You are on page 1of 16

FULL PAPER

Polyolefins www.mre-journal.de

A Monte Carlo Method to Quantify the Effect of Reactor


Residence Time Distribution on Polyolefins Made with
Heterogeneous Catalysts: Part I—Catalyst/Polymer Particle
Size Distribution Effects
João B. P. Soares* and Jazmín Romero

of a catalyst tested in the laboratory and


Polyolefins are commercially produced in continuous reactors that have a the same catalyst when scaled up to indus-
broad residence time distribution (RTD). Most of these polymers are made trial production.
with heterogeneous catalysts that also have a particle size distribution (PSD). This fact has been recognized for a
while. Soares and Hamielec[4] and Debling
These are totally segregated systems, in which the catalyst/polymer par­
and co-workers[5] were the first to inves-
ticle can be seen as a microreactor operated in semibatch mode, where the tigate the effect of RTD on the PSD of
reagents (olefins, hydrogen, etc.) are fed continuously to the catalyst/polymer polyolefins made with heterogeneous
particle, but no polymer particle can leave. The reactor RTD has a large influ­ catalysts, but they used distinct math-
ence on the PSD of the polymer particles leaving the reactor, as well as in ematical approaches. Soares and Hami-
polymer microstructure and properties, polymerization yield, and composi­ elec[4] discretized the catalyst PSD into
populations of several size classes, and
tion of reactor blends. This article proposes a Monte Carlo model that can
followed these populations as they were
describe how particle RTD in a single or a series of reactors can affect the transported down a series of continuous
PSD of polymer particles made under a variety of operation conditions. It is reactors having any RTD. Their model
believed that this is the most flexible model ever proposed to model this pheno­ could handle catalysts with any PSDs, and
menon, and can be easily modified to track all properties of interest during reactors with ideal, nonideal, or experi-
mentally measured RTDs. Theirs was a
polyolefin production in continuous reactors with heterogeneous catalysts.
versatile approach, but somewhat cum-
bersome from a computational point of
view, particularly when many reactors in
1. Introduction series needed to be model. Debling and co-workers[5] adopted
a more elegant approach, using population balances to model
Most commercial polyolefins are made in continuous reactors multistage olefin polymerization processes. Their model also
using heterogeneous catalysts. Even though the polymeriza- allowed for consideration of particle size selection effects in the
tion medium (slurry diluent, slurry bulk, gas phase) and reactor process, and the authors used it to simulate gas-phase vertical
configuration (autoclave, loop, fluidized bed, riser/downer and horizontal stirred tank reactors, loop reactors, and fluid-
multizone) can vary substantially from among polymer pro- ized-bed reactors. In subsequent publications, Debling et al.[6]
duction technologies, all of these reactors are ideally operated and Zacca et al.[7] extended their previous model to study the
near steady-state conditions and the catalyst/polymer particles production of multilayered products such as impact polypro-
follow residence time distributions (RTDs) that approximate pylene and bimodal polyethylene. Khang and Lee[8] proposed a
those of a continuous stirred tank reactor (CSTR) or a series model to predict the PSD of polyolefins made in fluidized bed
of CSTRs.[1] Contrarily, most olefin polymerization catalysts reactors, considering nonideal mixing, and derived analytical
are tested in laboratory reactors that are operated in either expressions that took into account both mass transfer limita-
semibatch or batch mode and, consequently, are exposed to tions and catalyst deactivation. Hatzantonis et al.[9] developed
a completely different RTD.[2,3] Failing to consider the drastic a population balance model to describe the PSD of polyolefins
differences in RTD between laboratory and industrial reactors made in gas phase reactors, including the effects of attrition,
may lead to significant discrepancies between the performance elutriation, and agglomeration for catalysts populations with
either uniform sizes or size distribution. Hatzantonis and
Prof. J. B. P. Soares, Dr. J. Romero Kiparissides[10] also used their population balance model to study
Department of Chemical and Materials Engineering the effect of RTD on the PSD of polyethylene made in fluidized
University of Alberta bed reactors, as well as to quantify the effect of mean particle
T6G 2R3 Edmonton, AB, Canada
E-mail: jsoares@ualberta.ca diameter on fluidization dynamics. Prasetya et al.[11] proposed
a dynamic population balance to model the impact of reactor
DOI: 10.1002/mren.201700031 RTD impact polypropylene copolymers. Covezzi and Mei[12]

Macromol. React. Eng. 2018, 12, 1700031 1700031  (1 of 16) © 2017 WILEY-VCH Verlag GmbH & Co. KGaA, Weinheim
www.advancedsciencenews.com www.mre-journal.de

used a population balance model to explain how the multi- fractions of polymer made on different reactors in the series
zone reactor could be used to achieve heterophasic and multi­ (for the case of heterophasic or “bimodal” reactor blends) and,
modal reactor blends with better homogeneity by alternating in some cases, intraparticle mass and heat transfer resist-
two CSTRs, one operated as a fast fluidized bed (riser) and the ances. As powerful as they are, most of these models require
other as a moving packed bed reactor (downcomer). Zoellner the solution of simultaneous differential-integral equations,
and Reichert[13] used a dynamic population balance approach and cannot handle complex polymerization kinetics, catalyst
to describe PSD evolution in a lab-scale ethylene gas phase particle nonuniformities, reactor RTDs or catalysts PSDs of
polymerization under mass transfer limitations, and validated unusual shapes, and inter- and intraparticle mass and heat
their results by video microscopy. Their model did not include transfer effects very effectively. The model we are proposing
RTD effects, since they assumed all particles stayed the same herein differs from the previously published methods because
time in the reactor, which is the right assumption for their lab it uses a Monte Carlo approach to simulate the polymer PSD
scale reactor, but they did not extend their model to continuous made in a series of reactors with RTDs of any shape, while
polymerization cases. Yiannoulakis et al.[14] used steady-state using catalysts with PSDs also of any shape. The approach
population balances to predict PSD of ethylene copolymers we adopted it to sample catalyst particles randomly from the
produced in fluidized bed reactors under moderate particle population entering the first reactor in the series, then follow
agglomeration conditions. They employed the polymeric flow each particle as it flows down the reactor train, while deter-
model[15] to describe heat and mass transfer limitations in the mining its residence time in each reactor by randomly sam-
polymer particles and, as a simplification, assumed that all pling it from the RTD of each individual reactor. This method
catalyst particles had the same size when they were fed to the allows not only for unprecedented flexibility regarding the
reactor. Yiagopoulos et al.[16] demonstrated that prepolymeri- shapes of reactor RTD and catalyst PSD, but it also makes it
zation could be used to effectively minimize the drastic mass easy to consider catalyst particles with nonuniform active site
and heat transfer limitations that may occur at the beginning concentrations, different polymerization kinetics parameters,
of the polymerization using the polymeric flow model, but did slow activation and induction times, prepolymerization effects,
not consider PSD or RTD effects in their model. Zacca and and volume-dependent reactor residence times, to men-
Debling[17] used a population balance model based on their pre- tion a few factors that can be investigated with this versatile
vious publication[5] to study particle overheating for an entire novel approach. Equally important, the computation time for
population of particles. They applied their model to different these simulations, one of the main concerns in Monte Carlo
reactor types and multistage configurations, and analyzed the methods, is not excessive even in a regular laptop computer,
effect of catalyst PSD, reactor RTD, and prepolymerization con- as we will discuss in more details below. Other phenomena
ditions. Neto and Pinto[18] used a population balance approach such as polymer powder bulk density, particle composition of
to describe the evolution of polypropylene PSD made in steady- polymer reactor blends (such as bimodal polyethylenes and
state slurry and bulk polymerization reactors. Their model also heterophasic impact polypropylene), influence of inter- and
predicted polypropylene microstructure, and used empirical intraparticle mass, and heat transfer resistances, electrostatics,
correlations to estimate product performance properties. Kim agglomeration, and particle melting may contribute to the PSD
and Choi[19] modeled particle segregation in fluidized bed reac- of polyolefins made in commercial processes, and will be the
tors using a multicompartment population balance approach, subject of future publications from our group.
employing size-dependent transfer constants to account for
the complex fluid dynamics of these reactors. Harshe et al.[20]
developed a generalized dynamic model based on a mixing cell 2. Model Development
approach coupled with population balances to model the PSD
of polypropylene made in gas phase reactors. Dittrich and Mut- The rationale for the proposed Monte Carlo method is illus-
sers[21] proposed a new model to study the effect of reactor RTD trated in Figure 1. The model assumes that polyolefins are
on the PSD of polyolefin particles made on horizontal stirred made on catalyst/polymer particles that act as completely seg-
gas-phase reactors that did not rely on the commonly used regated semibatch reactors. Reagents (monomer, comonomer,
CSTR-in-series approach, but rather on experimental RTDs. In hydrogen, etc.) are fed at a constant rate to the polymer particle,
a more sophisticated analysis, Fan et al.[22] combined popula- but the produced polymer stays in the particle “microreactor”
tion balances with a computational fluid dynamics model to until it leaves the reactor train. The polymerization reactors (the
compare the PSD of predicted and experimental polyethylene “macroreactors”, not the particle “microreactors”) may be of any
particles made in a fluidized bed reactor. Finally, Luo et al.[23] applicable type, such as stirred autoclaves, fluidized bed reac-
developed a steady-state population balance model to describe tors, plug flow reactors, each of them associated with a given
the PSD of polypropylene made in tubular reactors, consid- reactor RTD, which is assumed to be known a priori. This is not
ering flow type, particle dynamics, particle growth, and attri- an unrealistic assumption as most of industrial reactors have
tion; their model also included and empirical model to describe known (or, at least approximately known) underlying RTDs. The
single-particle heat and mass transfer effects. model also assumes that catalyst particles are spherical and that
All these previous modeling approaches—with the excep- the underlying catalyst particle size distribution (PSD) is known.
tion of the method proposed by Soares and Hamielec[4]—had The objective of our Monte Carlo method is to compute the
in common the use of population balances to describe how polymer PSD at the exit of each reactor in the train, as well
the RTD (ideal, nonideal, or empirical) of a single or a series as the fractions of polymer made on each reactor, taking into
of reactors influenced the PSD of the resulting polymer, the account the reactor RTD and the catalyst PSD, under a variety

Macromol. React. Eng. 2018, 12, 1700031 1700031  (2 of 16) © 2017 WILEY-VCH Verlag GmbH & Co. KGaA, Weinheim
www.advancedsciencenews.com www.mre-journal.de

Figure 1.  Schematic of the proposed Monte Carlo method.

of polymerization conditions. To achieve this goal, the model The polymerization rate, Rp, is given by Equation (1)
generates a population with N catalyst particles, wherein the

diameter of each particle is randomly sampled from the catalyst d[M ]
PSD. The model then follows the trajectory of each catalyst par- Rp = − = kp [M ]∑ [Pr ] = kp [M ][Y0 ]
dt (1)
r =0
ticle as it flows down the CSTR train according to the RTD of 
each reactor in the series.
At the beginning of the polymerization train, the model where [M] is the monomer concentration at the active sites, [Pr]
randomly selects a catalyst particle p with a diameter dp using is the molar concentration of living chains with length r, [Y0] is
the PSD provided for the catalyst. The duration (or time span) the total molar concentration of living chains in the reactor, and
that particle p spends in Reactor 1, tp1, is randomly sam- kp is the propagation rate constant.
pled from the RTD of reactor 1 (RTD1). The particle p will In semibatch reactors commonly used to collect polymeriza-
grow into a polymer particle with diameter dp1, following a tion kinetics data, the polymerization rate can be obtained by
predetermined polymerization kinetics mechanism. After dividing the monomer feed flow rate to the reactor, F, by the
exiting Reactor 1, particle p (which is now a polymer particle) reactor volume, VR
is transferred to Reactor 2; similarly, the residence time in
Reactor 2, tp2, is randomly sampled from the residence time F
Rp = (2)
distribution for Reactor 2 (RTD2), which may, or may not be, VR
the same as for Reactor 1, making the particle p grow to a 
diameter dp2. In similar way, particle p continues travelling For all practical purposes, [Y0] is equal to molar concentra-
down the reactor train, while the program keeps track of how tion of active sites in the reactor, since the fraction of polymer-
much polymer is made on each reactor, and how large the free active sites is negligible due to the high activity of coordi-
particle becomes. nation catalysts. [Y0] can be calculated using the chemical equa-
This algorithm is repeated for every catalyst particle in the tions in Table 1, as shown below.
population of N particles selected for the simulations. When The molar balance for the catalyst precursor added at the
the simulation is finished, a population of N polymer particles beginning of the polymerization in a semibatch reactor is
of different sizes was generated, which can be used for the
several calculations described later in this article. We used the d[C ]
numerical inverse method to randomly sample numbers from = −ka [ Al ][C ] (3)
dt 
a specified probability distribution.[24]
A crucial step in determining how much the polymer parti- Table 1.  Olefin homopolymerization mechanism for the determination
cles will grow as they flow down the reactor train is to define of chain growth.
the polymerization kinetics model for this system. Most olefin
polymerization kinetic curves can be modeled with relatively Description Chemical equation Rate constant
simple expressions involving only site activation, propagation, Activation C + Al →P0 ka
and catalyst deactivation (Table 1). Since chain transfer reac-
Propagation Pr + M →Pr + 1 kp
tions are assumed to have no effect on the polymerization rate,
they do not need to be included in the present treatment. More First-order Deactivation Pr → Cd +Dr kd
complex polymerization kinetic schemes can be easily intro- C, catalyst; Al, cocatalyst, P0, active site, Pr, living chain of length r, M, monomer.
duced in the proposed Monte Carlo approach without altering Cd, deactivated site, Dr, dead chain of length r, ka, activation rate constant, kp, prop-
the main conclusions drawn herein. agation rate constant, kd, deactivation rate constant.

Macromol. React. Eng. 2018, 12, 1700031 1700031  (3 of 16) © 2017 WILEY-VCH Verlag GmbH & Co. KGaA, Weinheim
www.advancedsciencenews.com www.mre-journal.de

where [C] and [Al] are the concentrations of catalyst and cocata- order Markovian models. Likewise, multiple site type catalysts,
lyst, respectively. such as heterogeneous Ziegler–Natta and Phillips catalysts may
Since the cocatalyst is generally present in large excess, the be modeled with similar polymerization kinetics equations.[1]
product ka[Al] = Ka may be assumed to be nearly constant, For the purposes of the following simulations, it suffices
which permits the straightforward solution of Equation (3) with to realize that the expression for Rp of copolymerization,
the initial condition [C](0) = [C0] Equation (8), is essentially the same as for homopolymeriza-
tion, Equation (7), except for the fact that pseudo kinetic con-
[C ] = [C0 ]exp(−K a t )  (4) stants for copolymerization replace the homopolymerization
kinetic constants.[1]
A molar balance for Y0 can then be developed It follows that for a catalyst that follows the general poly­
merization scheme listed in Table 1, the polymer yield Qi,p
d[Y0 ] for a particular polymer particle p produced in the ith reactor
= K a [C ] − kd [Y0 ] = K a [C0 ]exp(−K a t ) − kd [Y0 ] (5)
dt  of the train is given by the integration of Equation (7) or
Equation (8)
Equation (5) is a first-order linear differential equation that
can be solved with the initial condition [Y0](0) = 0 to obtain kp [M ][C o ] ti  − K a  1− d  t 
 k 

Q i,p =
k ∫ ti − 1


1 − e  K a   e − kd t d t

(10)
 k 
− K a  1− d  t 1− d
1− e  Ka  Ka 
[Y0 ] = [C0 ] e − kd t (6)
k where ti−1 is the time particle p exits Reactor i−1 and enters
1− d
Ka  Reactor i, and ti is the time particle p exits Reactor i in the
series. Equation (10) can be solved analytically (Equation (5.8)
Finally, substituting Equation (6) into Equation (2), we obtain an in Soares and McKenna[1]).
expression for olefin polymerization rate with a single-site catalyst[1] In this way, the polymer yield for particle p during the
whole polymerization process is obtained by simply adding the
 k 
−K a 1− d  t poly­mer yield in each of the n reactors
F 1 − e  Ka 
Rp = = kp [M ] [C0 ] e −kdt (7)
k Q ( p) = ∑ Q i ,p
n
VR 1− d (11)
Ka i =1 

The parameters Ka (or ka), kp and kd are estimated by fitting Finally, assuming that the particles are spherical, the diam-
experimental monomer uptake curves with Equation (7). eter Dp of each particle at the exit of the train is obtained as
Equation (7) can be easily modified for binary copolymeri- follows
zations. Assuming that the propagation rate depends only on
monomer type and not on the type of last monomer added to 6Q ( p) m
the polymer chain (Bernoullian model or 0th order Markov Dp = 3 (12)
πρ
model), it can be rewritten as 
where m is the average molar mass of the repeating unit and
ρ is the density of the polymer particle. Equation (12) assumes
  k  
1 − exp  −K a  1 − d  t  that the particle density is constant, which is probably a
  Ka  
R p = (kp,1 [M1 ] + kp,2 [M2 ]) [C0 ] exp ( −kd t ) reasonable assumption for most of the polymerization time,
k except for the few seconds that take place after initial particle
1− d
Ka break up and growth. Since the average residence time in usual
  kd   olefin polymerization reactors with heterogeneous catalysts
1 − exp  −K a  1 −  t 
  Ka   is in the order of tenths of minutes to hours, changes in par-
= kˆp [M ] [C0 ] exp ( −kd t ) (8) ticle density is not an important factor in these simulations. If
k
1− d required, however, it could be easily accounted for by defining
Ka
a ρ = f(t) function.

where [M1] and [M2] are the concentrations of monomers type 1
Table 2.  Polymerization rate constants and concentrations.
and 2, respectively (for instance, ethylene and 1-hexene), [M] is the
total monomer concentration, [M1] and [M2], and kˆp is the pseudo
Parameter Value
propagation rate constant for copolymerization, defined as
Ka 0.1–1 × 10−3 s−1

kˆp = f 1kp,1 + (1 − f 1 )kp,2 (9) kd 1 × 10−4–3 × 10−4 s−1



kp 2 × 105–1.0 × 105 L mol−1 s−1
and f1 is the molar fraction of monomer type 1 in the reaction [C*] 1 × 10−4 mol L−1
medium. A similar approach can be used to treat copolymers that
[M] 1 mol L−1
follow the terminal model (first-order Markov model) or higher

Macromol. React. Eng. 2018, 12, 1700031 1700031  (4 of 16) © 2017 WILEY-VCH Verlag GmbH & Co. KGaA, Weinheim
www.advancedsciencenews.com www.mre-journal.de

Figure 2.  Model catalyst particle size distribution (PSD): Normal distribution with an average of 20 µm and a standard deviation of 3 µm.

3. Results and Discussion catalyst PSD may be replaced with experimental values for
the PSD of an actual heterogeneous catalyst for olefin poly­
3.1. Effect of the Number of CSTRs in Series on Polymer PSD merization without affecting the conclusions of the following
simulations.
Table 2 lists the range of values for polymerization kinetic con- The first simulation assumes that the catalyst activates
stants, catalyst and monomer concentrations that will be used instantaneously (Ka = ∞) and does not deactivate (kd = 0).
in all simulations described below. These values correspond Figure 3 shows that, under these conditions, the cumulative
to typical estimates for olefin polymerization constants with PSD of the polymer made in one ideal CSTR deviates signifi-
Ziegler–Natta and metallocene catalysts. cantly from that obtained when the catalyst PSD is replicated
Figure 2 depicts the PSD of a model catalyst described by perfectly, as would be the case if the polymerization took place
a normal distribution with an average diameter of 20 µm and in a tubular plug flow reactor, or in an autoclave reactor oper-
a standard deviation of 3 µm. This choice is arbitrary; the ated in batch or semibatch mode. The polymer PSD approaches

Figure 3.  Comparison of the cumulative polymer PSD made under perfect replication conditions (plug flow, batch, or semibatch reactors), in one ideal
CSTR with τ = 1 h, and a) 10 ideal CSTRs each with τ = 6 min per reactor, and b) 20 ideal CSTRs each with τ = 3 min per reactor. (Simulation parameters:
kp = 1.5 × 105 L mol−1 s−1, Ka = ∞, kd = 0, [M] = 1.0 mol L−1, [C*] = 1 × 10−4 mol L−1, m = 42 g mol−1, and ρpol = 900 g L−1.)

Macromol. React. Eng. 2018, 12, 1700031 1700031  (5 of 16) © 2017 WILEY-VCH Verlag GmbH & Co. KGaA, Weinheim
www.advancedsciencenews.com www.mre-journal.de

Figure 4.  Comparison of the PSD of polymer made in 10 ideal CSTRs in series with τ = 6 min per reactor (numbered lines) and the PSD of polymer made
in one CSTR (dashed line) with τ = 60 min. (Simulation parameters: kp = 1.5 × 105 L mol−1 s−1, Ka = ∞, kd = 0, [M] = 1.0 mol L−1, [C*] = 1 × 10−4 mol L−1,
m = 42 g mol−1, ρpol = 900 g L−1.)

perfect replication when more CSTRs are added to the series: that activate more slowly will not grow as large as catalyst sites
Figure 3a illustrates the case for 10 CSTRs in series, and that activate instantaneously.
Figure 3b for 20 CSTRs. These results are expected, since the The deactivation rate constant has a similar effect on the
RTD of an infinite series of CSTRs is equal to the RTD of a polymer PSD. Figure 6 shows that, as kd increases, the poly­mer
plug flow reactor, and they confirm that the proposed Monte PSD becomes narrower and shifts toward lower averages
Carlo algorithm is working correctly. because some of the active sites deactivate before the catalysts
Figure 4 illustrates similar trends to those in Figure 3, but particles leave the reactor; consequently, the polymer particles
using the differential PSDs of polymers made in a single cannot grow as large as for catalysts that deactivate more slowly.
ideal CSTR or in 10 ideal CSTRs in series, with the same Differently from the slower activation in Figure 5 that caused
overall average residence time. The polymer particles exiting the polymer PSD to shift toward lower particle sizes, faster cata-
the last reactor in the series (labeled 10/10) have a consid- lyst deactivation has little to no effect in the fraction of particles
erably narrower PSD than those leaving the single CSTR with small diameters, but rather decreases the fraction of parti-
because the CSTR-in-series configuration has a much nar- cles with larger diameters in the population.
rower RTD. From these plots, it follows that since most olefin poly­
merization catalysts deactivate during the polymerization due
to impurities present in the reactor and/or via thermal deacti-
3.2. Effect of Polymerization Kinetic Constants on Polymer PSD vation mechanisms, the polymer PSD that results will be nar-
rower than the predicted using a simpler model that does not
It is illustrative to compare the effect slower catalyst activa- include catalyst activation and deactivation kinetics.
tion rates have on the polymer PSD. Figure 5 shows how A more unique case occurs when the olefin polymerization
decreasing Ka (see insert for polymerization kinetics plot) nar- catalyst is characterized by an induction period. Phillips-type chro-
rows the PSDs of polymers made in a single CSTR, shifting mium oxide catalysts generally behave in this fashion. In this case,
them to lower average particle diameters, since catalyst sites the induction period prevents the catalyst to start making polymer

Figure 5.  Effect of Ka on the PSD of polymers made in a single CSTRs with τ = 60 min. (Simulation parameters: kp = 1.5 × 105 L mol−1 s−1, kd = 0, [M] =
1.0 mol L−1, [C*] = 1 × 10−4 mol L−1, m = 42 g mol−1, ρpol = 900 g L−1.)

Macromol. React. Eng. 2018, 12, 1700031 1700031  (6 of 16) © 2017 WILEY-VCH Verlag GmbH & Co. KGaA, Weinheim
www.advancedsciencenews.com www.mre-journal.de

Figure 6.  Effect of kd on the PSD of polymer made in a single ideal CSTRs with τ = 60 min. (Simulation parameters: kp = 1.5 × 105 L mol−1 s−1, Ka = ∞,
[M] = 1.0 mol L−1, [C*] = 1 × 10−4 mol L−1, m = 42 g mol−1, ρpol = 900 g L−1.)

before a certain elapsed time. In the limiting case, if this time is sufficiently large before exiting the reactor, resulting in bimodal
larger than the residence time of a particular catalyst particle in the polymer PSDs, even though the catalyst PSD was unimodal and
reactor, that catalyst particle will leave the reactor without making followed the normal distribution shown in Figure 2. PSD bimo-
any polymer at all. Figure 7a shows how increasing the catalyst dality, however, starts to decrease and eventually disappears as
induction time affects the PSD of polymer particles made in a more CSTRs are added to the series (Figure 7b) because the resi-
single CSTR. It is interesting to notice how higher induction times dence time of individual particles converge toward the average
increase the fraction of very small particles that could not grow residence time when more CSTRs are added to the reactor train.

Figure 7.  Effect of induction time in the polymer PSD: a) Single ideal CSTR with τ = 120 min and different induction times, b) 1, 2, 4, and 8 CSTRs with
10 min of induction time. (Simulation parameters: kp = 1.2 × 105 L mol s−1, Ka = ∞, kd = 0, [M] = 1.0 mol L−1, [C*] = 1 × 10−4 mol L−1, m = 42 mol L−1,
ρpol = 900 g L−1.)

Macromol. React. Eng. 2018, 12, 1700031 1700031  (7 of 16) © 2017 WILEY-VCH Verlag GmbH & Co. KGaA, Weinheim
www.advancedsciencenews.com www.mre-journal.de

Figure 8.  Effect of Ka in the PSD of polymer particles made with a catalyst with an induction time of 5 min and varying Ka values in a CSTR with
τ = 120 min. (Simulation parameters: kp = 1.5 × 105 L mol−1 s−1, kd = 0, [M] = 1.0 mol L−1, [C*] = 1 × 10−4 mol L−1, m = 42 g mol−1, ρpol = 900 g L−1.)

We assumed instantaneous catalyst activation in the results original catalyst PSD, producing a polymer that has broad but
presented in Figure 7. If in addition to an induction time, the unimodal PSD. It is interesting to compare this result with the
catalyst has a slower activation rate, the profiles change again, obtained in the series of 20 CSTRs. In this case, the catalyst PSD
increasing the fraction of particles that experience no or very bimodality is nearly replicated in the polymer PSD. This obser-
little polymerization before exiting the reactor, as illustrated in vation has important implications for scale-up studies from lab-
Figure 8. oratory-scale reactors. Laboratory polymerization reactors used
to test new polymerization catalysts (or optimize existing ones)
are typically operated in semibatch mode, whereas the catalyst
3.3. Dealing with PSDs and RTDs with Unusual Shapes and liquid comonomers are fed in batch mode, while the gas-
eous monomers (ethylene or propylene) are fed continuously. In
Since the proposed Monte Carlo method adopts a procedure this situation, the PSD of the produced polymer particles would
of random sampling of a particular catalyst PSD, catalysts with resemble the PSD of the catalyst. If this catalyst is then trans-
arbitrary PSDs can be as easily modeled as those with more ferred to a continuous reactor, either in pilot plant or industrial
uniform PSDs. For instance, we will use the bimodal catalyst scales, the PSD of the formed polymer would be drastically
PSD with a large-diameter shoulder shown in Figure 9 in our different from that obtained in the laboratory tests. These dif-
next simulations. ferences could have important consequences in slurry reactors
When this catalyst is allowed to polymerize either in a single (due to differences in suspension viscosity, for instance) and in
CSTR or in 20 CSRTs in series, assuming instantaneous activa- gas phase reactors (due to distinct fluidization properties), and
tion and no deactivation, the model predicts the polymer PSDs also affect the postreactor processing of the reactor fluff. In fact,
shown in Figure 10. For the single CSTR case, the polymer PSD this is always a consideration for any catalyst PSD, as shown in
differs substantially from the catalyst PSD because the expo- the figures above, but this somewhat extreme case of a bimodal
nential reactor RTD evens out the bimodality and tailing of the catalyst PSD helps emphasize this phenomenon.

Figure 9.  Bimodal catalyst PSD with a large-diameter shoulder, having an average diameter μ =23 µm and a standard deviation σ = 7 µm.

Macromol. React. Eng. 2018, 12, 1700031 1700031  (8 of 16) © 2017 WILEY-VCH Verlag GmbH & Co. KGaA, Weinheim
www.advancedsciencenews.com www.mre-journal.de

Figure 10.  Particle size distribution of polymer particles made in a single CSTR (τ = 1 h) or in 20 CSTRs in series (τ = 3 min per reactor). (Simulation
parameters: kp = 1.5 × 105 L mol−1 s−1, Ka = ∞, kd = 0, [M] = 1.0 mol L−1, [C*] = 1 × 10−4 mol L−1, m = 42 g mol−1, ρpol = 900 g L−1.)

Figure 11.  Particle size distribution of polymer particles made in a single CSTR (τ = 1 h) or in 20 CSTRs in series (τ = 3 min per reactor). (Simulation
parameters: kp = 1.5 × 105 L mol−1 s−1, Ka = 1 × 10−3 s−1, kd = 1 × 10−4 s−1, [M] = 1.0 mol L−1, [C*] = 1 × 10−4 mol L−1, m = 42 g mol−1, ρpol = 900 g L−1.)

Figure 12.  Comparison of RTDs for ideal CSTRs and Levenspiel’s Model M. (Simulation parameters: τ = 30 min, e = 0.5, f = 1.5.)

Figure 11 illustrates how the PSD of polymer made with the simulation compares the effect of changing from an ideal
same bimodal catalyst would be affected if the catalyst activa- CSTR RTD to a nonideal RTD expressed by the classic Lev-
tion was not instantaneous and the catalyst activity decreased enspiel Model M shown in Figure 12. Levenspiel’s model
during the polymerization. M represents stagnant zones in a nonideal CSTR of total
Since the proposed Monte Carlo model uses random volume V as a combination of two ideal CSTR zones of
sampling of the RTD to determine how long a given cat- volume eV and (1−e)V, connected in series, with a forward
alyst/polymer particle will remain in the reactor, it can flow rate fν and a backwards flow rate (f−1)ν, where ν is the
handle equally well reactor RTDs with any shape. The next volumetric flow rate exiting the nonideal CSTR. The higher

Macromol. React. Eng. 2018, 12, 1700031 1700031  (9 of 16) © 2017 WILEY-VCH Verlag GmbH & Co. KGaA, Weinheim
www.advancedsciencenews.com www.mre-journal.de

Figure 13.  Effect of nonideal reactor RTD on the PSD of polymers made in a series of four CSTRs. (Simulation parameters: τ = 30 min per reactor, kp =
1.5 × 105 L mol−1 s−1, Ka = 1 × 10−3 s−1, kd = 1 × 10−4 s−1, [M] = 1.0 mol L−1, [C*] = 1 × 10−4 mol L−1, m = 42 g mol−1, ρpol = 900 g L−1.)

Figure 14.  Nonideal bimodal RTD with slight bimodality (τ = 30 min).

Figure 15.  Comparison of PSDs generated with ideal CSTRs and the RTD depicted in Figure 13. Effect of arbitrary reactor RTD on the PSD of polymers
made in a series of four CSTRs. (Simulation parameters: τ = 30 min per reactor, kp = 1.5 × 105 L mol−1 s−1, Ka = 1 × 10−3 s−1, kd = 1 × 10−4 s−1, [M] =
1.0 mol L−1, [C*] = 1 × 10−4 mol L−1, m = 42 g mol−1, ρpol = 900 g L−1.)

the f value, the higher the recirculation effect in the stag- series of CSTR becomes narrower as more reactors are added to
nant zone, and the higher the e value, the higher the size of the system.
the stagnant zone. A more extreme situation in introduced in Figure 14. In
Figure 13 shows that, because of the narrower reactor RTD this case study, the reactor RTD is bimodal, indicating by pass
of Model M, the polymer PSD also becomes narrower than and stagnant zones. The resulting polymer PSD is shown in
the PSD obtained in ideal CSTRs. This difference is more pro- Figure 15, demonstrating that our generic Monte Carlo algo-
nounced for the first reactor in the series, since the RTD of a rithm can model RTDs of arbitrary shapes very easily.

Macromol. React. Eng. 2018, 12, 1700031 1700031  (10 of 16) © 2017 WILEY-VCH Verlag GmbH & Co. KGaA, Weinheim
www.advancedsciencenews.com www.mre-journal.de

Figure 16.  Effect of propagation rate constant during prepolymerization on the PSD of polymer particles made in a single ideal CSTR with τ = 1 h.
(Simulation parameters: kp = 1.5 × 105 L mol−1 s−1, Ka = 1 × 10−3 s−1, kd = 1 × 10−4 s−1, [M] = 1.0 mol L−1, [M]prepoly = 0.1 mol L−1 , [C*] = 1 × 10−4 mol L−1,
m = 42 g mol−1, ρpol = 900 g L−1). Insert Rp profiles refer only to the prepolymerization stage.

3.4. Prepolymerization Effects Table 3. Polymer partition between prepolymerization and main poly­
merization CSTRs for simulations shown in Figure 16.
The proposed Monte Carlo model can also predict the effect of
prepolymerization on the PSD of the prepolymer particles, of Propagation constant % of mass prepolymer
course without including heat and mass transfer phenomena kp,prepoly = kp 3.30%
(in its current version) during the gentler particle fragmenta-
kp,prepoly = ½ kp 1.96%
tion that takes place in prepolymerization reactors.
Figure 16 shows the effect of changing the value of the prop- kp,prepoly = 1/8 kp 0.65%
agation constant during the prepolymerization stage (kp,prepoly),
which could be achieved, for instance, by changing the tem- tially, but among the prepolymerized samples, only slight differ-
perature of the prepolymerization reactor. Introducing a prepo- ences appear among the polymers, with increased prepolymer
lymerization stage substantially reduces the fraction of polymer fractions (Table 4) producing slightly narrower polymer PSDs.
particles with smaller diameters, but changing the value of Figure 18 investigates the prepolymerization effect from
kp,prepoly has only a small effect on the PSD of the polymer made another angle. In this case, we kept the monomer concen-
in the main CSTR reactor. Prepolymerizations with higher tration in the prepolymerization reactor constant, but varied
kp,prepoly values make more prepolymer (Table 3) and narrow the its average residence time. It is interesting to see that dif-
PSD of the whole polymer slightly, but this is a relatively small ferent prepolymerization times have a larger effect on the
effect. polymer PSD than when we change kp,prepoly or monomer
Figure 17 illustrates a similar effect for when the monomer concentration during the prepolymerization stage. In par-
concentration during prepolymerization changes but the value of ticular, longer prepolymerization times lead to narrow pol-
kp,prepoly remains the same. Very similar trends are observed: pol- ymer PSDs (Figure 19), which is a direct reflection of the
ymer PSDs without and with prepolymerization differ substan- narrower RTDs of two CSTRs in series having closer average

Figure 17.  Effect of monomer concentration during prepolymerization on the PSD of polymer particles made in a single ideal CSTR with τ = 1 h.
(Simulation parameters: kp,prepoly = ¼ kp, kp = 1.5 × 105 L mol−1 s−1, Ka = 1 × 10−3 s−1, kd = 1 × 10−4 s−1, [M] = 1.0 mol L−1, [C*] = 1 × 10−4 mol L−1, m =
42 g mol−1, ρpol = 900 g L−1). Insert Rp profiles refer only to prepolymerization stage.

Macromol. React. Eng. 2018, 12, 1700031 1700031  (11 of 16) © 2017 WILEY-VCH Verlag GmbH & Co. KGaA, Weinheim
www.advancedsciencenews.com www.mre-journal.de

Table 4. Polymer partition between prepolymerization and main poly­ more than one type of active site, nonuniformity of conditions
merization CSTRs for simulations shown in Figure 17. during catalyst preparation could, in principle, generate particles
with a distinct balance of active site types. These interparticle
Monomer concentration % of mass prepolymer nonuniformities would cause particles with different diameters
[M]prepoly = [M]/5 1.97% to have distinct apparent polymerization kinetic constants.
[M]prepoly = [M]/10 1.15% In the following simulations, we will assume that Ka, kd, and
[M]prepoly = [M]/100 0.17%
kp depend on the initial volume of the catalyst particles. For the
sake of argument, we will also assume that this dependency
is linear with volume, and that the particles with the smallest
residence times. Table 5 lists the fractions of prepolymer for diameters have the highest Ka, kd, and kp values, that is, the
these three simulations. smallest particles activate, propagate, and deactivate with the
highest frequencies.
The following expression was used to obtain a weighting
3.5. Catalyst Particles with Unequal Polymerization Kinetics factor, w(Vp), that depends linearly on particle volume

In this section, we use our Monte Carlo model to explore the


w (Vp ) = w min +
(Vp,max − Vp ) (w max − w min )
effect of catalyst particles with different polymerization kinetics Vp,max − Vp,min (13)
on polymer PSD. Such a situation could happen if the catalyst 
supporting procedure was not done uniformly, in which case
one could expect catalyst particles with different diameters to where, wmin and wmax are the minimum and maximum
have different catalyst loadings, cocatalyst/catalyst ratios, or weighting factors, respectively, Vp is the particle volume, and
donor/catalyst ratios. Since most supported olefin polymeriza- Vp,min and Vp,max are the minimum and maximum volumes for
tion catalysts, such as Ziegler–Natta and Phillips catalysts, have the catalyst particles fed to the first reactor in the series.

Figure 18.  Effect of prepolymerization time during prepolymerization on the PSD of polymer particles made in a single ideal CSTR with τ = 1 h.
(Simulation parameters: kp,prepoly = ¼ kp, kp = 1.5 × 105 L mol−1 s−1, Ka = 1 × 10−3 s−1, kd = 1 × 10−4 s−1, [M] = 1.0 mol L−1, [M]prepoly = 0.1 mol L−1 [C*] =
1 × 10−4 mol L−1, m = 42 g mol−1, ρpol = 900 g L−1).

Figure 19.  RTDs for simulations shown in Figure 17.

Macromol. React. Eng. 2018, 12, 1700031 1700031  (12 of 16) © 2017 WILEY-VCH Verlag GmbH & Co. KGaA, Weinheim
www.advancedsciencenews.com www.mre-journal.de

Table 5. Polymer partition between prepolymerization and main poly­ constants for all particles. We may assume, for instance, the
merization CSTRs for simulations shown in Figure 18. smaller particles have a higher concentration of active sites
that larger particles, which may indeed happen in the case of
Prepolymerization time % of mass prepolymer supported metallocene catalysts, since smaller particles are
τprepoly = 2.5 min 0.27% less prone to diffusion limitations during the catalyst sup-
τprepoly = 5.0 min 0.55% porting stage. In this case, we may propose the following
τprepoly = 10 min 1.15%
equation to account for this dependency

C *  (Vp ) = w (Vp ) C *  (15)


The weighting factor is used to adjust the polymerization 
kinetic constants for particles of different volumes with the Accordingly, Figure 21 shows that catalysts with nonuniform
generic equation active site distribution will produce polymers with narrow PSD,
since the smaller particles will grow at faster rates than the
k (Vp ) = w (Vp ) k larger ones.
 (14)
Even though it was not our intention to model the effect of
where k stands for Ka, kd, or kp. intraparticle mass transfer limitations in these case studies—
This is clearly an arbitrary factor, used to illustrate how although it would be relatively easy to combine the proposed
the proposed Monte Carlo approach can handle changes of Monte Carlo approach with models such as the polymeric flow
polymerization kinetic constants from particle to particle, and is or multigrain models—similar results would be observed:
not meant to represent any real catalyst system. In the following smaller particles, less prone to mass transfer limitations, would
simulations, we corrected all three kinetic constants by the same grow at faster rates than larger particles. The main difference, of
factor for the sake of simplicity, since we are not trying to simu- course, is that as the polymer particles grow, the relative impor-
late an actual catalyst system. A different set of weights, following tance of intraparticle mass transfer resistances decrease, while
dependencies very likely other than linear, could be developed for in our case the polymerization kinetics remains unchanged as
a particular catalyst if experimental data was collected using cata- the particles grow larger due to polymer formation.
lyst particles with different average diameters.
Figure 20 shows that when the values of the polymerization
constants of the smaller particles are higher than those for the 3.6. Biased Sampling and Fluidized Bed Reactors
larger particles, the PSD of the polymer exiting the reactors will
be narrower, which is a reasonable result, since smaller particles The proposed Monte Carlo approach also can handle a more
will grow faster, and larger particles will grow more slowly as they complex case, often ignored in previous modeling approaches:
flow down the polymerization reactors. In the simulations dis- the effect of particle size on the residence time of particles in
cussed in this section, the model parameters were chosen so that the polymerization reactor. The best example of this situation
the total mass of polymer made in the nonuniform catalyst and in refers to fluidized bed reactors, whereas larger and heavier par-
the comparative uniform catalyst case was the same, to allow for ticles are more likely to exit the polymerization from the bottom
an equitable comparison between these two cases. of the reactor increases.
We reach a similar conclusion if we assume that the con- We implemented a biased sampling method to handle this
centration of active sites depends on the volume of the cata- phenomenon, a weighting factor proportional to the volume of
lyst particle, but that they have the same apparent kinetic the particle and to the volume of the mean of the population is

Figure 20.  Effect of unequal polymerization kinetics on polymer PSD. Solid curves correspond to the PSD of polymer made in four ideal CSTRs in
series (with τ = 15 min per reactor) under equal polymerization kinetics, while the PSD made in the same reactor train with unequal poly­merization
kinetics is indicated with dashed lines. (wmin = 0.26 wmax = 1.26 kp = 1.5 × 105 L mol−1 s−1, Ka = 1 × 10−3 s−1, kd = 1 × 10−4 s−1, [M] = 1.0 mol L−1, [C*] =
1 × 10−4 mol L−1, m = 42 g mol−1, ρpol = 900 g L−1).

Macromol. React. Eng. 2018, 12, 1700031 1700031  (13 of 16) © 2017 WILEY-VCH Verlag GmbH & Co. KGaA, Weinheim
www.advancedsciencenews.com www.mre-journal.de

Figure 21.  Effect of unequal catalyst concentrations on polymer PSD. Solid curves correspond to the PSD of polymer made in four ideal CSTRs in series
(with τ = 15 min per reactor) under the same catalyst concentration, while the PSD made in the same reactor train with different catalyst concentra-
tions is indicated with dashed lines. (wmin = 0.43 wmax = 1.2 kp = 1.5 × 105 L mol−1 s−1, Ka = 1 × 10−1 s−1, kd = 1 × 10−4 s−1, [M] = 1.0 mol L−1, [C*] = 1 ×
10−4 mol L−1, m = 42 g mol−1, ρpol = 900 g L−1).

used to increase or decrease the polymerization of a given par- mean particle volume, tp is polymerization time, wmin and wmax
ticle: the smallest particles will increase their polymerization are parameters determining decrease or increase in residence
time tp by wmaxtp, while the largest particles will decrease them time, respectively. Figure 22 is a graphical representation of this
by wmintp, where wmax ≥ 1 and wmin ≤ 1. The closer the diameter weighting factor.
of the particle is to the mean diameter of the whole population, In the following simulations, we apply the biased sampling
the less affected is its polymerization time. method for all the reactors in the train, that is, Vp,min, Vp,max,
The biased sampling weighting factor is computed as follows and Vp,mean were recalculated for each reactor. In addition, we
assumed that wmin = 1 − w and wmax = 1 + w for a given value w.
 w max − 1  Figure 23 shows that narrower PSD should be expected if
w (Vp ) = 1 + (Vp,mean − Vp )   , if Vp < Vp,mean
Vp,mean − Vp,min  (16) smaller particle are likely to stay longer, while larger particle
 are likely to stay shorter times in the polymerization reactor.
Once again, this result is intuitively expected, but the proposed
 1 − w min  Monte Carlo method allows us to precisely calculate the effect
w (Vp ) = 1 − (Vp − Vp,mean )   , if Vp > Vp,mean (17)
Vp,max − Vp,mean  of this variable in the polymer PSD for any arbitrary biased
 sampling correction.

tp = w (Vp ) tp (18)
 3.7. Computational Time Considerations

where, Vp is the particle volume, Vp,min is the minimum particle One of the main criticisms Monte Carlo simulations face is
volume, Vp,max is the maximum particle volume, Vp,mean is the the relatively high computational time they require to obtain

Figure 22.  Plot of the biased sampling weight factor adopted for the following simulations.

Macromol. React. Eng. 2018, 12, 1700031 1700031  (14 of 16) © 2017 WILEY-VCH Verlag GmbH & Co. KGaA, Weinheim
www.advancedsciencenews.com www.mre-journal.de

Figure 23.  Comparison of the PSD predicted with the biased sampling method and an ideal CSTR RTD. The simulation considered four reactors in
series with a mean residence time of 15 min (1 h, combined). The solid curve corresponds to the PSD obtained for four ideal CSTRs, while the dashed
black and solid gray curves corresponds to the biased sampling method for w = 0.5 and w = 0.2, respectively. (kp = 1.6 × 105 L (mol s)−1, Ka = 1 ×
10−2 s−1, kd = 1 × 10−4 s−1, [M] = 1.0 mol L−1, [C*] = 1 × 10−4 mol L−1, m = 42 g mol−1, ρpol = 900 g L−1).

statistically valid results. The running time of our algorithm where the trajectory of many particles could be tracked simul-
increases linearly with the number of particles (N) and taneously, thus reducing the computation time by orders of
number of reactors (n). However, the number of reactors could magnitude.
be considered as a negligible factor, since it is much smaller
than the number of polymer particles required to represent a
polymerization system. For industrial applications, where typi-
cally there are 1 to 3 reactors in series (or a few more if one 4. Conclusion
decides to model a horizontal gas-phase reactor as a series of 4
to 6 CSTRs), the number of reactors is of little concern from a We have demonstrated that the proposed Monte Carlo model
computational point of view. is extremely versatile, being able to describe reactor systems
Table 6 compares the runtimes for different N and n values, with any arbitrary RTD for catalyst particles also having any
assuming a normal polymer PSD and exponential (ideal CSTR) PSD shape. Since our modeling approach randomly samples
reactor RTDs. All simulations were performed in a laptop PC the catalyst PSD and RTD to determine which particles stay
with Intel Core i3-4030U CPU at 1.90 GHz and 8.00 GB of which time in each of the reactors in series, these distribu-
RAM, which is a rather ordinary computer. The computation tions can take any shape. For the same reason, each catalyst/
time, even for the extreme case of 5 million particles in 60 reac- polymer particle can be considered individually, and be sub-
tors in series (which only has theoretical relevance, since no ject to distinct polymerization kinetics, selective residence
such industrial process exist) is still under 30 min, proving that time characteristics, and prepolymerization regimens. Con-
this approach is not computational onerous, even in limiting sidering that the computation time required for these calcu-
cases. lations is quite short for most cases of industrial relevance,
All sample sizes probed in this investigation, from 1 to 5 where only a few reactors in series need be considered, the
million, generate smooth PSDs that are statistically valid. It proposed approach seems to be ideal to treat practical cases
may be possible to use even smaller sample sizes, but the sta- for industrial applications.
tistical noise level will increase with decreasing sample size. In this article, we focused our attention on RTD effects on
Moving average or similar smoothing algorithms may be used polymer PSD, but these applications only scratch the surface of
to reduce the noise levels in these cases, further decreasing the many uses made possible by this methodology. In the next
the required computation time, but we have not studied this articles in this series, we will apply our Monte Carlo approach
type of correction in the present investigation. to study how reactor RTD may influence polymer powder
In addition, it should be mentioned that the proposed bulk density, particle composition of polymer reactor blends
approach is ideal for implementation in parallel computers, (such as bimodal polyethylenes and heterophasic impact poly-
propylene), and the influence of inter- and
Table 6. Run times of our Monte Carlo method considering different number of catalyst/ intraparticle mass and heat transfer resist-
polymer particles and reactors. ances on polyolefin production.
Number of particles 1 Reactor 6 Reactors 12 Reactors 60 Reactors
1 000 000 8s 45 s 1 min 6 min
2 000 000 17 s 1 min 2 min 13 min Conflict of Interest
5 000 000 36 s 3 min 6 min 27 min The authors declare no conflict of interest.

Macromol. React. Eng. 2018, 12, 1700031 1700031  (15 of 16) © 2017 WILEY-VCH Verlag GmbH & Co. KGaA, Weinheim
www.advancedsciencenews.com www.mre-journal.de

Keywords [10] H. Hatzantonis, C. Kiparissides, Comput. Chem. Eng. 1998, 22,


S127.
particle size distribution, polyethylene, polyolefins, polypropylene, [11] A. Prasetya, L. Liu, J. Lister, F. Watanabe, K. Mitsutani, G. H. Ko,
reactor residence time distribution Chem. Eng. Sci. 1999, 54, 3263.
[12] M. Covezzi, G. Mei, Chem. Eng. Sci. 2001, 56, 4059.
[13] K. Zoellner, K.-H. Reichert, Chem. Eng. Sci. 2001, 56, 4099.
Received: June 7, 2017 [14] H. Yiannoulakis, A. Yiagopoulos, C. Kiparissides, Chem. Eng. Sci.
Revised: August 17, 2017 2001, 56, 917.
Published online: October 9, 2017 [15] T. F. McKenna, J. B. P. Soares, Chem. Eng. Sci. 2001, 56,
3931.
[16] A. Yiagopoulos, H. Yiannoulakis, V. Dimos, C. Kiparissidis, Chem.
[1] J. B. P. Soares, T. McKenna, Polyolefin Reaction Engineering, Wiley- Eng. Sci. 2001, 56, 3979.
VCH, Weinheim 2012. [17] J. J. Zacca, J. A. Debling, Chem. Eng. Sci. 2001, 56, 4029.
[2] J. B. P. Soares, Macromol. React. Eng. 2014, 8, 235. [18] A. G. Mattos Neto, J. C. Pinto, Chem. Eng. Sci. 2001, 56, 4043.
[3] V. Touloupidis, J. B. P. Soares, in Multimodal Polymers with Sup- [19] J. Y. Kim, K. Y. Choi, Chem. Eng. Sci. 2001, 56, 4069.
ported Catalysts: Design and Production (Eds: A. R. Albunia, [20] Y. M. Harshe, R. P. Utikar, V. V. Ranade, Chem. Eng. Sci. 2001, 56,
F. Prades, D. Jeremic), Springer, 2016, in print. 5145.
[4] J. B. P. Soares, A. E. Hamielec, Macromol. Theory Simul. 1995, 4, 1085. [21] C. J. Dittrich, S. M. P. Mutsers, Chem. Eng. Sci. 2007, 62, 5777.
[5] J. J. Zacca, J. A. Debling, W. H. Ray, Chem. Eng. Sci. 1996, 21, 4859. [22] R. Fan, R. O. Fox, M. E. Muhle, in 12th Int. Conf. on Fluidization-New
[6] J. A. Debling, J. J. Zacca, W. H. Ray, Chem. Eng. Sci. 1997, 52, 1969. Horizons in Fluidization Engineering, Engineering Conferences Inter-
[7] J. J. Zacca, J. A. Debling, W. H. Ray, Chem. Eng. Sci. 1997, 52, 1941. national ( ECI ), New York, NY 2007, p. 993.
[8] D. Y. Khang, H. H. Lee, Chem. Eng. Sci. 1997, 52, 421. [23] Z. H. Luo, P. L. Su, X. Z. You, D. P. Shi, J. C. Wu, Chem. Eng. J.
[9] H. Hatzantonis, A. Goulas, C. Kiparissides, Chem. Eng. Sci. 1998, 2009, 146, 466.
53, 3251. [24] R. Saucier, Rep. ARL-TR-2168, 2000, 2.

Macromol. React. Eng. 2018, 12, 1700031 1700031  (16 of 16) © 2017 WILEY-VCH Verlag GmbH & Co. KGaA, Weinheim

You might also like