8 Macromol. Symp. 2008, 271, 814 DOI: 10.1002/masy.

200851102

Modeling and Simulation of Liquid Phase Propylene

Polymerizations in Industrial Loop Reactors
Eneida A. de Lucca,1 Rubens Maciel Filho,1 Pramo A. Melo,2 Jose Carlos Pinto*2

Summary: Liquid phase tubular loop polymerization reactors are widely used in the
polyolefin industries because of their capabilities to promote high mixing of
reactants in the reaction vessel and to allow for high heat transfer rates with the
cooling jacket due to their high aspect ratio. Previous works on this subject focused
on the modeling of the polymerization system, but only a few compared their results
with real industrial data. A literature review about the propylene production in loop
reactors shows that the validation of a distributed model with actual industrial data
is yet to be presented. A distributed mathematical model is presented for industrial
liquid phase loop polypropylene reactors and validated with actual industrial data for
the first time. The model is able to represent the dynamic trajectories of production
rates, MFI and XS values during grade transitions within the experimental accuracy.
The model indicates that the polymer quality can change significantly along the
reactor train and that manipulation of feed flow rates can be successfully used for
production of more homogeneous polymer products.

Keywords: liquid-pool; modeling; polymerization; propylene; Ziegler-Natta

Introduction but, as in previous works, validation with

industrial data was not presented. Recently,
Liquid phase tubular loop polymerization Reginato, Zacca and Secchi[5] used a
reactors are widely used in the polyolefin nonideal continuous stirred tank model to
industries because of their capabilities to describe this process and compared their
promote high mixing of reactants in the simulation results with actual plant data.
reaction vessel and to allow for high heat Production rates, slurry densities and melt
transfer rates with the cooling jacket due to flow indexes were used to validate the
their high aspect ratio. The first mathematical model developed.
models developed for the polymerization of This brief review shows that the valida-
propylene in loop reactors appeared in the tion of a distributed model for the propy-
1970s[1,2] and used continuous stirred tanks to lene production in loop reactors with
describe the mixing patterns in these industrial data is yet to be presented. A
systems. This approach was also used later distributed model may allow for a full
by Ferrero and Chioveta.[3] Zacca and analysis of the process behavior because it
Ray[4] were the first to propose a distrib- takes into account the distributed nature of
uted dynamic model to describe the poly- the system, the possibility of using different
merization of propylene in this process levels of macromixing to describe the
mixing patterns inside the reactor and the
1
Faculdade de Engenharia Qumica, Universidade
possibility of incorporating peculiar sedi-
Estadual de Campinas, Cidade Universitaria Zeferino mentation, mass transfer and heat transfer
Vaz, CP: 6066, Campinas, 13083-970, SP, Brazil aspects into the model formulation.[4]
2
Programa de Engenharia Qumica/COPPE, Universi- This paper presents a distributed
dade Federal do Rio de Janeiro, Cidade Universitaria,
CP: 68502, Rio de Janeiro, 21945-970 RJ, Brazil dynamic mathematical model for the poly-
E-mail: pinto@peq.coppe.ufrj.br merization of propylene in tubular loop

Copyright 2008 WILEY-VCH Verlag GmbH & Co. KGaA, Weinheim

Macromol. Symp. 2008, 271, 814 9

reactors. The model aims at describing the overall process dynamics. In addition to
the main process variables related to the that, the external thermal capacitance of
quality control of the industrial polymer- both the reactor tubular sections and the
ization process, such as the production rate, recycling pump were also taken into
the melt flow index (MFI) and the amount account. These modifications were moti-
of xylene solubles (XS) in the final polymer vated by previous experimental and mod-
resins. Hydrogen is used to control the eling results obtained in a lab-scale loop
average molecular weights (and, conse- reactor[6,7] that emphasized the need of
quently, the MFI) of the polymer resin. The adding the recycling pump into the model in
model implementation allows for the simu- order to describe appropriately the macro-
lation of a train of loop reactors with mixing behavior of the reactor as well as its
multiple feeds. It is shown for the first time thermal dynamics.
that a distributed model may be adequately As depicted in Figure 1, the loop reactor
tuned to describe grade transitions during considered here is composed of three
industrial operation of polypropylene loop distinct reaction sections, namely, the
reactors. It is also shown that MFI and XS tubular section 1, that connects the inlet
data may be properly simulated with the feed point to the exit stream and the
developed model. Therefore, the model can recycling pump; the recycling pump, that
be used for interpretation of process recycles most part of the polymer slurry and
operation data and for development of is connected to the tubular section 2; and
optimum operation strategies. the tubular section 2, that connects the exit
of the recycling pump to the inlet feed
point. The tubular sections were described
Mathematical Modeling according to the axial dispersion model and
the recycling pump is considered to be
As discussed in the previous section, several perfectly mixed. Other model assumptions
models have been proposed to describe the may be found elsewhere.[8]
polymerization of propylene in tubular Thus, according to the axial dispersion
loop reactors. The loop polymerization model, a molar balance for a given species
reactor investigated here consists of a pair in a tubular section of the loop reactor
of continuous tubular reactors whose ends results in:
are connected, at one side, to the inlet feed
and, at the other side, to the exit stream and @Cji z; t @Cji z; t
to a recycling pump, as presented in viz
@t @z
Figure 1.[6] It is important to emphasize @2 Cji z; t
that the recycling pump is not considered Dm
Rij z; t; (1)
@z2
here as a zero-capacitance vessel but,
instead, as a piece of equipment whose where Cji is the molar concentration of
behavior may play a fundamental part on species j (monomer, catalyst, cocatalysts,

Figure 1.
Loop reactor diagram.[6,7]

Copyright 2008 WILEY-VCH Verlag GmbH & Co. KGaA, Weinheim www.ms-journal.de
10 Macromol. Symp. 2008, 271, 814

hydrogen and moments of the polymer The energy balance for the recycling
molar mass distribution) in tubular section pump is given by
i, vz is the average flow velocity, Dm is the
h i dT p t
axial dispersion coefficient (assumed here
Vp rCp VrCP p
to be the same for all species present in the dt

tubular sections, which is equivalent to 
rCp Q2 T t exit
T p t
1
assuming that dispersion is determined by
the internal flow patterns), Rj is the overall DHr Vp Ri t
rate of consumption of species j, z and t are Up Tc
T p t: (4)
the space and time coordinates, respec-
tively. For Equation (1)(2), well-known
The energy balance is obtained by Danckwerts boundary conditions have
neglecting energy changes due to expansion been applied. Equation (1)(4) along with
work, viscous flow, external fields, radiation the boundary conditions constitute a set of
and heat of mixing. The energy balance for coupled, non linear parabolic partial and
each tubular section can be given by[10]: ordinary differential equations. The axial
coordinates of the tubular sections were

  @T j t; z
rCp rCp r discretized according to a polynomial
@t
approximation on finite elements. Conver-
j
@T t; z gence tests as well as other details regarding
rCp viz
@zj the numerical solution of the model
@2 T j t; z equations may be found elsewhere.[6,11]
Dt rCp DHr RM t; z One important point regarding the
@z2j
reactor model developed in this work is
4Ur that, in addition to the usual kinetic steps of
Tc
T j t; z; (2)
d site activation, chain initiation, propaga-
where T j is the reactor temperature in tion, transfer reactions, and deactivation
tubular section j. The term rCp in Equation (cf. Table 1 for the kinetic scheme and
(2) stands for the thermal capacitance of the Table 2 for the kinetic parameters), the
polymer solution flowing through

 tubular model also incorporates equations to
section j, whereas the term rCp r repre- account for the dynamic evolution of the
sents the overall thermal capacitance of the melt flow index and amount of solubles in
reactor externals (such as tube walls, xylene in the polymer resins produced.
fittings, valves, recycling pump etc). Dt is Models for these properties are taken from
the thermal dispersion coefficient, DHr is Latado et al.[9] The method of moments is
the reaction enthalpy and Ur is the reactor applied to population balances that
overall heat transfer coefficient. describe live and dead polymer chains in
The recycling pump is described as an order to allow for calculation of average
ideal continuous stirred tank reactor. The molecular weights.
mass balance for component j is given
below.
Table 1.
 Kinetic mechanism for the polymerization reaction.
dCjp t 
Vp Q2 Cj1 t Site Activation: Chain Transfer:
dt exit (3)
kA kTH

Cjp t
Vp Rj t; Sp A
! P0 Pi H2
! P0 Qi
Chain Initiation: kTM
Pi M
! P1 Qi
where Cjp is the molar concentration of kP kTS
P0 M
! P1 Pi
! P0 Qi
component j in the recycling pump, Vp is Chain Propagation: Site Deactivation:
the internal volume of the recycling pump, kP kD
Pi M
! Pi1 Pi
! SD Qi
and Q2 is the volumetric flow rate in tubular kD
P0
! SD
section 2.

Copyright 2008 WILEY-VCH Verlag GmbH & Co. KGaA, Weinheim www.ms-journal.de
Macromol. Symp. 2008, 271, 814 11
Table 2. In the industrial practice, grade transi-
Kinetic parameters for the liquid-phase polymeri-
zation of propylene with a fourth-generation Ziegler- tion zones should not be too long, as off-
Natta catalyst. spec polymer resins are usually produced
during these transitions. In order to mini-
Parameter Reference
mize the impact of grade transitions on the
5
12;000=Rg T 3 [12]
kA 7:04  10 e cm /gmol/s process productivity, it is usually sought to
kP 6:3  1011 e
10;000=Rg T cm3/gmol/s [12]
depart from a low MFI operation (low
kTH 2:22  1013 e
14;000=Rg T cm3/gmol/s [12]
hydrogen feed rates) and reach subsequent
kTM 2:76  1010 e
14;000=Rg T cm3/gmol/s [12]
kTS 1:72  106 e14;000=Rg T s-1 [4] higher intermediate MFI plateaus (by
kD 7:93  103 e12;000=Rg T s-1 [4] increasing the hydrogen feed rates) until
the attainment of the operation conditions
required to produce the resin with the
highest MFI. Figure 2 shows that model
predictions can be regarded as very good, as
Results and Discussion deviations between predicted and mea-
sured values are smaller than 10%, which
Model Validation is the inherent experimental error. Similar
Case studies presented herein report data results may be obtained for other process
and simulations concerning a real industrial conditions.
liquid phase propylene polymerization For the case of the amount of solubles in
process using loop reactors. Due to con- xylene, XS, Figure 3 illustrates typical
fidentiality agreements, process operational responses during MFI grade transitions.
parameters are not fully presented. Figure 2 XS values are influenced mostly by the
illustrates typical grade transitions at plant electron-donor concentrations (in this par-
site, during production of grades of very low ticular case, described as the ratio between
to very high molar masses. Grade transition TEA, tri-ethyl aluminum cocatalyst, and
was performed through modification of the electron-donor). The higher the concentra-
hydrogen feed concentrations according to tion of electron-donor, the lower is the
the industrial practice. As observed, the expected XS level in the produced polymer.
model is able to describe appropriately the Regarding Figure 3, one may see that the
operation in the whole range of MFI values model is capable of adequately reproducing
(within typical experimental errors), the XS amounts in the polymer for the
including the characteristic grade transition same operation conditions reported in
time, for multiple grade transitions per- Figure 2A. Again, model predictions can
formed sequentially. be regarded as very good, as deviations

Figure 2.
Measured and simulated dynamic MFI profile during Figure 3.
grade transitions. Measured and simulated dynamic XS profiles.

Copyright 2008 WILEY-VCH Verlag GmbH & Co. KGaA, Weinheim www.ms-journal.de
12 Macromol. Symp. 2008, 271, 814

Figure 4. Figure 5.
Validation of productivity data during the production Simulated monomer conversion dynamic profiles for
of polypropylene. a train of loop reactors.

between predicted and observed values are increased to reduce the residence times and
within the experimental accuracy (0.3%). safely allow for significant increase (about
Similar results can be obtained in all other 35%) of the polymer productivity.[13]
cases. Other variables are also of practical
Figure 4 presents the model validation interest. Simulated MFIs for the polymer
for the variable polypropylene production produced in the reactor train are presented
and considers the operation of two full- in Figure 6. As shown, the MFI of the
scale loop reactors in series. Operation polymer material can change quite con-
conditions are similar to the ones reported siderably along the train when the mono-
in Figure 2 for low MFI during reactor start- mer/hydrogen feed ratios are kept constant
up. It is possible to observe that model for all reactors. This can be explained in
simulations provide a very fair prediction of terms of the very distinct reaction condi-
the industrial process operation for both tions inside each of the reactors, due to
reactors. Low-amplitude measured data varying polymer concentrations (which
oscillations are due to the inherent fluctua- increase), catalyst concentrations (which
tions of the process instrumentation. decrease, due to catalyst decay) and
monomer concentrations (which decrease,
Process Simulation due to monomer consumption), which lead
The validated process model may now be to distinct hydrogen concentrations and
used to simulate and investigate plant
operation in order to increase process
performance and polymer productivity.
One possibility of increasing the polymer
productivity is by adding another large-
scale loop to the reactor train. Figure 5
explores this idea. Operation data refers to
those used in Figure 2 during production of
resins with low MFI. One may observe that
conversions levels up to near 60% may be
obtained, which can certainly be regarded
as a significant improvement, considering
the actual available process technology. If
one assumes that the maximum allowed Figure 6.
polymer concentration is equal to 55%, MFI dynamic profiles along the reactor train during
then the propylene feed flow rates can be grade transition.

Copyright 2008 WILEY-VCH Verlag GmbH & Co. KGaA, Weinheim www.ms-journal.de
Macromol. Symp. 2008, 271, 814 13

Figure 7.
Hydrogen dynamic profiles along the reactor train
Figure 8.
during grade transition.
MFI dynamic profiles along the reactor train when a
multiple hydrogen feed is applied.

rates of chain transfer inside each of the of the other two full-scale loop reactors
reactors, as shown in Figure 7. Therefore, it (although retaining the same length/dia-
sounds reasonable to propose the use of meter ratio). Monomer, catalyst, cocata-
distinct monomer/hydrogen feed ratios lysts and hydrogen are fed to the baby-loop,
(and also of hydrogen compositions) for but only hydrogen is also provided to the
each of the reactors, in order to allow for full-scale loop reactors. One may observe
production of custom-designed polymer that very different polymer grades may be
materials. obtained in each reactor, thus reinforcing
In order to explore this possibility, a the fact that hydrogen is a crucial variable
more realistic process simulation may be to determine the final product quality at
performed in such a way that any reactor in plant site. For this reason, hydrogen feed
the train may receive fresh monomer, rates should be manipulated independently
catalyst, cocatalysts, and hydrogen. Usually, at each reaction stage and properly
however, multiple feeds are restricted only designed with the help of validated simula-
to monomer and hydrogen, in order to tion models.
modify both process productivity and the
polymer grade. Besides, it is also known
that, in a typical industrial site, the catalyst is
firstly fed to a small size pre-polymerization
vessel known as baby-loop. In this first
reactor, catalyst particles are encapsulated
by a growing polymer mass at mild reaction
conditions. The control of the morphology
of the polymer particles is basically per-
formed in this step.[5] Pre-polymerization is
used to produce spherical polymer particles
in an early reaction stage, avoiding uncon-
trolled particle fragmentation in the large
scale reactors.
Figure 8 and 9 present MFI and hydro- Figure 9.
gen concentration profiles, respectively, for Hydrogen concentration dynamic profiles along the
the train of loop reactor composed by a reactor train when a multiple hydrogen feed is
baby-loop, whose total length is 1/58 of that applied.

Copyright 2008 WILEY-VCH Verlag GmbH & Co. KGaA, Weinheim www.ms-journal.de
14 Macromol. Symp. 2008, 271, 814

Conclusions Subscripts characters

1 related to tubular section 1
A distributed mathematical model is pre- p related to the recycling pump
sented for industrial liquid phase loop
polypropylene reactors and validated with Acknowledgements: The authors wish to thank
actual industrial data for the first time. The CNPq and FAPERJ for providing financial
model is able to represent the dynamic support and scholarships.
trajectories of production rates, MFI and
XS values during grade transitions within [1] B. A. Uvarov, V. I. Tsevetkova, Development of a
the experimental accuracy. The model Mathematical Model for Controlling the Yield of
indicates that the polymer quality can Propylene Polymerization in a Loop Reactor. Polim.
change significantly along the reactor train Protsessy Appar. 1974, 165168.
and that manipulation of feed flow rates can [2] D. M. Lepski, A. M. Inkov, Mathematical Modeling
be successfully used for production of more of Polymerization of Propylene in a Loop Reactor. Sb.
Tr. Vses. Obedin. Neftekhim 1977, 13, 3445.
homogeneous polymer products.
[3] M. A. Ferrero, M. G. Chiovetta, Preliminary Design
of a Loop for Bulk Propylene Polymerization. Polymer
Plastic Technology and Engineering 1990, 29, 263287.
[4] J. J. Zacca, W. H. Ray, Modelling of the Liquid Phase
Notation Polymerization of Olefins in Loop Reactors. Chemical
Engineering Science 1993, 48, 37433765.
[5] A. S. Reginato, J. J. Zacca, A. R. Secchi, Modeling
C molar concentration (gmol/cm3)
and Simulation of Propylene Polymerization in Non-
Cp heat capacity (cal/g/K) ideal Loop Reactors. AICHE Journal 2003, 49, 2642
D Dipersion coefficient (cm2/s) 2654.
d reactor tubes internal diameter [6] P. A. Melo, J. C. Pinto, E. C. Biscaia, Jr., Character-
Q volumetric flow rate (cm3/s) ization of Residence Time Distributions in Loop
R reaction rate (gmol/cm3/s) Reactors. Chemical Engineering Science 2001, 56,
t time (s) 27032713.
[7] P. A. Melo, E. C. Biscaia, Jr., J. C. Pinto, Thermal
T reactor temperature (K)
Effects in Loop Polymerization Reactors. Chemical
Ur reactor overall heat transfer coeffi- Engineering Science 2001, 56, 67936800.
cient (cal/cm2/s/K) [8] P. A. Melo, E. C. Biscaia, Jr., J. C. Pinto, The
Up recycling pump overall heat transfer Bifurcation Behavior of Continuous Free-Radical
coefficient (cal/s/K) Solution Loop Polymerization Reactors. Chemical
vz axial flow velocity (cm/s) Engineering Science 2003, 58, 28052821.
Vp volume of the recycling pump (cm3) [9] A. Latado, M. Embirucu, A. G. Mattos Neto, J. C.
Pinto, Modeling of End-Use Properties of Poly(propy-
z reactor tubular section length (cm)
lene/ethylene) Resins. Polymer Testing 2001, 20, 419
439.
[10] G. Froment, K. B. Bischoff, Chemical Reactor
Analysis and Design, 2nd ed., John Wiley and Sons,
New York 1990.
[11] E. A. Lucca, Modeling and Simulation of Industrial
Greek characters
Liquid Phase Loop Reactor for Polypropylene Pro-
DHr reaction enthalpy (cal/gmole) duction (in Portuguese), M.Sc. dissertation, UNICAMP,
r density (g/cm3) Campinas Sao Paulo, Brazil, 2007.
[12] J. J. Zacca, Distributed Parameter Modelling of the
Polymerization of Olefins in Chemical Reactors, Ph.D.
thesis, University of Wisconsin, Madison Wisconsin,
U.S.A. 1995.
[13] A. G. Oliveira, P. C. Muniz, P. A. Melo, J. C. Pinto,
Superscripts characters Steady-State Behavior of Slurry and Bulk Propylene
1 related to tubular section 1 Polymerization. Polymer Reaction Engineering 2003, 11,
2 related to tubular section 2 159180.

Copyright 2008 WILEY-VCH Verlag GmbH & Co. KGaA, Weinheim www.ms-journal.de