You are on page 1of 9

258 H. Patel et al.

Haresh Patel1 Research Article


Ramdhane Dhib1
Farhad Ein-Mozaffari1 Computational Fluid Dynamics Study of a
1
Department of Chemical Styrene Polymerization Reactor
Engineering, Ryerson
University, Toronto, Canada.
The computational fluid dynamics (CFD) approach was adopted to simulate ben-
zoyl peroxide (BPO)-initiated styrene polymerization in a laboratory-scale con-
tinuous stirred-tank reactor (CSTR). The CFD results revealed the effects of non-
homogeneity and the short-circuiting of the unreacted styrene and initiator on
the reactor performance. The study also investigated the effects of the impeller
speed and the residence time on the conversion and the flow behavior of the sys-
tem. The CFD simulation showed that intense mixing remained confined to a
small region near the impeller. With increasing impeller speed, it was found that
the perfectly mixed region near the impeller expanded, thus reducing non-homo-
geneity. Different contours were generated and exhibited the effect of the mixing
parameters on the propagation rate and styrene conversion. The monomer and
initiator conversions predicted with the CFD model were compared to those
obtained with a CSTR model. The CFD model accounts for the non-ideality
behavior of the polymerization reactor, and hence conversion predictions are
more realistic.

Keywords: Computational fluid dynamics, Mixing, Non-homogeneity, Polymerization,


Polystyrene, Simulations
Received: September 9, 2009; revised: October 17, 2009; accepted: October 26, 2009
DOI: 10.1002/ceat.200900440

1 Introduction ciency in a CSTR. Previously, several researchers [4–9] asserted


a strong relationship among the polymerization rate, the final
Mixing deserves particular attention in a highly viscous poly- polymer properties, and the nature of mixing in a polymeriza-
merization process because polymeric materials must be pro- tion process.
duced with some required specifications at the reactor stage Viscosity changes rapidly during polymerization. Conse-
itself [1]. In addition, the market competitiveness forces the quently, a non-ideal CSTR behavior is experienced, and it is
polymer industry to improve the conventional manufacturing not appropriate to represent it by an ideal model approach.
practice. Today, more emphasis is put on product quality. The Therefore, a model capable of incorporating the non-ideal flow
quality of a polymer formed in polymerization is strongly behavior is needed. Recent advancements in computational
affected by the nature of mixing and, hence, an in-depth un- power have made it possible to adopt the computational fluid
derstanding of the impact of mixing is inevitable. Insufficient dynamics (CFD) approach to build such non-ideal models for
mixing during polymerization leads to the formation of a polymerization reactors. Adopting the CFD approach to study
polymer with poor quality. Continuous stirred-tank reactors the effect of the mixing parameter on polymerization offers
(CSTR) are still widely used in polymerization. Short-circuit- some advantages. Early mathematical models on mixing were
ing, dead zones, and recirculation are the key mixing parame- formulated using complete homogeneity or complete segrega-
ters defining the degree of mixing in a CSTR [2, 3]. In fact, the tion conditions [6, 7, 10, 11]. However, polymerization pro-
degree of mixing significantly affects the reactant conversion, cesses are complex and cannot be represented by these simpli-
and the short-circuiting may lead to the loss of expensive fied conditions. In contrast to conventional models on mixing,
initiators. Low dispersion of the initiator also tends to lower CFD models are based on core transport equations. Flow visu-
initiator-monomer contacts and ends up in low initiator effi- alization of a polymerization process is not always possible
through experimental tests, but the flow domain can be simu-
– lated quite well using CFD. Besides, the mathematical and the
Correspondence: Assoc. Prof. F. Ein-Mozaffari (fmozaffa@ryerson.ca), experimental approaches cannot give the full details about the
Department of Chemical Engineering, Ryerson University, 350 Victoria spatial variations of the field variables. CFD can be exploited
Street, Toronto M5B 2K3, Ontario, Canada. to study the effect of mixing on reactive processes. For in-

www.cet-journal.com © 2010 WILEY-VCH Verlag GmbH & Co. KGaA, Weinheim Chem. Eng. Technol. 2010, 33, No. 2, 258–266
Polymerization 259

stance, Randick [12] used CFD to study CSTR behavior and Thus, CFD is an important tool for the design, scale-up, and
showed that feeding in the highly turbulent region around the optimization of the industrial reactors.
impeller reduces the formation of by-products in competitive-
consecutive reactions and increases the reaction yield. Kolha-
pure and Fox [13] carried out a CFD analysis of a tubular low- 2 Reactor Geometry
density polyethylene (LDPE) reactor and found that the effect
of mixing became more important for high monomer feed A CFD package (FLUENT 6.3) was used to simulate BPO-initi-
temperature and an appropriate injection point of the ini- ated styrene polymerization in a laboratory-scale CSTR
tiator. They showed that imperfect mixing disfavored mono- equipped with a 45° pitched-blade impeller. Fig. 1 shows the
mer conversion and increased the polydispersity index. Mahl- reactor specifications used in this study. In order to employ
ing et al. [14] used CFD to investigate the effect of the size of the CFD method for a flow domain analysis, the first task is to
the initiator injection pipe on the initiator mixing in a high- define the computational domain and to create the geometry
pressure LDPE tubular reactor. They found that a narrower of the flow domain. A well-mixed styrene-initiator mixture
tube diameter improved the initiator mixing. Since there was enters the reactor continuously. Under thermal effect, styrene
no consideration of the polymerization kinetics in this study, reacts and produces the polymer which, together with the
the effect of viscosity on the initiator mixing might not have unreacted styrene, leaves the reactor through the outlet pipe.
been evaluated properly. Similarly, Tosun and Bakker [15], To impose isothermal flow conditions inside the reactor, the
Read et al. [16], Bakker et al. [17], and Wells and Ray [18] used inlet and the wall temperatures were kept constant at 120 °C.
the CFD approach to investigate an LDPE polymerization The inlet initiator concentration was 0.01 mol/L for all BPO-
reactor. They reduced the computational time, but they sacri- initiated polymerization simulations. The residence time was
ficed the details achievable from CFD. Furthermore, Zhou et varied from 150 to 326 min, corresponding to flow rates from
al. [19] employed CFD to predict the initiator consumption 1.609 × 10–3 to 7.405 × 10–4 m3/h. The impeller speed was var-
and the molecular weight distribution of the LDPE polymer in ied from 100 to 1000 rpm. The reactor operating conditions
2D tubular and 3D autoclave reactors. However, they assumed are summarized in Tab. 1.
constant physical and transport properties in their simulations.
Hence, the effects of viscosity and density on the flow patterns
and polymer quality were not accounted for. Meszena and 3 Model Development and Solution
Johnson [20] and Zhu et al. [21] exploited CFD to simulate a Strategy
polymerization reactor, but they used the mass-averaged veloc-
ity field extracted from the continuity and momentum equa- Once the reactor geometry was generated, the geometry vol-
tions to solve the species transport equations. The velocity ume was divided into the tiny control volumes or cells using
field, satisfying the mass conservation law, might result in Gambit (FLUENT Inc.). The process of making small control
overshoot or undershoots for the modified species transport volumes is called grid generation or discretization, and these
equations. control volumes are called cells. The cells consist of faces and
Benzoyl peroxide (BPO)-initiated styrene polymerization is nodes. The tetrahedral cells were used to generate the unstruc-
a polymerization chain reaction that consists of three main tured grid for the simulation of the CSTR reactor (see Fig. 2).
steps, namely initiation, propagation and termination. Some- The advantage of using an unstructured grid is that a complex
times, due to viscosity rise during the polymerization, the ter- geometry can be meshed easily with the various shapes avail-
mination reaction becomes diffusion controlled, leading to the able for the cells. The accuracy of the solution depends
well-known gel effect. This effect may be due to poor micro- strongly on the grid resolution of the flow domain. The mesh
mixing on the molecular level and may give rise to reactor for the CFD calculations was fine-tuned on the basis of the ve-
instability. Micro-mixing affects both propagation and ter-
mination reaction rates. These effects are usually consid-
ered in the formulation of the two reaction rate constants,
in which diffusion control takes over and auto-acceleration
polymerization becomes a function of conversion. How-
ever, macro-mixing cannot be treated in the same way and
needs a different approach. In order to understand this be-
havior, it is necessary to analyze the mixing patterns in the
polymerization reactor. Despite the high importance of
mixing in a BPO-initiated styrene polymerization reactor,
no attempts have ever been made to simulate the reactor
model to extract the details about the non-homogeneity in
the CSTR. The present work explores efficient ways to con-
trol the product quality, by ensuring good mixing in a
BPO-initiated styrene polymerization CSTR, using the
CFD package. CFD enables us to compute velocity
domain, viscosity variations, and monomer and polymer
concentrations throughout the polymerization reactor. Figure 1. Polymerization reactor.

Chem. Eng. Technol. 2010, 33, No. 2, 258–266 © 2010 WILEY-VCH Verlag GmbH & Co. KGaA, Weinheim www.cet-journal.com
260 H. Patel et al.

Table 1. Polymerization process operating conditions. ~


in which ~
s is the stress tensor defined as:
  2 
Parameter Operating conditions ~
~
s ˆ l …∇~ t† ‡ …∇~ t†T ∇~t (3)
3
Inlet temperature [°C] 120
where P is the pressure, g is the gravitational force and ~
F is the
Wall temperature [°C] 120 external force.
Residence time [min] 150, 200 and 326 at 100 rpm
Species balance
Impeller speed [rpm] 100, 500 and 1000 for 200 min
∇…q~
tWj † ˆ ∇ Jj ‡ Sj (4)
Inlet stream BPO conc. [mol/L] 0.01
where Jj is the diffusive flux defined by:
locity profile calculated close to the impeller, which is a zone
of large velocity gradients. A minimum grid size in the mesh ~
Jj ˆ qDj ∇Wj (5)
was achieved by reducing the cell size to a final value below
which the changes in the velocity profile were insignificant. where Dj is the diffusivity and Sj is any source related to the jth
Mesh refinement near the impeller was accomplished using species. The transport equations are coupled.
the mesh growth factor function. This factor controls the mesh The properties of the polymerizing mixture depend on the
density by allowing the mesh elements to grow slowly as a mass fraction of each individual species, but the initiator is an
function of the distance from the impeller blade to the vessel exception. Therefore, for the simulation purpose, although the
walls. The original 3D mesh of the model had 372 571 compu- species mass fractions of initiator, monomer and polymer were
tational cells. To verify the grid independency, the number of known, viscosity and density of the mixture were considered
cells was increased from 372 571 to 562 435. The additional to depend on the monomer and polymer mass fractions only.
cells did not change the velocity magnitude in the regions of The viscosity significantly affects the flow behavior in the reac-
high velocity gradients by more than 0.78 %. Thus, 372 571 tor. The viscosity of the mixture was calculated from the fol-
cells were employed in this study. lowing correlation suggested by Kim and Nauman [23]:
 : 1:2
0:6
l ˆ l0 = 1 ‡ l0 c =35000 (6)

where c_ is the shear rate. The zero shear viscosity (l0) of the
reaction mass was estimated using the correlation below [23]:

ln…l0 † ˆ 11:091 ‡ 1109=T ‡ MwP


0:1413
‰12:032wP

19:501wP2 ‡ 2:92wP3 ‡ 1327wP ‡ 1359wP2 ‡ 3597wP3 =TŠ
(7)

where MwP is the polymer molecular weight, wP is the polymer


mass fraction in the mixture and T is the temperature. The
shear rate in Eq. (6) was calculated using Metzner-Otto’s cor-
relation [24]:

c_ ˆ Ks N (8)
Figure 2. Discretized domain of the reactor.
where Ks is a weak function of impeller type, which was as-
signed a value of 11.0 for the pitched-blade impeller [25]. N is
Governing transport equations were discretized in order to
the impeller speed. The density of the mixture was estimated
get numerical values of the flow field variables (pressures,
from the correlation reported by Soliman et al. [26]:
species mass fractions, and velocities) on the cell-centered
nodes. Assuming steady-state conditions, the transport equa-
q ˆ …1174:7 0:918T †…1 wP † ‡ …1250:0 0:605T †wP (9)
tions of continuity, momentum, and species balance are writ-
ten below [22]:
The monomer diffusivity was assumed constant (2.0·10–9 m2/s)
Continuity and was taken from Soliman et al. [26]. The user-defined func-
tions (UDF) were utilized to link Eqs. (6)–(9) to the FLUENT
…∇q~
t† ˆ 0 (1) solver.
The species transport equation has a reactive source term
where q is the density and ~
t is the velocity vector. which represents the production of the species and, therefore,
formulation of the reactive source term is required. Styrene
Momentum
  polymerization has been widely studied. Hui and Hamielec
∇…q~t† ˆ
t~ ~
s ‡ qg ‡ ~
∇P ‡ ∇~ F (2) [27] experimentally studied styrene thermal polymerization.

www.cet-journal.com © 2010 WILEY-VCH Verlag GmbH & Co. KGaA, Weinheim Chem. Eng. Technol. 2010, 33, No. 2, 258–266
Polymerization 261

Lately, Gao and Penlidis [28] and Dhib et al. [29] thoroughly sumption rates. The source terms should be provided to the
studied initiator decomposition and chemical initiation ki- FLUENT code in terms of unit mass per unit volume:
netics for several polymerization scenarios and tested kinetic
models against a wide range of data. Their modeling approach Sm ˆ RP × Mwm (19)
was comprehensive and took into account the overwhelming
diffusion effects in the auto-acceleration region. Under ther- Si ˆ Ri × Mwi (20)
mal effect, the initiator fragments and styrene polymerizes
according to the reaction mechanism below: where Mwm and Mwi are the molecular weights of the mono-
mer and the initiator, respectively. Furthermore, all rate
Thermal initiation
constants involved in the reaction are assumed to obey the
Kth
3M ! 2R1 (10) Arrhenius equation. The kinetic parameters of the rate con-
stants are given in Tab. 2.
Chemical initiation FLUENT 6.3 was used to solve the discretized governing
transport equations in a steady-state domain. FLUENT does
Kd
I ! 2Rin (11) not have the option to incorporate the polymerization reaction
kinetics and was therefore modified by incorporating a UDF
Ki
Rin ! R1 (12) coded in C language. The UDF code was written for the source
terms previously defined and linked to the species transport
Propagation model coded in FLUENT. The UDFs are defined using
Kp DEFINE macros which are supplied by FLUENT. The details
Rr ‡ M ! Rr‡1 r≥1 (13) can be found in the FLUENT User’s Guide Documentation.
The source term was linearized with respect to the monomer
Termination mass fraction. Species transport equations were solved for ini-
Ktc tiator and monomer. The polymer mass fraction was simply
Rr ‡ Rs ! Pr‡s r; s ≥ 1 (14)
obtained by subtracting the monomer and initiator mass frac-
tions from unit 1.
where M, I, R , and P refer to the styrene monomer, initiator,
The multiple reference frames (MRF) technique was used to
live polymer radical, and dead polymer, respectively. Kth, Kd,
model the geometry of the impeller in the mixing vessel [31].
Ki, Kp, and Ktc refer to the rate constant for thermal decompo-
A rotating frame was used for the region containing the impel-
sition, initiator decomposition, intermediate initiation, propa-
ler, while a stationary frame was used for regions that are sta-
gation and termination reaction, respectively.
tionary, containing the tank walls. The governing equations of
In BPO-initiated polymerization, live radicals are produced
the flow domain inside the rotating frame were solved in the
by thermal and chemical initiation reactions. Employing the
frame of the enclosed impeller, while those outside the rotating
steady-state hypothesis leads to equating the overall initiation
frame were solved in the stationary frame. A steady transfer of
rate with the termination rate, which gives the total live radical
information was made at the MRF interface as the solution
concentration as follows:
progressed. This method facilitates incorporation of the impel-
s ler motion even with a complex geometry and has been found
2Kth ‰M Š3 ‡2fKd ‰I Š to yield flow field predictions comparable to those obtained
‰R Š ˆ (15)
Ktc using the time-dependent sliding mesh (SM) impeller model
[32, 33]. Therefore, the MRF model is usually employed to
where f is the initiator efficiency. It stands for the fraction of reduce computational requirements. Saeed et al. [3] and Ford
the initiator amount that participates in the reaction. Thus, et al. [34] successfully utilized the MRF technique to model
the polymerization rate per unit volume is given by: continuous-flow mixing processes.

RP ˆ Kp ‰M Š‰R Š (16)
Table 2. Values of kinetic parameters (K = A e–E/RT).
or
Frequency Value Activation Value Source
s factor A energy E
2Kth ‰M Š3 ‡2fKd ‰I Š
RP ˆ Kp ‰M Š (17) Ad 3.816·109 m3/s Ed 2.73254·104 cal/mol Villalobos
Ktc
et al. [30]

The initiator (BPO) consumption rate per unit Ap 2.170·107 m3/kmol s Ep 7.75923·103 cal/mol Gao and
volume is given by: Penlidis [28]
Ath 2.190·105 m6/kmol2 s Eth 2.74400·104 cal/mol Villalobos
Rini ˆ Kd ‰I Š (18) et al. [30]
Atc 8.200·109 m3/kmol s Etc 3.47129·103 cal/mol Gao and
However, the species transport equation requires
Penlidis [28]
source terms for the monomer and initiator con-

Chem. Eng. Technol. 2010, 33, No. 2, 258–266 © 2010 WILEY-VCH Verlag GmbH & Co. KGaA, Weinheim www.cet-journal.com
262 H. Patel et al.

At the feed stream, the inlet-velocity boundary condition each steady-state simulation of the polymerization reactor was
was considered. The inlet velocity and styrene mass fraction achieved after 5–6 days.
were supplied. At the reactor outlet, zero normal gradients
were employed for all variables. Therefore, there was no diffu-
sive flow normal to the boundary. Assuming symmetrical 4 Results and Discussion
boundary conditions, hence zero normal gradients and no
penetration were assumed for all variables on the liquid sur- A commercially available CFD package (FLUENT 6.3) was
face. No-slip and no-penetration conditions were imposed on used to investigate the effect of mixing on the polymerization
the transport equations on the tank walls. A fixed temperature of styrene in a CSTR. CFD simulations of the BPO-initiated
was employed on the vessel walls. A single-phase laminar re- styrene polymerization were not easy to converge. The difficul-
gime was assumed in all simulations with a Reynolds number ties arose due to the following two reasons:
of approximately 20 for the highest impeller speed and for the Since bulk polymerization involves simultaneous decompo-
lowest viscosity. The reactor was considered isothermal. A sec- sition of monomer and initiator, two species transport equa-
ond-order upwind discretization scheme was used to calculate tions were required to simulate the reactor. Both thermal and
the face fluxes in the momentum and species transport equa- chemical initiation reactions instantly contributed to the for-
tions. The PRESTO scheme was employed for the pressure dis- mation of live polymer radicals. Besides, the monomer trans-
cretization, and velocity-pressure coupling was solved using port equation is coupled with the initiator transport equation,
the SIMPLE algorithm. The governing transport equations thus making the polymerization rate highly nonlinear.
were integrated over small control volumes, and the integra- The coupled nature of the initiator and monomer transport
tion resulted in a system of linear algebraic equations, which equations and the highly non-linear source term were addi-
was solved using the Gauss-Seidel iterative method in combi- tional complications compared to thermal polymerization
nation with the algebraic multi-grid (AMG) method. alone [35]. Moreover, in comparison with other species, the
The reactor model comprises a continuity equation, three initiator is provided in very small quantities and hence any
momentum equations, a pressure correction, and two species truncation error impacted significantly on the convergence.
transport equations. In addition, the model has property up- Nevertheless, no specific rule had to be established to get the
date equations for density and viscosity. A special solution solution converged. Thus, one has to proceed with the trial
strategy was undertaken to ensure convergence, which was not and error approach to get the desired solution for a given flow
easily achievable for the highly coupled nonlinear transport domain. Still, the convergence was not straightforward and
equations of the polymerization system. First, the momentum each simulation needed a special treatment for the under-re-
equations were solved for pure styrene in the flow domain to laxation factor to get the problem solved.
get a partially converged solution for the flow. Once the non- The study of BPO-initiated polymerization of styrene is
reactive flow field solution was obtained for all the transport divided into two parts: the effect of the impeller speed and the
equations, the species source terms were introduced and the effect of the residence time on the styrene conversion.
complete model was solved. The under-relaxation factor for
the species transport equation was increased gradually as the
simulation progressed. The solution was considered converged 4.1 Effect of the Impeller Speed
when the residuals for the continuity, momentum, and species
equations were smaller than 1 ×10–6. However, only residual Three simulation runs were conducted to investigate the effect
monitoring did not ensure the convergence of the solutions, as of the impeller speed on the styrene conversion. The impeller
the field variables of interest were still changing with the itera- speed was varied from 100 to 1000 rpm. The residence time,
tion, even though the residual was dropped below the recom- the BPO concentration at the inlet, and the temperature were
mended criteria. Hence, it became necessary to monitor the kept constant at 200 min, 0.01 mol/L, and 120 °C, respectively.
field variables themselves, to assure that they were not chang- Fig. 3 shows the contours of the styrene mass fractions for an
ing with the iteration near the convergence. The monitoring of impeller speed of 100 rpm. The CFD simulation results showed
the field variables on a single node may be misleading. There- that the polymerization reactor was not perfectly mixed and
fore, the field variables were monitored on a surface made of a the styrene concentration was not uniform. Two separate
group of nodes. It was assured that convergence was reached regions are clearly visible. A fresh styrene stream entered the
for each simulation by monitoring the surface integral quan- reactor at a point far from the impeller. In this region, the con-
tity of the field variables, such as the sum of the velocity mag- vective flow was not high enough to sweep away the newly
nitudes on the top liquid surface, the unreacted styrene entered styrene. Thus, a styrene-rich region formed near the
outflow from the outlet, and the momentum coefficient about reactor top, whereas in the vicinity of the impeller there was a
the impeller axis. Computations were carried out using the mixture rich in polymer.
supercomputing facilities of High-Performance Computing Conversion contours were also generated in order to investi-
Virtual Laboratory (HPCVL). Each simulation was run in par- gate the reaction progress at different locations in the reactor.
allel with 24–35 dual-core SUN Ultra-Spark IV, 1.8-GHz Sun Fig. 4 shows the contour of the percentage of styrene converted
Micro-Systems central processing units (CPUs). The grid was in the vertical plane for an impeller speed of 100 rpm. A con-
partitioned into 24–35 parts and one CPU was assigned to version not exceeding 4 % took place in the inlet pipe. How-
each partition. On average, a single simulation run required ever, it is worth noting that up to 20 % conversion took place
around 300,000 iterations to converge. The convergence for in the segregated region. The polymer formed in this region

www.cet-journal.com © 2010 WILEY-VCH Verlag GmbH & Co. KGaA, Weinheim Chem. Eng. Technol. 2010, 33, No. 2, 258–266
Polymerization 263

Figure 5. Contour of the initiator mass fraction at N = 100 rpm,


Figure 3. Contour of the styrene mass fraction at N = 100 rpm, V/Q = 200 min, T = 120 °C, and [If ] = 0.01 mol/L.
V/Q = 200 min, T = 120 °C, and [If ] = 0.01 mol/L.
a)

Figure 4. Contour of the percent conversion at N = 100 rpm, V/Q


= 200 min, T = 120 °C, and [If ] = 0.01 mol/L.
b)

has a high polydispersity index and therefore dominates the


overall product quality.
Additionally, the contours of the initiator mass fraction in
Fig. 5 indicate that the initiator concentration is not uniform
in the regions far from the impeller. Consequently, one may
expect a wider distribution of molecular weight of the polymer
and a variable polydispersity index in the upper region. An-
other important parameter affecting the polymer characteris-
tics is the propagation rate. The simulation results presented
in Fig. 6 show the influence of the impeller speed on the prop-
agation rate. It is worth noting that the region having a high
propagation rate was larger at an impeller speed of 100 rpm
than at 500 rpm.
Fig. 7 shows the plot of initiator and monomer conversions
versus impeller speed. The falling conversion trend can be at- Figure 6. Contour of the propagation rate (kg/m3 s) at V/Q =
tributed to two reasons. Firstly, as the impeller speed increased, 200 min, T = 120 °C, and [If ] = 0.01 mol/L: (a) N = 100 rpm and
more unreacted styrene was pushed out of the reactor. This (b) N = 500 rpm.

Chem. Eng. Technol. 2010, 33, No. 2, 258–266 © 2010 WILEY-VCH Verlag GmbH & Co. KGaA, Weinheim www.cet-journal.com
264 H. Patel et al.

tor is fully mixed as is shown in Fig. 8. It is obvious that the


reactor achieved complete homogeneity at the impeller speed
of 1000 rpm. Therefore, the conversion obtained using CFD
should match the CSTR prediction at 1000 rpm impeller
speed. However, referring to Fig. 7, the CFD-predicted conver-
sions were lower than the CSTR ones at 1000 rpm, indicating
that styrene short-circuiting was taking place in the reactor.

Figure 7. Conversion versus impeller speed at V/Q = 200 min,


T = 120 °C, and [If ] = 0.01 mol/L.

flow movement resembles a short-circuiting of unreacted sty-


rene at high impeller speeds in this type of vessel geometry.
The identification of the non-ideal flow (e.g. short-circuiting
and dead zones) in continuous-flow mixing vessels through
CFD has been discussed by Saeed et al. [3] and Ford et al. [34].
Secondly, the decrease in conversion can be more appropri-
ately explained by looking at the contours of the propagation
Figure 8. Contour of the styrene mass fraction at N = 1000 rpm,
rate shown in Fig. 6 for the impeller speeds of 100 and
V/Q = 200 min, T = 120 °C, and [If ] = 0.01 mol/L.
500 rpm. It is obvious that the propagation rate is higher in
the upper styrene-rich region of the reactor in both cases.
Since, this styrene-rich region shrinks at higher impeller 4.2 Effect of Residence Time
speeds, the average propagation rate slows down and results in
low polymer formation. This phenomenon can be called The residence time is a predominant factor affecting the con-
monomer dilution effect on the propagation rate. Both short- version. Three simulations at constant impeller speed, temper-
circuiting and monomer dilution effects lead to a decrease in ature, and initial BPO concentration were performed with resi-
conversion at higher impeller speeds. Similarly, the experimen- dence times of 150, 200, and 326 min. The monomer and
tal data reported by Erdogan et al. [36] confirmed the CFD initiator conversions are plotted versus residence time in
predictions of decreasing conversion at higher impeller speed. Fig. 9. The CFD model predicted a higher monomer conver-
The conversion predicted using CFD was compared to that sion than the CSTR model. The difference between the mono-
obtained with the CSTR model in Fig. 7. The difference mer conversions is likely due to the fact that the CSTR model
between the two predictions is explained below. Considering
the CSTR model for BPO-initiated polymerization given by:
for initiator
Q 
0ˆ Kd ‰Ie Š ‡ ‰I Š ‰Ie Š (21)
V f
for monomer
s
2fKd ‰Ie Š ‡ 2Kth ‰Me Š3 Q
0ˆ Kp ‰Me Š ‡ …‰Mf Š ‰Me Š† (22)
Kt V
where [If ] and [Ie] are the initiator concentration in the feed
and at the reactor exit, respectively, and Q/V is the inverse of
the residence time. Eqs. (21) and (22) were numerically solved
to calculate the species conversion for a given residence time.
The conventional CSTR mole balance implicitly assumes per-
fect mixing without accounting for the effect of impeller speed.
As mentioned earlier, at N < 500 rpm the reactor is not fully
mixed; the conversion is higher because of the monomer dilu- Figure 9. Conversion versus residence time at N = 100 rpm,
tion effect. At higher impeller speed (N = 1000 rpm), the reac- T = 120 °C, and [If ] = 0.01 mol/L.

www.cet-journal.com © 2010 WILEY-VCH Verlag GmbH & Co. KGaA, Weinheim Chem. Eng. Technol. 2010, 33, No. 2, 258–266
Polymerization 265

does not take into account the possibility of non-homogeneity. Symbols used
A longer residence time allowed the system viscosity to
increase (Fig. 10); therefore, less homogeneity and monomer A [m3/kmol s] frequency factor, also in [m6/kmol2 s]
dilution were attained in the reactor at a fixed impeller speed. CP [J/kg K] specific heat
As mentioned earlier, the monomer dilution leads to a Dm [m2/s] diffusivity
decrease in monomer conversion. The presence of non-homo- E [J/mol] activation energy
geneity contradicted the hypothesis made to develop the CSTR ~
F [N/m3] external forces
model. G [m/s2] gravitational acceleration
[Ie] [kmol/m3] initiator concentration at the reactor
exit
[If ] [kmol/m3] initiator concentration in the feed
J [kg/m2s] diffusive flux
Ks [–] Metzner-Otto’s constant
K [–] rate constant
[Me] [kmol/m3] monomer concentration
[Mf ] [kmol/m3] monomer feed concentration
Mw [kg/kmol] molecular weight
N [s–1] impeller speed
P [Pa] pressure
Q [m3/s] volumetric flow rate
r [–] number of monomers in live radical
[R ] [kmol/m3] radical concentration
s [–] number of monomers in live radical
Sj [kg/m3s] source term for the jth species
Tw [K] wall temperature
V [m3] reactor volume
Figure 10. Contour of the viscosity (kg/m s) of the reactive mass
Wj [–] mass fraction of the jth species
at N = 100 rpm, T = 120 °C, and V/Q = 326 min.
Greek symbols
5 Conclusions q [kg/m3] mixture density
c_ [s–1] shear rate
CFD was employed to investigate the effect of mixing on sty- l [Pa s] dynamic viscosity
rene polymerization with BPO initiator in a laboratory-scale l0 [Pa s] zero shear viscosity
CSTR. The highly coupled nature of the transport equations
for the polymerization process raised difficulties in achieving
convergence. Simulations were run on a supercomputing facil- References
ity using 24–35 CPUs for each run. The CFD results showed
that the region close to the impeller was almost perfectly [1] B. W. Brooks, Ind. Eng. Chem. Res. 1997, 36, 1158.
mixed and the regions far from the impeller remained [2] S. Saeed, F. Ein-Mozaffari, J. Chem. Technol. Biotechnol.
unmixed, due to high reaction mass viscosity. Taking into 2008, 83, 559.
account the short-circuiting and the non-homogeneity, the [3] S. Saeed, F. Ein-Mozaffari, S. R. Upreti, Ind. Eng. Chem. Res.
CFD model gave better conversion predictions than the CSTR 2008, 47, 7465.
model. Various contours showing the variations of mass frac- [4] M. Harada, K. Tanaka, W. Eguchi, S. Nagata, J. Chem. Eng.
tion, propagation rate, viscosity, and conversion were gener- Jpn. 1968, 1, 148.
ated to make flow visualization possible. These contours [5] W. M. Cole, AIChE Symp. Ser. 1975, 72, 51.
helped to extract information about the polymer quality in dif- [6] G. Tosun, AIChE J. 1992, 38, 425.
ferent regions of the reactor. The CFD simulations also showed [7] Y. A. Prochukhan, K. S. Minsker, A. A. Berlin, R. K. Bakhito-
the impact of the impeller speed and the residence time on the va, N. S. Yenikolopyan, Polym. Sci. USSR 1988, 30, 1317.
propagation rate and conversion. [8] M. F. Kemmere, J. Meuldljk, A. A. H. Drinkenburg, A. L.
German, J. Appl. Polym. Sci. 2001, 79, 944.
[9] J. Heidarian, M. G. Nayef, M. A. W. D. Wan, Ind. Eng. Chem.
Acknowledgements Res. 2004, 43, 6048.
[10] J. Y. Kim, R. L. Laurence, Korean J. Chem. Eng. 1998, 15, 273.
The authors would like to thank the Natural Science and Engi- [11] C. M. Villa, J. O. Dihora, W. H. Ray, AIChE J. 1998, 44, 1646.
neering Research Council of Canada (NSERC) and the High [12] J. J. Randick, M.A.Sc. Thesis, University of Missouri, Rolla
Performance Computing Virtual Library (HPCVL) Canada for 2000.
providing financial support for this research. [13] N. H. Kolhapure, R. O. Fox, Chem. Eng. Sci. 1999, 54, 3233.
The authors have declared no conflict of interest.

Chem. Eng. Technol. 2010, 33, No. 2, 258–266 © 2010 WILEY-VCH Verlag GmbH & Co. KGaA, Weinheim www.cet-journal.com
266 H. Patel et al.

[14] F. O. Mahling, A. Daib, N. Kolhapure, R. O. Fox, 10th Eur- [26] M. A. Soliman, T. Aljarboa, M. Alahmad, Polym. Eng. Sci.
opean Symposium on Computer Aided Process Engineering, 1994, 19, 1464.
Florence 2000, 427. [27] A. W. Hui, A. E. Hamielec, J. Appl. Polym. Sci. 1972, 16, 749.
[15] G. Tosun, A. Bakker, Ind. Eng. Chem. Res. 1997, 36, 296. [28] J. Gao, A. Penlidis, J. Macromol. Sci. Rev. Macromol. Chem.
[16] N. K. Read, S. X. Zhang, W. H. Ray, AIChE J. 1997, 43, 104. Phys. 1996, C36, 199.
[17] A. Bakker, A. H. Haidari, L. M. Marshall, Chem. Eng. Prog. [29] R. Dhib, J. Gao, A. Penlidis, Polym. React. Eng. 2000, 8, 299.
2001, 97, 31. [30] M. A. Villalobos, A. E. Hamielec, P. E. Wood, J. Appl. Polym.
[18] G. J. Wells, W. H. Ray, DECHEMA Monogr. 2001, 137, 49. Sci. 1993, 50, 327.
[19] W. Zhou, E. Marshall, L. Oshinowo, Ind. Eng. Chem. Res. [31] J. Y. Luo, A. D. Gosman, R. I. Issa, Inst. Chem. Eng. Symp.
2001, 40, 5533. Ser. 1994, 136, 549.
[20] Z. G. Meszena, A. F. Johnson, Macromol. Theory Simul. 2001, [32] A. Brucato, M. Ciofalo, F. Crisfi, G. Micale, Chem. Eng. Sci.
10, 123. 1998, 53, 3653.
[21] L. Zhu, K. A. Narh, K. S. Hyun, Int. J. Heat Mass Transfer [33] N. G. Deen, T. Solberg, B. H. Hjertager, Can. J. Chem. Eng.
2005, 48, 3411. 2002, 80, 1.
[22] R. B. Bird, W. E. Stewart, E. N. Lightfoot, Transport Phenom- [34] C. Ford, F. Ein-Mozaffari, C. P. J. Bennington, F. Taghipour,
ena, John Wiley, New York 2002. AIChE J. 2006, 52, 3562.
[23] D. M. Kim, E. B. Nauman, J. Chem. Eng. Data 1992, 37, 427. [35] H. Patel, M.A.Sc. Thesis, Ryerson University, Toronto 2007.
[24] A. B. Metzner, R. E. Otto, AIChE J. 1957, 3, 3. [36] S. Erdogan, M. Alpbaz, A. R. Karagoz, Chem. Eng. J. 2002,
[25] R. J. Wilkens, C. Henry, L. E. Gates, Chem. Eng. Prog. 2003, 86, 259.
99, 44.

www.cet-journal.com © 2010 WILEY-VCH Verlag GmbH & Co. KGaA, Weinheim Chem. Eng. Technol. 2010, 33, No. 2, 258–266

You might also like