2H)[5]
a hindered settling process. Crystals that have grown to the required
size settle towards the bottom of the vessel and are discharged It consists of a trough (1) with internal agitation provided by a
through the outlet (6) at predetermined intervals or continuously. helical impeller (2) and a cooling jacket (3). The shape of the
Fines are removed through a port (9) of the cyclone separator (7). The trough is semi‐cylindrical. The feed (4) is introduced at one end of
mother liquor is recirculated (8) back to the crystallizer. the trough and crystal slurry is removed from the opposite end (5).
A helical agitator rotates inside the trough to aid the growth of
Standard Messo Crystallizer (Figure 2E)[5] crystals by lifting them and allowing them to fall back through the
The mother liquor is fed (1) just below the impeller (2) and inside solution. The system is kept in gentle agitation and crystals are
the guide draft tube (3). Two liquor flow circuits are created by conveyed along the trough.

VOLUME 92, DECEMBER 2014 THE CANADIAN JOURNAL OF CHEMICAL ENGINEERING 2139
Stirred Tank Adiabatic Crystallizers (Figure 2 I‐K) and Stirred Tank use two fluid model with an assumption of interpenetrating media.
Cooling Crystallizers This approach has been improved in the last decade where
the liquid phase is modelled using Lagrangian approach and the
The Figure 2I‐K are examples of stirred tank crystallizers. Figure 2I
Newton’s law of motion is applied for individual particles. The
shows a typical batch adiabatic crystallizer, which consists of a
approach is called Discrete Particle Modelling (DPM). Thus, in
tank (1), propeller (2) and baffles (3). Vacuum (4) is applied at the
published literature, various combinations of CFD, PBM and DPM
top facilitating the vaporization of the solvent and results into
have employed such as CFD alone, CFD ¼ PBM, and CFD þ PBM
evaporative cooling of solution. Figure 2J‐K shows stirred tank
þ DPM. The published literature has been summarized in the
crystallizers equipped with disc turbine and PBT respectively. The
Table 1. A brief description is given below.
clearance of the agitator is decided depending on the application
Kramer et al.[12] carried out CFD simulations for the design of
under consideration. Baffles are required in all the cases to prevent
evaporative forced circulation crystallizer and found very important
vortex formation. The stirred tanks can also be used as cooling
information on the supersaturation profiles present in the crystal-
crystallizers. The cooling medium is passed through a jacket. Such
lizer. Thus, they have shown that only a part of the crystallizer
a mode of cooling is likely to result into deposition on the inside
volume is used for growth. This paper has made an important
surface and results into reduction in the cooling rate. Therefore, in
contribution to model development. Further, experimental valid-
such crystallizers, close clearance impellers such as helical ribbons
ation of the results has not been presented.
are employed.
Song et al.[13] carried out CFD simulations of an industrial draft
tube baffled (DTB) crystallizer using a two phase model to study the
Previous Work CSD of KCl crystals. The objective of the study was to reduce the
Ruying et al.[6] have presented a mathematical model based on the energy consumption for an existing crystallizer and increase the
Mixed Suspension Mixed Product Removal (MSMPR) model for the quality of KCl product. Three different impeller shapes (Rushton
design of DTB crystallizers with elutriators. The authors have turbine, PBT, and propellers) were studied. The effects of impeller
claimed agreement within 10 % between model predictions and shape, rotational speed, off bottom clearance, feed volumetric flow
experimental CSD data. Wöhlk et al.[7] have provided a systematic rate and fluid properties (fluid viscosity) were investigated.
classification of various types of industrial crystallizers. They Although propeller was found to be the best impeller with less
considered agitated tank crystallizers, forced circulation crystal- power consumption, the authors found that the flow patterns
lizers, draft tube baffled crystallizers and fluidized bed crystallizers generated from the existing propeller did not give the desired CSD.
and focused on crystal suspension and selection of supersaturation Hence, a new type of propeller was designed which gave optimum
levels in industrial crystallizers. The design conditions given by CSD. Therefore, the authors have pointed out the importance of the
these authors allow control of CSD in batch crystallizers while for effect of flow patterns on CSD and recommended more research.
continuous crystallizers they advise a caution in regard to the Wantha and Flood[14] performed two phase CFD simulations
choice of rotating machinery such as circulatory pumps and with population balance modelling. They observed that the
impellers which reduce secondary nucleation and hence the overall parameters in the population balance are not constant throughout
performance. the vessel and that further understanding of hydrodynamics is
Jager et al.[8] considered a draft tube baffled evaporative very important for the prediction of CSD. They have recommended
crystallizer and developed a measurement technique for the online the use of the evaporative type DTB crystallizer, since it has
measurement of CSD and derived a dynamic process model so that narrow size distribution, have been widely used in past, provides
the necessary process inputs can be given to the crystallizer to good turn down ratio, has low operating costs and is simple to
get good CSD. The authors have concluded that the available operate, start and shutdown. In their study, the crystallization of
knowledge is insufficient to understand the effects of process ammonium sulphate was studied. Plewik et al.[15] studied the
parameters on CSD. They have further suggested the use of hydrodynamics and CSD in draft tube (DT) and fluidized bed (FL)
hydrocyclone separation to improve CSD. Mersmann[9] has crystallizers. They found that hydrodynamic conditions in
presented comprehensive design procedures for crystallizers based industrial crystallizers were far from ideal and increase mixing
on the mass and energy balances and highlighted the importance of energy dissipation and cause increase in crystal destruction. It was
hydrodynamics on the CSD. He has suggested empirical correla- further reported that the axial velocity distribution was important
tions for secondary nucleation relating to CSD. The correlations for both DT and FL crystallizers, since the proper axial velocity
take into account impeller crystal interactions and crystal‐crystal distribution would ensure proper CSD, negligible settling and less
collisions which are responsible for secondary nucleation. Further, destruction of crystals. Hydrodynamic conditions in industrial
Mersmann has provided correlations and suggestions for the scale‐ crystallizers were also reported to be far from ideal. Internal
up of parameters like crystal growth rate, nucleation rate, attrition backward circulation loops disable even distribution of particles,
rate, minimum impeller speed for suspension, macromixing, and increase mixing energy dissipation, and cause increase in crystals
specific power input for micro‐mixing, heat transfer and CSD. The destruction. Their study, however, was restricted to only three
correlations for CSD have direct proportionality with growth rate crystallizers and no detailed hydrodynamic study (all components
and residence time. Mersmann has also deliberated on the problems of velocities and turbulence parameters) was made in the said
in the scale‐up of lab‐scale results to large scale crystallizers. The crystallizers. Zhu and Wei[16] explained the difficulties in scale‐up
procedures of Mersmann have been useful in industrial practice. of crystallizers and how CFD can help to optimize the crystallizer
In a crystallizer, the spatial variation of the extent of super- design by fixing important equipment design parameters at the
saturation depends upon the corresponding spatial variations in process development stage or for scaling up existing processes.
the rates of macromixing. In turn, the rate of macromixing depends The authors claim that chemical development in laboratory
upon the mean and turbulent flowfield.[10,11] vessels permits plenty of surface area for heat transfer and
In addition to the supersaturation field, the CSD also depends approximate well to perfectly mixed conditions for mass transfer.
upon the rates of aggregation and break‐up. Therefore, PBM is Al Rasheed et al.[17] used an Eulerian Model and performed CFD
being coupled with CFD. The CFD simulations, in majority of cases, simulations of Oslo crystallizer with elutriation leg or classifier.

2140 THE CANADIAN JOURNAL OF CHEMICAL ENGINEERING VOLUME 92, DECEMBER 2014
 A B C

D E F

G H I

J K

Figure 2. Schematic representation of the 11 industrial crystallizers: (A) APV Kestner (1) feed inlet (2) pump (3) heat exchanger (4) crystallizer inlet (5)
crystallizer (6) liquid level (7) vapor liquid dispersion (8) baffle (9) crystal slurry outlet (10) recirculation loop (11) vapor outlet (12) heating medium inlet (13)
heating medium outlet (B) Swenson Evaporative (1) feed inlet (2) pump (3) heat exchanger (4) crystallizer inlet (5) crystallizer (6) liquid level (7) vapor liquid
dispersion (8) crystallizer outlet (9) crystal slurry outlet (10) recirculation loop (11) vapor outlet (12) heating medium inlet (13) heating medium outlet (C)
Oslo‐Krystal evaporative (1) inlet (2) pump (3) heat exchanger (4) crystallizer inlet (5) crystallizer (6) liquid level (7) vapor liquid dispersion (8) conical
downflow (9) crystal slurry outlet (10) recirculation loop (11) vapor outlet (12) heating medium inlet (13) heating medium outlet (D) Oslo cooling (1) feed
inlet (2) pump (3) heat exchanger (4) inlet pipe (5) crystallizer (6) crystal slurry outlet (7) cyclone separtor (8) recirculation (9) fines (E) Standard Messo (1)
feed inlet (2) impeller (3) guide tube (4) ejector tube (5) baffle (6) crystal slurry outlet (7) to vacuum (8) slots between ejector and baffle (9) vaporization
section (F) Swenson DTB (1) feed inlet (2) impeller (3) draft tube (4) baffle (5) pump (6) elutriation leg (7) crystal slurry outlet (8) to vacuum (G) Cerny Direct
Contact (1) feed inlet (2) coolant inlet (3) draft tube (4) impeller (5) crystal slurry outlet (6) coolant layer (7) cyclone separator (8) mother liquor from coolant
(9) pump (10) heat exchanger (11) cooling medium in (12) cooling medium out (H) Swenson Walker (1) Trough (2) Helical impeller (3) cooling jacket (4)
feed inlet (5) crystal slurry outlet (6) cooling medium inlet (7) cooling medium outlet (I) BV Crystallizer (1) tank (2) impeller (3) baffles (4) to vacuum (J) DT
crystallizer (K) Pitched Blade Turbine Crystallizer (J) and (K): (1) tank (2) impeller (3) baffles.

VOLUME 92, DECEMBER 2014 THE CANADIAN JOURNAL OF CHEMICAL ENGINEERING 2141
They found that there were major differences in velocity profiles Continuity equation of the kth Phase
and distribution of solids in fluidized bed reactors (FBR) (Oslo
crystallizer as considered by the authors). When FBR is scaled up @
ðr 2k Þ þ r  ðrk 2k vk Þ ¼ Sm ; ð1Þ
with conservation of outlet velocity as the scaling factor, @t k
compression of bed and circulation loop disappearance occur.
This results increase in velocity of fluid. Certain data collected where the kth phase refers either to the continuous liquid phase or
using CFD technique is impossible to obtain in any other way the discrete solid phase.
provide information about suspension flow behaviour. Momentum equation
There are two aspects which need to be addressed (a) coupling of
CFD and PBM, with a suitable method for solving PBM and (b) @
ðr 2k vk Þþ r  ðrk 2k vk vk Þ ¼ 2krp þ rk 2k g þ 2k mk r2 vk þ FI :
application of this methodology to large scale crystallizers. The @t k
work on part (a) has gained momentum in the last decade, Baldyga ð2Þ
and Orciuch[18] presented the work on coupling of CFD and PBM,
which was used to simulate precipitation of barium sulphate. In Equation (1), Sm is a source term that captures the aggregation
Later, many researchers[19–22] have worked on precipitation. and break‐up processes. The right side of Equation (2) describes
Though not exactly a crystallization operation, precipitation is the details of which have been given by Murthy and co‐
very similar to crystallization as solid particles are being formed in workers.[34,32] Murthy et al.[34] has considered the following
a liquid phase which also involves the same two steps of nucleation formulation of drag force: the drag force exerted by the dispersed
and growth. Kramer et al.[12] and Kougoulos et al.[23] modelled phase on the continuous phase is calculated as
an industrial crystallizer with the help of compartmental
modelling. Logashenko et al.[24] employed CFD and PBM but FD ¼ K sl ðvc  vd Þ; ð3Þ
they utilized 1D CFD model. Falola et al.[25] and Rane et al.[26] have
simulated crystallization by coupling CFD and PBM. However, part where Ksl is the fluid‐solid exchange coefficient and can be written
(b) has been sparsely done. Some authors[15,17] have worked on in the following general form:
large scale crystallizers and have shown only flow field of fluid or
classification of solids. 2s rs f
K sl ¼ ; ð4Þ
Sen et al.[27] attempted to model a continuous flowsheet for a ts
pharmaceutical actice pharmaceutical ingredient (API) comprising
of crystallizer, filter, dryer, and blender. They coupled CFD with a where f is defined differently for the different exchange‐coefficient
2D PBM. The particles scale information (e.g., collision kinetics models. The drag function due to Syamlal and Obrien[35] gives a
and velocity) was obtained from DEM simulation, while PBM somewhat better prediction when compared with the other models
provided the process‐scale information (e.g., distributions of and it is defined as
particle attributes such as particle size). The information was
interchanged using gPROMS software. The authors could qualita- C D Res 2l
tively track the trend of different variables of interest for each unit f¼ ; ð5Þ
24v2r;s
operation. Grof et al.[28] also coupled PBM‐DEM to crystallize an
where the drag coefficient, derived by Dallavalle,[36] is given by
API. The PBM utilized the breakage kernel and the daughter
distribution obtained from DEM to predict the final crystal size. The 0 12
simulation results were compared with the experimental results
B 4:8 C
and the authors found a good match. Table 1 summarizes the C D ¼ @0:63 þ qffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiA : ð6Þ
attempts of various researchers working in the field of crystalliza- Res =v2r;s
tion employing CFD and PBM.

This model is based on measurements of the terminal velocities

MATHEMATICAL MODELLING of particles in fluidized or settling beds, with correlations that
are a function of the volume fraction and relative Reynolds
Computational Fluid Dynamics number:[37]
Ranade et al.,[29] Joshi et al.,[30] Joshi et al.,[31] and Murthy and
Joshi[32] have given the details pertaining to governing equations, rl ds j
vs 
vl j
Res ¼ ; ð7Þ
method of solution and appropriate precautions for the imple- ml
mentation of CFD. Further, several examples of relationship
between the fluid mechanics and design objectives have been where the subscript l is for the lth fluid phase, s is for the sth solid
describes in the published literature. For instance, heat trans- phase, and ds is the diameter of the sth solid phase particles/
fer,[33] mixing,[10,11] and solid suspension.[34] Therefore, it was crystals.
thought desirable to undertake CFD for understanding the fluid Energy equation
mechanics in crystallizers.
" #
@ X
Governing Equations ðrEÞ þ r  vk ðrE þ pÞ ¼ r  kef f rT  hj Jj þ tef f  vk þ Sh :
@t j
The ensemble‐averaged mass and momentum transport equations
for each phase are given by the following equations. Nomenclature ð8Þ
section provides a description of all the variables appearing in the
differential equations. The first three terms on the right‐hand side of Equation (8)
represent energy transfer due to conduction, species diffusion and

2142 THE CANADIAN JOURNAL OF CHEMICAL ENGINEERING VOLUME 92, DECEMBER 2014
Table 1. Previous work

Impeller
Crystallizer Geometry speeds and Impeller
Author type Details other details type Grid Size Model Remarks

Kramer Forced 200 L; Inlet low — 26  260 k‐e model; Compared one
et al.[11] circulation T ¼ 0.3 m; rate:3 L/sec FLUENT, compartment and
crystallizer H ¼ 1.5 m SPEEDUP five compartment
program model was used. 2D
simulation as
crystallizer is axis
symmetric. No
comparison with
experimental results
Rousseaux Stirred tank T ¼ 0.15 m; 1200, Common 75  60  24 k‐e model; Studied effect of feed
et al.[39] crystallizer D ¼ 0.12 m; 1500 rpm impeller FLUENT location, rotating
2.5 L 4.5 speed and reagent
concentration
Zheng Continuous STR T ¼ 0.27 m; 200–500 rpm Ruston 36  72  90 Standard Compared crystal mean
et al.[41] (precipitation H ¼ T; turbine (r,u,z) k‐e model; size by varying
of barium b ¼ T/10; In‐house impeller speed, feed
sulfate) D ¼ T/3; code position, feed
C ¼ T/2 concentration and
residence time. Also
studied the effect of
draft tube
Logashenko Stirred tank 0.2  0.2 m Impeller speed Common 11,500 (2D); In‐house Performed 2D and 3D
et al.[13] crystallizer simulated impeller 3,000 (3D) code simulations;
by source term Neglected crystal‐
crystal, crystal‐wall
collision
Sha Stirred tank 1/4th tank; 600, 1000, 4‐bladed 34,400 k‐e Studied effect of
et al.[42] crystallizer T ¼ 0.15 m; 1400 rpm; propeller turbulence impeller speed,
with 4‐baffles H¼T Particle model, SM; product removal
with and sizes: 50, 100, CFX location and effect of
without draft 300, 500, 700 crystallizer geometry
tube and 900 mm
Zhu and Stirred tank 200 ml & 200 rpm for Two blade varied from Standard Only flow field is given
Wei[26] crystallizer 300 L 200 ml; and PBT 10,000 to k‐e model
with and w/o 200 and 120,000
draft tube 600 rpm
for 300 L
Wantha and Draft Tube T ¼ 1.05 m; — — 70000–82000 Standard k‐e Studied the
Flood[25] Crystallizer H ¼ 3.1 m; model; characteristics of flow
(DTB) 1.05 m3 ANSYS CFX field, effect of
momentum source
strength and effect of
fines removal flow
rate on flow
characteristics
Plewik Fluidized Bed T ¼ 0.68 m; FL: — — — Standard k‐e Presented only axial
et al.[16] (FL), Double H ¼ 0.83 m; model; velocity, volume
Draft Tube DDT: Fluent fraction and fluid
(DDT) and H ¼ 0.93 m; flow profiles though
Draft Tube DTM: the three crystallizers.
Magma H ¼ 0.83 m CFD data was
(DTM) analyzed for different
geometry of tank and
found to be within
10 % deviation with
LDA data

VOLUME 92, DECEMBER 2014 THE CANADIAN JOURNAL OF CHEMICAL ENGINEERING 2143
viscous dissipation, respectively. Sh includes the heat of chemical present to provide information useful for engineering and design
reaction, and any other volumetric heat sources. purposes. The moments mj of the CSD are computed according to
Species transport equation the following moment transport equation, considering the first
The conservation equation for the ith species: four moments:
 
@ Dt @     
ðrY i Þ þ r  ðrvk Y i Þ ¼ r  rDm þ rY i þ Ri þ Si ; ð9Þ rmj þ r  rumj  Gef f rmj ¼ r B
ag;j þ B

ag;j  D
br;j  D

br;j
@t Sct @t
þ rJ j0 n_ 0 þ rjGmj1 ; j ¼ 0; 1; 2
where Yi is the mass fraction of each species, Sct is the turbulent
Schmidt number, Dm is molecular diffusion, Dt is the turbulent ð15Þ
diffusivity, Ri is the net rate of production of species i by chemical where B
br;j ; D

br;j ; B

ag;j ; D

ag;j are, respectively, the birth and death
reaction and Si is the rate of creation by addition from the due to break‐up and aggregation of crystals. J is the nucleation rate
dispersion phase plus any user‐defined sources. The default Sct is and G is the growth rate.
0.7. Joshi et al.[38] have discussed the current status of CFD The moments mj of the number density are calculated by the
simulations. quadrature approximation

Turbulence Modelling X
N
mj ¼ wi V ji ; j ¼ 1; 2; …; 2N  1: ð16Þ
For the continuous phase, the RNG k  e model was applied. The i¼1
governing equations for the RNG k  e model are
The quadrature approximation of order N of the number
  density nðV; r; t Þ is defined by its N weights wi ðr; t Þ and N
@ m
ðrl 2l kÞ þ r  rl 2l kvk ¼ r  2l ml þ t rk þ rl 2l ðP  eÞ; abscissas Vi, and can be calculated by its first 2N moments
@t sk
m0 ; m1 ; m2 ; …; m2N1 . The number of scalars (moments) 2N for
ð10Þ QMOM is significantly lower than the number of classes M for the
  discrete methods. After writing this relationship in matrix form, it
@ m e  
is easy to show that the roots of the polynomials correspond to the
ðrl 2l eÞ þ r  rl 2l evk ¼ r  2l ml þ t re þ rl 2l C 1e P  C 2 e ;
@t se k eigen values of the Jacobi matrix.[42] This procedure is known as
ð11Þ the product difference algorithm.[43] Once the weights and

abscissas are known the transport equations for the moments
where C 2e is given by can be solved.
For a given set of moments, the most likely CSD is obtained
based on the statistically most probable distribution for turbulent
 C m hs ð1  h=ho Þ flames.[44] The QMOM has been utilized for modelling
C 2e ¼ C 2e þ : ð12Þ
1 þ bhs crystallization.[18]
P is the rate of production of turbulence given by The number density function n(V) is expressed as
!
mt   X
N1
P ¼ rv þ rvT : rv: ð13Þ nðVÞ ¼ exp i
: ð17Þ
r Ai V
i¼0

C 2e ¼ 1:68; C 1e ¼ 1:42; s k ¼ 1:0, and s e ¼ 1:2 are constants.

The equation for the kth moment is now written as
Turbulent eddy viscosity, mt is computed from
Z !
1 X
N1
k2 mk ¼ V k exp Ai V i dL k ¼ 0; 1; …N  1: ð18Þ
mt ¼ rC m ; ð14Þ
e 0 i¼0

where Cm ¼ 0.0845 and the difference between the RNG k  e For N moments, the coefficients Ai can be found by a globally
model and the standard k  e model lies in the additional sink term convergent Newton‐Raphson method to reconstruct the CSD.
in the e equation.
Growth and Nucleation
The crystal growth rate G is, for most systems, exponentially
Population Balance Modelling dependent on the supersaturation, if two dimensional nucleation is
As observed earlier, the CSD is an essential criterion for assessing assumed as a growth model for all the crystal faces. However, the
the properties and quality of the product. To predict the CSD, the end result of the crystal size obtained in a crystallizer is not a matter
population balance equations needs to be solved. In the present only of the growth rate, but also of the nucleation rate, J (how many
work, we have used the quadrature method of moments (QMOM) crystals take part in crystal growth), and the attrition rate (how
method for the crystallization processes. The CSD can then be easily crystals break, and how small are the broken fragments).
expressed in terms of these moments. The QMOM method was first The nucleation rate is also a function of the supersaturation, and is
used by McGraw.[39,40] The method requires a relatively small affected by supersaturation to a far greater degree than the growth
number of scalar equations to track the moments of population rate. Thus it is important to maintain a suitably appropriate
with small errors.[41] This approach is deemed sufficient for the supersaturation field in the crystallizer.

2144 THE CANADIAN JOURNAL OF CHEMICAL ENGINEERING VOLUME 92, DECEMBER 2014
 The crystal growth rate, G, and the nucleation rate, J, are number densities of particles of volumes Vi and Vj, respectively.
calculated as functions of the supersaturation ratio S defined in Two physical mechanisms are behind the calculation of the relative
Equation (21), following[19] velocity. The first is that of turbulent mixing. Assuming that the
crystal sizes lie in the inertial range of turbulence, and the
G ¼ kg ðS  1Þ2 ; ð19Þ
tij of the two crystals
turbulence is isotropic, the mixing velocity u
J ¼ kn ðS  1Þ 2:77
; ð20Þ can be expressed as
 1=2
where kg and kn are constants. The key parameter for determining
tij ¼ u
u
2i þ u

2j ; ð26Þ
these kinetics is the local supersaturation, S, which reflects the
thermodynamic driving force for crystallization and is defined as where the turbulent velocity in the inertial sub‐range of isotropic

i ¼ 1:43ðedi Þ1=3 . The expression for the probability
turbulence is u
Solute mole fraction of aggregation is given by
S¼ ; ð21Þ
Solubility
8 h   i1=2 9
>
< þ 2
þ 3 >
=
Solubility ¼ 0:000919T  0:199: ð22Þ   0:75 1 x ij 1 x ij 1=2
Pag V i ; V j ¼ exp C 1  3 We ;
>
: ðr2 =r1 þ 0:5Þ 1=2
1 þ xij
ij >
;
The moments are used to calculate the mass‐average crystal
mean size, d32, as follows: ð27Þ

P
N where C1 is a constant of order unity, xij  di =dj , r1 and r2 are the
m3 vi densities of the primary and secondary phases, respectively, and
d32 ¼ i¼1 ; ð23Þ the Weber number is defined as
PN
m2 vi
i¼1  2

tij
r1 di u
Weij  : ð28Þ
where vi is the volume of the ith cell. s

Aggregation Model
The birth rate of group j due to aggregation of group i and group k
The aggregation kernel is modelled by Luo.[45] The general crystal is
aggregation kernel is defined as the rate of particle volume formation
as a result of binary collisions of particles with volumes Vi and Vj: X
N X
N

ag;j ¼ 1
B wi wk ðV i þ V k Þj : ð29Þ
      2 i¼1 k¼1
a V i ; V j ; r ¼ vag V i ; V j ; r Pag V i ; V j ; ð24Þ
   
where vag V i ; V j ; r is the frequency of collision and Pag V i ; V j is The death rate of group j due to aggregation with other crystals is
the probability that the collision results in aggregation. The
frequency, based on projected area, is given by X
N X
N

ag;j ¼
D V ji wi wk aðV i ; V k ; rÞ: ð30Þ
  p 2  i¼1 k¼1
vag Vi; Vj; r ¼
ij ðrÞ;
di þ d2j ni nj u ð25Þ
4
Breakage model
where di  ð6V i =pÞ1=3 is the diameter of the particle of volume Vi The breakage kernel encompasses both the breakage frequency
assuming spherical shape, u
ij is the velocity of the ith particle and the PDF of crystals formed from breakage.[45] The general
th
relative to the j particle which has volume Vj, and n
i and n

j are the breakage rate per unit volume is usually written as

Table 2. Geometrical details of crystallizers

Swenson‐Walker Cerny direct contact Oslo cooling Swenson evaporative APV Kestner Oslo‐Krystal

Diameter (m) 0.16 0.193 0.46 0.36 0.36 0.35

Height (m) 1.28 0.94 0.575 0.375 0.735 1.05
Inlet Diameter 0.06 0.13 0.06 0.072 0.072 0.06
Inlet velocity (m/sec) 3 1.55 4.136 3.663 3.663 4.136
Impeller Diameter 0.158 0.108 — — — —
Impeller speed (rpm) 1760 60 — — — —
Swenson DTB Standard Messo Batch Vacuum Disc Turbine Pitched blade Turbine

Diameter (m) 0.4 0.44 0.50 0.5 0.5

Height (m) 1.2 1.005 0.64 0.5 0.5
Inlet Diameter — — — — —
Inlet velocity (m/sec) — — — — —
Impeller Diameter 0.0375 0.0375 0.16 0.16 0.16
Impeller speed (rpm) 2249 2249 343 636 451

VOLUME 92, DECEMBER 2014 THE CANADIAN JOURNAL OF CHEMICAL ENGINEERING 2145
Vbr ðV; V 0 ; rÞ ¼ g ðV 0 ; rÞbðV jV 0 Þ; ð31Þ Z1  
ð1 þ jÞ2
Vbr ðV; V 0 Þ ¼ K exp bj11=3 dj; ð32Þ
j 11=3
jmin
where the parent particle suffering breakage has volume V0 and the
daughter particle has volume V. In the above expression, g ðV 0 ; rÞ is where the model parameters are assumed to be as below:
the spatially non‐uniform breakage frequency, and bðV jV 0 Þ is the
normalized daughter particle distribution function. For binary h   i
breakage, the bðV jV 0 Þ must be symmetrical with respect to VV0 ¼ 0:5. K ¼ 0:923e1=3 d2=3 e; b ¼ 12 f 2=3 þ 1  f 2=3  1 sr1 e2=3 d5=3 :
The general form is the integral over the size of eddies l hitting the
particle with diameter d (and volume V). The integral is taken over The birth rate of group j crystals due to breakage of larger
the dimensionless eddy size j ¼ l=d. The general form is crystals is

Figure 3. Grid and surface mesh for all 11 crystallizers (A) APV Kestner (B) Swenson Evaporative (C) Krystal evaporative (D) Oslo cooling (E) Standard Messo
(F) Swenson DTB (G) Cerny Direct Contact (H) Swenson Walker (I) BV Crystallizer (J) DT crystallizer (K) Pitched Blade Turbine Crystallizer.

2146 THE CANADIAN JOURNAL OF CHEMICAL ENGINEERING VOLUME 92, DECEMBER 2014
 A B

C D

Figure 4. Variation of velocity and TKE profiles with dimensionless radial distance for APV Kestner Crystallizer. (A) Axial velocity (B) Radial velocity (C)
Tangential velocity (D) Turbulent Kinetic Energy (TKE) (1) z ¼ 0.45, (2) z ¼ 0.285, (3) z ¼ 0.0875.

X
N Z1
population balance model utilizes velocity (Equation (25)) from

br;j ¼
B wi V j g ðV i ; rÞbðV jV i ÞdV: ð33Þ CFD for aggregation while Equation (32) shows general breakage
i¼1
0 rate per unit volume which include turbulent energy dissipation (e)
obtained from CFD.
The death rate of group j crystals due to breakage into smaller
crystals is Geometrical Details of Crystallizers
The crystallizers are classified into two types, viz., crystallizers
X
N

br;j ¼
D wi V ji g ðV i ; rÞ: ð34Þ equipped with impeller to facilitate fluid motion, and crystallizers
i¼1 which do not have any moving part and the desired flows are
generated using some other method of energy supply. The details of
In a crystallizer the supersaturation is dependent on mixing, the geometries are given in Table 2. Each crystallizer has a volume
which in turn controls nucleation and crystal growth. The of 100 liters.

A B

C D

Figure 5. Variation of axial, radial and tangential velocity and TKE profiles with dimensionless radial distance for Swenson Evaporative Crystallizer. (A) Axial
velocity (B) Radial velocity (C) Tangential velocity (D) Turbulent Kinetic Energy (TKE) (1) z ¼ 0.45, (2) z ¼ 0.3, (3) z ¼ 0.15.

VOLUME 92, DECEMBER 2014 THE CANADIAN JOURNAL OF CHEMICAL ENGINEERING 2147
Grid Sensitivity crystallizers as measured at different axial positions. In order to
The total grid count of each crystallizer is around 1 million. The compare the performance of these crystallizers, the velocities are
impeller speed (for Cerny Direct Contact, Swenson Draft Tube, normalized with inlet velocity or tip velocity (in the case of flow
Standard Messo and Batch Vacuum crystallizers) or the inlet generated by impellers) and the radial distance is normalized with
velocity (APV‐Kestner, Swenson Evaporative and Oslo Krystal the radius of the crystallizer. The axial distance was normalized
crystallizers) is given in such a way so as to maintain a constant with the height of the crystallizer.
power to volume ratio in all the crystallizers. A tetrahedral grid was Velocity and Turbulence Profiles
created using GAMBIT 2.2.3 and single phase simulations were
performed using water as fluid employing ANSYS FLUENT 13.0 Figure 4A–D shows the velocity and turbulent kinetic energy
(ANSYS Inc., Canonsburg, PA 15317, USA). Figure 3A–K shows (TKE) profiles for APV Kestner Crystallizer. The flow patterns
the grids of the crystallizers under consideration. Impeller rotation represent a sinusoidal wave‐like pattern which is the same at all the
was modelled using sliding mesh technique. The simulations were three positions as shown in Figure 2A. In this type of crystallizer,
carried out till steady state was reached, which was decided by the flow enters in a tangential direction. Hence, tangential
mixing time (Joshi and Pandit).[16] Renormalized standard k‐e velocities are dominant as depicted in Figure 4C. The axial and
model has been used, owing to its simplicity and suitability for radial velocities are low in magnitude (Figure 4A‐B). Tangential
isotropic flows with high Reynolds number. velocities are higher at the walls (about 130 % of the inlet
velocities). Further, at a dimensionless radial distance of 0.5
Method of Solution
from the centre the tangential velocities are of the same order of
No slip conditions were imposed at all the walls. All the magnitude as the inlet velocity, however, in opposite direction.
computational work has been carried out using the commercial Axial and radial velocities are around 10 % and 6 % of the inlet
software Fluent 13. The second order implicit scheme has been velocities, respectively. The TKE profiles are about 10 % of the
used for time discretization. In the case of realizable k  e mean kinetic energy at the inlet.
turbulence model, quadratic upwind interpolation for convective Figure 5A–D shows the velocity and TKE profiles for Swenson
kinetics (QUICK) discretization scheme was used for the turbu- Evaporative crystallizer at three axial positions (Figure 2B). Similar
lence parameters. The QUICK formulation has a third order to APV Kestner crystallizer, Swenson Evaporative crystallizer
accuracy which helps to mitigate the unfavourable effect of shows sinusoidal axial flow of about 15 % strength of inlet velocity
artificial diffusion. All the discretized equations were solved in a (Figure 5A). Radial velocities (Figure 5B) are weak as compared to
segregated manner with the Pressure Implicit with Splitting of axial or tangential velocities (6 %). This crystallizer also has a
Operators (PISO) algorithm. All the solutions were considered to be tangential inlet and hence, the tangential velocities are dominant
fully converged when the sum of residuals was below 1  105. All (Figure 5C). Figure 5D shows high magnitudes of TKE near the inlet
the computations have been performed on an SGI dual processor of the crystallizer (z ¼ 0.45 m) with a value of 3.5 % and lower
computer node (8 computer cores with 4 GB RAM). magnitudes near the bottom.
Figure 6A–D shows the radial profiles of axial, radial and
tangential velocities and TKE at various axial positions
RESULTS AND DISCUSSION (Figure 2C) of a Krystal Crystallizer. The axial velocities
The performance of various crystallizers was evaluated on the (Figure 6A) have highest magnitudes at axial positions 3 and 4
basis of axial, radial and tangential velocities in individual for a particular radial position. These peaks signify the entry to

A B

C D

Figure 6. Variation of axial, radial and tangential velocity and TKE profiles with dimensionless radial distance for Krystal Crystallizer. (A) Axial velocity (B) Radial
velocity (C) Tangential velocity (D) Turbulent Kinetic Energy (TKE) (1) z ¼ 0.075, (2) z ¼ 0.3, (3) z ¼ 0.5, (4) z ¼ 0.7, (5) z ¼ 0.95.

2148 THE CANADIAN JOURNAL OF CHEMICAL ENGINEERING VOLUME 92, DECEMBER 2014
 A B

C D

Figure 7. Variation of axial, radial and tangential velocity profiles and TKE profiles with dimensionless radial distance for Oslo Krystal Crystallizer. (A) Axial
velocity (B) Radial velocity (C) Tangential velocity (D) Turbulent Kinetic Energy (TKE) (1) z ¼ 0.1, (2) z ¼ 0.2, (3) z ¼ 0.3, (4) z ¼ 0.4, (5) z ¼ 0.5.

the crystallizer which is in upward direction. The bottommost inlet velocities. As the direction of flow is predominantly axial
part of the crystallizer shows highest peak at r/R ¼ 0.05 with and radial, the magnitudes of tangential velocities for the same
negative velocities. This suggests that flow is predominantly axial positions are relatively quite low (3 % of inlet velocities) as
downward at this position owing to the conical down flow which seen in Figure 6C. Figure 6D shows TKE profiles for Krystal
separates the nucleation and growth zones, while at other radial crystallizer, which are around 6 % of the mean kinetic energy at
positions the axial velocity magnitudes are lower. Figure 6B the inlet.
shows the trends for the radial velocities for the same axial Figure 7A–D shows the radial profiles of axial, radial and
positions. Similar to the axial velocity profiles, peaks can be seen tangential velocities and TKEat various axial positions (Figure 2D)
at positions 3 and 4, where the mother liquor travels in radially in Oslo cooling crystallizer. Distinctly downward axial velocity
upward direction. It is observed that the bottom of the crystallizer profiles can be observed at all the axial positions. The flow reversal
have lower magnitudes of radial velocities around 10 % of the is observed at a dimensionless radial distance of 0.75 (i.e.,

A B

C D

Figure 8. Variation of axial, radial and tangential velocity and TKE profiles with dimensionless radial distance for Standard Messo Crystallizer. (A) Axial velocity
(B) Radial velocity (C) Tangential velocity (D) Turbulent Kinetic Energy (TKE) (1) z ¼ 0.2, (2) z ¼ 0.3, (3) z ¼ 0.455, (4) z ¼ 0.75, (5) z ¼ 0.9.

VOLUME 92, DECEMBER 2014 THE CANADIAN JOURNAL OF CHEMICAL ENGINEERING 2149
 A B

C D

Figure 9. Variation of axial, radial and tangential velocity and TKE profiles with dimensionless radial distance for Swenson Draft Tube Baffled Crystallizer. (A)
Axial velocity (B) Radial velocity (C) Tangential velocity (D) Turbulent Kinetic Energy (TKE) (1) z ¼ 0.5, (2) z ¼ 0.65, (3) z ¼ 0.75, (4) z ¼ 1.

r/R ¼ 0.75). The magnitudes of negative velocities are highest near axial positions, the axial velocities are uniform with slight
the inlet pipe (r/R ¼ 0–0.1) while upward velocities are prevalent variations. The radial velocities (Figure 8B) are negligible
in the near wall region (r/R ¼ 0.75–1). Figure 7B shows the radial compared to the axial velocities. Positions 3 and 5 show significant
velocity profiles at various axial positions. The design of Oslo radial velocities because of impeller position and flow reversal at
crystallizer being dominated by axial velocities, we can hardly see the top respectively. Position 2 also shows significant radial
any radial velocity component apart from the outflow at position 5. velocity in inward direction, which is attributed to the suction
Likewise, the tangential velocities (Figure 7C) are also minimal in present at the bottom of draft tube. The tangential velocities
comparison with the axial velocities. Figure 7D shows TKE is high (Figure 8C) at par with and have similar trend as axial velocities.
only at position 5. Figure 8D shows the high magnitude of TKE at position 3 and
Figure 8A–D shows the velocity profiles and TKE profiles for negligible elsewhere.
the Standard Messo Crystallizer. The type of impeller used is the Figure 9A–D shows the velocity profiles and TKE profiles of
propeller and hence axial velocity magnitudes are high near the Swenson Draft Tube baffled (DTB) crystallizer. The trends are
impeller as observed in Figure 8A (position 3). At all the other similar to those of Messo Crystallizer with high axial, radial and

A B

C D

Figure 10. Variation of axial, radial and tangential velocity and TKE profiles with dimensionless radial distance for Cerny Direct Contact Crystallizer. (A) Axial
velocity (B) Radial velocity (C) Tangential velocity (D) Turbulent Kinetic Energy (TKE) (1) z ¼ 0.082, (2) z ¼ 0.552, (3) z ¼ 0.843.

2150 THE CANADIAN JOURNAL OF CHEMICAL ENGINEERING VOLUME 92, DECEMBER 2014
 A B

C D

Figure 11. Variation of axial, radial and tangential velocity and TKE profiles with dimensionless radial distance for Swenson Walker Crystallizer. (A) Axial
velocity (B) Radial velocity (C) Tangential velocity (D) Turbulent Kinetic Energy (TKE) (1) z ¼ 0.16, (2) z ¼ 0.64, (3) z ¼ 1.12.

tangential velocities (Figure 9A–C) at axial position near impeller negligible compared to location 2. The radial velocity (Figure 10B)
(varying between 15 % and 30 % of inlet velocities). The velocity in the crystallizer is maximum at location 3 where flow reversal
magnitudes are negligible (as compared to near impeller) away from takes place and the solution travels in downward direction in the
the impeller and other positions. The TKE profiles (Figure 9D) are annular region. The tangential velocity (Figure 10C) in the
similar to the tangential velocity profiles with upward velocities with crystallizer is maximum at the impeller line, as it a close clearance
magnitudes highest at position 2 (z ¼ 0.65 m) near the impeller. impeller is employed to impart stirring motion. Turbulent kinetic
Figure 10A–D shows velocity and TKE profiles of Cerny Direct energy (Figure 10D) is dominant at locations 2 and 3, while
Contact Crystallizer. Since the coolant is fed to the crystallizer negligible at location 1.
below the draft tube strong axial velocities are observed at Figure 11A–D shows velocity and TKE profiles of Swenson
location2 (z ¼ 0.552), axial velocities at location 1 and 3 are Walker crystallizer. In Swenson Walker crystallizer the impeller is

A B

C D

Figure 12. Variation of axial, radial and tangential velocity and TKE profiles with dimensionless radial distance for Batch Vacuum Crystallizer. (A) Axial velocity
(B) Radial velocity (C) Tangential velocity (D) Turbulent Kinetic Energy (TKE) (1) z ¼ 0.07, (2) z ¼ 0.165, (3) z ¼ 0.2, (4) z ¼ 0.3, (5) z ¼ 0.4.

VOLUME 92, DECEMBER 2014 THE CANADIAN JOURNAL OF CHEMICAL ENGINEERING 2151
 A B

C D

Figure 13. Variation of axial, radial and tangential velocity and TKE profiles with dimensionless radial distance for Disc Turbine Crystallizer. (A) Axial velocity
(B) Radial velocity (C) Tangential velocity (D) Turbulent Kinetic Energy (TKE) (1) z ¼ 0.07, (2) z ¼ 0.165, (3) z ¼ 0.2, (4) z ¼ 0.3, (5) z ¼ 0.4.

across the entire length of the crystallizer, hence negligible positions while negative velocities are at lower axial positions.
variation is seen in the velocity and kinetic energy profiles. Radial velocities are 10 % at the top positions (position 4 and 5).
Interestingly, the radial velocities dominate the flow as observed in Tangential velocities are highest at the bottom with 9 % of the tip
Figure 11B. The radial velocities are up to 60 % of the tip velocity velocities. As seen in Figure 12C, the tangential axial positions are
and have a uniform trend for all positions. low in magnitude. Figure 12D shows that the TKE is high near the
Figure 12A–D shows the velocity and TKE profiles of Batch impeller with upward magnitudes near the centre decreasing
Vacuum Crystallizer. Here as depicted from Figure 12A axial away with radial distance.
velocities dominate the flow, owing to the propeller impeller. Figure 13A–D shows the radial profiles of axial, radial and
Near the impeller, velocities are high (40 % of tip velocity) tangential velocities, and TKE at various axial positions of a
especially for axial positions 2 and 3. Further, radial velocities Disc turbine crystallizer. Five different positions from the bottom
(Figure 12B) also depict positive velocities for higher axial of the crystallizer were chosen. Disc turbine creates a radial flow;

A B

C D

Figure 14. Variation of axial, radial and tangential velocity and TKE profiles with dimensionless radial distance for Pitched Blade Turbine Crystallizer. (A) Axial
velocity (B) Radial velocity (C) Tangential velocity (D) Turbulent Kinetic Energy (TKE) (1) z ¼ 0.07, (2) z ¼ 0.165, (3) z ¼ 0.2, (4) z ¼ 0.3, (5) z ¼ 0.4.

2152 THE CANADIAN JOURNAL OF CHEMICAL ENGINEERING VOLUME 92, DECEMBER 2014
 A B

C D

E F

G H

I J

Figure 15. Crystal Size Distribution for all 11 crystallizers (A) APV Kestner (B) Swenson Evaporative (C) Krystal evaporative (D) Oslo cooling (E) Standard Messo
(F) Swenson DTB (G) Cerny Direct Contact (H) Swenson Walker (I) BV Crystallizer (J) DT crystallizer (K) Pitched Blade Turbine Crystallizer.

VOLUME 92, DECEMBER 2014 THE CANADIAN JOURNAL OF CHEMICAL ENGINEERING 2153
hence, radial velocities are dominant as seen in Figure 13B. Strong similar to Messo crystallizer with the difference being CSD
radial velocities of round 60 % of tip velocities near the impeller extending up to 5000 mm. Cerny Direct contact crystallizer is
can be seen. The radial velocities are lower as we go away from the shown in Figure 15G, where majority of the crystals are around
impeller. Figure 13C shows that the tangential velocities, which 2000 mm with CSD spread up to 3200 mm. Figure 15H shows the
are higher near the impeller and at par with radial velocities. TKE CSD for Swenson Walker crystallizer which gives relatively low
profiles are shown in Figure 13D, as expected high TKE is found spread amongst all the crystallizers with a uniform distribution
near impeller and dies out as measurement location is shifted from 0 to 1500 mm. Figure 15I shows the CSD for Batch Vacuum
away from it. Crystallizer. Majority of the crystals are in the range of 500–
Figure 14A–D shows the radial profiles of axial, radial and 1000 mm with a small fraction of coarser crystals up to 2000 mm.
tangential velocities and TKE at various axial positions of PBT Figure 15J shows the CSD for Disc Turbine Crystallizer with
Crystallizer. Since PBT is axial upward flow impeller, all the peak at 800 mm and spread up to 2500 mm. Figure 15K shows
three components of the velocity namely axial, radial and CSD for PBT Crystallizer. The trend is similar to the Batch
tangential have the velocity magnitudes higher in the bottom Vacuum Crystallizer with the maximum size of particles being
most position (z ¼ 0.017 m) and goes on decreasing in the 1200 mm and the size range varying from 0 to 3000 mm. The
upward direction with the negative velocity magnitudes at the stirred tank crystallizers (Batch Vacuum, Disc Turbine and PBT)
top portion. This shows clear evidence of recirculation loop all show the distribution skewed towards smaller size crystals as
over the tank and good mixing. This is evident from the radial opposed to Messo, Swenson DTB and Cerny direct contact
velocities (Figure 14B) which are downward in the top portions crystallizers.
and upward in the bottommost portion. Maximum axial
velocities contribute to around 40 % of the tip velocities while
maximum radial velocities are around 20 % of tip velocities and
tangential velocities are 25 % of tip velocities. The minimum Comparison of Crystallizers
velocities are around 10 %, 5 %, and 4 % of tip velocities, In the present work APV Kestner, Swenson Evaporative, Krystal
respectively. The TKE profiles of Figure 14D show higher and Swenson Walker crystallizers show least spread of the CSD,
magnitudes near the walls at all positions from bottom to top. while Messo and Swenson DTB showed large spread of CSD. This
This suggests good mixing and hence good solid suspension due can be attributed to the dead zones in the two crystallizers, which
to good turbulence kinetic energy. are evident from the schematic diagrams of the crystallizers
(Figure 2E‐F). Oslo and Cerny Direct Contact crystallizer show
comparable CSD as axial velocity is dominant in both the
Crystal Size Distribution of All 11 Crystallizers (CSD) crystallizers and aids in classification of the crystals. Crystallizers
Figure 15A and Table 3 show the CSD of APV Kestner with impeller and no draft tubes (Batch Vacuum, Disc Turbine, and
Crystallizer. Although the CSD is spread across 1200 mm the PBT) produce similar and comparable results with increasing
peak can be seen at around 600 mm, which constitutes of 45 % of crystal sizes.
total CSD. Figure 15B shows Swenson Evaporative Crystallizer
shows least distribution (up to 1000 mm) of CSD with peak at
CONCLUSIONS
500 mm. In case of Krystal evaporative crystallizer, the trend is
similar to the Swenson evaporative crystallizer, the difference Hydrodynamic studies of eleven crystallizers have been carried out
being the spread being more, i.e., up to 1200 (up to 1000 mm). and comparison has been made between the crystallizers. The CFD
For Oslo crystallizer (Figure 15D) the CSD varies within 500– simulations have shown that Swenson Evaporative, Krystal and
2600 mm, comprising of majorly 1800 mm crystals. Figure 15E Batch Vacuum show relatively low CSD. Additional work is in
shows CSD for Messo Crystallizer which produces coarser progress for analyzing the role of macromixing on the field of
crystal sizes of around 1000–4000 mm with peak at 3000 mm. supersaturation and hence the performance.
The Messo CSD differs from previous crystallizers as the
distribution is skewed towards coarser crystals. Figure 15F NOMENCLATURE
shows the CSD of Swenson Draft Tube Crystallizer which is
A Coefficient for moments
Bag birth due to aggregation, defined in Equation (29)
Bbr birth rate due to break‐up, defined in Equation (33)
Table 3. The consumption of 11 crystallizers on the basis of mean and C impeller clearance from the tank bottom (m)
CSD Cm, Ce1, Ce2 turbulence model parameters in the k‐e model
Mean crystal Crystal Standard Dag death rate due to aggregation, defined in Equa-
Sr. No. Crystallizer size (mm) deviation tion (30)
Dbr death rate due to break‐up, defined in Equation (34)
1 APV Kestaner 573 298 D impeller diameter (m)
2 Swenson Evaporative 532 264
DT disc turbine
3 Krystal 616 328
Dm molecular diffusion (m2/s)
4 Oslo Cooling 1155 753
5 Standard Messo 1448 1037
Dt the turbulent diffusivity (m2/s)
6 Swenson DTB 1840 1232 d particle size (m)
7 Cerny Direct Contact 1333 912 d32 mean crystal size (m)
8 Swenson Walker 537 302 f defined in Equation (5)
9 Batch Vacuum 608 394 G growth rate, defined in Equation (19)
10 DT crystallizer 1157 689 g acceleration due to gravity (m/s2)
11 PBT crystallizer 1100 748 H liquid height (m)

2154 THE CANADIAN JOURNAL OF CHEMICAL ENGINEERING VOLUME 92, DECEMBER 2014
J nucleation rate, defined in Equation (20) [8] J. Jager, S. De Wolf, W. Kapwijk, E. J. De Jong, “A New
k turbulent kinetic energy (m2/s2) Design for On‐line CSD‐Measurements in Product Slurries,”
keff the effective conductivity Proc. 10th Sym. Industrial Crystallization, Amsterdam 1989.
kg growth constant [9] A. Mersmann, Chem. Eng. Proc. 1988, 23, 213.
kn nucleation constant [10] J. B. Joshi, Y. T. Shah, Chem. Eng. Commun. 1981, 11, 165.
kt the turbulent thermal conductivity
[11] N. K. Nere, A. W. Patwardhan, J. B. Joshi, Ind. Eng. Chem.
m momentum, defined in Equation (2)
Res. 2003, 42, 2661.
n number density
N impeller rotation speed (1/s) [12] H. J. M. Kramer, J. W. Dijkstra, P. J. Verheijen, G. M. Van
p pressure (Pa) Rosmalen, Powder Technol. 2009, 108, 185.
P power dissipation (W) [13] X. Song, M. Zhang, J. Wang, P. Li, J. Yu, Proc. AIChE J.,
PBT pitched blade impeller Separations Division, New York 2009, 497a.
r radial coordinate (m) [14] W. Wantha, A. Flood, Chem. Eng. Commun. 2008, 195, 1345.
R radius of the vessel (m) [15] R. Plewik, P. Synowiec, J. Wójcik, A. Kus, Chem. Eng. Res.
S supersaturation Des. 2010, 88, 1194.
t time (s)
[16] Z. Zhu, H. Wei, Science Asia 2008, 34, 97.
T tank diameter (m)
T temperature (K) [17] M. Al Rasheed, J. Wójcik, R. Plewik, P. Synowiec, A. Kus,
uinlet inlet velocity (m/s) Chem. Eng. Proc. Process Intensif. 2013, 63, 7.
ur radial velocity (m/s) [18] J. Baldyga, W. Orciuch, Chem. Eng. Sci. 2001, 56, 2435.
utip tip velocity (m/s) [19] Z. Jaworski, A. W. Nienow, Chem. Eng. Sci. 2003, 91, 167.
v velocity (m/s) [20] A. A. Öncül, K. Sundmacher, D. Thévenin, Chem. Eng. Sci.
V volume of the reactor (m3) 2005, 60, 5395.
w weights [21] S. Rigopoulos, Prog. Energy Combust. Sci. 2010, 36, 412.
z axial coordinate (m)
[22] J. Cheng, C. Yang, Z. S. Mao, Chem. Eng. Sci. 2012, 68,
469.
Greek Letters [23] E. Kougoulos, A. G. Jones, M. Wood‐Kaczmer, Chem. Eng.
e turbulent energy dissipation (m2/s3) Res. Des. 2005, 83, 30.
m viscosity (Pa s) [24] D. Logashenko, T. Fischer, S. Motz, E. D. Gilles, G. Wittum,
mt turbulent eddy viscosity (Pa s) Comput. Vis. Sci. 2006, 9, 175.
r density (kg/m3) [25] A. Falola, A. Borissova, X. Z. Wang, Comput. Chem. Eng.
sk Prandtl number for turbulent kinetic energy 2013, 56, 1.
se Prandtl number for turbulent energy dissipation rate
[26] C. V. Rane, K. Ekambara, J. B. Joshi, D. Ramkrishna, AIChE J.
v frequency collision, defined in Equation (24)
2014, 60, 3596.
V break‐up rate per unit volume, define in Equation (31)
[27] M. Sen, A. Chaudhury, R. Singh, J. John, R. Ramachandran,
Int. J. Pharm. 2013, 445, 29.
Subscripts and Superscripts [28] Z. Grof, C. M. Schoellhammer, P. Rajniak, F. Štepánek, Int. J.
ag aggregation Pharm. 2011, 407, 12.
br break‐up [29] V. V. Ranade, J. B. Joshi, A. G. Marathe, Chem. Eng. Commun.
eff effective 1989, 81, 225.
i, j, k axis indexes of space coordinates [30] J. B. Joshi, N. K. Nere, C. V. Rane, B. N. Murthy, C. S.
t turbulent Mathpati, A. W. Patwardhan, V. V. Ranade, Can. J. Chem.
Eng. 2011, 89, 23.
[31] J. B. Joshi, N. K. Nere, C. V. Rane, B. N. Murthy, C. S.
REFERENCES Mathpati, A. W. Patwardhan, V. V. Ranade, Can. J. Chem.
[1] W. J. Genck, Chem. Eng. Proc. 2003, 36. Eng. 2011, 89, 754.
[2] A. S. Myerson, Handbook of Industrial Crystallization, [32] B. N. Murthy, J. B. Joshi, Chem. Eng. Sci. 2008, 63, 5468.
Butterworth‐Heinemann, Boston 1993. [33] J. B. Joshi, M. M. Sharma, Y. T. Shah, C. P. P. Singh, M. Ally,
[3] W. Omar, J. Urlich, Cryst. Res. Tech. 2003, 38, 34. G. E. Klinzing, Chem. Eng. Commun. 1980, 6, 257.
[4] J. Herman, M. Kramer, P. J. Jasens, Chem. Eng. Tech. 2003, 3, [34] B. N. Murthy, R. S. Ghadge, J. B. Joshi, Chem. Eng. Sci. 2006,
247. 62, 7184.
[5] J. W. Mullin, Crystallization, 4th edition, Butterworth‐ [35] M. Syamlal, T. J. O’Brien, AIChE Symp. Ser. 1989, 22.
Heinemann, Oxford 2001. [36] J. M. Dallavalle, Micromeritics, Pitman, London 1948.
[6] Q. Ruying, C. Zude, Chin. J. Chem. Eng. 1986, 36, 80. [37] J. R. Richardson, W. N. Zaki, Trans. IChemE 1954, 32, 35.
[7] W. Wöhlk, G. Niederjaüfner, G. Hoffmann, “Recovering of [38] J. B. Joshi, V. V. Ranade, Ind. Eng. Chem. Res. 2003, 42, 1115.
Cover Salt in Secondary Aluminium Industry,” Proc. of Int. [39] R. Mcgraw, Aerosol Sci. Technol. 1997, 27, 255.
Sym. Crystallization and Precipitation, Saskatoon, SK 1987, [40] M. R. Bhole, J. B. Joshi, D. Ramkrishna, Chem. Eng. Sci. 2008,
p. 99. 63, 2267.

VOLUME 92, DECEMBER 2014 THE CANADIAN JOURNAL OF CHEMICAL ENGINEERING 2155
[41] D. L. Marchisio, R. D. Virgil, R. O. Fox, J. Colloid Interface Sci.
2003, 258, 322.
[42] W. H. Press, S. A. Teukolsky, W. T. Vetterling, B. P. Flannery,
Numerical Recipes, Cambridge University Press, Cambridge
1992.
[43] R. G. Gordon, J. Math. Phys. 1968, 56, 655.
[44] S. B. Pope, “Probability Distributions of Scalars in Turbulent
Shear Flow,” Turbulent Shear Flows 2, L. J. S.</Ed>
Bradbury, Ed., Springer‐Verlag, Berlin 1980, p. 7.
[45] H. Luo, F. Svendsen, AIChE J. 1996, 42, 1225.
[46] W. Zheng, M. Zaisha, Y. Chao, S. Xiangqian, Chin. J. Chem.
Eng. 2006, 14, 713.
[47] Z. Sha, P. Oinas, M. Louhi‐Kultanen, G. Yang, S. Palosaari,
Powder Technol. 2001, 121, 20.

Manuscript received May 30, 2014; revised manuscript received

August 27, 2014; accepted for publication August 28, 2014.

2156 THE CANADIAN JOURNAL OF CHEMICAL ENGINEERING VOLUME 92, DECEMBER 2014