Chemical Engineering Science 66 (2011) 5474–5486

Contents lists available at ScienceDirect

Chemical Engineering Science

journal homepage: www.elsevier.com/locate/ces

Controlled peroxide-induced degradation of polypropylene in a twin-screw

extruder: Change of molecular weight distribution under conditions
controlled by micromixing
P.D. Iedema a,n, K. Remerie b, M. van der Ham b, E. Biemond b, J. Tacx b
a
Universiteit van Amsterdam, the Netherlands
b
Sabic Europe, Geleen, the Netherlands

a r t i c l e i n f o a b s t r a c t

Article history: Controlled degradation of polypropylene (PP) by peroxide was carried out in a laboratory twin-screw
Received 24 December 2010 extruder ZSK 18 and the change in Molecular Weight Distribution (MWD) was measured using Size
Received in revised form Exclusion Chromatography–Differential Viscosimetry (SEC–DV). The MWD results were compared to
6 May 2011
MWD predictions from a kinetic model developed and validated in earlier work (Iedema et al., 2001)
Accepted 14 June 2011
assuming ideal mixing. Clear deviations were observed – the measured MWD was broader – that could
Available online 11 August 2011
only be explained by unaccounted heterogeneity in the extruder. Incorporating the relatively narrow
Keywords: Residence Time Distribution (RTD) in the twin-screw extruder did not lead to MWD broadening. In
Polymer processing contrast, the exponential RTD of a Continuously Stirred Tank Reactor (CSTR) yielded a MWD widening
Reaction engineering
that was too extreme. A new micromixing model, based on the striation thinning model by Ottino
Mathematical modeling
(1980), was constructed partly based on Monte Carlo sampling using a monomer scission probability
Mixing
Polypropylene (Tobita, 1996). This model was adapted to the geometry of the extruder entrance and the peroxide feed
Twin-screw extruder practice consisting of introducing a few per thousand peroxide-rich PP particles among pure PP
particles. This micromixing model indeed allowed obtaining very good matches between measured and
modeled MWD. Under different experimental conditions with respect to initial PP quality and amount
of peroxide added, with a constant value for the striation thinning parameter the errors between
measured and predicted MWD were around 5%.
& 2011 Elsevier Ltd. All rights reserved.

1. Introduction become even more distributed by such macroscopic inhomogene-

The controlled degradation of Polypropylene (PP) using per-
oxides is extensively used in the Polyolefin industry to optimize
both its processing capabilities and the end use properties desired
by their customers. At the same time processing conditions form
an interesting situation for Chemical Engineering from a selectiv-
ity point of view, since fast reactions of peroxide radicals are
paired to a relative slow diffusion of these highly reactive initiator
radicals through the polymer melt. It is interesting, since possibly
the reactions proceed under spatially inhomogeneous conditions.
In the area of Polymer Reaction Engineering this problem has
been addressed for macroscopically inhomogeneous systems
(Wells and Ray, 2005; Schmidt et al., 2005), where flow condi-
tions create compartments with different conditions. Distributed
micro-structural properties such as the molecular mass, them-
selves possessing a certain spread in properties, are seen to
n
Corresponding author. Tel.: þ31 20 525 6484; fax: þ31 20 525 5604.
E-mail addresses: piet@science.uva.nl, p.d.iedema@uva.nl (P.D. Iedema).
ities. The same holds for inhomogeneities at a more microscopic
level, as shown by our earlier publications on the controlled
degradation of Polypropylene in a Static Mixer Reactor (SMR)
(Fourcade et al., 2001a,b). Both macroscopic and microscopic
inhomogeneities originate from the joint effect of flow and
diffusion, so attempting the detailed mathematical description
of these phenomena using Computational Fluid Dynamics (CFD) is
one serious manner in solving the problem (McKenna et al.,
1998). However, this approach is still highly challenging as it is
faced with a number of serious problems, like the non-Newtonian
rheology of the polymer melt and the small scale of especially
micromixing phenomena.
Dealing with micromixing in the framework of CFD requires sub-
grid modeling, which has been applied to polymerization of low-
density Polyethylene (ldPE) using probability density functions to
describe the chemistry (Tsai and Fox, 1996; Kolhapure et al., 2005).
Accounting for micromixing phenomena has not frequently been
applied in Polymer Reaction Engineering. An early example using
compartment modeling to describe the micromixing effect, again for
ldPE, is given in Marini and Georgakis (1984), and a more recent one

0009-2509/$ - see front matter & 2011 Elsevier Ltd. All rights reserved.
doi:10.1016/j.ces.2011.06.071
 P.D. Iedema et al. / Chemical Engineering Science 66 (2011) 5474–5486
by Nele and Soares (2002), who suggest micromixing to be the cause
of enhanced branching in metallocene-catalyzed Ethylene polymer-
ization. Micromixing has scarcely been applied to polymer degrada-
tion. Note that the mixing addressed in a recent paper on PP
degradation in a twin-screw extruder (Suresh et al., 2008) is
concerned with macroscopic mixing. In a series of publications on
the controlled degradation of PP in a Static Mixer Reactor (SMR) we
did systematically address the micromixing phenomenon. There we
followed – as we will do in the present publication – the approach
developed by Ottino (1980) and Fields and Ottino (1987) using the
concept of ‘‘striation thinning’’. According to it the complex
3-dimensional flow and convective diffusion problem is reduced
to a one-dimensional problem with a limited number of parameters,
like the striation thinning rate, a. At one point in the whole
procedure we ‘averaged’ the 3D-flow (Newtonian) as computed by
CFD and thus inferred the parameter a in a computational manner.
In addition, a was experimentally determined by Laser Induced
Fluorescence (LIF). When finally applied to experiments with con-
trolled PP degradation in the SMR we employed a as an empirical
parameter, obtained by fitting full MWD’s on measured ones (SEC-
DV). Micromixing turned out to have a strong impact on the MWD
shape, which deviates considerably from the shape obtained under
ideal mixing conditions.
The objective of this paper is to demonstrate the effect of
micromixing on PP degradation in a twin-screw extruder. The
experiments performed in the ZSK 18 are closer to the industrial
practice than those with the SMR in our earlier work that has not
been used in industry for PP degradation until today. However,
here again a measurable effect of micromixing is not unimagin-
able, since the addition of the peroxide happens in solid form by a
few pp particles possessing a peroxide coating, which may lead to
locally high initiator radical concentrations (see Fig. 1). The
modeling approach is comparable to the one followed in our
earlier work, although we replaced the deterministic approach to
compute MWD’s by an extremely fast Monte Carlo sampling
method (Tobita, 1996).
This paper is structured as follows.
First, a brief description of the laboratory extruder setup and
the materials and measuring equipment used is given. The
modeling chapter starts with the kinetic equations used culmi-
nating in the scission ‘density’ that is the central parameter of the
Monte Carlo scission algorithm. This parameter possesses a
distribution of values according to the RTD in a Continuously
Stirred Tank Reactor (CSTR). Then the way the laminar micro-
mixing model is applied to the twin-screw extruder is explained
in terms of a 3-dimensional description of the entrance section of
the extruder. Subsequently, the partial differential equations
describing micromixing and reaction and their boundary condi-
tions are presented. The solution of these equations provides the
profiles of scission probabilities across the striations, which
represent the input for the MC scission simulation procedure.
Section 4 starts with a presentation of a quantitative measure
of agreement between measured and predicted MWD. Then the
Initially selected monomer unit
nleft nright
ns,left (from exp (-ρs)) > nleft ns,right (from exp (-ρs)) < nright
no scission scission
Fig. 1. Scission procedure. A linear chain is sampled from the weight distribution
pw
n ; on this chain an arbitrary monomer unit is selected; lengths left and right are
compared to sampled values from the scission probability function psn ¼ exp(  rsn).
Shortest pair of lengths decide whether scission event(s) took place and are added
to yield final length.
5475
effect of RTD in a CSTR and a twin-screw extruder on the scission
probability distribution is shown as a reference to compare the
micromixing results with. Finally, the effect of micromixing on
the MWD is presented. Proper values for the striation thinning
parameter a are obtained from the best fits of the MWD. These
prove the broadening of the MWD due to micromixing as
compared to ideal mixing.
2. Experimental
Polypropylene with molecular weights Mw ¼370, 470, 590 and
880 kg/mol and Mn ¼70, 105, 130, 245 kg/mole, respectively, were
used as starting base powder obtained from Sabic Europe. This
starting powder material was of high purity and contained no
additives. The base powder was compounded adding 0.2 wt%
Irganox B225 as antioxidant and Luperox 802PP40 as peroxide
to control the molecular weight. The stabilizer and Luperox
(concentrations of 0–0.5 wt%) were mixed with the pure PP base
powder at SABIC-Europe using a Henschell mixer. This mixture
was dosed to the hopper of a ZSK18 co-rotating twin-screw
extruder with a length/diameter ratio of 40. In all experiments a
throughput of 2.5 kg/h was maintained at a rotational speed of
200 rpm. The temperature of the barrel is measured at various
places along the screw axis as shown in Fig. 4 and increases to
240 1C. The temperature profile shown is assumed to be kept
during all experiments. The temperature of the polymer melt at
extruder exit was also measured and turns out to be slightly
higher: 242 1C. These temperatures were chosen to ensure high
peroxide conversion. The residence time distribution was mea-
sured by introducing a PP grain with black colorant into the
extruder and measure the time of first coloring of the exiting melt
(td in Eq. (14)) and observing the color intensity change. Typically,
under experimental conditions the residence time varied from 37
to 70 s (see Fig. 7).
3. Modeling
3.1. Kinetics and ideally mixed batch reactor and CSTR
The kinetic model we use has been described in our previous
publications. The reaction mechanisms involved are listed in Table 1,
the pre-exponential factors and activation energies, if available, are
listed in Table 2. In the previous work we employed a deterministic
population balance approach to compute the MWD. Here, however,
we decided to utilize a Monte Carlo sampling method, for reasons of
computational speed. The method has been introduced by Tobita
(1996). The method requires knowledge of the scission probability
of each monomer–monomer bonding (probably at the secondary
radical sites, Camara et al., 2006), also called the scission density,
denoted by, rs. The Monte Carlo sampling procedure will be
explained for both a batch reactor and a CSTR. The latter reactor
type, as it has a much broader residence time distribution (RTD)
Table 1
Reaction mechanisms.
Reaction Equation
Initiation kd
I!2I
Recombination of initiator radicals kc
I þ I !IH
Hydrogen abstraction followed by scission kht n
Pn þ I !Pm

þ Pn-m þ IH
Transfer to polymer followed by scission ktr Um
Pn þ Pm !Pn þ Pml

þ Pl
Disproportionation termination ktd
Pn þ Pm

!Pn þ Pm
5476 P.D. Iedema et al. / Chemical Engineering Science 66 (2011) 5474–5486

Table 2
Kinetic coefficients.

Reaction Coefficient Pre-exponential factor Activation energy/R (K) Units

Decomposition Luperox 802PP40 kd 7.7  1015 18,600 s1

Recombination reaction initiator radicals kc 2  1017 6000 m3k mole  1 s  1
Hydrogen abstraction ki 2.7  105 – m3k mole  1 s  1
Transfer to polymer ktr 1.5  104 5998 m3k mole  1 s  1
Termination by disproportionation ktd 5  104 10 m3k mole  1 s  1

than a twin-screw extruder, might therefore seem a less relevant the latter equality using Eq. (2c). Finally, the scission probability
reference system, but we will nevertheless take it into consideration, now follows as the sum of two contributions, one from the
for it enables to clearly demonstrate the effect of RTD on the hydrogen abstraction and one from the transfer to polymer
molecular weight distribution. Before discussing the MC scission reaction, both supposed to be instantaneously followed by scis-
procedure, we will describe how the scission density is derived from sion (see Table 1). For the batch reactor it is given by the integral
the kinetic equations governing both ideally mixed reactor types. between 0 and the residence time in the extruder, t
The balance for the initiator concentration, with initiator feed Z t
concentration I2f and average residence time for a CSTR, tCSTR , rs ¼ ðkht I þ ktr l0 Þdt ð4aÞ
0
reads
As regards the CSTR we may formulate the average scission
dI2 1
¼ 2kd I2 þ ðI I Þ ð1aÞ probability in terms of the average residence time,tCSTR , as
dt tCSTR 2f 2
rs ¼ ðkht I þktr l0 ÞtCSTR ð4bÞ
Here, t represents the residence time of a fluid element in the
extruder. In a batch reactor without inflow and outflow the initiator Note that the manner in which the scission probability is used
concentration is seen to decay at a first-order rate according to in the sampling procedure CSTR requires a little more explana-
tion, to be given in a following section.
I2 ðtÞ ¼ I2 ð0Þexpfkd ðtÞtg ð1bÞ
I2(0) is the initial initiator concentration. The dissociation coefficient 3.2. Monte Carlo sampling to describe random scission
kd coefficient becomes a strong function of residence time, if the
temperature profile imposed is accounted for, due to the fairly large The procedure starts by sampling a primary polymer from the
activation energy for initiator dissociation (Table 1). In a CSTR at chain length weighted length distribution of PP, pw n . This is
steady-state Eq. (1a) yields for the initiator concentration equivalent to sampling one monomer unit out of all monomer
I2f units present in polymer chains. Then, an arbitrary monomer unit
I2 ¼ ð1cÞ
1 þ kd tCSTR on this chain is selected and the segments to the right and left of
this unit, having lengths nr and nl, are checked for an eventual
The change of initiator radicals concentration follows as the
scission point (Fig. 1). In order to detect scission events the
dynamic balance between production through initiator dissocia-
lengths at RHS and LHS are to be compared to lengths from a
tion and consumption by the hydrogen abstraction reaction:
scission length distribution that is based upon the scission prob-
dI 1  ability, rs. The probability that the unit adjacent to the initially
¼ 2kd I2 kht I m1 kc ðI Þ2  I ð2aÞ
dt tCSTR selected one has not undergone scission and hence is connected
For the batch reactor this yields equals (1  rs); for the subsequent one it is (1  rs) (1 rs); and
for the ns-th one to be connected the scission length distribution

dI function is
¼ 2kd I2 kht I m1 kc ðI Þ2 ð2bÞ
dt
ps ðns Þ ¼ ð1rs Þns rs  expðrs ns Þ ð5Þ
Here, all quantities are functions of residence time, except for the
total number of monomer units in polymer chains, m1, which Eq. (3) expresses the probability that given the scission
remains unchanged. Also, if the batch reactor operates non- probability rs, the length of a scission fragment at RHS or LHS is
isothermally, kinetic coefficients are a function of residence time. exactly ns. Note that this is a chain length weighted distribution,
In the CSTR case we gets since it is based on sampling of monomer units. Furthermore, the
qffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi distribution of fragments according to Eq. (3) would be found in

kht m1 þ1=tCSTR þ ðkht m1 þ 1=tCSTR Þ2 þ 4kc kd I2 practice, if scission would occur to linear chains of infinite length.
I ¼ ð2cÞ
2kc In the present case of PP with a given MWD undergoing scission,
length values sampled from the exponential distribution of Eq. (5)
The macroradical concentration results from production by
are compared to the RHS and LHS lengths as obtained in the above
hydrogen abstraction and consumption by disproportionation
described manner. If the scission length is shorter than the chain
termination
part at a certain side, scission was a real event, and the length of
dl0 1 that chain part assumes the shorter value. If it is longer, no
¼ kht I m1 ktd ðl0 Þ2  l , ð3aÞ
dt tCSTR 0 scission took place, and the length of the fragment at that side
which in the batch case gives retains its original value. Thus, a population of chain lengths
sampled from the original length distribution is processed yield-
dl0
¼ kht I m1 ktd ðl0 Þ2 ð3bÞ ing a population of chains having undergone scission.
dt Now, the procedure for a CSTR is slightly more complicated
and for the CSTR than that for the batch reactor, since we have to account for the
sffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi residence time distribution of individual chains in the former
kht I m1 reactor type. Chains with a longer residence time are longer
l0 ¼ ð3cÞ
ktd exposed to the conditions causing scission. The scission probability
 P.D. Iedema et al. / Chemical Engineering Science 66 (2011) 5474–5486
as expressed by Eq. (4a) is also valid for a CSTR, but as the (macro-
)radical concentrations I and l0 now have assumed constant
values, so the scission density becomes a linear function of
residence time
Z t
rs ðtÞ ¼ ðkht I þktr l0 Þdt ¼ ðkht I þ ktr l0 Þt ð4cÞ
0 aðX,tÞx
Thus, instead of the constant value for the scission probability
in a batch reactor, we have a distribution of scission probabilities
according to the distribution of residence times, t, which in the
case of a CSTR has an exponential shape
Pðrs Þ ¼ ðkht I þ ktr l0 Þt expðt=tCSTR Þ ð4dÞ
This implies that the MC scission procedure for the CSTR
requires an extra sampling step: that from the residence time
distribution, yielding a value for t and subsequently one for rs.
Note that sampling of t happens independently from sampling of
the length, n. Furthermore, the value obtained for rs holds for the
length comparisons at both RHS and LHS. The MC scission
procedure for batch reactor and CSTR proceeds very fast; a
population of 1 million chains takes around 1 s of CPU-time.
The method is compared with the deterministic procedure based
on the Galerkin-hp method of PREDICIs (Wulkow, 1996) and
identical results were obtained.
3.3. Application of laminar micromixing model to twin-screw
extruder
The melting of solid particles under the conditions at the
extruder entrance is an extremely complex problem to describe
exactly. The micromixing model as we will discuss in this section
is a simplification of the real process. This model must be
regarded as an engineering approach to quantitatively describe
the competition between a fast dissociation reaction and relatively
slow diffusion of low-molecular compounds. We do not claim that it
is a correct 3D description of the rheologically complex flow
around a melting and deforming object.
Laminar micromixing approaches such as developed by Ottino
(1980) and Baldyga et al. (1998) employ a Lagrangian framework
for the description of simultaneous deformation and diffusion in
fluid elements. In the lamella-mixing model, the flow is seen as a
population of microflow elements Sx, in which the flow field is
homogeneous. In each of these microflow elements the mixture
has a lamellar structure, which is described by so-called striations
with a certain thickness (Fig. 2). This approach enables to describe
the complex folding and deformation process in 3 dimensions
(x1, x2, x3 in Fig. 2) in terms of a 1-dimensional (x) convective
x3
x1
x2
SX
Fig. 2. Microflow element with an ideal lamellar structure of the mixture and
definition of striation thickness, s.
5477
diffusion equation. At Sx scales, the evolution of the concentration
of a species, s, while following Sx in the flow (Lagrangian
description), is given by the convection diffusion equation with
reaction source term
@Cs @Cs @ 2 Cs
¼ Ds 2 þ Rs ð6Þ
@t @x @x
where x is the direction perpendicular to the striations, Cs is the
concentration of the species, Rs is the rate of production of the
species from chemical reaction, Ds is the diffusion coefficient, and
a is the rate of striation evolution (thinning in the case of mixing)
in Sx. The lamella-mixing model employs non-adjustable para-
meters that have a real, physically clear meaning to describe the
basic interplay between diffusion, reaction and hydrodynamics at
small scales. The chemical source term is obtained from the
scission model.
As regards the application of the micromixing model to the
situation of molten Polypropylene in a twin-screw extruder we
employ the same principle as in our previous publications con-
cerning the Static Mixer Reactor. Most importantly, this implies
that diffusion is supposed to be limited to the low-molecular
species, initiator and initiator radicals, the Cs in Eq. (6). Polymer
liquid layers are assumed to be deforming and thinning, but
diffusion of polymer chains in and between striations is thought
to be negligible. The previous section has shown that the scission
probability rs is the key variable determining scission, so the
most important question to be answered in this section is how
micromixing affects this variable. Eq. (4a) expresses the scission
probability in terms of (macro-)radical concentrations I and l0,
hence mathematically speaking one desires the solution of the
micromixing equation (6) for these two species. The remainder of
this section, therefore, is devoted to defining the parameters and
boundary conditions involved in Eq. (6) and subsequently solving
the differential equation.
According to the lamellar micromixing model, mixing is
described as the one-dimensional convection and diffusion in a
striation layer of decreasing thickness. Hence, we have to define
the starting situation of a striation as related to the 3D-situation
at the extruder entrance; it is depicted in Fig. 3. In the experi-
mental twin-screw extruder peroxide is introduced as peroxide-
rich polymer grains with average diameter dp and peroxide
weight fraction wp(0). The initial values of the parameters
describing the striation, its thickness, s0, and the fraction of it
having this high concentration of diffusing species, d, are related
Fig. 3. Starting point of a striation as related to the 3D-situation at the extruder
entrance.
5478 P.D. Iedema et al. / Chemical Engineering Science 66 (2011) 5474–5486
to the real geometrical situation as shown in Fig. 3. First, one
should realize that the fraction d equals the ratio between the
volume flows of peroxide-rich polymer grains and bulk polymer
grains. Thus, the overall weight fraction averaged over all grains
follows as:
wp ð0Þ ¼ dwp ð0Þ
Now, Fig. 3 shows that we may consider the micromixing
process around each peroxide-rich particle, while traveling
through the extruder, to take place inside a sphere around it,
having a diameter equal to the average separation distance
between the particles. Note that the concentration development
in the expanding spheres is the result of diffusion and convection
in deforming and folding fluid elements as described by the
lamellar micromixing model. With this picture in mind we decided
to take half the inter-particle separation as the initial striation
thickness, s0. Realizing that the ratio between the volumes of the
particles and the surrounding spheres again is expressed by d, we
find the value for the inter-particle separation and hence s0 to be
related to the power  1/3 of this fraction
s0 ¼ d
ð1=3Þ
dp =2
For instance, if the particle diameter dp equals 0.75 mm, with a
volume (flow) fraction d ¼2.5  10  3, the average inter-particle
separation becomes 5.5 mm and s0 ¼ 2.8 mm. It should be noted
that the extent of degradation is controlled by changing the
average weight fraction, wp ð0Þ, which in practice simply is carried
out by adjusting the number of peroxide-rich particles in the feed
stream. This implies that while varying the average weight
fraction, wp ð0Þ, the initial particle weight fraction, wp ð0Þ, remains
unchanged. However, changing the number of particles does alter
the volume fraction, d, and hence the initial striation thickness.
Increasing the average weight fraction thus corresponds to a
decrease of inter-particle distance, which under these circum-
stances implies a better mixing situation. More general, the
adopted concept as described above is logically in line with the
obvious fact that more and finer grains corresponds to better
mixing.
Next, the solution of Eq. (6) under proper boundary conditions
is discussed. These apply to the diffusing species, i.e. the peroxide
species, I, and the peroxide radical species, I . At the low-
concentration part of the striation, x ¼s, a no-flux boundary
condition is imposed
@I @I
¼ 0; ¼0
@x @x
The boundary conditions at the high-concentration side of the
striation, x ¼0, are obtained from mass balances over the volume
fraction d of peroxide-rich polymer, in which a diffusion flux term
to the remaining polymer appears. The boundary conditions for
I and I thus become
@I2 @I2
dsðtÞ ¼D þ dsðtÞkd I2 at x ¼ 0
@t @x
@I 
@I n o
dsðtÞ ¼D þ dsðtÞ 2kd I2 kht I m1 kc ðI Þ2 at x ¼ 0
@t @x
The left-hand side of these equations represents the concen-
tration changes over time in the peroxide-rich layer. The first
term right-hand side is the flux of the peroxide (or peroxide
radical species) into the polymer at x¼0 and the second term
refers to the production and consumption of peroxide and per-
oxide radicals in that layer. The term ds(t) refers to the absolute
thickness of the peroxide-rich layer given by the product of its
volume fraction d and the striation thickness s, decreasing over
time. Obviously, at the starting time, t¼ 0, the absolute thickness
equals ds0. Note again that variations in the amount of peroxide
added in the micromixing model is implemented by changing the
initial striation thickness s0, reflecting that the average distance
between peroxide layers is the manipulated quantity.
For reasons of computational efficiency in solving the differential
equations given by Eq. (6) we have performed two variable
transformations. The first one was already proposed by Ottino
ð7Þ
(1980). It consists of transforming the space variable x into a
dimensionless variable x at a scale between 0 and 1 based on the
actual value, s, of the striation thickness. In addition, the time
variable t is transformed into a new variable, t, called ‘warped
time’. The second transformation is meant to increase the resolution
inside the thinning striation in the most interesting, peroxide-rich
side by implementing a finer numerical solution grid for dimension-
less space, x. The two transformations of Eq. (6) result into a set of
coupled ordinary differential equations (ODE). As the procedure is
practically identical to that of our previous publication (Fourcade
et al., 2001b), we summarize it in the Appendix.
3.4. Monte Carlo scission procedure in the micromixing model
ð8Þ Solving the micromixing equation (6) through its numerical
implementation as described in the Appendix leads to solutions
for the profiles in time and space of the radical concentrations
I and l0. This is the outcome of the concept that PP chains at
initially different distances from the peroxide-rich regions experi-
ence a different history of being exposed to these scission
enforcing radicals. Numerically, this heterogeneity is expressed
by striation nodes, i, each with different time profiles of I and l0
and consequently, scission probability rs. In line with Eq. (4d) this
leads to different chain length distributions after scission as
computed using the MC scission procedure based on the scission
density as the key variable. Accounting for the proportions of the
different striation regions (Eq. (B.11) in Appendix), this leads to an
overall chain length distribution, which is the one to be compared
with the experimentally measured distributions.
4. Results
In the previous section we have presented models describing
scission of PP for different mixing situations. In this section we
want to compare the results of these models in terms of
Molecular Weight Distribution to those obtained from experi-
ments with the ZSK-18. This requires defining a quantitative
ð9Þ manner to rate how well a computed MWD matches to a
measured one. Furthermore, one of the questions to be answered
is which mixing regime is applicable to the twin-screw extruder
conditions. We have to describe the strategy to determine this
and consequently identify the proper model. After this, the
remainder of the section is devoted to comparing the results of
this model to the experimental MWD.
ð10Þ
4.1. Quantifying a match between MWD’s
ð11Þ The quantitative evaluation of the match between MWD’s is
based on the representation of the semi-logarithmic MWD usual in
Gel Permeation Chromatography, where a normalized concentra-
tion PMM2 is plotted versus log10 M, if M is the weight of a polymer
molecule with n monomer units of molecular weight Mmonomer.
This MWD has exactly the same shape as the distribution of
monomer units per chain, Pnn2, versus log10 n. The normalization
enforces the surface under the MWD curve to equal 1, so that the
normalization factor follows as the summation:
nX n o
¼1
10 10 1 nX ¼1
pn n2 log ðn þ1Þlog ðnÞ  pn n ð12Þ
n¼1
lnð10Þ n ¼ 1
 P.D. Iedema et al. / Chemical Engineering Science 66 (2011) 5474–5486
The error e between a calculated distribution pcn and an experi-
mentally measured distribution pen consequently is expressed as
 
 pcn pen 
e ¼ n2 lnð10ÞPn ¼ 1 c
 Pn ¼ 1 e

 ð13Þ
n¼1 p n n n¼1 p n n
This error lies between 0 and 1.
4.2. Strategy for finding the relevant mixing regime
The task for the model is to properly describe the observed
change in MWD of degraded PP. It is not obvious on beforehand
that mixing will play a role at all. For instance, the experiments
performed by Berzin and Vergnes (2000), also in a twin-screw
extruder turned out to produce MWDs that could perfectly be
described by relatively simple random scission kinetics and ideal
mixing in a batch reactor. Note also, that the kinetic rate
coefficients we use, were found using a kinetic model developed
by ourselves (Iedema et al., 2001) and Berzin’s MWD’s; this model
and these coefficients formed the basis for our earlier work on PP
degradation in an SMR (Fourcade, 2001). The first question now
is: can the MWDs measured by us equally well be explained by
ideal mixing as Berzin’s data?
4.3. The ideal mixing reference in relation to thermal
inhomogeneities in axial direction
At this point it is necessary to specify the ‘‘ideal mixing’’
reference. It is decided that this reference is to be the ideal plug
flow reactor without any axial or transversal dispersion of
temperature or concentrations. In that case all polymer material
has the same residence time in the extruder and there is no
residence time distribution. This implies that the batch reactor
model described above applies to this situation, with the scission
probability computed from Eq. (4a). As the temperature profile
the measured temperature in the 6 ZSK compartments has been
adopted, an example is shown in Fig. 4. Hence, the reference
system is not an isothermal one. Now, one might wonder whether
the shape of the axial temperature profile already would influence
the shape of the MWD in the reference system. We have checked
this and found that the MWD resulting from any temperature
profile could exactly be reproduced by an MWD at a certain
effective ‘‘average’’ processing temperature in an isothermal
system. This implies that the heterogeneity in axial direction
does not alter the MWD shape and cannot explain any broadening
of the MWD in PP degradation.
The MWD has been calculated using the batch model to serve
as the reference for all the various experimental conditions
applied, as shown in Figs. 12–15. We observe a systematical
250
200
Temperature °C
150
100
50
0
0 5 10 15 20 25 30
Residence time in extruder
Fig. 4. Temperature profile in extruder.
5479
deviation from the experimental MWD: the latter ones are
broader than those calculated assuming ideal mixing. In terms
of the error e defined above in Eq. (13) the deviation amounts to
around 17%. Hence, we conclude that our data – in contrast to e.g.
Berzin’s data – are not well represented by an ideal mixing model.
4.4. Thermal inhomogeneities in lateral direction
In the previous section the effect of the thermal inhomogene-
ities in axial direction has been discussed. In this paragraph we
will pay attention to such inhomogeneities in lateral direction,
which is perpendicular to the screw axes. The importance of this
aspect has been put forward by Berzin and Vergnes (2006) and
Suresh et al. (2008). The latter authors have developed a detailed
flow model of a twin-screw extruder and computed lateral
temperature profiles, indicating that maximum temperature dif-
ferences in lateral direction may amount up to 20 K. Similar
information is available in Berzin and Vergnes (2006).
We will address the effect of lateral temperature inhomogene-
ities using a simple compartment approach. According to this
concept the extruder is a system of ideally mixed reactors in
parallel, where polymer is exposed to the same initial peroxide
concentration and subjected to the same shape of temperature
profile as depicted in Fig. 4, but at various absolute levels. No
interchange or interaction exists between the compartments.
Obviously, here our objective is not to give a realistic description
of transport phenomena in an extruder, such as developed by
Suresh et al. (2008). It does, however, allow us to check whether
the existence of any temperature difference in principle could
lead to MWD broadening.
One series of computations have been performed with the
measured temperature profile (Fig. 4) as the lowest in one
compartment and with DT higher levels in 10 further compart-
ments, such that the highest level is 30 K higher than the
measured temperature everywhere along the extruder. Thus, in
the coldest compartment the temperature increases from 200 1C
at time 15 s (Fig. 4) to 240 1C at the end and the hottest from
230 1C to 270 1C. This must be considered as fairly extreme in
view of the data concerning lateral temperature differences in
Suresh et al. (2008) and Berzin and Vergnes (2006), around 20 K
at maximum, but here this approach serves well our goal to check
for MWD broadening. The computations were based on the
experimental shift of PP with an MW of 930 kg/mole to a product
with an MW of 125 kg/mole. The result is shown in Fig. 5 as the
MWD indicated with the average temperature T melt ¼ Tmeas þ15. It
turns out to be close to the MWD for ideal mixing and to possess
Tmelt = Tmeas+15, per = 98%
1.2
Tmelt = Tmeas−5, per = 78%
1 Ideal mixing, per = 93%
Tmelt = Tmeas−15; ηper = 58%
Concentration
0.8
Measured
0.6 SEC-DV Virgin PP
Mw = 125 kg/mole Mw = 930 kg/mole
0.4
0.2
0
104 105 106
Molar Mass
35 40 45 Fig. 5. Effect of thermal heterogeneities in lateral direction. No MWD broadening
and stronger MW shift due to higher conversion (98%) for average 15 K higher
melt temperature compared to ideal mixing case (93%). Broadening occurs at
lower temperatures and conversions and weaker shift.
5480 P.D. Iedema et al. / Chemical Engineering Science 66 (2011) 5474–5486

the same shape, so no broadening is observed in this case. The

1.2
small shift to lower MW is caused by the higher peroxide Batch reactor
conversion, Zper, being 98% instead of 93% in the ideal mixing
1 SEC-DV
case. More important, however, is to notice that it is the Mw = 125,000
Virgin PP

Concentration
difference in conversion due to the temperature differences 0.8
between the compartments that leads to different scission prob-
abilities. In this case the conversion level in all compartments is 0.6
high and consequently the differences in scission intensity small,
CSTR
which does not lead to significant broadening. The measured 0.4
MWD of the PP product after the considerable strong peroxide
shift in this experiment suggests a high peroxide conversion 0.2
indeed. Our conclusion here is that eventual higher temperature
0
regions can have only a minor effect on the conversion that is 103 104 105 106 107
already at a high level in view of the high overall temperature Molar Mass
level as measured in the extruder. Hence, such thermal hetero-
geneities may be very important for the product quality in itself, Fig. 6. Effect of residence time distribution on MWD; comparison of batch reactor
and CSTR.
but they will not lead to significant heterogeneities in shift
intensity and consequently neither lead to MWD broadening.
From the foregoing one may expect that stronger differences in
conversion in principle still may lead to significant broadening.
This is only possible, if the overall temperature and conversion
are lower, which in the experimental example taken would imply
that parts of the polymer melt in the extruder on average are at
lower than the measured temperature. Since this is not very
probable the issue of conversion differences under these condi-
tions is of mainly academic interest. However, our simple com-
partment model does allow us to check the effect of such lower
conversion. Fig. 5 shows the results of these exercises: one for, on
average, 5 K and one for 15 K lower than the measured tempera-
ture. Now, for these conditions we do observe considerable
broadening, but indeed, as expected, at the expense of much
lower conversions: 78% and 53%, respectively. It is also clear that
the lower peroxide conversion leads to considerably smaller shift
than the experimentally observed one. Our conclusion from this
exercise is that indeed thermal heterogeneities may lead to MWD Fig. 7. Residence time distribution in a twin-screw extruder yielding distribution
broadening, but only if the overall conversion is low. Hence, we of scission probabilities.
observe that the temperature differences cannot explain the
observed MWD broadening in our experiments.

1.2
4.5. The effect of residence time distribution on MWD 44 s
38.5 s
SEC-DV

1
At this point we will investigate the effect of residence time
Concentration

0.8
distribution. From an academic point of view one might be 58 s
interested in the results for the case of most extreme RTD: the
0.6
CSTR. The model equations applying to the CSTR have been
derived in the previous section, yielding the exponential distribu- 0.4
tion of scission probabilities as expressed by Eq. (4d). The MWD
for PP scission in a CSTR is shown in Fig. 6. One observes the 0.2
resulting MWD to be much broader than the one from the batch
reactor, but also much broader than all the experimental MWD 0
104 105 106
(the sample with average molecular weight Mw ¼125,000). The Molar Mass
error is even larger than that in the batch case: around 40%.
Hence, we conclude that RTD does have a strong impact on the Fig. 8. MWD computed for fractions of low, average and high residence time in a
twin-screw extruder. Overall MWD coincides with that for average t without
shape of the MWD indeed. However, the RTD of a CSTR is
visible broadening. Comparison with measured MWD.
evidently far from being representative for a twin-screw extruder.
The next issue is: what is the influence of a more realistic RTD
on the MWD of PP degradation in a twin-screw extruder. From
in which this range of t results, according to Eq. (4a). The MWD is
Poulesquen and Vergnes (2003) we adopt a general expression for
calculated for fractions of low (38.5 s) and high (58 s) t within
such an RTD
this range, the result is shown in Fig. 8. The overall MWD was
 
ðttd Þ2 ðttd Þ computed from all residence time fractions. This MWD turned out
EðtÞ ¼  3 exp ð14Þ to practically coincide with the one for the average t of 44 s,
2 ðttd Þ=3 ðttd Þ=3
without any visible broadening. Thus, we observe that where the
with t the average residence time and td the minimum value, in RTD of a CSTR leads to important broadening, the RTD of a twin-
the laboratory ZSK18 being 44 and 38.5 s, respectively. Fig. 7 screw extruder turns out to be too narrow for significant
shows this RTD together with the variation in scission probability, broadening.
 P.D. Iedema et al. / Chemical Engineering Science 66 (2011) 5474–5486 5481

4.6. Effect of micromixing on scission probability kmole/m3 Concentration peroxide radicals

10-8
The micromixing model as described above has been applied
to compute the effect of micromixing on the MWD of PP 10-10
degradation in a twin-screw extruder, using a proper parameter

Concentration
set. The experimental conditions in the ZSK18 fix the initial
10-12 Stri
striation thickness as half the average distance between perox- atio
n no
ide-rich PP particles in the entrance section, s0, which depends on des
1-3
0
the amount of peroxide added and the volume fraction peroxide- 10-14
rich PP through Eq. (8). The striation thinning parameter, a, is not
known on beforehand, but will act as the fitting parameter in 10-16
order to match the calculated MWD to the measured ones. The
number of nodes in the striation is taken such that an increase of
10-18
this number does no longer change the overall MWD resulting. In 0 5 10 15 20 25 30 35 40 45
our previous publication this was seen to be at 11 nodes, in the Residence time (s)
present case 30 nodes were required. Note that as a consequence
kmole/m3 Concentration macroradicals
of using a Monte Carlo scission procedure the number of stria- 10-5
tions does not influence the required CPU time, as long as the
total size of the population of chains sampled – in the order of
millions – remains unchanged. Sampling using a larger number of Stri
atio
different scission probabilities simply implies fewer chains n no
10-10

Concentration
des
1-3
sampled per striation node, still yielding sufficiently smooth 0
distributions.
Before showing the resulting MWD under micromixing condi-
tions, first the consequences of micromixing for the key variable 10-15
scission probability will be presented. In Figs. 9 and 10 the
calculated concentration profiles over the extruder in all the
striations as defined above are shown. Discontinuities are due
to the discontinuous temperature profile imposed, see Fig. 4. 10-20
0 5 10 15 20 25 30 35 40 45
Employing these concentration profiles in the integration of
Residence time (s)
Eq. (4c) the variation of scission probability across the striations
Fig. 10. Concentrations versus residence in twin-screw extruder in all striations of
peroxide radicals (upper plot) and macroradicals (lower plot). Discontinuity caused
kmole/m3 Peroxide concentration by discontinuous temperature profile applied (see Fig. 4). Striation thinning rate:
100 a ¼0.06 s  1.

2500
Distribution of ρs
Concentration

10-2 Str from RTD in twin-

iati
on
nod 2000 screw extruder
Normalized frequency

es 1
-30

1500
Distribution of ρs due
10-4
to micromixing
1000

Exponential
10-6 -2 500 distribution of ρs
10 10-1 100 101 from RTD in CSTR
Residence time (s)
0
kmole/m3 Peroxide concentration 0 0.2 0.4 0.6 0.8 1 1.2 1.4 1.6 1.8 2
100 Scission probability ρs x 10-3

Fig. 11. Frequency distribution of scission probability for RTD of ideally mixed (no
micromixing) in twin-screw extruder and CSTR as compared to scission frequency
distribution due to micromixing (striation thinning rate: a ¼ 0.06 s  1).
Concentration

10-2
Stri
atio
n no
des
1-3 is determined. It is now interesting to compare the spread in
0
scission intensity as a result from RTD in a CSTR (exponential) and
10-4
in a twin-screw extruder (Fig. 7) under ideal mixing conditions to
the scission probability variation due to micromixing. We observe
that the scission intensity is extremely narrowly distributed if the
10-6 RTD in the twin-screw extruder is accounted for, as compared to
0 5 10 15 20 25 30 35 40 the exponential RTD that leads to a much broader (equally
Residence time (s) exponential) distribution in scission in the CSTR. Quite remark-
Fig. 9. Peroxide concentration versus residence in twin-screw extruder in all
able is the spread in scission density due to micromixing, which
striations, double-logarithmic (upper plot) and semi-logarithmic (lower plot). appears to be in between the two ideally mixed cases. It is still
Striation thinning rate: a ¼ 0.06 s  1. much wider than the spread caused by RTD in the extruder.
5482 P.D. Iedema et al. / Chemical Engineering Science 66 (2011) 5474–5486

Ideal mixing
SEC-DV α > 0.2 s-1, ideal mixing 1.2 SEC-DV
1.2 Mw = 125 kg/mole Micromixing model Mw = 225 kg/mole
0.08
ε = 0.056
1 0.06 1
εideal = 0.16
Concentration

Concentration
0.05 Virgin PP
0.8 0.8
Mw = 125 kg/mole
0.04
ε = 0.025 Virgin PP
0.6 0.03 0.6 εideal = 0.17 Mw = 930 kg/mole

0.4 -1 0.4
.01s
α=0
0.2 0.2
0
103 104 105 106 107 0
103 104 105 106 107
Molar Mass
Molar Mass
Fig. 12. Influence of striation parameter a on shape of MWD. For a 4 0.2 s  1 no
further changes are observed, MWD shape is identical to that of ideal mixing in a
batch reactor (Fig. 6). 1.2 SEC-DV
Ideal mixing
Micromixing model
Mw = 510 kg/mole
1 ε = 0.056
Although this RTD in the extruder is present and adds to the εideal = 0.15

widening due to micromixing, we decided to carry out all
micromixing calculations at the average value of the residence
time, since the additional broadening is expected to be practically
invisible, as was already noticed from Fig. 7.
4.7. Effect of micromixing on MWD
In the previous sections it was shown that the kinetic model in
combination with the assumption of ideal mixing did not lead to a
good match between the calculated and measured MWD. Now,
the variability in scission probability due to micromixing lying in
between that resulting from the RTD in a twin-screw extruder
and in a CSTR, indeed turns out to produce good fits, errors e in
the order of 5%. Therefore, the remainder of this paper is devoted
to a discussion on the key micromixing parameter in the actual
context of PP degradation. The experimental parameters varied in
this study were the Melt Flow Index (MFI) of the starting PP
material and the amount of peroxide added; in one experiment
the number of rounds per minute of the extruder was increased.
A higher starting MFI by definition implies lower viscosity and
different flow conditions at which the phenomena take place.
Also, a larger amount of peroxide leads to a higher extent of
degradation and a stronger variation of rheological conditions.
One might expect both parameters to affect the micromixing
phenomena as well. In the micromixing model the only freely
adjustable parameter with respect to flow and mixing to our
disposal is the striation thinning parameter, a. Clearly, the initial
MWD of an experimental sample is used as the initial MWD in the
model calculations, so this variation is accounted for, but still an
additional effect on the striation thinning rate a might be
possible. Furthermore, as explained in the previous paragraph,
the variation in peroxide amount is not expressed in terms of
initial feed concentrations, but rather as the average distance
between peroxide-rich particles in the feed, appearing as the
initial striation thickness s0 in the micromixing model. In this
manner the micromixing model already accounts for changes in
peroxide feed amounts. Nevertheless, as the a-value might be a
function of the rheological conditions referred to above, in
addition to the effect accounted for through s0 the a-value
possibly varies as well with peroxide amount. What we found,
however, was that the a-values giving the best fits, that is the
lowest error e according to Eq. (13), did not considerably change
neither with initial MWD nor with peroxide amount and the
resulting extent of degradation. Using a constant a-value of 0.06
produced errors between 2.5% to 8.5%, as goes from Figs. 13–16.
For some samples slightly better (1%) fits could be obtained, if a
Concentration
Mw = 160 kg/mole
0.8 ε = 0.064
εideal = 0.15
0.6 Virgin PP
Mw = 930 kg/mole
0.4
0.2
0
103 104 105 106 107
Molar Mass
Fig. 13. MWD of degraded PP with initial Mw ¼ 930 kg/mole (MFI 0.15) with
(a) 0.486 and 0.193 wt% peroxide, yielding Mw ¼ 125 and 225 kg/mole (MFI-values
134 resp. 15.7), and (b) 0.314 and 0.007 wt% peroxide, yielding Mw ¼ 160 and
510 kg/mole (MFI-values 44 resp. 1.4). MWD deviation between SEC-DV and
micromixing model (a ¼0.06 s  1): e ¼ 2.5–6.4% (ideal mixing e ¼ 16%).
was allowed to vary, but only minor variations in a were observed
(o0.01) rather than a clear trend in this parameter. Hence, it was
decided to employ the model with one a-value in combination
with the initial striation thickness, s0, depending the peroxide
amount fed through Eq. (8). We noted, besides, that the absolute
level of a-value producing the best fit was strongly correlated to
the value of s0 applied.
In order to still get an idea of the quantitative sensitivity of the
MWD-shape to the value of a, we varied it around 0.06 down-
wards to 0.01, yielding broader distributions, and upward leading
to narrower MWD, ultimately to the ones obtained for ideal
mixing, see Fig. 12. One observes that above values of 0.2 the
MWD coincides with that of ideal mixing. Evidently, a has to vary
by at least one order of magnitude to arrive at visible differences
in MWD. This is in line with our finding that changing flow
conditions by modifying the extent of degradation did not
noticeably change a. In other words, this parameter might change
a little, but not strong enough to change MWD. This turned out
also to be the case for a changed rotational speed of the extruder,
which was in a few experiments increased from the standard
200 rpm to 600 rpm. Again, no noticeable change in the best fit
yielding a-value was obtained.
In Table 3 the whole range of experimental conditions in terms
of initial weight average molecular weight Mw and MFI as well as
the amounts of peroxide added in weight percentage of the
polymer throughput in the extruder is listed. In Figs. 13–16 the
MWD as computed with both the ideal mixing and the micro-
mixing model with a constant a-value of 0.06 s  1 are compared
to the ones measured with SEC-DV. The deviation between model
and experiment as calculated by Eq. (13) is shown in the figures
 P.D. Iedema et al. / Chemical Engineering Science 66 (2011) 5474–5486 5483

1.2 SEC-DV
1.2 SEC-DV
Micromixing model
Micromixing model Ideal mixing Ideal mixing
1 Mw = 350 kg/mole 1 Mw = 280 kg/mole
ε = 0.084 ε = 0.025

Concentration
εideal = 0.17 εideal = 0.045
Concentration

0.8 Mw = 120 kg/mole 0.8

ε = 0.074 Mw = 140 kg/mole
εideal = 0.19 0.6 ε = 0.040
0.6 εideal = 0.095

0.4 Virgin PP 0.4 Virgin PP

Mw = 590 kg/mole Mw = 380 kg/mole

0.2 0.2

0 0
103 104 105 106 107 103 104 105 106 107
Molar Mass Molar Mass

Fig. 16. MWD of degraded PP with initial Mw ¼ 380 kg/mole (MFI 3.1) with 0.23
SEC-DV and 0.03 wt% peroxide, yielding Mw ¼ 140 and 280 kg/mole (MFI-values 119 resp.
1.2
Micromixing model 14.7). MWD deviation between SEC-DV and micromixing model (a ¼ 0.06 s  1):
Ideal mixing
e ¼ 2.5–4% (ideal mixing e ¼ 4.5–9.5%).
1 Mw = 500 kg/mole
ε = 0.049
Concentration

0.8 εideal = 0.090 Table 3

Mw = 180 kg/mole
ε = 0.049 MWD fits of ideal mixing and micromixing model (with a ¼ 0.06 s  1) to SEC-DV
0.6 εideal = 0.16 measurements of PP samples degraded in ZSK 18.

Virgin PP Initial Mw MFIa Final Mw MFIa wt% peroxide e eideal

0.4
Mw = 590 kg/mole
380,000 3.06 280,000 14.65 0.03 0.025 0.045
0.2 140,000 119.1 0.23 0.04 0.095
480,000 1.23 200,000 20.98 0.13 0.065 0.10
0 145,000 66.23 0.26 0.039 0.11
103 104 105 106 107 590,000 0.49 500,000 0.95 0.012 0.049 0.09
Molar Mass 350,000 5.63 0.07 0.084 0.17
180,000 39.6 0.23 0.049 0.16
Fig. 14. MWD of degraded PP with initial Mw ¼590 kg/mole (MFI 0.49) with 120,000 135.11 0.45 0.074 0.19
(a) 0.45 and 0.07 wt% peroxide, yielding Mw ¼120 and 350 kg/mole (MFI-values 930,000 0.15 510,000 1.4 0.045 0.056 0.15
135 resp. 5.6), and (b) 0.23 and 0.0012 wt% peroxide, yielding Mw ¼ 180 and 225,000 15.71 0.193 0.056 0.16
500 kg/mole (MFI-values 40 resp. 0.95). MWD deviation between SEC-DV and 160,000 44.07 0.314 0.064 0.15
micromixing model (a ¼0.06 s  1): e ¼ 4.9–8.5% (ideal mixing e ¼9–19%). 125,000 134.07 0.486 0.025 0.16

a
MFI: Melt Flow Index.
1.2 SEC-DV
Micromixing model Ideal mixing
Mw = 200 kg/mole
1
ε = 0.065
(around 10%), while the fits of the micromixing model are equally
εideal = 0.10 around 5%. Note again that the best fit between modeled and
Concentration

0.8 experimental MWD obtained by allowing freedom in all para-

Mw = 145 kg/mole
ε = 0.039 meters employed never was lower than 2%. We suspect that this
0.6 εideal = 0.11 deviation is close to the total experimental error in measuring
Virgin PP the MWD.
0.4
Mw = 480 kg/mole

0.2
0
103 104 105 106
Molar Mass
Fig. 15. MWD of degraded PP with initial Mw ¼480 kg/mole (MFI 1.2) with 0.26
and 0.13 wt% peroxide, yielding Mw ¼ 145 and 200 kg/mole (MFI-values 66 resp.
21). MWD deviation between SEC-DV and micromixing model (a ¼0.06 s  1):
e ¼ 3.9–6.5% (ideal mixing e ¼10–11%).
and listed in Table 3. We clearly observe that assuming a
micromixing effect strongly improves the MWD fit as a conse-
quence of the MWD broadening involved with the extra hetero-
geneity due to this effect. Quantitatively speaking, the
improvement consists of a reduction of the modeling deviation
from 16% for the ideal mixing to only 2.5–7.5% for the micromix-
ing model. In addition, one may observe that the impact of
micromixing is strongest for the samples with highest initial Mw
that obviously underwent the strongest degradation: those with
starting Mw of 930 and 590 kg/mole. For the samples with lower
initial Mw the ideal mixing assumption leads to smaller errors
5. Conclusion
107 We have performed a series of controlled degradation experi-
ments on a ZSK 18 twin-screw extruder with Polypropylene of
varying Mw under addition of various amounts of peroxides and
have measured the initial and final MWD with SEC-DV. Models
assuming ideal mixing while introducing different degrees of
heterogeneity by accounting for different distribution function
of residence time did reveal a strong impact on MWD. In absence
of RTD the model predicted MWD turned out to be too narrow,
even accounting for the typical steep temperature profile existing
in the extruder. Assuming a realistic RTD for a twin-screw
extruder (Fig. 7) did not lead to noticeable broadening, but the
exponential RTD of a CSTR did indeed strongly widen the MWD;
the latter produced an MWD that was too broad. However,
neither of the several RTD options produced a good fit between
model and experiments. Clear deviations between the MWD’s
were visible, while the calculated (Eq. (13)) deviation in surface
overlap amounted to 15% for the batch reactor and to 40% for the
CSTR. The effect of thermal heterogeneities in lateral direction
5484 P.D. Iedema et al. / Chemical Engineering Science 66 (2011) 5474–5486

(perpendicular to screw axis) has also been investigated using a Ds diffusion coefficient of species s Eq. (5)
simple compartment model. This turns out to lead to MWD I2 initiator concentration
broadening only at lower overall conversion levels, which then IK initiator radical concentration
lead to weaker MW shift than experimentally observed. Hence, in kc initiator radical combination coefficient
the present case thermal inhomogeneities could not explain the kd initiator dissociation coefficient
observed broadening. kht hydrogen abstraction coefficient
Micromixing represented a further possible phenomenon ktd termination by disproportionation coefficient
introducing heterogeneity to be explored. Based on the principles ktr transfer to polymer coefficient
of an earlier developed, fully deterministic micromixing model M, Mmonomer molecular weight of polymer, resp, monomer
(Fourcade et al., 2001a,b) we constructed a partly Monte Carlo n number of monomer units in polymer chain
sampling based model. Ottino’s striation thinning approach is pn concentration of polymer with n monomer units
applied to the geometrical conditions of a twin-screw extruder, PM concentration of polymer with weight M
where peroxide is fed by a few (one in a few thousand) PP Rs reaction rate of species s Eq. (5)
particles with 40 weight % peroxide. This yields the initial s0 initial striation thickness Eq. (7)
striation thickness as half the average distance between such wp peroxide weight fraction Eq. (6)
particles in the extruder entrance section. The variation of the
scission probability over the total striation thickness (calculated
at 30 striation nodes) turned out to be considerable (see Fig. 10). Greek letters
Thus, as expected, the effect of this variation on the MWD
calculated was also significant, allowing matches between mea- a striation thinning rate Eq. (5)
sured and modeled MWD of very good optical agreement, while d volume fraction peroxide-rich part Eq. (7)
surface deviations reduced to around 5%. It turned out that all the l0 total concentration of macroradicals
measured MWD could be reproduced with the micromixing m1 total concentration of monomer units in polymer
model within this error using a constant value for the key rs scission probability
parameter, the striation thinning rate, a ¼0.06 s  1. This was so t, t, td residence time (RT), average RT, minimum RT in
despite the different conditions under which the experiments extruder
were carried out: several qualities of the starting material (Mw
varying from 380 to 930 kg/mole) and varying peroxide amounts
(between 0.012 and 0.486 wt%). Neither did one experiment at Appendix. Variable transformations
higher rotational speed (600 rpm instead of 200 rpm) alter the
a-value yielding the best fit. The transformations discussed in this Appendix are meant to
The broader experimental MWD’s as compared to those reduce the computational effort of solving the differential equa-
calculated assuming ideal mixing can only lead to the conclusion tion describing micromixing, Eq. (5) of the main text. Thus,
that more heterogeneous conditions in the extruder must exist solutions are found for the profiles of I and l0 that finally yield
than those resulting from the temperature profile and the narrow the scission probability rs under micromixing conditions.
RTD. We consider the good fit between the predicted MWD from
the micromixing model to the SEC-DV data as an (indirect) proof A. Co-ordinate transformation according to Ottino
that micromixing phenomena are the cause of the extra hetero-
geneity. If micromixing thus is observed to affect MWD and A1. Formulation of the problem
henceforth product properties, the question arises, whether this A co-ordinate transformation for the space and time coordi-
phenomenon might be utilized as an extra handle to control nates, x and t, was applied as described by Ottino (1980). New
properties. For instance, if one would desire a broad or even variables x and t were defined as follows:
bimodal product (see Fig. 11), it might be realized by even x x
x¼ ¼ eat ðA1Þ
‘‘worse’’ micromixing (lower a). Now, the present study did not st s0
reveal how to manipulate micromixing, for instance, by creating Z 
t
stronger or weaker striation thinning (a-value), so more research D 0 D e2at 1
t¼ dt ¼ ðA2Þ
in this direction is required to investigate this issue. Further 0 st0 2 s0 2 2a
research should also make clear whether micromixing plays a role where the striation thickness at time t, st, is
in other polymer degradation systems. In the present study we Z t
saw the impact of micromixing to become stronger with a greater st ¼ s0 exp aðt 0 Þdt 0 ¼ s0 eat ðA3Þ
MWD shift. In reactive modification of Polyethylene smaller shifts 0

(even to higher MW, due to branching) are observed, so the In terms of the new variables x and t, the convection diffusion
impact of micromixing might be weaker. On the other hand, the equation Eq. (6) of the main text is transformed into the following
grafting on Polypropylene (or another polymer) of other polymer form:
chains, where grafting, scission and diffusion are in competition,
@Cs @ 2 Cs
might be even more sensitive to micromixing. In general, polymer ¼ þ KRs ðA4Þ
@t @x
2
systems where selectivity is an important issue might be very
interesting to investigate with respect to micromixing. with
s0 2
K¼ ðA5Þ
Nomenclature 2as0 2 t þ D
This form is evidently much simpler than the original one, no
Latin letters longer containing a product of a time/space dependent variable
and a derivative. It should be noted that x defines a dimensionless
Cs concentration of species s Eq. (5) space scale between 0 and 1 based on the actual value of the
dp particle diameter striation thickness t is called warped time. Eq. (A4) implies that in
 P.D. Iedema et al. / Chemical Engineering Science 66 (2011) 5474–5486 5485

the warped time scale t and space scale x the apparent rate of resolution at the high peroxide concentration side, we apply a
reaction is changed by a factor K. The no flux boundary conditions mesh transformation, g, which transforms variable x into a new
Eq. (9) of the main text become variable xn, to solve Eq. (A4) on equidistant grid nodes in the
@I2 @I xn-space, which corresponds to non-equidistant grid nodes in the
¼ 0 and ¼0 at x ¼ 1 ðA6Þ x-space
@x @x
g : ½0,1-½0,1
Boundary condition Eqs. (9) and (10) of the main text,
respectively, become x/xn ¼ gðxÞ ðB1Þ

@I @I2 In terms of variables x and t, Eq. (A4) becomes

n

d ¼ þK dkd I2 at x ¼ 0 ðA7Þ
@t @x @Cs @Cs @2 Cs
¼ g 00 ðxÞ n ðxÞ þ ðg0ðxÞÞ2 n2 ðxÞ þ KRs ðB2Þ
n o @t @x @x
@I @I
d ¼ þ K d 2kd I2 kht I m1 kc ðI Þ2 at x ¼ 0 ðA8Þ
@t @x where g 0 ðxÞandg 00 ðxÞstand, respectively, for the first and the
second derivative of g. The following functions were considered
for the mesh transformation g:

A2. Solution x ¼ g 1 ðxn Þ ¼ axn2 þ ð1aÞxn ðB3Þ

The solution of Eqs. (A4)–(A8) is obtained using the method of
where a is a parameter between 0 and 1. The density of nodes at
lines. The striation is discretized in N nodes (Fig. A1) and the
the vicinity of the peroxide-rich layer is related to a in such a way
diffusion term is approximated using centered finite differences.
that higher values of a correspond to higher density. All the
We define I2i and Ii as the concentration of the two (diffusing)
simulations in this paper have been performed with a value of
peroxide species I2 and I at node i, while the (non-diffusing)
i 0.7 for a. The functions taking part in Eq. (B2) are thus
macroradical concentration, l0, at node i is l0 .
The coupled set of ordinary differential equations (ODE) ð1aÞ þðð1aÞ2 þ 4axÞ1=2
gðxÞ ¼
resulting from the discretization of Eq. (A4) is 2a
@li0 i
g 0 ðxÞ ¼ ðð1aÞ2 þ4axÞ1=2
¼ KRðl0 Þ
@t g 00 ðxÞ ¼ 2aðð1aÞ2 þ 4axÞ3=2 ðB4Þ
@I2i @2 I2i
@t ¼ 2 þKRðI2i Þ i ¼ 1,N ðA9Þ
@x The boundary conditions, Eqs. (A6)–(A8), respectively, become
@Ii @2 Ii
@t ¼ þ KRðIi Þ @I @I
@x2 n
n ¼ 0 and n ¼0 at x ¼ 1 ðB5Þ
i
@x @x
Here Rðl0 Þ,RðI2i Þ
and RðIi Þ
represent the chemical reaction terms for
these species as defined above. Obviously, no diffusion term is @I @I n
d ¼ g 0 ð0Þ n þK dkd I2 at x ¼ 0 ðB6Þ
appearing in the equations for the macroradical. The diffusion term @t @x
2
@2 I2i =@x is calculated with the finite difference approximation
@I @I n o
¼ g 0 ð0Þ n þ K d 2kd I2 kht I m1 kc ðI Þ2
n
@2 I2i1 d at x ¼ 0 ðB7Þ
¼ 2 ðI2i1 2I2i þI2i þ 1 Þ ðA10Þ @t @x
2 h
@x
The resulting set of ODE’s to be solved in the (xn,t) space is
and similarly that for the initiator radicals. The derivative now
w.r.t. x terms in Eq. (A7) is approximated by the following finite
@li0 i
difference: @t ¼ KRðl0 Þ
 @I2i i @Ii i @2 I2i
@I2  3I0 4I21 þ I22 @t ¼ g 00 ðx Þ @x2n þ g 0 ðx Þ2 þ KRðI2i Þ i ¼ 1,N ðB8Þ
¼ 2 ðA11Þ @xn2
@x x ¼ 0 2h @Ii i @Ii i 2 @2 Ii
@t
00
¼ g ðx Þ @xn þ g ðx Þ 0
n2 þ KRðIi Þ
@x
where I20 is the peroxide concentration at x ¼0; a similar expression
is valid for I0 . with the finite difference approximations for the peroxide con-
centration derivatives (and similarly for initiator radicals)
B. Transformation to non-equidistant space grid @I2i
@xn
¼ 1
ðIi þ 1 I2i1 Þ
2h 2
@2 I2i
ðB9Þ
In the derivation above diffusion is dealt with using a centered n2 ¼ 1
h2
ðI2i1 2I2i þ I2i þ 1 Þ
@x
finite difference approximation in combination with an equidi-
stant mesh of the space co-ordinate x. In order to increase the The solution of Eqs. (B8) yields profiles over warped time, t, of
the initiator radical and macroradical concentrations I and l0 ,
respectively, at all striation nodes i. Since it is the scission
1 k-1 k k+1 7
probability rs we are interested in, we infer the proper expression
to compute it for all nodes from Eq. (4a) of the main text as
0 h 1
Z tb Z 1
i i
ris ¼ ðkht Ii þktr l0 Þdt ¼ KðtÞðkht Ii þ ktr l0 Þdt ðB10Þ
0 0
1 k-1 k k+1 7 Here tb is the residence time in the extruder (in real time, t) and
K(t) is the transformation function as defined by Eq. (A4). Note
1
0 hk that obviously the integrals over real time, t, and warped time, t,
should lead to identical scission probabilities.
Peroxide-rich polymer Virgin polymer Now, the scission probabilities for each striation node i by
Fig. A1. Meshes of the striation with 7 nodes. Upper part: equidistant mesh, lower applying the Monte Carlo scission procedure lead to a chain
part: mesh from 2nd transformation with higher resolution near peroxide-rich layer. length distribution after scission for each striation node, Pni . From
5486 P.D. Iedema et al. / Chemical Engineering Science 66 (2011) 5474–5486
this, finally, the overall chain length distribution is computed
using hi, the length of the ith element in the grid (see Fig. A1)
X Marini, L., Georgakis, C., 1984. The effect of imperfect mixing on polymer
Pn ¼ Pni hi ðB11Þ
i ¼ 1,N
References
Baldyga, J., Rozen, A., Mostert, F., 1998. A model of laminar micromixing with
application to parallel chemical reactions. Chem. Eng. J. 69 (1998), 7–20.
Berzin, F., Vergnes, B., 2000. Modelling of peroxide initiated controlled degradation
of polypropylene in a twin screw extruder. Polym. Eng. Sci. 40, 344–356.
Berzin, F., Vergnes, B., 2006. Modeling of reactive systems in twin-screw extru-
sion:challenges and applications. C. R. Chim. 9, 1409–1418.
Camara, S., Gilbert, B.C., Meier, R.J., van Duin, M., Whitwood, A.C., 2006. EPR
studies of peroxide decomposition, radical formation and reactions relevant to
cross-linking and grafting in polyolefins. Polymer 47, 4683–4693.
Fields, S.D., Ottino, J.M., 1987. Effect of striation thickness distribution on the
course of an unpremixed polymerization. Chem. Eng. Sci. 42, 459–465.
Fourcade, E., Wadley, R., Hoefsloot, H.J.C., Green, A., Iedema, P.D., 2001a. Laminar
striation thinning in static mixer reactors from computational fluid dynamics
and laser induced fluorescence. Chem. Eng. Sci. 56, 6729–6741.
Fourcade, E., Hoefsloot, H.C.J., van Vliet, G., Bunge, W., Mutsers, S.M.P., Iedema, P.D.,
2001b. The influence of micromixing on molecular weight distribution during
controlled Polypropylene degradation in a static mixer reactor. Chem. Eng. Sci.
56, 6589–6603.
Iedema, P.D., Willems, C., Van Vliet, G., Bunge, W., Mutsers, S.M.P., Hoefsloot, H.C.J.,
2001. Using molecular weight distributions to determine the kinetics of
peroxide-induced degradation of Polypropylene. Chem. Eng. Sci. 56,
3659–3669.
Kolhapure, N.H., Fox, R.O., Daiss, A., Mahling, F.O., 2005. PDF simulations of
ethylene decomposition in tubular LDPE reactors. AIChE J. 51 (2), 585–606.
quality in low density Polyethylene vessel reactors. Chem. Eng. Sci. 30 (6),
361–375.
McKenna, T.F., Cokljat, D., Wild, P., 1998. CFD modeling of heat transfer during
gas phase olefin polymerisation. Comput. Chem. Eng. Vol. 22 (Suppl),
285–292.
Nele, M., Soares, J.B.P., 2002. Long-chain branching with metallocene catalysts: is a
purely kinetic mechanism for terminal branching sufficient? Macromol.
Theory Simul. 11 (9), 939–943.
Ottino, J.M., 1980. Lamellar mixing models for structured chemical reactions and
their relationship to statistical models; macro- and micromixing and the
problem of averages. Chem. Eng. Sci. 35, 1377–1391.
Poulesquen, A., Vergnes, B., 2003. A study of residence time distribution in co-
rotating twin-screw extruders. part i: theoretical modeling. Polym. Eng. Sci. 43
(12), 1841–1848.
Schmidt, C.U., Busch, M., Lilge, D., Wulkow, M., 2005. Detailed molecular structure
modeling—a path forward to designing application properties of ldPE. Macro-
mol. Mater. Eng. 290 (4), 404–414.
Suresh, A., Chakraborty, S., Kargupta, K., Ganguly, S., 2008. Low-dimensional
models for describing mixing effects in reactive extrusion of polypropylene.
Chem. Eng. Sci. 63 (14), 3788–3801.
Tobita, H., 1996. Random degradation of branched polymers. 1. Star polymers.
Macromolecules 29, 3000–3009.
Tsai, K., Fox, R.O., 1996. PDF modelling of turbulent-mixing effects on initiator
efficiency in a tubular LDPE reactor. AIChE J. 42 (10), 2926–2940.
Wells, G.J., Ray, W.H., 2005. Mixing effects on performance and stability of low-
density polyethylene reactors. AIChE J. 51 (12), 3205–3218.
Wulkow, M., 1996. The simulation of molecular weight distributions in polyreac-
tion kinetics by discrete Galerkin methods. Macromol. Theory Simul. 5,
393–416.