SN E T

E C H N I C A L

O T E

Dynamic Modeling and Simulation of Low Density Polyethylene

Production A Comparative Study
Ilknur Disli1, Achim Kienle1,2*
Institut fur Automatisierungstechnik, Otto-von-Guericke-UniversitatMagdeburg, Postfach 4120,
39106 Magdeburg, Germany; *achim.kienle@e-technik.uni-magdeburg.de
2
Max-Planck-Institut fur Dynamik komplexer technischer Systeme, Magdeburg, Germany
1

Abstract. This study deals with the modelling and simulation of low density polyethylene (LDPE) production. LDPE
is one of the most widely produced polymers which appears in the form of simple goods in our everyday life (e.g.
food and pharmaceutical packaging, carrier bags). It is
produced in a complex industrial process which takes
place under extreme operating conditions (at pressure of
ca. 2000-3000 bar) and may lead to nonlinear dynamics
due to highly exothermic addition polymerization reactions. In principle, the process is represented by a distributed system with an external coordinate (the reactor
length > 1000m) and various internal coordinates (the
chain length of the polymer molecules, short and long
chain branching and the number of double bounds), which
can have strong effect on product properties. In this contribution a detailed reference model is introduced and
possible model simplifications are discussed systematically
from an on-line optimization and control point of view.

Introduction
Low density polyethylene (LDPE) is one of the most
widely produced polymers which appears in the form of
simple goods in our everyday life (e.g. food and pharmaceutical packaging, carrier bags). It is produced in a
complex industrial process which takes place under
extreme operating conditions (at pressure between
2000-3000 bar). It can show intricate nonlinear dynamic
behaviour due to the high exothermicity of the chemical
reaction in combination with some internal recycle
loops [7, 8], which requires some stabilizing control.
Usually, such a plant receives the raw materials from
upstream processes which may cause load changes in
the LDPE plant. It is important to maintain the product
quality in the face of such load changes. Furthermore, in
such a plant, typically more than 15 different grades of
product with different product properties (density, melt
flow index etc.) may be produced.

In order to follow frequently changing market demands it is necessary to follow an optimum grade transition policy during manufacturing different polymer
while maintaining a profitable operation.
These issues emphasize the significance of control
tasks for the LDPE plant and subsequently the significance of suitable process models which can be employed for control purposes. First attempts on optimization based control were based on steady state models
only [4]. However more recently it is possible to use
dynamic optimization methodswithin the framework of
nonlinear model predictive control (NMPC) based on
dynamic models. Although first steps have been made
in this direction [9], the size of the mathematical model
arising from the distributed nature of the LDPE plant
remains a challenging issue.
The objective of this study is to provide a benchmark model of the LDPE plant with a reasonablemodel
size which can be useful for optimal control approaches.
Therefore first a detailed reference model developed in
our group [2, 1] is reviewed and then possible model
simplifications are introduced systematically. The reference model of the LDPE plant is represented by a distributed system with an external coordinate (the reactor
length > 1000 m) and various internal coordinates (the
chain length of the polymer molecules, short and long
chain branching and the number of double bounds),
which can have strong effect on product properties.
This model describes the heat transfer in detail in
addition to including material recycles, both of which
are shown to have influence on the plant dynamics. The
experience gained on the process via reference model is
used to make necessary model simplifications. The
simulation with the suggested simple model are carried
out to analyse the effects of heat transfer, the types of
initiators and modifier on the state profiles and product
properties. Results of the simple and reference models
are compared and conclusions are drawn.

SNE Simulation Notes Europe Print ISSN 2305-9974 | Online ISSN 2306-0271
SNE 22(3-4), 2012, 1141-146 | doi: 10.11128/sne.22.tn.10145

SNE 22(3-4) 12/2012 141

A Kienle et al.

TN

Dynamic Modeling and Simulation Polyethylene Production

1 Process Description
The flow diagram of the process, which is shown in
Figure 1, includes a tubular reactor, some peripheral
units such as compressors, separators, heat exchangers,
mixers, two recycle streams. The tubular reactor of
ethylene polymerization (i.e. LDPE production) operates under high pressures between 2000 and 3000 bar.
Reactor temperature along the reactor varies between
400 K and 600 K. Such a reactor is usually very long
( > 1000 m ) and has a small diameter with a large ratio
of length to diameter such as 25000.
The feed stream to the reactor consists of high purity
ethylene as monomer and a suitable modifier (chain
transfer agent). Most often a mixture of peroxides is
used as initiator and fed at four injection points along
the reactor. Feed stream to the reactor is first preheated
then its pressure is raised to the required value through
the compressors. The addition polymerization reactions
shown in Table 1 are initiated by decomposition of
initiators around the injection positions. Typically ethylene polymerization is highly exothermic such that the
heat released during the process is required to be removed partially by cooling jackets . Right after each
initiator injection point there exist four cooling jackets.
Furthermore, in order to keep the reactor temperature below the allowed maximum value, the conversion
in the reactor is kept low by adjusting the initiator flow
rates in addition to using the available cooling capacity.
The polymer product, LDPE, is withdrawn from the last
separator whereas the unreacted ethylene and modifier
are recycled from both separators to the inlet of the
reactor to be mixed with the fresh feed.

reaction mechanism is considered. In peripheral units

quasi-stationary energy balance and well mixed fluids
are assumed. The chemical reactions of free radical
addition polymerization are shown in Table 1.
The polymer product, LDPE, includes polymer molecules of different chain lengths. It is necessary that the
chain length distribution of the product should be described. For this purpose, the method of moments is
used by including only the first 3 moment balances of
dead and live polymer molecules.

Figure 1. Flow diagram of LDPE plant.

The tubular reactor is represented by 16 modules in

series. Each module comprises a part of reactor and a
countercurrent coolant cycle around it. The model describes the heat transfer in detail in addition to including
material recycles, both of which are shown to have
influence on the plant dynamics. Model equations based
on momentum, mass and energy balances are summarized for one reactor module as follows:
Reactor:
+ ( , )

+ ( , )

142 SNE 22(3-4) 12/2012

2 Reference Model
A detailed reference dynamic model of the LDPE production plant has been developed in our group [2, 1].
The process is represented by a distributed system with
an external coordinate (the reactor length of 2 km) and
various internal coordinates (the chain length of the
polymer molecules, short and long chain branching and
the number of double bounds), which can have strong
effect on product properties.
In the reactor it is assumed that: (i) ethylenepolyethylene mixture is homogeneous; (ii) there exist
only liquid phase flowing in plug-flow mode without
axial mixing; (iii) physical properties are function of
temperature, pressure and composition; (iv) quasistationary pressure dynamics is valid; (v) a detailed

<
)

+ ( , )

Coolant cycle:

(2)
(3)

+ ( , )

(1)

( , ).

Wall:

< , = 1, . ,

(4)

=
)

(5)

There z is the axial coordinate; L is the total length of a

reactor section and t is the time; is the axial velocity of
the reaction fluid; is the stoichiometric coefficient; j is
the index for reaction; r is the rate of reaction; Hr is
the heat of reaction;
is the mass flow through the
cross section of the tubular reactor with diameter d; ,
Cp, U are density, heat capacity and heat transfer coefficient at different layers.

TN

A Kienle et al.

Dynamic Modeling and Simulation Polyethylene Production

Hafele et al. [1] further studied the influence of the heat

transfer through reactor wall
and the influence of material
recycles on the plant dynamics.
It has been shown that: (i) the
heat transfer through thick
reactor wall dominates the time
constant of the reactor; (ii) by
including the material recycles
the time constant is significantly
Table 1: Reaction mechanism : I represents the initiator, I* is the initiator radical,
increased and in some cases
M is the monomer, R is the growing or live polymer radical, P is the dead polymer,
they give rise to intricate nonn and m denote the degree of polymerization.
linear behaviour.
The state vector xi represents weight fraction of monoIn general, the detailed model represents the steady state
mer, initiators, modifier, and related radicals, moments
and dynamic behaviour of the industrial LDPE plant
of living and dead polymer chains. The state vectors T,
very well.
TW, TC, Tamb, represent temperatures of reaction fluid,
However the size of the model is required to be rereactor wall, cooling medium and ambient air respecduced in order to carry out on-line optimization and
tively.
control studies within a reasonable computation time.
The balance equations for one reactor module result
Therefore in the following section a simplified version
in a set of 30 partial differential and algebraic equations
is proposed.
(PDAEs). The peripheral units are represented by dif3 Simple Model
ferential algebraic equation systems (DAE). The simulation of the dynamic behaviour of the process is done
The simple model of the LDPE plant is based on the
within the flow-sheet simulation package DIVA [6]. As
mass and energy balances. Main simplifications are
a priory to that the PDAE system is transformed to a
done on the plant flow-diagram, the reaction mechanism
DAE system by using the adaptive method of line with
and the energy balance in addition to some other model
finite difference scheme.After discretization of the spaassumptions.
tial coordinate (60 grid points for one module) the final
Simplified plant flow-diagram. The simplified flow
DAE system consists of approximately 30000 dynamic
diagram of the LDPE plant is shown in Figure 3. It
and state variables [1].
comprises a tubular reactor with four modules in series,
This model was validated successfully by showing
one flash unit and one recycle stream. The operating
very good agreement with steady state data available
conditions are the same as in the case of detailed model.
from industrial partners. The steady state temperature and
conversion profiles are shown Figure 2.

Figure 2. Detailed model - Dimensionless steady state

temperature and conversion profiles along the reactor.

Simplified reaction mechanism. The reaction mechanism of the simple model includes the main reactions
which are highlighted in grey colour in Table 1 whereas
the detailed plant model considers all reactions listed
there. Kinetic parameters are given in Table 2 .
Simplified energy balance. It is assumed that (i) the
overall heat transfer coefficient U has a constant value
in each reactor module ; (ii) the cooling temperature Tc
is constant along the whole cooling jacket. These lead to
neglecting the energy balance equations on the coolant
medium and the reactor wall side.

SNE 22(3-4) 12/2012 143

A Kienle et al.

Dynamic Modeling and Simulation Polyethylene Production

Additionally, the following assumptions hold: one

phase (liquid) flow; no axial dispersion (plug flow);
constant pressure drop; constant physical properties;
negligible time delay between peripheral units; constant
temperature in the recycle; ideal separation in the flash
unit. Mass and energy balances for both flash and mixer
units are assumed to be quasi-stationary.
3.1 Model Equations
The model equations for one reactor module, which
comprises a part of the reactor between two initiator
injection points and one counter current coolant cycle
around it, are given as follows:
=

<

< , = 1, ,

(6)

) (7)

The boundary conditions:

Ci(0,t)=Ci,in(t) T(0,t)=Tin(t)
The initial conditions:
T(0,t)=Tin(t)
Ci(z,0)=Ci,0(z)
The state vector Ci represents concentrations of monomer, initiators, modifier and related radicals, moments
of living and dead polymer chains. The state vector T
represents the reactortemperature. The method of moments is used to represent the progress of the reaction in
terms of the leading moments of the chain length distribution of the live and dead polymer chains. These
moments are defined by the following equations:

(8)

where i and i are the ith moments of the living polymer and the dead polymer chains respectively.
0th moment corresponds to the total concentration of
the polymer. 1st and and 2nd moments are used to characterize the molecular weight distribution, e.g. the number average chain length, NACL, also known as degree
of polymerization, DPn, and the polydispersity, PD:
=
=(

/
)/( )

(9)
(10)

If one is interested only in the stability control of the

plant then the reactor model with 0th moment is satisfactory. On the other hand, the product quality information
(e.g. NACL and PD) is crucial for grade transition control problems.

144 SNE 22(3-4) 12/2012

TN

It requires to include at least 3 leading moments (0th,

1 and 2nd) in the model equations. Accordingly the
size of the model will increase.
The simple model formulation (with 3 leading moments) consists of a set of 14 PDAE per reactor module.
After transforming this PDAE system into a DAE system, simulations are carried out within the flow-sheet
simulation package DIVA [6]. The results of the simple
model are presented in the following section.
st

3.2 Results
Results obtained from simulations of the simple model
are presented for 4 different cases.
In Case I, II and III, the effects of the overall heat
transfer coefficient and the types of initiators on steady
state profiles are shown. In these cases the reaction
mechanism includes only the main reactions listed in
Table 1. Case IV includes additionally the reaction
related to the modifier consumption and aims to show
the influence of modifier on product properties. The
results obtained in each case are compared to those of
the reference model.
Case I.
In Case I, it is assumed that: (i) one type of initiator
with a constant flow-rate is fed at each injection point;
(ii) the heat transfer coefficient U has the same constant
value in all four modules. The steady state profiles of
temperature and concentration along the reactor are
illustrated in Figure 4. The reference model profiles are
represented by red solid lines, the state profiles of the
simple model-Case I by blue dashed-dotted lines. Results of the Case I are qualitatively in agreement with
the detailed model but there exists a large deviation
from reference state profiles.
This is not unexpected due to many simplifying assumptions made . In the next step these assumptions
will be modified by considering the behaviour of the
reference model.
Case II
In Case II, the effect of the heat transfer coefficient U
on the steady state profiles is shown. The detailed model
results indicate that U decreases along the reactor. Consequently, for the simple model, U will be specified at a
different mean value in each reactor module but in a
decreasing manner hrough the end of the reactor. The
assumption of using one type of initiator in eachmodule
is still valid. The resulting profiles of Case II are illustrated in Figure 4 by pink dashed lines. It can be noted

TN

A Kienle et al.

Dynamic Modeling and Simulation Polyethylene Production

that some improvement is obtained for the temperature

profile of the simple model especially through the end
of the reactor. However, a considerable large offset still
remains between the states of simple and reference
models.
Case III

Conversion

NACL

Polydispersity

Simple model (Case III)

(no modifier)

0.876

17.18

2.81

Simple model (Case IV)

(With modifier)

0.876

0.99

0.49

Table 3. Product quality values relative to the

In Case III, a mixture of two types of initiators is used at

reference model
each injection point, which is a more realistic approach.
Additionally kinetic parameters of initiator
decomposition reaction are taken from the
reference model. The condition on the
parameter U is the same as in Case II: U is
kept constant at a different value in each
reactor module. The results of Case III are
demonstrated in Figure 4 by black solid
lines. It is shown that almost perfect
agreement is obtained quantitatively between the state profiles of simple and reference models.
At this point, a question arises about the
Figure 3. Simplified flow diagram of LDPE plant.
product quality. Conversion,NACL and
PD, are given in Table 3. Although the
ethylene conversion approximately agrees
for the reference model and the simple
model (Case-III), there is a considerable
deviation in NACL and PD values. This is
due to neglecting all side reactions of the
reaction mechanism up to this point. Side
reactions have functions to stop a growing
polymer chain or to cause branching along
a polymer chain etc. In the next step, the
most important side reaction will be consider in the reaction mechanism of the
simple model.
Table 2. Kinetic rate constants: a Kim and Iedema [3],b Lee and Marano [5]
Case IV
In Case IV, the conditions are the same as
in Case III except that the reaction mechanism is extended by including the chain
transfer reaction to the modifier which is
responsible to regulate the chain length of
the polymer molecules. The state profiles
obtained in Case IV are the same as in Case
III. Moreover, NACL and PD values are
improved considerably (see Table 3). Especially NACL with a realtive value of
0.99 is very close to that of thereference
model.

Figure 4. Dimensionless steady state temperature and

conversion profiles along the reactor : Red continuous line - Detailed model;
Dashed line - Simple model-I; Dashed and dotted line- Simple model-II;
Black continuous line Simple model-III.

SNE 22(3-4) 12/2012 145

A Kienle et al.

Dynamic Modeling and Simulation Polyethylene Production

TN

As a result of these four case studies, it can be concluded that the simple model with the conditions applied
in Case IV represents the LDPE production process
reasonably well close to the reference states and product
properties. It can be used in principle research for grade
transition.

which has a much smaller size than reference model

works fine except some small deviations and has potential for stabilizing control studies. Considering that it
also provides product properties at least in a qualitative
way, the simple model can be used in principle research
for grade transition control.

Model Sizes.

References

There may be some differences in the model sizes

among the cases studied above. This depends on (i) the
number of initiators - whether one type or a mixture of
few types of initiators is used ; (ii) whether the modifier
reaction is included or not. (iii) the number of leading
moments included. However, any version of the simple
models has a much smaller size then the reference model.
The comparison of the model sizes for reference and
simple models are given in Table 4. The order of the
ODAE system for the simple model-Case IV is 5600
which is reduced by a factor of 5 comparing to the order
of 30000 in the case of reference model. If one is interested in only stability control of the plant, then one
should use a simple model with only 0th moments,
which will result in a model with an order of 2400.
Reference
Model

Simple
model

Simple
model

[2,1]

Case IV

0th moment

Number of
reactor module

16

Order of
PDAE sys

30
/ module

14
/ module

Equidistant
grid points

60
/module

100
/ module

100
/ module

Oder of the
ODAE sys

30000

5600

2400

Table 4. Comparison of model sizes.

4 Conclusions
In this study a detailed reference model was introduced
and possible model simplifications were discussed systematically from an on-line optimization and control
point of view. The simulation with the suggested simple
model were carried out to analyse the effects of the
overall heat transfer coefficient, the types of initiators
and modifier on the state profiles and product properties. The simple model was validated by comparing with
the reference model. It was shown that the simple model
146 SNE 22(3-4) 12/2012

[1] M. Hafele, M., Kienle, A., Boll, M., and Schmidt, C.-U.
(2006). Modeling and analysis of a plant for the production of low-density polyethylene. Comput. Chem.
Engng., 31, 5165.
[2] M. Hafele, M., Kienle, A., Boll, M., Schmidt, C.-U., and
Schwibach, M. (2005). Dynamic simulation of a tubular
reactor for the production of low-density polyethylene
using adaptive method of lines. J. Computational and
Appl. Math., 183(2), 288300.
[3] Kim, D.-M., and Iedema, P.(2004). Molecular weight
distribution in low-density polyethylene polymerization ;
impaxsct of scission mechanism in the case of a tubular
reactor. Chem. Eng. Sci., 59, 20392052.
[4] Kiparissides, C., Verros, G., and MacGregor, J. (1993).
Mathematical modeling, optimization and quality control
of high-pressure ethylene polymerization reactors. J. of
Macromol. Sci.-Rev. Macromol.Chem. Phys., C33(4),
437527.
[5] Lee, K., and Marano, J. (1979). Free-radical polymerization: sensitivity of conversion and molecularweights to
reaction conditions. ACS Symp.Ser., 104, 221251.
[6] Mangold, M., Kienle, A., Mohl, K., and Gilles, E.(2000).
Nonlinear computation in diva - methods and applications. Chem. Eng. Sci., 55, 441454.
[7] Pushpavanam, S., and Kienle, A. (2001. Nonlinear behavior of an ideal reactor separator network with mass
recycle. Chem. Eng. Sci., 56, 28372849.
[8] Ray, W., and Villa, C. (2000). Nonlinear dynamics
found in polymerization processes - a review. Chem.
Eng. Sci., 55, 275290.
[9] Zavala, V., and Villa, C. (2008). Large-scale nonlinear
programming strategies for the operation of low-density
polyethylene tubular reactors. In Proc. ESCAPE-18,
Lyon, France, 629634.
Submitted MATHMOD 2009: November 2008
Accepted MATHMOD 2009: January 2009
Submitted SNE: January 2010
Accepted: February 5, 2010