Download as pdf or txt
Download as pdf or txt
You are on page 1of 28

Pergamon Chemical Engineerin9 Science, Vol. 51, No. 21, pp.

4859 4886, 1996


Copyright 1996 Elsevier Science Ltd
Printed in Great Britain. All rights reserved
P I I : S0009-2509(96)00258-8 0009 2509/96 $15.00 + 0.00

REACTOR RESIDENCE TIME DISTRIBUTION EFFECTS ON


THE MULTISTAGE POLYMERIZATION OF OLEFINS--I.
BASIC PRINCIPLES AND ILLUSTRATIVE EXAMPLES,
POLYPROPYLENE

JORGE J. ZACCA, JON A. DEBLING and W. H. RAY*


Department of Chemical Engineering, University of Wisconsin-Madison, Madison, WI, U.S.A.

(First received 7 August 1995; revised manuscript received and accepted 27 February 1996)

Abstract--Manypolyolefinsare produced by catalyzed olefinpolymerization over solid catalysts. A variety


of processes are currently in use including loop, continuous-stirred tank, horizontal stirred bed and
fluidized-bedtechnologies. In each of these continuous processes, the reactor residence time distribution is
unique and affects the resulting distribution of polymer properties. In this paper we present a population
balance modeling approach for modeling multistage olefin polymerization processes using catalyst resi-
dence time as the main coordinate. The catalyst may possess a broad particle size distribution and contain
multiple active sites with a comprehensive kinetic scheme involving multiple monomers, Diffusion limita-
tions during reaction may also be a consideration. A variety of reactor systems can be modeled each with
unique residence time distributions and having different internal and external residence time distribution.
In addition, the modeling allows the consideration of particle size selection effects within the process. These
effects are implemented in a model of a fluidized-bedreactor. The population balance model is illustrated
with examples of propylene polymerization in the four major commercial processes listed above, with a high
activity TiCL~/MgC12 catalyst. Comparisons of total production capacity, catalyst yield distribution,
polymer particle size and porosity distribution are made. It is found that significant differences exist
between the differentindustrial processes. It is also found that the performance of the fluidized-bedreactor
is highly sensitive to fluidization conditions, particle size selection effects and the catalyst particle size
distribution. Differences in product quality are discussed and issues such as fines generation and particle
sticking are addressed. Copyright 1996 Elsevier Science Ltd

Keywords: Population balances, olefin polymerization, residence time distribution, propylene,


Ziegler-Natta, fluidized bed, stirred bed, loop, multistage polymerization.

I. INTRODUCTION complex and may involve multiple sites on the cata-


One of the most important methods for the produc- lyst surface, reactions with cocatalyst and donor,
tion of polyolefin resins is by solid-catalyzed olefin poisoning and site transformation effects. In addition,
polymerization. In this process, small catalyst par- the polymerization may involve chain transfer
ticles (10-200 pm diameter) are continuously fed into agents such as hydrogen for molecular weight control
a reactor that is operated at controlled temperature, and several monomers may be simultaneously poly-
pressure and composition. The catalyst particles merized. Further complicating this process is the fact
themselves are porous and composed of small sub that diffusion resistances may be present during the
fragments containing active metal. As the monomer growth of the polymer particle.
diffuses through the porous catalyst pellet, it polymer- Traditionally, there have been two approaches
izes via a coordination-insertion mechanism on taken to model these systems depending on the scope
active sites present on the catalyst surface. As the of interest. When the primary interest is in under-
polymerization proceeds, the catalyst fragments are standing the influence of diffusion on reaction rate,
dispersed and the particle expands in size, ultimately composition and molecular weight distribution, de-
growing into a final polymer particle ranging between tailed reaction diffusion models for individual cata-
100 and 5000 pm diameter [see Hutchinson (1990) for lyst-polymer particles are studied. Several of these
more details]. models have been published in the literature (Debling
Industrially, many different catalysts have been and Ray, 1995; Ferrero and Chiovetta, 1987; Floyd
and are currently used for olefin polymerization. and Ray, 1986; Galvan and Tirrel, 1986; Hutchinson
The most significant groups include supported or et al., 1992; Sarkar and Gupta, 1991). Single-particle
unsupported Ziegler-Natta, chromium oxide and models can provide some useful information about
recently Metallocene-based systems. The kinetics a continuous process by running the model repeatedly
that control the polymerization process are quite to account for reactor residence time distribution,
reactors in series and catalyst particle size distribu-
tion. However, this can become time consuming and
* Corresponding author. computationally intensive.

4859
4860 J. J. ZACCAet al.
When the main interest is in studying the macro- state of the reactor media and in the mechanical
scale properties of the process such as dynamics of the operation of the unit. The choice of process is deter-
reactor unit, control, scaleup of pilot plant data, mined by economics, feedstock availability, catalyst
grade transitions and average polymer properties, and desired range of products to be produced. Signifi-
a macroscopic dynamic model of the reactor vessel is cant improvements in catalyst design over the past
usually studied. Various models have been developed years have led to improvements in process design and
for several commercially important reactors such as simplification. Modern supported catalysts are able to
the loop (Zacca and Ray, 1993), continuous-stirred produce polymer with high yields and stereospecific-
bed (Choi and Ray, 1988; Demel, 1989), fluidized bed ity (for polypropylene) such that removal of catalyst
(Choi and Ray, 1985a; Lagemann, 1989; McAuley, residue and atactic polymer from the resin is no lon-
1991; Talbot, 1990) and continuous-stirred tank (Ar- ger required. Control of polymer particle morphology
riola, 1989). However, the main limitation of most of via catalyst morphology has also allowed the produc-
these models is that they can only calculate average tion of high bulk density, spherical particles directly in
properties of the polymer product leaving the reactor, the reactor. An overview of the main methods for the
such as molecular weight, composition and yield and production of polyolefin resins is shown in Fig. 1.
do not account for the distributed nature of the prob- Comprehensive reviews have also been published
lem as a result of reactor residence time distribution (Choi and Ray, 1985b; 1985c).
effects.
To capture rigorously these effects, population bal- 2.1. Slurry-phase processes
ances on the growing particles are necessary. Popula- Slurry-phase processes may involve either an inert
tion balances have been used in chemical engineering diluent such as iso-butane or heptane, or condensed
to study particulate processes (Hulburt and Katz, monomer such as propylene. In either case the cata-
1964; Ramkrishna, 1985) and are well known in the lyst particles are suspended and well mixed in the
field of emulsion polymerization where computing the liquid medium. Monomer concentrations are high
latex particle size distribution is important (Min, and the liquid provides good removal of the heat of
1976; Prindle, 1989; Rawlings, 1985). However, the polymerization from the polymer particles. The two
application of population balances in catalyzed olefin main reactors for slurry-phase olefin polymerization
polymerization is not very common. Caracotsios are the loop reactor and continuous-stirred tank.
(1989) has used the concept of residence time distribu- Slurry-phase processes are very attractive for high-
tion to study average yield and degree of polymeriz- crystalline homopolymer products such as poly-
ation for propylene polymerization in a horizontal propylene and polyethylene but have limited applica-
stirred bed reactor at steady state. Choi et al. (1994) tions for low-density, amorphous copolymers. For
have applied population concepts to a continuous- these products, the liquid phase can dissolve the
gas-phase reactor at steady state for ethylene amorphous copolymer making the resin tacky in na-
copolymerization and studied the effects of catalyst ture and hard to process. Also the solubilization of
deactivation, catalyst particle size distribution and amorphous material may cause operational problems
catalyst feed rate on polymer particle size distribution, such as increases in slurry viscosity and reduction of
comonomer content and average molecular weight. the reactor heat transfer coefficient.
Polymer particle radius was chosen as the internal
coordinate. Kim (1995) has used population balances 2.2. Gas-phase processes
to simulate the unsteady behavior of a CSTR for Gas-phase processes for olefin polymerization in-
propylene polymerization with two active sites. In clude the vertical stirred bed reactor (VSBR), horizon-
Kim's work, polymer volume was used as the internal tal stirred bed reactor (HSBR) and fluidized-bed reac-
coordinate and a quasi-steady-state assumption was tor (FBR). They provide an attractive environment for
applied to the total number of active sites. olefin polymerization with no liquid inventory. To be
In this paper we present a population balance competitive with liquid-phase processes, gas-phase re-
modeling approach for modeling multistage olefin actors must operate close to the dew point of the
polymerization processes using catalyst residence monomer mixture to obtain high monomer concen-
time as the main coordinate. The catalyst is allowed to trations and high yields. Today a full range of polymer
possess a broad particle size distribution and contain products ranging from polyethylene and polypropy-
up to two active sites with a simple kinetic scheme lene (homopolymer and impact polypropylene) to
involving up to two monomers and including diffu- random copolymer products are produced in gas-
sion. A variety of reactors can be studied each with phase processes.
unique residence time distribution. The internal resi- In both the ttuidized-bed and stirred-bed reactors,
dence time distribution can also be different from the the catalyst morphology must be closely controlled to
external distribution. prevent particle overheating and melting due to the
poor heat transfer characteristics of the gas. Over-
heating is most serious for large catalysts with high
2. PROCESSES F O R O L E F I N P O L Y M E R I Z A T I O N activity, precisely the desired catalyst characteristics
Many processes have been developed for the pol- for maximizing polymer yield and minimizing fines
ymerization of olefins. They differ in both the physical production.
Multistage polymerization of olefins--I 4861
I II
[ Particulate Processes ~

I I
SlurryPhase Gas Phase

I I I I
LiquidDiluent LiquidMonomer Fluidized Vertical Horizontal
bed stirred bed stirred bed

i
I I I I
Loop Stirred Loop Stirred
Tank Tank

not considered in this paper

I I
Tube Stirred
Tank

Fig. l. Classification of polyolefin processes.

2.3. Solution processes is the subject of a future paper in this series (Debling
Solution processes include free radical initiated poly- et al., to appear); however, here we consider only
merization under supercritical conditions and cata- the homopolymerization stage for purposes of
lytic processes where the solvent temperature is high illustration.
enough to dissolve the polymer material. Although
these processes are not particulate in nature, the con- 3. BASICMODELING EQUATIONS
cepts of population balances may also be applied to
fluid elements of the reactor volume. Although not 3.1. Probability distributions
explicitly considered in this paper, the equations de- Since population balance models inherently involve
veloped for particulate processes may be easily ex- statistical distributions, it is prudent here to define the
tended to the solution reactors. probability density functions (also known as 'distribu-
tions') used in this paper. The distributions presented
2.4. Illustrative examples below may be defined on a variety of distinct bases
To illustrate the concepts discussed in this paper, depending on the property of interest. In this work we
four major processes for propylene homopolymeriz- focus on three types of bases: (1) mass of the catalyst,
ation are used as comparative examples. These in- (2) number of particles, and (3) weight of the polymer.
clude the loop reactor, vertical and horizontal We denote the type of distribution by a superscript.
stirred bed reactors and fluidized-bed reactor. A superscript 'w' denotes weight of catalyst basis, 'n'
Typical operating conditions are shown in Table denotes number of particles basis and 'p' denotes
1 and process schematics are given in Figs 2-5. weight of polymer basis. The distributions may also
Note that the processes shown also include the be defined at the reactor inlet, reactor outlet or
possibility of a second copolymerization stage used inside the reactor. We denote these by another super-
for making impact polypropylene. The second stage script. A superscript 'R' denotes inside the reactor,
4862 J. J. ZACCAet al.
Table 1. Processes for propylene homopolymerization
BASF* Amoco/Chiso* Union Carbide t MontelP
Process Novolen - UNIPOL SPHERIPOL
Medium Gas Gas Gas Slurry-liquid
monomer
Reactor VSBR HSBR FBR Two loops in series
Size 25-50 m 3 n/a n/a 45 m 3 each loop
Catalysts used TiCI3 TiCI4/MgCI2 TiCI4/MgCI2 TiCI4/MgCI2
TiC14/MgClz
Temperature (C) 50-100 57-93 n/a 70
Pressure (atm) 20-35 19 23 n/a 35-40
Reported 10,000-20,000 Up to 30,000 Up to 40,000 Up to 50,000
yield (g/g cat)
*Jaggard (1989), Ross and Bowles (1985), Schill and Buchner (1988).
*Caracotsios (1992), Brockmeier and Koizumi (1991), Hattori (1986). Brockmeier (1987), Brockmeier
(1991).
~Hussein and Nemzek (1989), Sawin and Baas (1985).
~Di Drusco and Rinaldi (1984), Galli and Ali (1987), Simonazzi et al. (1991).

Purge

Hydrgen 5 ~'~'4~= ~ CW~tI~= !


Ethylene
Propylene
Catalyst
Ni~ogen
Cocatalyst ~i Steam

Prepolyme

Propylene
Ethylene ~ Finishing
Fluidized
Bed Reactor
Loop Reactors

Fig. 2. Himont spheripol loop process.

a superscript 'F' denotes at the reactor feed and In this work all the distributions are normalized,
a superscript '0' denotes at the reactor outlet. Sub- therefore;
scripts are used to denote the type of distribution.
The fraction of particles with a given property , in
any space of interest is denoted by the distribution, f.
This is equivalent to stating that the probability of
min
t~ d e =
fT;?mln min
f,, d e dv = 1,

where the minimum and maximum terms in the inte-


(2)

finding a particle with property in the range to grals define the region of feasibility for the variable of
+ de is fd~. The probability that a randomly interest.
chosen particle in the entire space of interest has
a certain property ~ and property v, is termed the 3.2. Population balance modeling of the catalyst and
~ioint" probability. This is denoted by the joint distri- polymer particles
bution f~. v. Note the comma separating the two prop- Consider a population of particles within a con-
erties of interest. The probability of finding a particle tinuous-reactor process. A population balance over
with property ~, 'from the subset' of particles having the mass fraction of catalyst particles in the reactor,
a property v, is termed the 'conditional' probability having a property ~ and initial catalyst radius teat at
and denoted by the conditional distribution f~l,. The chronological time t, fw......
,R
can be made:
relationship between the two distributions is easily
found by wR ~ wR
g(Wcat, R f,',-o.,)+ ~-~(rW~at,Rf,',-o.,)

= qcat.r f w.....
,F w,O
-- qcat,of ...... (3)
Multistage polymerization of olefins--I 4863

To Recycle

Cyclone
7---~ Condenser

cTR

Catalytic System
Hydrogen
Propylene

To Recycle

Cyclone
[ CW ~ Condenser

Fines 2nd REACTOR

Hydrogen
Propylene
Ethylene ~ Polymer
I Product

_ J

Fig. 3. BASF vertical stirred bed processes.

where f~,R ..... is the probability density function on Expanding and choosing the internal coordinate to be
catalyst weight fraction basis within the reactor, the catalyst residence time, z, eq. (3) can be rewritten
fw,O
..... is the probability density function on catalyst as
weight fraction basis at the reactor outlet, f~,'~, is the
probability density function on catalyst weight frac- we 0 0 ? w,R
tion basis at the reactor feed stream, re is the rate of
change of property ~ with time and Wc~t,R, q~,,r,
q~t,o, are the mass of catalyst in the reactor, catalyst = q~at.v f ~ ' ~ , - q~.t.o f~,,o,. (5)
mass feed rate, and catalyst mass exit rate from the
reactor, respectively. Substitution of eq. (4) into eq. (5) leads to
The property ~ in eq. (3) is known as the internal
coordinate. In the above equation it has been assumed
t h a t there is no particle breakage or agglomeration It ~ t ( f ...... ) + & f ...... )
is furthermore assumed t h a t the reactor fluid phase is
well mixed so that the continuous fluid phase is uni- q.t,r ( f['~,- f~'~,) -- q~.t.o(f~'~ , -- f~.'~,)
form. These assumptions are reasonable for all pro- (6)
Wear,R
cesses considered here and necessary to apply the
concepts of residence time distributions (Shinnar and From eq. (6) we note the following. For well-mixed
Rumschitzki, 1989) systems where f~;~, = f~;o the unsteady-state popu-
A balance on total mass of catalyst in the rcactor lation balance takes the form
yields
d W~at. R
w.R - ( r w , R~_ - , / ...... ), (7)
dt q~t,F - - qcat,O. (4) ~.t ( f ...... ) + ?Jr ........ _ ~(f~,;,., "w,o
4864 J. J. ZACCAet al,

Powder
Gas
Separator
Propylene
Cocatalyst

Catalyst

L i
Purge
I Column

Propylene ~ r "~
Ehtylene ~-- ~ ~ Nitrogen
Steam
GasLock

\/
REACTOR Hydrogen Extruder }

t ' Pelletized
Product
Fig. 4. AMOCOhorizontalstirredbed processes.

Purge Purge

Catalyst
Feeder

Prop Propylene
Hydrogen Cocatalyst Ethylene
Hydrogen

Product
Product
Fig. 5. Unioncarbidefluidized-bedprocesses.
Multistage polymerization of olefins--1 4865
where the mean catalyst residence time ? is based on the properties of the entire distribution must be con-
the inlet feed rate and is time dependent: served,

-~ = ~(t) = wo.,,R(t)
qeat.r(t) " (8)
;-
rain
f,.dv = ?rain
fcd~ = 1. f13)

At steady state q~t.v = qat.O and eq. (6) is simplified It is sometimes more convenient to base the probabil-
to ity distributions in terms of weight fraction of polymer
particles or number fraction polymer particles instead
d wR 1 wF of catalyst particles. Derivations of some important
d r (f";"') = f ( f~';" - f~;0o). (9)
properties from the three bases are presented in this
work.
Equation (9) can also be written as a differential
equation with initial condition (Zacca, 1995) shown in 3.4. Reactor residence time distribution functions
eq. (10): In order to apply the population balance model to
specific processes, the residence time distribution
d w.R 1 w o
d--~( f ...... ) = - - f " ; ~ " z functions must be derived or measured. We illustrate
this with the four polypropylene processes introduced
(lO)
earlier. For a more complete review of the determina-
= ~f,2, , tion and application of residence time distributions to
other reactor design problems, the interested reader is
referred to the work of Shinnar (1987).
where again the mean catalyst residence time is more
appropriately based on inlet feed rate to the reactor.
3.4.1. Vertical stirred bed reactor (BASF). When
Equation (10) above is used to relate the exit age
the system is well mixed with no particle size selectiv-
distribution to the internal age distribution of the
ity at the exit there is no dependence of residence time
process for a system at steady state.
on polymer particle size or catalyst radius, therefore
Many processes consist of a train of reactors in
the internal and external distribution functions are
series, each operating at identical conditions of pres-
identical. Therefore, the solution of eq. (10) can be
sure, temperature and composition. The catalyst par-
easily recognized as the residence time distribution
ticle passes from each reactor adding an additional
function of a CSTR. The catalyst residence time distri-
polymer in each stage. The exit age distribution of the
bution of the BASF-type reactor is modeled in this
catalyst particles after the nth reactor of a process is
manner:
given by a convolution of the exit age distributions of
n reactors, For example, the exit age distribution after
two reactors in series is

fw, o~ Mean catalyst residence time in the reactor can be


~,rv,l~ r
, ~
cat/ ~ l rw':tz--O,ra,)
d t . r e a t ~, f~.'~:(O,r~t)dO.
calculated from known parameters such as the weight
(11) of the polymer bed Wb(g), the catalyst feed rate
qeat.v(g/s) and mean catalyst yield in the reactor YR
It is useful to remark at this point that the formulation (g polymer/g catalyst):
of the population balance model in terms of catalyst
residence time has several advantages. First, the We,,, R Wb
= - - ~15)
modeling equations are simpler and do not involve qcat, V qcat,F(1 + 17R)'
growth terms that appear when the main coordinate
where the mean yield in the reactor is given by
is particle volume or radius. Catalyst mass never cha-
nges with time. Secondly, residence time is a logical
choice, since all of the kinetic reactions are time de-
pendent.
YR =
f/Tfo
~a,. i.
f ......
w , R Y d z d r .... (16)

Here Y is the yield of polymer in g polymer/g catalyst


that is formed at residence time z from a particle
3.3. Derivation of polymer properties from the resi-
originating from a catalyst of size r~,. Yield at time
dence time distribution
may be a function of catalyst radius if monomer
To calculate the expected distribution of any prop-
diffusion plays a role in the polymerization. Since
erty when the distribution of another property v is
mean yield in the reactor is a function of the residence
known, the following relationship is used as shown by
time distribution, eqs (14)-(16) comprise a nonlinear
Rawlings (1985):
set of equations that must be solved for catalyst mean
dv residence time. Thus, catalyst mean residence time is
f(~) = [f~(v)-d-~]~=~(~. (12) a function of the overall production in the reactor,
catalyst feed rate and catalyst activity in the reactor.
This formula may be used when there is a one-to-one It is interesting to note that in cases where the
relationship between the two properties ~ and v. Since external distribution is equivalent to the internal
4866 J. J. ZACCA et al.

distribution (i.e. a well-mixed system with no exit follows:


selection) the catalyst mean residence time is the same
as the polymer mean residence time: R e , - QR,~
Qo (19)

f Wcat,a _ fp = Wb. (17) vL


qcat, F qo Pe = - - . (20)

3.4.2. Horizontal stirred bed reactor ( A M O C O / Blenke (1979) by means of superposition principles
Chisso). When the process consists of several reactors proposed an expression for the outlet residence time
operating in series and under the same conditions distribution of a loop reactor. Unfortunately, this
of pressure, temperature and composition, the resi- expression is inconsistent and the residence time dis-
dence time distribution is found by convolution of tribution neither integrates to 1 as expected, nor does
the residence time distributions of each reactor as it converge to the corresponding CSTR limit as the
illustrated in eq. (11). The solution to this problem recycle ratio tends to infinity. After careful reexamina-
for n fully backmixed reactors in series having distinct tion of Blenke's expression the following corrected
mean residence times is given by (Ogunnaike and Ray, equation for the outlet residence time distribution of
1994) a loop reactor was rederived (Zacca, 1995):

o
= i=l exp . (18)
=3 lz(l + Rec)~q~7 I i : i,,, \ l~--R~eJ
j-1
j@i

The horizontal stirred bed reactor (AMOCOIChisso)


is a cylindrical reactor that contains a series of
paddles attached to a shaft for agitation of the powder
bed. The gas phase is expected to be well mixed but 4(1 +Rec)f-~ot ]
the powder flow acts as if it is flowing through be-
tween 3 and 5 fully backmixed CSTRs in series
(Caracotsios, 1992). In this paper we assume four The maximum and minimum integers used in eq. (21)
CSTRs is sufficient to model this process. The volume are given by

if floor (l+Rec)~~ --j] > 0


iml n (22)
1 otherwise

ceiling [(1 + Rec)~ot + J ] if ceiling (1 + Rec)~Gi + J > 0


imax = (23)
1 otherwise

ceiling \ ~/---~-) if ceiling ( 2 ~ >_-2,


\V 10 )
j= (24)
2 otherwise

of each reactor zone is equivalent and therefore the where the 'floor' of the argument is defined as the
mean residence time of each zone will steadily de- largest integer smaller than the argument and the
crease from zone to zone to reflect the cumulative 'ceiling' of the argument is the smallest integer larger
production of polymer from the previous zones. For than the argument.
each reactor zone, the mean residence time must be In the above equation, the mean residence time is
calculated by solving eqs (15), (16) and (18). Again, we based on a total mass basis, i.e.
assume no dependency of residence time on particle
size. riot = WR PR Va (25)
qF qv
3.4.3. Loop reactor (Himont). The two most impor-
tant mixing properties of a loop reactor are its recycle However, it can be shown that this quantity is identi-
ratio and the Peclet number which are defined as cal to the mean residence time based on the catalyst
Multistage polymerization of olefins -1 4867
mass as indicated in eq. (8). At steady state a total the reactor:
mass balance around the unit is given by

qr = q0. (26) FR = 7 f o of ......


w,R Y dr dr~,t

The total mass outflow is calculated as (35)

f l + 17o~
qo = t ~ l q ~ a t , r " (27)
\ ~,o /
B = B(r,r~at)= ~ Y(r.r~,t)dz. (36)
Therefore,

(1 + 17o\ 3.4.4. Fluidized-bed reactor (Union Carbide). In


qF = / ~ - / q c a t . F" (28) fluidized-bed reactors the catalyst residence time dis-
\ 's,o /
tribution can be size dependent due to particle selec-
The total mass of material in the reactor can be tion effects along the reactor height. In this work, we
calculated as use a selection factor approach (Kunii and Levenspiel,
1991) in order to account for size selection effects. Let
(I + ITR)WCa~,R us define a particle size selection factor, ip(rp), such
WR (29) that
WS, R

From eq. (25), the mean residence time is


,ffTo = lis(rp)fv~ R. (37)

Equation (37) relates the probability (weight fraction


r i o t = , (1 +' 17R)
- , -Ws,
- r .o _
~,~ (30) polymer) of a polymer particle of radius rp being
(1 + 17o) w~.R
found in the exit product stream to the probability of
that size of particle existing in the reactor. For a given
Also, from an overall mass balance around the reactor
catalyst size, it is possible to convert particle size
it can be shown that
distributions from a weight fraction of polymer basis
to particle size distributions on a weight fraction of
I -~ Y R 1 + 17o
(31) catalyst basis. Also using the selection factor defini-
Ws, R Ws, 0 tion one can relate the outlet and internal particle size
distributions on a weight fraction of catalyst basis as
where Ws,R, Ws,0 stand for the solids content inside
follows:
and at the outlet of the reactor. Therefore,
( 1 -~- Yo \ wR
f , o , = f. (32) (38)
= I
Thus, for the case of a loop reactor the outlet resi-
A transformation of variables between particle size,
dence time distribution on a total mass basis equals
rv, and residence time, r, is possible through the rela-
the outlet distribution on a catalyst mass basis:
ftot. o = f~,.o. tions:
The reactor mixture density can then be calculated
.,R dr
as S,';:,L =

1 Ws, R 1 -- Ws, R
- - + (33) -w,o w,o dr
Pg Pp Pl ,lrr ..... = -- (39)
f ...... drp
Therefore, by combining eqs (25)-(33), the mean cata-
Therefore, it is possible to rewrite eq. (38) as:
lyst residence time can be expressed in terms of the
solids content in the reactor,
'1 + 17o)
f~wL' = , ~ / t~(rv) .fl~i;,~,. (40)
I VR 1 w~.o
= (W.~.R/pp) + (1 -- Ws.R)/pl (1 + ]70)qeat, r '
From the population balance equation (10) it is pos-
(34)
sible to write :
By specifying the outlet solids content, the internal d w.o
solids content can be calculated by eq. (31). Since .fr.r~at
-- .1 ...... ) =
it would be too cumbersome to find an analytical d r ( "~'"
expression for the internal residence time distribution,
we apply integration by parts to eq. (16) making f~.: R ,= o - f w:vg (41)
use of eq. (10) in order to get the average yield inside
4868 J. J. ZACCAet al.
Solving for the outlet residence time distribution obtained:
(Zacca, 1995) one obtains
~b(rp) = exp [b(rp - ro,~t)] (46)
~,,o ~( r p)f ....
w,F /' ~ ~(rp) d'r'~ (42) where b is a parameter that depends mainly on fluidiz-
f ...... ~, exp k ~ ,) ation conditions, rp,~t is the particle cutting radius
corresponding to a selection factor of ~k(rp) -- 1. rp,~t
wb
is not a free parameter but is established by imposing
q~,t.v(1 + Yo) the condition that the internal particle size distribu-
Wb tion integrates to 1. In our work we use the same slope
(43) 'b' as used in Kang's work so that the particle selec-
q~t,v(1 + JPlt,
f ........ f ~ fw' Yd~dr~t)
rain dO J ' teat tion sensitivity is roughly similar. It should be noted
rp = r,(~, ro~,). (44) that the range of particle sizes studied in Kang's work
is similar to that expected in solid-catalyzed olefin
The overall outlet residence time distribution of the polymerization.
process is given by
3.5. Polymerization kinetics
~r cat,
f~,o ma~

Jf~,,o dr c a t .
t , rml~ (45) 3.5.1. Kinetic model. The kinetic mechanisms used
~t,min
in this paper are based on a generalized Ziegler-Natta
The selection factor needs to be experimentally deter- kinetic scheme (Chen, 1993). The following reactions
mined by sampling the reactor internal and outlet are expected for the Ziegler-Natta multi-site catalyst.
particle size distributions. Not much experimental Notation is given at the end of this paper. Note that in
work has been done in his area. Kang et al. (1989) this paper we only consider time-dependent reaction
have measured the selection factor of a pilot scale mechanisms that are first order with respect to cata-
fluidized bed of mineral particles under different par- lyst sites. It is possible to model other complex reac-
ticle diameters and fluidization conditions. They were tions with cocatalyst, donor or reactions with hydro-
able to correlate the selection factor to the difference gen to fractional orders by application of pseudo-rate
between particle diameter and the mean particle dia- constants.
meter in the bed. Polynomial fits were performed
under different experimental conditions. The data of Site activation ~p k~ ~k
Kang et al. are shown in Fig. 6 with an exponential
data fit. Initiation P~ + Mi e~,, ~k
Pl.i
In this work, we model the selection phenomenon
in a similar method as Kang et al. Since for
the application considered here, larger particles Propagation ~ak J + Mi k~'u' ~k
, Pn+ l,i
are selected over smaller ones, the selection factor
must be a nonnegative monotonically increasing
Site deactivation [ak . k~j C a + Dk.
function of particle size. An exponential correlation n,J
is used to relate the selection factor and particle
radius, giving the same shape of the curve as Kang Site transformation ~ k P,, v l
Pn.j ~ P.,j

Chain transfer [ak, j + H2 V,,, ~k + ~)k.


I Kanget al. (1989) 10.3 /
-- a = 0.8481; b =3.5269 x

~2
3
ag= a * exp( b * (dp - draeart) / The kinetic scheme gives rise to a system of linear
equations on the catalyst sites of the following form:

d_~C= K(~
dt
(47)

(2 = [~,, ~o, ~o,,Y,


o i = 1. . . . . N . . . . k = 1. . . . . Nsite , (48)

where 'K' represents a matrix involving all the kinetic


terms. It can be shown (Lancaster and Tismenetsky,
1985) that for any generic kinetic matrix K e R ' ~
having a certain number of nz distinct eigenvalues, the
0 i I i i i I I
solution of system (47) can be expressed as
-300 -200 -100 0 100 200 300 400 500 C(t) = exp(Kt)(2(0) (49)
~ . dmean(r,,icrons)
n~ ml-- 1
(2(0 = ~ ~ Z,j tiexp(2,t)(2(O), (50)
Fig. 6. Size selection factor vs particle diameter. i=1 1=0
Multistage polymerization of olefins--I 4869
where rn~ represents the multiplicity of the kth distinct The rate of consumption of each monomer (g/g/s) can
eigenvalue of matrix K and Z o are time-independent be calculated by
matrices in ~R"". If K has only distinct eigenvalues the
previous equation reduces to kk v k ] T*WT
2 ( pijPo,J)l~[Mi]s Mwi. (56)
C.(t) = Pexp (At)a - t (2(0) (51)
Monomer concentrations used in this paper are the
exp(At) = diag [exp (2x t). . . . . exp (,t,t)], (52) concentrations of the monomer at the catalyst site.
These are not necessarily the same concentrations in
where P is a matrix of right-hand eigenvectors of K. the bulk phase of the reactor. Assuming thermodyn-
Equation (51) provides an analytic expression for any amic equilibrium between phases, one can calculate
site concentration profile corresponding to a given the concentration at the active site as a function of the
kinetic scheme. For the purpose of this paper we now fluid-phase concentration. For example, Hutchinson
make the following simplifications. (1990) relates the two concentrations by a sorption
factor )'i. The sorbed monomer concentration is re-
It is assumed that the catalyst is fully activated at lated to the bulk monomer concentration by
time t = 0. This is a reasonable assumption since
the time scales for activation by aluminum alkyl are [ M i L = ?,[M,]~. (57}
of the order of minutes. For propylene polymeriz-
For liquid-phase reactors, the methods of H utchinson
ation, the catalyst system is usually precontacted in
are used to calculate the sorption equilibria for each
a previous stage so that cocatalyst, donor and cata- monomer. The concentration of monomer in the bulk
lyst may complex for high isotacticity polymer. fluid is calculated by the Benedict Webb-Rubin
The long-chain hypothesis can be applied such that equation of state. For all liquid-phase reactors, the
the rate of chain transfer is small compared to the concentration of monomer at the catalyst site is less
total rate of propagation and initiation.
than the bulk fluid. For gas-phase reactors, the
The initiation step proceeds at the same rate as the method of Chen (1993) is used to calculate monomer
propagation step. sorption factors. Chen uses the theory of Heuer et al.
There are two types of sites and only site of type (1989) and Scbotte (1982) to calculate the solubility of
1 may transform into site of type 2 which sub- monomer in a multicomponent mixture. Sorption
sequently deactivates.
from the gas phase can result in higher monomer
The QSSA for the live chains may be applied. concentrations at the catalyst sites compared to the
The fraction of vacant sites is small compared to gas phase. The BWR equation of state is also used to
the occupied sites. calculate the gas-phase composition.

With the above assumptions the concentration of sites 3.5.2. Kinetic parameters. Many kinetic models de-
of type k having a chain ending in monomer of type scribing olefin polymerization with Ziegler Natta
1 or 2, respectively (i.e. a copolymer system) is given catalysts have appeared in the literature (Burfield et
by al., 1972; Carvalho et al., 1989, 1990: Chen, 1993:
Chien, 1987; van der Burg et al, 1993). With them
]20,1
k M kk a broad scatter of kinetic parameters has been pro-
posed. This lack of agreement is due partly to intrinsic
i~kz = /)k k~,[M2]~ differences in the proposed kinetic schemes, inconsist-
(53) ent volume basis for determining kinetic rate con-
k~,~[M1], + k kP21[M2]s"
stants, nonconsideration of sorption effects, and the
The concentration of active sites of type k can be great variety of polymerization conditions and cata-
found by lyst types. All these factors combined make the task of
kinetic parameter evaluation a difficult one.
In this paper, simulations are chosen to reflect
- dt
- = - kd1Y1
p o -- kU Pol
a modern high-activity TiCI4/MgCI2 supported cata-
lyst to compare the different processes. The basic
d/~2
(54) kinetic parameters were taken from data published on
= ktl 2/~~ - ka2 I~o.
-2
dt an industrial catalyst used for propylene polymeriz-
ation in liquid propylene (Di Drusco and Rinaldi,
Analytically, this can be expressed as
1984). The main goal was to establish kinetic para-
fi~(t) =/)~(0)exp(-- (k,~ + k~2)t) meters that match overall productivities and deactiva-
tion levels for an industrial catalyst system and not to
~ ( t ) = ~ ( O ) ( k ~ + k~ )
kt22 - k 2 / { e x p ( - k 2 t ) validate a generalized kinetic mechanism. To this end
a two-site kinetic model was fitted to their data with
the following assumptions:
- - exp [ -- (kd1 + k~2)t]} +/]~(0) e x p ( - k~t).
The total mol fraction of Ti available as active sites
(55) is equal to 40% (Karol, 1984). The initial amounts
4870 J. J. ZACCAet al.
of sites of type 1 and 2 are 80.64 and 19.36%, Table 2. Kinetic parameters for two-site model used in the
respectively (Chien, 1987), reflecting precontacting simulations (base case)
of the catalyst system.
Kinetic parameter Value
Site 1 does not deactivate but transforms to site
type 2. Site type 2 deactivates with time. This model k~11 (l/mol/s) 342.8
is designed to reflect a change in oxidation state of k211 (1/mol/s) 34.28
the catalyst from Ti 3 + to Ti 2 +. Based on typical k~2 (l/s) 2.385 x 10-4
time scales for site transformation (Chien, 1987; k2 (l/s) 7.950 x 10- 5
XTi* 0.40
Rincon-Rubio et al., 1990) it was assumed that the
~o~(o) 0.8064
kinetic constant for site transformation of site 1 to ho~(o) 0.1936
site 2 is three times the kinetic constant for site
deactivation of the second site.
Site of type 2 was assumed to be 10 times less active
60-
than site 1 for propylene polymerization (Barbe et 1
Data of Di Druseo and Rinaldi ( 1988 ) I
al., 1987; Kissin, 1985). -- 2 Sitc Model used in thisW o r k
J
50
The final kinetic parameters at 70C used in this paper
can be seen in Table 2. In this work the same catalyst
system was used for all the simulations. We do not 40
make any attempt to change the kinetic parameters to
reflect differences in the catalysts used by each pro-
~a0
ducer. By keeping a consistent set of kinetic rate
coefficients, we are able to compare better the perfor-
mances of each process. 20
The fit of the kinetic model to the experimental data
can be seen in Fig. 7. Figure 8 shows the active-site
variation with time. Site 2 is clearly indicated as the 10

stable site.

3.6. Catalyst and particle effects o 1 2 3 4 5


Equation (56) represents the kinetic rates of con- Time ( hr )
sumption of the monomers evaluated at a point along
the particle radius. In the absence of monomer diffu- Fig. 7. Fit of kinetic data to two-site model.
sion resistances or temperature gradients, eq. (56) may
be used directly. However, in cases when significant
diffusion resistance or particle temperature increases 0.a5 J
occur, the net rate of production averaged across the
growing particle radius can be related to the kinetic 0.30 4/
rate at bulk reactor conditions by an efficiency factor
/~ob,
pi
/~pi~, (58) o.25

where (i = (i(z, rcat) and the index i is over the mono- ~ 0.20
meric species.
The total rate of polymerization at any residence -~
time can be found by ~ o.Is

dY N.o, 0.10
d-~- = ~ /~p'('" (59)
i=1 """'"'",..,,............
0.05
Yield of polymer is then calculated by
N=on t
f t ~ t
2 4 6 8 10
Time ( hr )
The effectiveness factors are functions of many vari-
ables including particle size, diffusivity and the kinetic Fig. 8. Active site fraction vs time.
rates of consumption. To solve rigorously the popula-
tion balance problem with diffusion and temperature
gradients in the particle would require solving The effectiveness factors are approximated by simple
a coupled set of nonlinear integral equations involv- functional forms that are obtained by fitting the data
ing residence time, yield, particle radius and catalyst from a detailed model of particle growth and morpho-
size. To avoid this arduous task, a simpler more logy. The heterophasic muitigrain model of particle
effective method is used here to capture these effects. growth (Debling and Ray, 1995) has been used in this
Multistage polymerization of olefins I 4871
work. Note that this formulation allows a generic systems where the residence time distribution is inde-
approach to integrating the particle characteristics pendent of catalyst size the number fraction and
into the population balance model without having to weight of catalyst yield distributions are equivalent.
include particle radius as an additional coordinate. This is not true of the fluidized-bed reactor but is true
In addition to the particle diffusional and temper- of the other reactors considered in this work when
ature effects, one must also consider that catalysts are there is no size selection upon exit. Note also in cases
never monodisperse but have a certain particle size where there is catalyst deactivation the catalyst yield
distribution. A log normal particle size distribution will reach a finite value. Thus, although the domain of
(weight of catalyst basis) is often used to represent the residence time is infinite, the domain of yield is not.
catalyst particle size distribution (since it does not For solid-catalyzed olefin polymerization yield dis-
permit negative particle sizes as a normal distribution tribution on a weight of catalyst basis is more useful
does): than the other bases.
1 [ -- (In (rcat/fcat)) z ] 3.7.2. Polymer particle porosity. The phenomenon
f r W j f = rcatO'cat N ~ exp 20.2at A (61)
of particle growth and development of morphology is
quite complex. During the polymerization reaction,
f~.f~:rdr = 1. (62) several forces act upon the growing polymer particle
to shape its ultimate morphology at the end of the
This implies the log weighted mean size is given by reaction. In the early stages of particle growth, con-
trolling the rate of polymerization may be essential to
avoid excessive fragmentation of the catalyst (Sim-
~"m~"fwf ln(rcat)drcat = In (fc~t). (63)
dr m onazzi et al., 1991). Hutchinson (1990) suggests that
monomer diffusion limitations in the early stages of
We assume that the fraction of metal and the kin- particle growth can lead to concentration gradients in
etic rate coefficients are independent of catalyst par- lhe growing polymer particle. This causes stresses
ticle size. The distribution of catalyst particle sizes along the particle radius and increases in porosity
affects the overall polymer particle size distribution causing particle fragmentation and poor bulk density
and also the effectiveness factors for diffusion and product.
temperature effects. As the polymerization proceeds, it is believed that
the porosity of the granule decreases with yield, and
3.7. Calculation of polymer properties can decrease far below that of the initial catalyst
3.7.1. Polymer yield. Using eq. (12) the relationship porosity. Bukatov et al. (1982) measure the porosity of
between yield Y in g/g catalyst and catalyst residence polypropylene produced over a TIC13 A1C13 catalyst
time r in the homopolymer stage can be found. Note system with mercury porosimetry. Porosity is re-
the dependency on catalyst particle size. ported to decrease from 0.45 to 0.17 after a polymer
On a weight of catalyst basis, yield of 2000 g/g polymer. Tang (1986) reports in-
creasing bulk density with polymerization yield for
wo d~
fr~',~, = f~, ;,., ~--~, (64) polyethylene produced with a TiC14/MgCI2 catalyst
system, consistent with internal pore volume changes.
where in the above equation the f ~ ; o is evaluated at Simonazzi et al. (1991) suggest that with increasing
the appropriate residence time that gives the desired yield, all of the particle void space is progressively
yield. filled with polymer.
The polymer yield distribution on a catalyst weight Changes in polymer particle porosity are also no-
basis can then be calculated as ticeable with copolymer systems. Chen (1993) has
conducted random copolymerization studies with
ethylene and propylene over an unsupported
f~,o =
/,7
cat. in
f rw,o
..... dr .... (65)
TIC13 AIC13 catalyst in gas phase. Micrographs of the
copolymers produced show a great deal of fusing and
On a number of particles basis this is given by
agglomeration within the particle leading to elimina-
tion of void space. Furtek (1993) indicates linear low-
fT;o =
f r . . . . . (lrca t
~ (66)
.~,. i. rcatlrcat
density polyethylene produced with a Metallocene
catalyst is essentially nonporous. Nonporous ethy-
(.r.~t max w, F
" fr , lene propylene copolymer particles are reported by
lr~at = ] ~drcat. (67)
) r~,.mi, rcat Hoel et al. (1994).
While it is difficult to determine precisely the rela-
On a weight of polymer basis, tionship between yield of homopolymer and particle
porosity, it seems clear that the particle becomes less
f~.o
U
(I + 17o) .... ,. (1 + Y ) f r ..... dr ....
p,O
(68) porous with time. Polymer composition, catalyst
structure and operating conditions play an important
Note in the absence of diffusion limitations (polymer role. In this work we assume an exponential relation-
yield is independent of the initial catalyst size) and for ship between porosity and homopolymer yield for
4872 J. J. ZACCA et al.
a polymer particle as shown in eq. (69). This relation- On a weight of polymer basis,
ship seems to fit the general trend found by Bukatov
et al. (1982) reasonably well. Other correlations for 1 f ....... (1 + Y) f , ,p.o
different systems can be applied as determined by the fp;o = (I + lTo)J, ....... ..... dr=at. (80)
experiment. The parameters will be varied in the pa-
For particle size, it is most useful to calculate the
per to study the effects of this assumption on the
polymer weight distribution since, in practice, poly-
resulting particle size distribution:
mer particle size is often calculated on a weight basis,
e = (~o -- e~i.)exp(- flY) + ~i.. (69) i.e. from mechanical sieving. However, it is also pos-
sible to calculate the number particle distribution by
Porosity distributions resulting from particles having particle imaging methods, or by computer imaging/
different yields are then found from the residence time microscopy methods.
distributions by
3.8. Statistical properties
,~.o ~ o dz d Y (70) In addition to the probability distributions, mean
f ...... = f~';" d'Y de and standard deviations are calculated by the stan-
dard formulae
de
d--Y = (eml, - eo)flexp(- flY). (71) m,

On a catalyst weight basis this is given by


=
ff ~min
~fcd~ (81)

a = (~ -- 02f~d~. (82)
f2/, o = r~,o
' ~,ritdr cat" (72) rain

~t, in These values are determined by application of


On a number of particles basis this is given by Simpson's rule to approximate the integrals.

(.r..,,~o~ fw, O 4. M O D E L I M P L E M E N T A T I O N
. / e , rca t
f~.o= | ~ dr<~t. (73)
J r,at,mi~ teat ~/rcat Simulations were conducted with the software
package Mathcad Plus 5.0*. The integrals were nu-
On a weight of polymer basis, merically solved by a successive Romberg (poly-
nomial interpolation) algorithm. The convergence of
ff'(e) (1 + 1 Yo) f ' ....... (1 + Y)f~;o dr~.t (74) trapezoidal approximations is accelerated by extra-
cat.rain
polation of intermediate estimates and the corres-
Weight of polymer basis is most useful since the ponding subinterval widths.
measurement of porosity is most likely carried out on If an integrand function f(x) is such that it cannot
the total mass or volume of polymer product, for be properly approximated by polynomials, the integ-
example, through mercury porosimetry. ration method will become inefficient in terms of
precision and computational speed. In those cases
3.7.3. Polymer particle size, Polymer particle a variable transformation s(w) often improves numer-
radius can be related to polymer yield by eq. (75) with ical integration by changing the shape of the function
porosity given by eq. (69) above: to be integrated and by shifting the integration do-
main to the interval (0,1).
= ( 1 - e0~1/3(1 Peal Y ) 1/3. (75)
rp Feat ~ -i--~-~-i8 ff ~k -I--- ~ dw
P.,,
The expected distribution of particle sizes from the s(0) = xl, s(l) = x~. (83)
process can be related to the residence time distribu-
tion as Many functional forms for the transformation s(w) are
possible depending on the specific shape of the integ-
,~.o -w s dr dY (76) rand function f(x). The only requirement is that s(w)
f'" ..... = J " ;" d'Y dr~ be a monotonically increasing function in the interval
(0, 1).
dr, ( 1 - eo~r~, [po., r~ de-] Another problematical situation arises when the
d-'--Y = \ 1 - e J3r~ L pp + (1 - ~o)r~a, d-Y_] (77) integrand approaches a delta-like function. In that
case the function 'spiky' behavior tends to cause nu-
~,0 merical errors in the integration routines. A way
= f,,w,O
..... dr .... (78)
cir. l. around that problem is to divide the integration inter-
val into two regions. The first one (xx,x*) corres-
On a number of particles basis this is given by ponds to an arbitrarily chosen interval in whichf(x) is
well behaved. The second one is the interval (x*, x2)
~,o= ;?7 ....3 frp,rc.t
-o 1
or .... (79) where f(x) changes abruptly over the integration vari-
,i, I. rcat//'cat able x. This is shown in Fig. 9.
Multistage polymerization of olefins 1 4873
Table 3, Simulations conditions for base case
f~x)
Gas-phase Loop
reactors

Temperature (C) 70 70
Pressure (atm) 25 35
Monomer conc (mol/1) 1.267 9.917
Sorption factor 1.474 0.479
Reactor
Bed weight (t) 17.04
Volume (m3} 50 45
Solids fraction wt% 50
Catalyst feed rate (g/s) 0.2 0.2
ac~t 0.25 0.25
rc,t (pm) 33.4 33.4
b x 103 (l//~m) 7.054

I J D
In the case of the gas-phase reactors the gas is not
x1 x* x2 x ideal at the pressure and temperature used in the
simulations. For propylene the concentration at the
Fig. 9. Delta-like-behavior of integrand functions. catalyst site is higher than in the bulk fluid as in-
dicated by the sorption factor. Bed weight was cal-
culated assuming a reactor volume of 50m 3 with
The numerical integration is then split over the two 32 m 3 solids volume and 0.415 bed voidage. A poly-
intervals: mer density of 0.91 g/cm 3 was assumed for polypropy-
lene.
f ( x ) dx = f ( x ) dx + f ( x ) dx. (84) For the loop reactor a recycle ratio of 5 was chosen
i 1 to represent the low recycle ratio case. It has been
previously reported that a recycle ratio of below 30
The first integral in eq. (84) is evaluated in the usual
marks the onset of significant deviations from an ideal
way. The second integral requires a variable trans-
CSTR behavior (Zacca and Ray, 1993). For the high
formation aiming at a non-stiff integrand function.
recycle ratio case, we assume a CSTR model since
That is accomplished by transforming the integration
a typical recycle ratio for these processes is about 100.
variable, x, over to the residence time domain, r:
Note in the loop reactor the concentration of mono-
mer at the catalyst surface is roughly 1/2 that of the
f ( x ) dx = J~x') f i r ( x ) ] dr. (85) bulk phase due to monomer sorption effects.
For the base case simulations, the log mean catalyst
Notice that in the residence time domain all the func- particle radius was 33.4 ~m and catalyst spread stan-
tions are relatively well behaved. dard deviation, ac,~ was 0.25 in eq. (61). We have
Whenever variable transformations such as the one chosen to use the same catalyst size and distribution
in eq. (12) are used, it is necessary to solve numerically for each process so as to compare better the effects of
for the residence time of the catalyst as a function of the different reactor residence time distributions on
the property of interest. In this work that task was the product properties (see Fig. 10). We have also
accomplished through a Levenberg-Marquardt neglected heat and mass transfer effects for the simula-
quasi-Newton method. tions studied. In practice, the catalyst size distribution
The computational speed corresponding to the cal- is an important consideration for industrial processes.
culation of all the distributions and statistical Gas-phase processes typically operate with smaller
variables of interest was reasonably fast. Total simula- catalyst particles to avoid overheating and melting
tion times ranged from 20 rain to 2 h on a 90 MHz under the high rates of polymerization. Prepolymeriz-
Pentium computer. ing the catalyst under mild conditions, prior to injec-
tion into the main reactor is also used for the same
5. RESULTS purpose. The population balance approach presented
here can be easily extended to study the effects of
5.1. Process operatin9 conditions continuous prepolymerization and catalyst particle
Simulation conditions for the reactors used in the size distribution on the performance of multistage
simulations are shown in Table 3. The operating con- processes and is the subject of another article (Debling
ditions are typical of those reported in the literature. et al., 1996b).
For all cases a temperature of 70C was used and the
same kinetic parameters were used for each reactor. 5.2. Residence time distributions
Sorption factors were also used so that each process Figure I 1 shows the reactor residence time distri-
could be directly compared on a consistent basis. No butions for the gas-phase reactors studied in this
adjustment to the kinetic parameters was necessary. paper, on a catalyst weight basis. The residence
4874 J. J. ZACCAet al.
0.14 Table 4. Mean residence times for the base case
-- t~eat= 0. I0
Reactor Stage ? (h)
......... ac~ = 0.25
0.12 -

- - - a~.~ = 0.50 VSBR 1 2.096


0.10 - HSBR 1 0.846
2 0.571
3 0.475
2 0.08- 4 0.424
%
FBR 1 2.064
~ 0.06-
! Loop (high recycle ratio) l 1.016
2 0.695
0.04 - ../
Loop (low recycle 1 0.995
7- ratio = 5) 2 0.635
// /

0.00 "/"'' ' C7~7~"7---q. . . . r-


0 2O 40 60 80 100 120 would expect for 4 CSTRs in series. This is because
Catalyst Radius ( microns ) the HSBR is modeled as 4 CSTRs in series with equal
volumes, not equal residence times. As the catalyst
Fig. 10. Catalyst particle size distributions considered in the moves through the HSBR, the cumulative product
simulations (log mean catalyst radius = 33.4/~m). overflow increases from stage to stage. This causes the
mean residence time, each in stage, to form a strictly
1.6e-4 monotonically decreasing sequence. For a given bed

1.4e-4
~ VSBR
HSBR 1
FBR
weight, however, the HSBR has a higher mean resi-
dence time (as given by the sum of the mean residence
times of each stage) than the VSBR as a result of the
1.2e-4 RTD differences.
The residence time distribution of the fluidized-bed
\ reactor (FBR) is a hybrid distribution that is not easily
~ 8.0e-5 modeled by n CSTRs in series. At short residence
times, the residence time distribution curve of the
6.0e-5 FBR lies in between the VSBR and the HSBR. How-
ever, at longer residence times, the FBR residence
4.0e-5 time distribution exhibits a long tail. These results are
easily explained by inspection of Fig. 12. In Fig. 12,
2.0e-5
the residence time distribution of three catalyst par-
0.0e+0
ticles sizes (10, 33.4, 75 #m) are shown. As a result of
0 10000 20000 30000 40000 50000 particle size selection, small catalyst particles are seen
Residence Time ( s ) to remain inside the reactor for extremely long peri-
ods of time while most of the larger ones leave the
Fig. 11. Base case residence time distributions for the gas- reactor very soon. The residence time distribution for
phase processes. the mean catalyst radius is seen to be narrower than
the overall residence time distribution. The overall
time distribution of the vertical stirred bed reactor residence time distribution of the FBR is complex and
(VSBR) is the typical exponential curve of the CSTR. can be seen as a combination of the individual resi-
An important consequence of this is that a significant dence time distributions of the different catalyst par-
fraction of the catalyst particles leaves the reactor at ticles from the feed.
very early times. These particles are still very active Figure 13 compares the residence time distribution
and have little polymer yield. A narrower residence for a single-loop reactor as the recycle ratio is
time distribution is found for the horizontal stirred changed. For low recycle ratios the residence time
bed reactor (HSBR) as a result of the 4 CSTRs in distribution is oscillatory in nature and presents
series for the powder flow. One major advantage of a time lag due to the first pass of material through the
the HSBR over the VSBR is the reduced fraction of tubular section. Peaks of material leave the reactor
catalyst particles with very short or very long resi- after each circulation time. As the recycle ratio is
dence times in the reactor. Catalyst particles with long increased the oscillations disappear and the residence
residence times in the reactor have been significantly time distribution tends toward that of a CSTR. Fig-
deactivated and are not productive. A summary of the ure 14 shows the effect of recycle ratio on two-loop
mean catalyst residence times for operation under reactors in series. As in the case of a single reactor the
base case conditions is shown in Table 4. RTD is time delayed and oscillatory for low recycle
It is interesting to note that the residence time ratios. Complex peaks in the RTD result from non-
distribution of the HSBR is not as narrow as one synchronized interactions between the peaks of each
Multistage polymerization of olefins--I 4875
1.8e-4 single reactor. As the recycle ratio is increased the
-- Total RTD
residence time distribution of two continuous-stirred
1.6e-4 /,\ tank reactors in series is obtained. Note that even
1.4e-4 though the residence time distributions of one loop at
high and low recycle ratios are completely different,
1,2e-4

1.0e-4
~ 8.0-5
/1' the residence time distributions of the two-loop pro-
cess are very similar, particularly at longer residence
times. For the range of simulation parameters studied,
it would appear that two loops in series can be simply
modeled as two CSTRs with little error.
6.0-5

4.0-5 5.3. Production and catalyst yield


2,0e-5
Increasing the feed rate of catalyst increases the
production rate (ton/yr) of all reactors mainly because
0.0e+0 - i r i i i i i of two factors: (a) at higher catalyst feed rates (reduced
5000 10000 15000 20000 2 5 0 0 0 3 0 0 0 0 35000 40000 mean residence times) deactivation effects are pro-
Residence Time ( s )
gressively less important and catalyst particles
retain their activity; and (b) higher catalyst concen-
Fig. 12. Catalyst residence time distribution for the FBR
base case (log mean catalyst radius = 33.4/~m). trations are available for polymerization inside of
the reactor. On the other hand, reduced mean resi-
dence times result in lower catalyst mean yields
(cf. Figs 15 and 16). For the VSBR, for example,
1.2e-3 the mean residence time drops from 2.096 to 1.312 h
....... ~Sh RecydeR a ~ o
- - LowRecyclRatio= 5 by increasing the feed rate from 0.2 to 0.4 g/s.
1.0e-3 Production rate was calculated by assuming normal
operation at 0.95 service factor. Although catalyst
8.00-4 feed rate was varied over a wide range of flow rates
(0.05-1.6 g/s), practical process operational issues
6.0e-4
obviously restrict this. This sensitivity may be differ-
ent for each type of process. In addition, if catalyst
~3 yield is insufficient, residue content may be too high in
4.0e-4
the end product.
As a consequence of the higher mean residence time
2.0~4
of the HSBR over the VSBR, a production 'boost' is
noted. This yield enhancement, however, is a function
0.0e+0 I [ i i of catalyst deactivation rate and mean residence time.
2500 5000 7500 10000
Time(See)
When catalyst feed rate is changed from 0.05 to 1.6 g/s
the relative production rate of the HSBR over the
Fig. 13. Base case catalyst residence time distribution for VSBR increases from 1.175 to 1.255.
one-loop reactor.

45000

...... High Recycle Ratio -- VSBR


Low Recycle Ratio = 5 40O00
\ ......... HSBR
1.6e-4 -- - - - 2 Loops - High Recycle
FBR
35000
\
\\
\
30000 \
1.2e-4 \
\
25000

\,,
8.0~5 -~ z000o
v

4.0e-5
/ ~. 15000

10000

500O
i
/

0.0e+0 0 i i i i i m._ .__~

0 5000 10000 15000 20000 0.0 0.2 0.4 0.6 0.8 1.0 1.2 1.4 1.6
Time (~) Catalyst Feed Rate ( g / s )

Fig. 14. Base case catalyst residence time distribution for Fig. 15. Effect of catalyst feed rate on mean catalyst yield of
two-loop reactors in series, the product.
4876 J. J. ZACCAet al.

800000 1.4e-4
I ~ VSBR

~ VSBR
HSBR
2 Loops - High Recycle
FBR
1.2e-4 f"
HSBR
FBR
2 Loops - High Recycle
J i
600000 /////
/// 1.0e-4
///////
,," 8.0e-5 J

"
-~ 400000

/ ..... '~ 6.0e-5

1 200000 ///
,/
.......
.... . . . .
4.0e-5

2.0e-5
I ~ .... ~'x\
\
\
\
\
0 -g----"-7 r r r r \
0.0 0.2 0.4 0.6 0.8 1.0 1.2 1.4 1.6 O.Oc+O ---r r r~ -
Catalyst Feed Rate ( g / s ) 10000 20000 30000 413000 50000
Yield(g/g)
Fig. 16. Effect of catalyst feed rate on polymer production
rate. Fig. 17. Catalyst yield distribution for base case.

Figures 15 and 16 illustrate the advantage of the weight basis which is the natural method for calcu-
liquid pool loop process (two loops at high recycle). lation of yield. Note also that all curves tend to the
The extremely high monomer concentrations give this limiting yield as a result of catalyst deactivation which
process the highest production rates (assuming the is obtained at infinite residence times.
same catalyst intrinsic activity). Among the gas-phase Again the main difference between the VSBR and
reactors the HSBR has the highest production rate, as the HSBR is the increased weight fraction of catalyst
a result of its higher mean residence time at identical particles that exit the VSBR with low yields. The
total reactor volumes. It is interesting to note (cf. narrow yield distribution of the HSBR reflects its
Fig. 15) that at low catalyst flow rates (high mean residence time distribution. For the FBR, particle
residence times) the FBR performs worse than the selection effects cause the smaller catalyst particles to
VSBR. As the catalyst flow rate is increased (de- remain in the reactor for long residence times and
creased mean residence times) the FBR production therefore reach values very close to the maximum
surpasses the VSBR. Similar effects of catalyst flow yield. The yield distribution of the FBR, therefore, is
rate on the product yield are found. This result is due sharply increasing close to the maximum yield, for the
to particle selection effects in the FBR. Under the base case simulation parameters. In the case of two-
conditions of the base case, particle selection hurts the loop reactors (high recycle) the maximum yield is
FBR performance by withdrawing the large catalyst larger because of higher monomer concentration and
particles from the reactor prematurely, thus reducing the distribution is broader.
overall polymer yield. However, as the catalyst feed The effect of recycle ratio on the yield distribution
rate is increased and the mean residence time of the of a single-loop reactor and two-loop reactors in
catalyst decreases, the particles are not able to grow as series is shown in Figs 18 and 19. For low recycle
large in size. Thus, particle selection effects do not ratios the oscillatory nature of the RTD is reflected in
appear to suppress production as significantly as at the yield distribution. At high recycle ratios the yield
the lower feed rates. This effect is extremely interesting distribution of a CSTR is obtained. The yield distribu-
since many models of the fluidized bed assume either tion of two loops in series at low recycle shows the
a continuous-stirred tank (Choi and Ray, 1985a; same interaction pattern as the residence time distri-
McAuley, 1991; Talbot, 1990) or several stirred tanks bution curve. It is interesting to note that practically
in series for the powder phase. Thus, particle selection no particles leave the two-loop reactors process with
effects must be considered. yields less than about 7000 g/g for the low recycle
The influence of the outlet solids content on the case.
production and yield of two-loop reactors in series By varying the feed rate of catalyst to the VSBR
was found to be as expected. Higher production and and hence changing the mean residence time of the
yields are obtained with higher solids level. Of course, catalyst, the shape of the yield distribution changes
there is a practical limit to the solids content as dramatically (cf. Fig. 20). For high catalyst flow rates
dictated by process operation and safety. (very low residence times) catalyst deactivation is not
very significant and, therefore, the catalyst weight
5.4. Polymer properties yield distribution tends to approach the exponential
5.4.1. Yield distribution. Figure 17 shows the dis- decay shape of the residence time distribution curve.
tribution of yields obtained by each of the continuous [When deactivation is not considered at all (not
processes studied. These curves are based on catalyst shown) the yield distribution curve actually mimics
Multistage polymerization of olefins--I 4877
1.64 1.6e-4
-- Low Raeyele Ratio = 5 !-- qcat = 0.05 g/s ; ~ = 5.979 hr
....... High R~eyele Ratio .... qeat = 0.Hi g/s ; ~= 3.471 hr [ I
1.4e-4 1.4e-4 - - - - qcat = 0.20 g/s ; ,w= 2.096 hr [ I
\ r .... q.,=0,0,, .... ,,,2~p /
1.2e-4 1.2e-4

= 1.0e-4 = 1.0e-4

~ 8.0e-5 ~ 8.0e-5

g
6.0e-5 6.0e-5

4.0e-5 4.0e-5

2.0e-5 2.0e-5

00e+0 0.0e+0
0 1O1300 20000 30000 40000 50000 60000 0 5000 10000 15000 20000 25000
Catalyst Yield ( g / g ) Ca a yst Yield I g / g )

Fig. 18. Catalyst yield distribution for a single loop reactor Fig. 20. Effect of catalyst feed rate on mean catalyst yield of
(base case). the VSBR.

are related to the presence of a second stable site. In


4e-5 -- 2 Loops, Low Recycle Ratio = 5 [
..... 2 Loops, High R~eycle Ratio this case the mean residence time of the catalyst is
1
longer than the deactivation time constant of the first
site, which causes a positive increasing slope to the
3e-5
curve. However, the mean residence time is shorter
than the deactivation times for the second site causing
the downward trend at high yields. Thus, this curve
2e-5 represents a hybrid of the two extreme cases. Perhaps
other kinetic schemes with more than two sites can
lead to other exotic distributions.

le-5
5.4.2. Porosity distribution. Porosity distributions
(weight of polymer basis) at base case conditions are
related to the yield distributions and shown in Fig. 21.
Oe+O !" Since the productivities of the loop reactor are the
10000 20000 30000 40000 50000 60000 highest, polymer from this reactor has the lowest
Catalyst Yield ( g / g ) porosity. This can be important for the addition of an
ethylene-propylene copolymer phase for impact
Fig. 19. Catalyst yield distribution for two-loop reactors in copolymer as well as the effect on monomer diffusion
series (base case).
limitation, monomer devolitization and additive addi-
tion in downstream units. Note that the selection
the residence time distribution curve scaled for yield.] factor for the FBR leads to a sharp peak at the lowest
For very low catalyst feed rates (high residence times) porosity. Figures 22 and 23 show the effect of the
catalyst deactivation effects dominate and a large recycle ratio on the porosity distributions of a single-
proportion of the catalyst leaves the reactor with little loop reactor and two-loop reactors in series, respec-
remaining activity. The shape of the yield distribution tively. Again the oscillatory nature of the residence
curve then tends to approach infinity at the maximum time distribution is apparent at low recycles. In the
catalyst yield. For intermediate feed rates, transitional case of two loops in series the damped oscillations
curves are found. might not even be detected experimentally.
The results of Fig. 20 are explained as follows. The
shape of the yield distribution curve for the VSBR is 5.4.3. Particle size distribution. Steady-state weight
dependent on the relative relationship between cata- particle size distributions for the base case of the
lyst deactivation and mean catalyst residence time in VSBR, HSBR, FBR and two loops (high recycle) are
the process. When the catalyst mean residence time is shown in Fig. 24. On a polymer wcight basis the PSDs
shorter than the time constant for deactivation, the for the VSBR and HSBR are very similar. This is
curve is a decay type that asymptotically approaches because the weight contribution of the particles leav-
the maximum yield. When the catalyst mean residence ing the VSBR at small times is not very significant.
time is greater than the decay times, the curve is Particle size selection effects in the FBR tend to nar-
positive exponential and tends to infinity at the max- row the particle size distribution compared to the
imum yield. Transitional curves showing a maximum VSBR and HSBR gas-phase reactors, and reduces the
4878 J. J. ZACCA et al.

20 3.0e-3
-ff
I I - VSBR -- VSBR
]] ....... HSBR /,, ......... HSBR
t'~ ] ~ FBR
2.5e-3

151 f'. o= 2.0e-3

~ 1.5e-3

~" 1.0e-3

//\ \,

0 ~ i i
5,0e-4

O.Oe+O
J
0.15 0.20 0.25 0.30 0~35 0.40 0.45 0.50 ) 5OO 1000 1500 2000 2500 3000
Particle Porosity ( void fraction ) Particle Radius ( microns )

Fig. 21. Particle porosity distribution for base case. Fig. 24. Weight particle size distribution for base case.

10

I -- Low Recycle Ratio = 5


......... High Recycle Ratio fraction of large size particles. For two-loop reactors
in series the PSD is broader and larger particle sizes
can be obtained.
Particle size distributions may be measured on
a polymer weight basis by mechanical sieving
6 or a particle number basis by scanning electron
microscopy. A comparison of the number and weight
basis particle size distributions shows that these
distributions are not the same. The weight basis al-
g.
ways gives a higher mean size than the number basis
(cf. Fig. 25).
2 For a single-loop reactor the number particle size
distribution (cf. Fig. 26) indicates that with lower
recycle ratios narrower particle size distributions are
0 obtained and no particles will leave the reactor with
0.15 0.20 0.25 0.30 0.35 0.40 0.45 0.50 short residence times. This is not true of the loop at
Polymer particle Porosity ( void fraction )
high recycle ratios. On a polymer weight basis, how-
ever, Fig. 27 shows that the particle size distributions
Fig. 22. Particle porosity distribution for single-loop reactor.
of the loop at low and high recycle ratios are practic-
ally indistinguishable from each other. For two loops
,"~"~. [ -- Low Recyde Ratio ='~q in series, the particle size distributions at low and high
........ High Recycle Ratio
recycle ratios are even closer.
8 The effect of varying the spread of the catalyst
particle size distribution on the polymer particle size
distribution of the VSBR has the obvious effect of
6 broadening the polymer particle size distribution
(Fig. 28). No changes in production or yield are noted
for the loop, VSBR and HSBR since diffusion was not
considered. Slightly higher weight average particle
sizes are obtained for the catalyst with larger spread
parameters since the log normal average size was
2 preserved, not the catalyst weight average size.
Particle size selection effects cause the external par-
ticle size distribution of the FBR to be different than
0 the internal particle size distribution as illustrated in
0.15 0.20 0.25 0.30 0.35 0.40 0.45 0.50 Fig. 29. Size selection causes the external particle size
Polyr~r particle Porosity ( void fraction ) distribution to produce larger particles than the inter-
nal distribution. The particle size distribution from
Fig. 23. Particle porosity distribution for two loops in series. the FBR is also narrower than the other reactors.
Multistage polymerization of olefins--I 4879

- - VSBR- Wright Flzcdon 4.0c-3


2.0e-3 ....... VSBR - N u m l ~ r Fraeti<m -- ~== 0.25j
..O- 2 Loq~- l-lighR~yde - Ntna~rFriction 3.5e-3 .... o" a= 0.10
--II.-- 2 Loops- High R ~ y d e - Weight Fraction --- O'c= 0.50

3.0e-3
1.5e-3
g
'~ 2.5e-3
u.
== 2.0e-3
1.0e-3
1.5e-3 . : ~ ~ ~

1.0e-3 t
5.0e-4

5.0e-4 1

O.Oe+O
.~'"~ - _-
0 500 1000 1500 2000 2500 3000 3500 4000
500 1000 1500 2000 25C0 3000 particle Radius ( microns )
Particle Radius( microns )

Fig. 28. Effect of catalyst particle size distribution on poly-


Fig. 25. Comparison of number and weight particle size mer particle size distribution of the VSBR (log mean catalyst
distribution for base case. radius = 33.4 #m).

1.6e-3
-- LowRceydRatio= 5 3.0e-3
1.40-3

1.2e-3 2.5e-3

1.0o-3
~ 2.0-3
8.0o-4
1.5-3

z 4.00-46"00-41" "a
1.0-3

2.0e-4
5.0e-4
O.Oe+O i i

0 500 1000 1500 2000 2500 0.0e+0 """'/ F


Particle Radius ( microns ) 500 1000 1500 2000
Particle RaclkL~ ( microns)
Fig. 26. Number particle size distribution for a single-loop
reactor (base case). Fig. 29. Comparison of the internal and external particle
size distribution of the for the FBR (base case).
1.4e-3

1.2e-3
Varying the spread of the catalyst particle size dis-
tribution has a pronounced effect on the production
1.0e-3
of the FBR (cf. Table 5). For the same log mean
catalyst particle size, the production and yield from
8.0e-4
g the FBR is considerably higher for the narrow cata-
lyst particle size distribution than the broad size
6.0e-4 distribution. By inspection of Table 5, it is seen
g. that the mean yield of the polymer inside of the
4.0e-4 reactor is larger than in the product stream for the
broad catalyst distributions. However, as the catalyst
2.0e-4
size distribution becomes narrower, the outlet catalyst
yield exceeds the internal yield and the performance of
O.Oe+O
the unit surpasses the VSBR. Simulations for three
0 500 1000 1500 2000 2500 3000
sizes of monodispersed catalysts were also performed
P a ~ c l e Radius (microns)
and reveal the outlet mean yield was always higher
Fig. 27. Polymer weight particle size distribution for a single- than the internal yield and productivities were always
loop reactor (base case). higher than cases for broader catalyst particle size
4880 J. J. ZACCAet al.
Table 5. Summary of results for the FBR

Effect fP (h) f (h) W~,t (g) Y (g/g) Yo (g/g) rp.cut(#m) Capacity


(t/yr)
Catalyst PSD
Narrow (goat= 0.1) 1.787 2.254 1623.0 10,494 13,238 967.5 79,322
Base case (trcat =0.25 ) 2.122 2.064 1485.8 11,453 11,150 833.4 66,806
Wide (O'cat = 0 . 5 ) 2.844 1.821 1310.9 13,009 8318 594.1 49,841
Monodisperse (rear= 16.7/zm) 1.819 2.699 1943.5 8,763 13,009 499.8 77,948
M o n o d i s p e r s e (teat = 33.4 vm) 1.712 2.725 1962.2 8,678 13,819 1000.1 82,799
Monodisperse (r~,t=66.8/~m) 1.630 2.780 2001.9 8,508 14,514 2008.1 86,966
Catalyst flow rate
qc,t= 0.05 g/s 6.438 5.783 1040.9 16,351 14,699 890.8 22,019
qcat=0.2 g/s (base case) 2.122 2.064 1485.8 11,453 11,150 833.4 66,806
qcat= 1.6 g/s 0.512 0.595 3427.6 4966.8 5769.8 701.3 276,574
Selection parameter (b)
b = 0.007054 #m- 1 (base case) 2.122 2.064 1485.8 11,453 1,150 833.4 66,806
b = 0.003527 #m -1 2.000 2.118 1524.7 11,076 1,837 894.2 70,926
b =0.001764/~m -1 1.987 2.131 1534.2 11,090 1,908 947.6 71,348

distributions. Monodispersed catalysts with larger 5e-3


sizes led to higher production rates. Simulations with
........ o = 0 . I 0
the largest catalyst size (66.8#m radius) reveal the - - - - a = = 0.50
FBR production rate exceeded that of the HSBR 4e-3

reactor.
The above results are explained by understanding .g
the role of size selection in the FBR. Consider a mono- ,.~ 3e-3

dispersed catalyst polymerizing in an FBR. Since size


selection implies larger polymer particles are preferen-
tially removed from the reactor, the catalyst particles ~ 2e-3

do not leave the reactor until they have grown suffi-


ciently in size and yield. Thus, the mean catalyst yield /
le-3
in the product leaving the reactor is always higher //
than inside the reactor. To maintain the proper vol-
ume of powder in the reactor, the concentration of 0e+0 J J J ~ ~ p I I
catalyst in the reactor is then higher than that of 500 1000 1500 2000 2500 3000 3500 4000
a CSTR. Thus, size selection aids the FBR by provid- Particle Radius ( microns )
ing a 'boost' in production. Larger catalyst particles
lead to larger 'boosts' in production since the selection Fig. 30. Effect of catalyst particle size distribution on the
factor is sensitive to the size of the polymer particles polymer weight particle size distribution of the FBR (log
mean catalyst radius = 33.4 #m).
[cf. eq. (46)] which are proportional to catalyst size
according to eq. (75). Increased cutting diameters and
outlet yields cause production levels to increase with Since size selection appears to have significant im-
catalyst radius. plications on the performance of the FBR, simulations
By broadening the particle size distribution of were performed by varying the sensitivity of selection
the catalyst, size selection begins to suppress the by changing the selection parameter 'b'. The effect of
productivity of the FBR. This is because larger poly- changing 'b' on the residence time distribution of the
mer particles no longer imply high yield catalyst FBR is shown in Fig. 31. As particle size selection
particles. Large polymer particles may also be gener- becomes less sensitive, more particles leave the reactor
ated from large catalyst particles with low yield. with shorter residence times. Under no size selection
It would then seem reasonable that utilization of the residence time distribution approaches the CSTR
large catalyst particles with narrow particle size case. The effects of size selection on the production of
distribution would be beneficial to the productivity the FBR are shown in Fig. 32. We find an optimal
of the FBR (Fig. 30). However, practical issues value of b for which the production of the FBR is
limit the maximum catalyst size that can be employed maximized, given the base case conditions and cata-
in the reactor. The potential for particle overheating lyst size distribution. At high values of the selection
greatly increases with catalyst size, for example parameter b, particle selection effects tend to decrease
(Hutchinson and Ray, 1987), and monomer diffusion production due to the withdrawal of large particles
limitations also become more significant (Hutchinson, with low polymer yield. At very low values of b some
1990). fraction of the particles will leave the reactor with
Multistage polymerization of olefins--I 4881
1.4e-4
--'" b=7.054x I04Coasec~se) I
i

1.8e-3 I .... ]
----- b = 3 , 5 2 7 x 10 d /
1.20-4 ...... b = 1.7635x 10 .3 ] 1.6e-3 - - - Largepor~ty changes(fl = 8 109x]O~) j
- - b=0
1.40-3
1.0e-4
:'k\ .~ 1.2e-3

~ 8.0e-5 ~ ,.oe-3

6.0e-5 ~ 8.0e-4
/ f
,~ a, 6.00-4
4.0e-5
4.00-4
It
2.00-5 I / / ~ 2.0e-4

/ ~ o.oo+o ! "*'#
-
I i i

O, Oo+O i j i 0 50O 1000 1500 2000 2500


0 5000 10000 15000 20000 25000 PartideSi~ (micmns)
ResidenceTime(s)
Fig. 33. Effectof porosity changes on the particle size distri-
Fig. 31. Effectof size selection parameter'b'on the residence bution of the VSBR reactor.
time distribution of a fluidized-bed reactor.

5.5. Effect of varyin 9 the kinetic parameters


1.07
To study the effects of changing the kinetic rate
constants has on the behavior of the processes
1.06 studied, simulations were performed by varying the
rate of site transformation reaction. Results are shown
1.05
for the VSBR in Fig. 34. Changing the site trans-
1.04 formation rate (by possibly operating at higher co-
catalyst concentrations) has a similar effect as chang-
1.03
,,:.. ing catalyst feed rate on the yield distribution. There is
1.02 a maximum yield based on catalyst deactivation and
it can be quite low for rapid catalyst deactivation.
1.01 Often, researchers use a one site plus simple first-
order decay to model their catalyst system. Metal-
1.00
locene catalysts may also be modeled in this manner.
0.99 Simulations were performed with the VSBR using
a one-site model fit to the data of Di Drusco and
0.98 i I [ I I I
Rinaldi (1984) and shown in Table 6 (see also Fig. 35).
0 1 2 3 4 5 6 7
Small differences are noted in the overall yield and
Size Selection P a r a m e t e r "b" x 10 3
production rates of the reactor compared to the two-
Fig. 32. Effect of size selection parameter 'b' on the relative site model of Fig. 34. However, there is no maximum
production of the FBR with respect to the VSBR. in the yield distribution curve because of the absence
of the second stable site.

short residence times and low polymer yield and the 5.6. Statistical properties of the distributions
FBR production reaches the levels of the VSBR. Statistics for the base case are shown in Table 7. It
Table 5 also illustrates the fact that for nonwell is important to recognize that different bases (catalyst
mixed reactors the mean residence time on a polymer weight, number of particles, weight of polymer) show
weight basis differs from the one calculated on a different results. It is important to choose the proper
catalyst weight basis. This is an important difference basis for calculation purposes. For instance, average
to keep in mind whenever average residence times of yield is more naturally calculated on weight of cata-
a reactor are calculated. lyst basis while average particle size is usually ex-
Additionally, in this paper we consider varying the pressed on a polymer weight basis. The standard
rate at which polymer particle porosity changes with deviation is the variable of choice to evaluate the
polymer yield. By varying this sensitivity it can be spread of the different distributions. Due to particle
seen that changes in porosity can have a significant size selection phenomena in case of the fluidized-bed
effect on the particle size distribution as shown in reactor, number-based statistical properties differ
Fig. 33. Often, however, researchers neglect these cha- from catalyst-weight-based statistical properties.
nges and attempt to correlate polymer yield with Among the gas-phase reactors the HSBR is the one
weight mean polymer particle size. The results of with the highest catalyst weight average yield since its
Fig. 33 indicate this may not be a valid measurement average residence time is the highest at base case
unless the changes in polymer porosity are small. conditions. The HSBR is the reactor with the smallest
4882 J.J. ZACCAet al.
porosity a n d yield distribution spreads because of its particle size distribution a m o n g all reactors, yet its
n a r r o w residence time d i s t r i b u t i o n relative to the yield distribution is broad. Low recycle ratios n a r r o w
o t h e r gas-phase reactors. However, as a consequence the distributions with respect to the well mixed case
of size selection the F B R possesses the n a r r o w e s t for the loop reactor. F u r t h e r n a r r o w i n g takes place
with the a d d i t i o n of one more reactor in series.

2.0e-4
i 1~ lqR= 1.193x 10~ (1 is)
kit2= 2.385 x 10 4 (1 Is) 6. CONCLUDING REMARKS
[ kal2= 4.770 x 10 4 (1 Is) In this paper, we have developed a general framework
1.6e-4 [/ k~I~= 2.385 x 10.3 ( 1/ s)
for p o p u l a t i o n balance modeling of solid-catalyzed
//' olefin processes using a comprehensive multisite
1.2e-4 kinetic model a n d including catalyst particle size
z

~ 8.0e-5 2.0e-4
-- No Deactwation
--- kal = 9.055 x 10"~sq
......... kdl = 1.811 x 104 S"l
4.0e-5 ---- kal = 3.622 x 10"+s"t
1.5e-4
I
II
0.0e+0
5000 10000 15000 20000 25000 30000 35000 40000
1.0e-4
i
t
Yield ( g/g ) 1
/
I ,;
Fig. 34. Effect of site transformation reaction on catalyst
yield distribution for the VSBR.
// /,,,"
5,0e-5 // J/

Table 6. Kinetic parameters used for the one site model

Kinetic parameter Value O.Oe+O ~


0 5000 10000 15000 20000 25000 30000 35000 40000
kill (l/mol/s) 269.1
Weld(g/g)
kd~ (l/s) 1.811 10 -4
0.40
To'0t 10 Fig. 35. Effect of catalyst deactivation rate on catalyst yield
distribution: one-site model for VSBR.

Table 7. Summary of results for the base case simulations

VSBR HSBR FBR 1 loop 1 loop 2 loops 2 loops


(high rec) (low rec) (high rec) (low rec)

Mean residence time (h) 2.096 2.317 2.064 1.016 0.979 1.711 1.630
Production (t/h) 67,620 83,500 66,806 119,000 120,600 173,900 174,124
N* avg mean yield (g/g) 11,270 13,940 13,925 19,860 20,120 29,020 29,060
N yield STD* (g/g) 5791 3269 6256.2 12,760 11,490 11,300 10,082
W: avg mean yield (g/g) 14,240 14,700 13,324 28,060 26,680 33,420 32,501
W yield STD (g/g) 4516 2922 5520.8 11,350 11,030 9722 9212
Cat s avg mean yield (g/g) 11,270 13,940 11,170 19,860 20,120 29,020 29,060
Cat yield STD (g/g) 5791 3269 6277.6 12,760 11,490 11,300 10,082
N avg mean porosity 0.386 0.36 0.363 0.328 0.323 0.265 0.264
N porosity STD 0.053 0.028 0.053 0.091 0.079 0.069 0.0591
W avg mean porosity 0.358 0.353 0.386 0.271 0.279 0.239 0.243
W porosity STD 0.039 0.025 0.051 0.07 0.069 0.053 0.0505
Cat avg mean porosity 0.386 0.36 0.367 0.328 0.323 0.265 0.264
Cat avg porosity STD 0.053 0.028 0.047 0.091 0.079 0.069 0.0591
N avg radius (/~m) 774 858 836.5 896.4 917.1 1035 1040
N radius STD (/~m) 259.5 227.5 158.4 315.7 285.9 295.7 285.9
W avg radius (/~m) 1029 1050 914.2 1238 1212 1304 1294
W radius STD (#m) 285.2 273.1 151.3 353.7 346.8 350.2 345.0

*N = Particle number based.


t STD = Standard deviation.
; W = Polymer weight based.
SCat = Catalyst weight based.
Multistage polymerization of olefins--I 4883
distributions and monomer diffusion. Catalyst resi- transition studies may be done by extending the
dence time was used as the internal coordinate and modeling work here.
the internal and external distributions may be differ- In addition to the obvious role of catalyst deactiva-
ent. The modeling framework may be applied to tion, a secondary effect is to dilute the differences
a process consisting of a train of reactors in series. The between the natural residence time distributions of
steady-state modeling of multistage processes for the different processes. We find the largest differ-
propylene polymerization has been shown. ences amongst the processes for cases where the resi-
It is shown that common industrial processes for dence time of the process is shortest (little catalyst
propylene polymerization can produce a product with deactivation).
a wide distribution of polymer properties. Processes Besides the properties calculated in this work, other
with narrow residence time distributions produce important properties can also be generated from the
a product with narrow property distributions. Signifi- residence time distribution. Polymer architecture
cant differences among the processes were noted. properties such as composition and molecular weight
With no diffusion resistances considered, catalyst par- distributions can be predicted. This will be addressed
ticle size distribution does not affect the production in a future publication (Zacca et al., 1996}.
and yield of the VSBR, HSBR or loop. However, Residence time distributions can have a pro-
catalyst particle size distribution does have a signifi- nounced effect on the properties of multi-layered
cant effect on the performance of the FBR. To im- products produced in a series of reactors. In each
prove the performance of the FBR it would appear reactor, a completely different polymer product may
advantageous to use a very narrow catalyst particle be formed by changing the composition of the reactor
size distribution. For monodispersed catalysts, larger fluid, temperature and pressure adding to the pre-
particle sizes lead to higher productivities in the FBR viously formed polymer. In a subsequent paper of this
compared to smaller particles. Larger polymer par- series, we look at the case of impact polypropylene
ticles are often desired for better fluidization, minimiz- produced by several commercial processes (Debling
ing fines carryover and product handling. However, et al., 1996a). It is seen that the residence time distri-
larger catalyst particles may be more prone to over- bution of the process leads to varying amounts of
heating and sticking in the reactor. Issues such as copolymer in the final product. This has implications
catalyst feed position, removal of fines, the presence of for polymer particle sticking, diffusion, composition
a disengagement zone will clearly have an effect on the distributions and can affect product quality. Also, as
performance of the fluidized bed. For the simulations a result of site transformation effects, catalyst particles
presented here, some conditions for maximizing pro- that remain in the homopolymer stage for long peri-
duction in the FBR were noted. Further optimization ods of time are essentially in the Ti 2 + state. Thus,
with the modeling work presented here is easily pos- when they enter the copolymer reactor, they produce
sible. a more ethylene-rich copolymer.
From the results of this study, it is clear that the size
selection features of the FBR greatly affects its perfor- Acknowled(4ements The authors are indebted to the Na-
mance. Thus, there is value in considering the rational tional Science Foundation, the Industrial Sponsors of the
design of size selection functions. Some loop processes University of Wisconsin Polymerization Reaction Engineer-
have settling legs which can favor larger particles; the ing Laboratory (UWPREL), O.P.P. Petroqu[mica S. A. and
Conselho Nacional de Desenvolvimento Cientlfico e Tec-
VSBR is an expanded bed and by adjustment of the nol6gico (CNPq) for their financial support of this research.
withdrawal tube one can achieve some degree of size
selection. A recent patent by Mitsubishi Chemical
Industries describes the use of external size selection NOTATION
equipment to return smaller particles to the reactor b size selection parameter, 1//~m
(Takashima et al., 1985). The modeling approach de- (7 vector of normalized catalyst site concen-
veloped here can be used as a tool to simulate and trations, dimensionless
design practical size selection equipment in order to C~ normalized concentration of dead sites,
improve process performance. dimensionless
It should be noted that simulation studies shown ~'p normalized concentration of potential
here are for reactors at steady-state model where sites, dimensionless
processes are usually maintained. However, many /)k normalized concentration of dead chains
processes are designed for the production of multiple of length n at site k, dimensionless
grades during a set cycle time. Minimizing the amount loop axial dispersion coefficient, cm2/s
of 'offspec' product produced during the transition f generic distribution representation,
to a different molecular weight or comonomer com- 1/property
position is important and the subject of several [H2] concentration of hydrogen at the sites,
articles (Debling et al., 1994; McAuley and Mac- gmol/l
Gregor, 1992). It is interesting to note that population k~ apparent rate constant for activation of
balance modeling can be used to study the effect of site type k, 1/s
changing grades on the complete distribution of prod- kcllj
k rate constant for chain transfer at site
ucts produced in the reactor environment. Grade type k, 1/mol/s
4884 J. J. ZACCAet al.
k~j apparent rate constant for deactivation e~i. minimum catalyst porosity, dimensionless
of site type k, 1/s 2 eigenvalue of matrix K
k~o, rate constant for initiation of site type k, /~. normalized zeroeth moment of polymer
l/mol/s chains ending in monomer i at site k,
k~,~ rate constant for propagation of mono- dimensionless
mer i at a chain ending with monomer /]ko normalized zeroeth moment of all poly-
j on site type k, 1/mol/s mer chains at site k, dimensionless
k~,' apparent rate constant for site trans- mathematical constant (3.14159 ... )
formation from type k to 1, 1/s PR, Peat, reactor, catalyst, polymer and liquid-
K matrix of apparent kinetic rate constants, Pp, Pt phase densities, respectively, g/cm 3
1/s a distribution normalized standard devi-
L loop tube length, cm ation
[Mi]b concentration of monomer i in the bulk acat catalyst particle size distribution spread
fluid, gmol/1 parameter,/~m
[Mi]~ sorbed concentration of monomer i at ~O size selection factor, dimensionless
the catalyst sites, gmol/l z catalyst residence time in reactor, s
MWl molecular weight of monomer i, g/gmol .~tot,.~, ~p total, catalyst and polymer weight mean
MWT, molecular weight of Titanium, g/gmol residence times, respectively, s
Nsites number of sites ~ effectiveness factor for monomer i,
Nmon number of monomers (dimensionless)
P matrix of right-hand eigenvectors of K
Pe Peclet number, dimensionless Acronyms
normalized concentration of sites of type CSTR continuous-stirred tank reactor
k and chain length n ending in monomer FBR fluidized-bed reactor
j, dimensionless HSBR horizontal-stirred bed reactor
normalized concentration of vacant sites PSD particle size distribution
of type k, dimensionless RTD residence time distribution
q, qeat, qp total, catalyst and polymer mass flow VSBR vertical stirred bed reactor
rates, respectively, g/s
QO loop volumetric discharge rate, cma/s Superscripts and subscripts
QRec loop volumetric recycle rate, cm3/s n number fraction of particles
rcat catalyst radius, #m p weight fraction of polymer
rp particle radius,/~m tot total mass fraction
rp, c u t particle cutting radius, #m w weight fraction of catalyst
r, rate of change of property with time, F reactor feed
property/s O outlet of reactor
Rec loop recycle ratio, dimensionless R inside the reactor
kinetic rate of mass consumption of catalyst radius
Feat
monomer i, g/g/s rp particle radius
A obs
Rpi observed rate of mass consumption of y polymer yield
monomer i, g/g/s e particle porosity
t chronological time, s catalyst residence time
loop tubular velocity, cm/s mean value
v~ reactor volume, cm 3 mol fraction with respect to the total
x~i fraction of titanium atoms that are cata- amount of catalyst
lyst sites, dimensionless
Ws solids weight fraction, dimensionless
REFERENCES
WTi weight fraction of titanium in the cata-
lyst, dimensionless Arriola, D. J., 1989, Modeling of addition polymerization
systems. Ph.D. Thesis, University of Wisconsin-Madison.
WR, Wb, Barbe, P. C., Cecchin, G. and Noristi, L. 1987, The catalytic
Wear, R total material, bed and catalyst weight in system Ti-complex/MgCl2, In Advances in Polymer
the reactor, respectively, g Science 81, 1. Springer, Berlin.
Y polymer yield in g polymer/g catalyst Blenke, H., 1979, Loop reactors. Adv. Biochem. Enong. 13,
Zij time-independent matrix 121-214.
Brockmeier, N., 1987, Cost savings achieved with latest poly-
propylene process technology. AIChE Annual Meeting,
Greek letters New York.
fl parameter used in porosity calculation Brockmeier, N., 1991, The AMOCO/Chisso gas phase poly-
Yi sorption factor for monomer i, dimen- propylene copolymer technology. SPE Polyolefins VII
International Conference, Houston.
sionless Brockmeier, N. and Koizumi, T., 1991, Design challenges for
e polymer particle porosity, dimensionless a gas-solid polymerization reactor that approaches plug-
eo initial catalyst porosity, dimensionless flow. Reactor Engineerin 9 Conference, Santa Barbara.
Multistage polymerization of olefins I 4885
Bukatov, G. D., Zaikovskii, V. 1., Zakharov, V. A., Furtek, A. B., 1993, Ultra strength polyethylene resins pro-
Kryukova, G. N., Fenelonov, V. B. and Zagrafskaya, R. V., duced in a fluid-bed process utilizing Metallocene-based
1982, The morphology of polypropylene granules and its catalysts. MetCon, Houston.
link with the titanium trichloride texture. Polym. Sci. Galli, P. and Ali, S., 1987, The Spheripol process: a versatile
U.S.S.R. 24, 599-608. technology for advanced polypropylene property mater-
Burfield, D. R., McKenzie, I. D. and Tait, P. J. T., 1972, ials. AIChE Annual Meeting, New York.
Ziegler-Natta catalysis: a general kinetic scheme. Polymer Galvan, R. and Tirrell, M., 1986, Orthogonal collocation
13, 302 326. applied to analysis of heterogeneous Ziegler Natta pol-
Caracotsios, M., 1992, Theoretical modelling of Amoco's gas ymerization. Comput. Chem. Engng 10, 77-85.
phase horizontal stirred bed reactor for the manufacturing Hattori, N., 1986, Gas phase polypropylene process. Chem.
of polypropylene resins. Chem. Engng Sci. 47, 2591-2596. Economy Engng Rev. 18, 21 26.
Carvalho, A. B. M., Gloor, P. E. and Hamielec, A. E., 1989, Heuer, T., Peuschel, P., Ratzsch, M. and Wohlfarth, C., 1989,
A kinetic model for heterogeneous Ziegler-Natta (co)pol- Acta Polymerica 40, 320.
ymerization. Polymer 30, 280 297. Himmelblau, D. M. and Bischoff, K. B., 1968, Process Analysis
Carvalho, A. B. M., Gloor, P. E. and Hamielec, A. E., 1990, and Simulations--Deterministic Systems, Wiley, New York.
A kinetic model for heterogeneous Ziegler-Natta (co)pol- Hoel, E. L., Cozewith, C. and Byrne, G. D., 1994, Effect of
ymerization. Part 2: stereochemical sequence length distri- diffusion on heterogeneous ethylene propylene
butions. Polymer 31, 1294-1311. copolymerization. A.I.Ch.E J. 40, 1669-1684.
Chen, C. M., 1993, Gas phase olefin copolymerization with tlulburt, H. M. and Katz, S., 1964, Some problems in particle
Ziegler-Natta catalysts. Ph.D. Thesis, University of technology. Chem. Engng. Sci. 19, 555-574.
Wisconsin Madison. Hussein, F. D. and Nemzek, T., 1989, UNIPOL P P - - l n -
Chien, J. C. W., 1987, Polymerization of olefins with magne- novation through combined technologies. AIChE Spring
sium chloride-supported catalysts, In Advances in Poly- National Meeting, Houston.
olefins- The WorM's Most Widely Used Polymers, pp. Hutchinson, R. A., 1990, Modelling of particle growth in
255- 280. Plenum Press, New York. heterogeneous catalyzed olefin polymerization. Ph.D.
Choi, K. Y. and Ray, W. H., 1985a, The dynamic behavior of Thesis, University of Wisconsin-Madison.
fluidized bed reactors for the solid catalyzed gas phase Hutchinson, R. A., Chen, C. M. and Ray, W. H,, 1992,
polymerization of propylene. Chem. Engng Sci. 40, Polymerization of olefins through heterogeneous cataly-
2261 2279. sis, X. Modeling of particle growth and morphology.
Choi, K. Y. and Ray, W. H., 1985b, Recent developments in J. Appl. Polym. Sci. 44, 1389-1414.
transition metal catalysed olefin polymerization a sur- Hutchinson, R. A. and Ray, W. H., 1987, Polymerization of
vey, 1. Ethylene polymerization. J. Macro. Sci. Revs. Mac- olefins through heterogeneous catalysis, VII. Particle igni-
ro. Chem. 25, 1 55. tion and extinction phenomena. J. Appl. Polym. Sci. 34,
Choi, K. Y. and Ray, W. H., 1985c, Recent developments in 657 676.
transition metal catalysed olefin polymerization a sur- Hutchinson, R. A. and Ray, W. H., 1991, Polymerization of
vey, 11. Propylene polymerization. J. Macro. Sci. Revs. olefins through heterogeneous catalysis, IX. Experimental
Macro. Chem. 25, 5 7 97. study of propylene polymerization over a high activity
Choi. K. Y. and Ray, W. H., 1988, The dynamic behavior of MgCl2-supported Ti catalyst. J. Appl. Polym. Sci. 43,
continuous stirred bed reactors for the solid catalyzed gas 1271-1285.
phase polymerization of propylene. Chem. Engng Sci. 43, lhm, S. K, Kang, K. S., Chu, K. J. and Chang, H. S., 1989,
2587 -2604. Kinetics of ethylene-propylene copolymerization over
Choi, K. Y., Zhao, X. and Tang, S., 1994, Population balance MgCl2-supported catalysts, In Catalytic Olefin Polymeriz-
modeling for a continuous gas phase olefin polymerization ation. Proceedings of the International Symposium on
reactor. J. Appl. Polymer Sci. 53, 1589 1597. Recent Developments in Olefin Polymerization Catalysts,
Debling, J. A., Han, G. C., Kuijpers, F., VerBurg, J., Zacca, J. Tokyo, pp. 263 275.
and Ray, W. H., 1994, Dynamic modeling of product Jaggard, J. F. R., 1989, Vertical stirred bed gas phase techno-
grade transitions for olefin polymerization processes. logy current status and future. AIChE National Meeting,
A.I.Ch.E.J. 40, 506-520. Houston.
Debling, J. A. and Ray, W. H., 1995, Heat and mass transfer Kang, S., Yoon P. and Lee, W. K., 1989, Steady state particle
cffects in multistage polymerization processes: impact growing with segregation in a fluidized bed granulator,
polypropylene. Ind. Engng Chem. Res. 34, 3466-3480. d. Chem. Engng Japan, 22, 272.
Debling, J. A., Zacca, J. J. and Ray, W. H., 1996a, Reactor Karol, F, J., 1984, Studies with high activity catalysts for
residence time distribution effects on the multistage pol- olefin polymerization. Catal. Rev.-Sci. Engng, 26, 557 -595.
ymerization of olefins -111: multilayered products--im- Kim, J. Y., 1995, Modeling of addition polymerization sys-
pact polypropylene. Chem Engng Sci., in press tems. Ph.D. Thesis, University of Massachusetts.
Debling, J. A., Zacca, J. J. and Ray, W. H., 1996b, Reactor Kissin, Y. V., 1985, lsospecific Isospeeific Polymerization of
residence time distribution effects on the multistage pol- Olefins with Heterogeneous Ziegler Natta Catalysts.
ymerization of olefins. Part IV: prepolymerization and Springer, Berlin.
particle overheating. Chem Engng Sci., in press. Kunii, D. and Levenspiel, O., 1991, Fluidization Engineering,
Demel. P., 1989, Modelling, simulation and control of a con- 2rid Edn. Butterworth-Heinemann, Boston.
tinuous stirred bed reactor for the gas phase polymeriz- Lagemann, B., 1989, Modelling, simulation and control of
ation of olefins. M.S. Thesis, University of Wisconsin- a fluidized bed reactor for the gas phase polymerization of
Madison. olefins. M.S. Thesis, University of Wisconsin-Madison.
Di Drusco, G. and Rinaldi, R., 1984, Polypropylene Lancaster, P. and Tismenetsky, M., 1985, The Theory qf
process selection criteria. Hydrocarbon Processing 55, Matrices, 2nd Edn. Academic Press, New York.
113 117. McAuley, K., 1991, Modelling, estimation and control of
Ferrero, M. and Chiovetta, M. G., 1987, Catalyst fragmenta- product properties in a gas phase polyethylene reactor.
tion during propylene polymerization: Part 1. The effects Ph.D. Thesis, McMaster University.
of grain size and structure. Polymer Engng Sei. 27, McAuley, K. and MacGregor, J.F., 1992, Optimal grade
1436 1447. transitions in gas phase polyethylene reactors. A.I.Ch.E. ,l.
Floyd, S. and Ray, W. H., 1986, Polymerization of olefins 38, 1564.
through heterogeneous catalysis III. Polymer particle Min, K. W., 1976, Modelling and simulation of emulsion
modeling with an analysis of intraparticle heat and mass polymerization reactors. Ph.D. Thesis, University of Wis-
transfer effects. J. Appl. Polym. Sci. 32, 2935-2960. consin-Madison.
4886 J. J. ZACCAet al.
Ogunnaike, B. and Ray, W. H., 1994, Process Dynamics, Shinnar, R. and Rumschitzki, D. 1989, Tracer experiments
Modeling, and Control. Oxford University Press, New York. and RTD's in heterogeneous reactor analysis and design.
Prindle, J. C., 1989, Dynamics and stability of emulsion A.I.Ch.E J. 35, 1651-1658.
polymerization reactors. Ph.D, Thesis, University of Wis- Simonazzi, T., Cecchin, G. and Mazzullo, S., 1991, An out-
consin-Madison. look on progress in polypropylene based polymer techno-
Ramkrishna, D., 1985, The status of population balance logy. Prog. Polym. Sci. 16, 303-329.
models. Rev. Chem. Engng. 3, 45-95. Takashima, R., Goko, N., Uehara, Y., Nishihara, Y. and
Rawlings, J. B., 1985, Simulation and stability of continuous Yamamoto, H., 1985, U.S. Patent No. 4,492,787 issued to
emulsion polymerization reactors. Ph.D. Thesis, Univer- Mistsubishi Chemical Industries.
sity of Wisconsin-Madison. Talbot, J., 1990, The dynamic modelling and particle effects
Rincon-Rubio, L. M., Wilen, C.-E. and Lindfors, L.-E., on a fluidised bed polyethylene reactor. Ph.D. Thesis,
1990, A kinetic model for the polymerization of propylene Queen's University.
over a Ziegler-Natta catalyst. Eur. Polym. J. 26, 171-176. Tang, S., 1986, Studies on olefin polymerization, In Catalytic
Ross, J. F. and Bowles, W. 1985, An improved gas phase Polymerization of Olefins (Edited by Keii, T. and Soga, K.),
polypropylene process. Ind. Engng Chem. Product Res. pp. 165-179. Elsevier, New York.
Develop. 24, 149-154. van der Burg, M. W., Chadwick, J. C., Sudmeijer, O. and
Sarkar, P. and Gupta, S. K., 1991, Modeling of propylene Tulleken, H. J. A. F., 1993, Probabilistic multi-site model-
polymerization in an isothermal slurry reactor. Polymer ling of propene polymerization using Markov models.
32, 2842-2852. Makromol. Chem. Theory Simul. 2, 399-421.
Sawin, S. and Baas, C., 1985, Unipol PP: a gas-phase route Zacca, J. J., 1995, Distributed parameter modelling of the
to polypropylene. Chem. Engng. 92, 42-43. polymerization of olefins in chemical reactors. Ph.D. The-
Schotte, W., 1982, Ind. Enong Chem. Process Des. Dev. 21, sis, University of Wisconsin-Madison.
289. Zacca, J. J., Debling, J. A. and Ray, W. H., 1996, Reactor
Schill, D and Buchner, O., 1988, BASF Gas Phase Novolen residence time distribution effects on the multistage pol-
Process: a Better Route to Polypropylene Impact ymerization of olefins--II: polymer properties--bimodal
Copolymers. BASF Aktiengeseilschaft. polypropylene and linear low density polyethylene. Chem.
Shinnar, R., 1987, Use of residence and contact time distribu- Engng Sci., submitted.
tions in reactor design, In Chemical Reaction and Reactor Zacca, J. J. and Ray, W. H. 1993, Modeling of the liquid
Engineering (Edited by Carberry, J. and Varma, A.), pp. phase polymerization of olefins in loop reactors. Chem.
63-149. Marcel Dekker, New York. Engng Sci. 48, 3743-3765.

You might also like