1st IFAC Workshop on Control of Systems Governed by

Partial Differential Equations

September 25-27, 2013. Paris, France

Modelling and Control of Extruder Barrel Temperature Field

Slavomír Lipár, Pavol Noga, Gabriel Hulkó

* Slovak University of Technology in Bratislava, Center for Control of Distributed Parameter Systems
Nám. slobody 17, 81231 Bratislava, Slovakia (Tel:+421 2 5249 7193; e-mail: slavomir.lipar@ stuba.sk).

Abstract: Modelling and control of the extruder temperature field is discussed, a process governed by a
nonlinear partial differential equation. Utilizing finite element approximation a discretized input/output
representation of the system is created with the inputs being powers of heaters and output the extruder
temperature field. Local linearization is applied at the operating point and the controller is designed based
on the lumped-input and distributed-parameter-output systems approach, using the time-space decoupling
of system’s dynamics.
Keywords: distributed parameter systems, extrusion, finite element method, nonlinear partial differential
equations, temperature control.

the barrel has to be preheated to the desired temperature

1. INTRODUCTION
profile. Insufficient preheating would cause the material not
Extrusion is the most common plastics and rubber processing melting where it is supposed to melt, consequent immense
technology. It is being used to pre-process the plastic mixture pressure rise, which eventually results in screw seizing.
for subsequent operations (e.g. injection moulding and blow Higher temperatures on the other hand may be beneficial in
moulding) as well as for the production of finished profiles. order to facilitate material flow in the barrel and lowering
The corresponding device is called an extruder. Several production costs, however though, overheating results in too
configurations are known, most commonly single- and twin- soft output material, or worse, rendering it useless due to
screw extruders with various modifications and irreversible heat induced degradation. The cooling channel
enhancements, Rauvendaal (2001). (Fig. 1. item 11) serves the protection of the screw drive
preventing it from overheating.
This paper deals with the control of a single screw extruder
barrel temperature control. The barrel temperature field has a Regarding the temperature profile, most works focus on the
major influence on the product quality as well as the process melt exit temperature only and temperatures in the remaining
itself. To get a closer insight to the problem, an experimental zones are set relatively to this value as in Previdi, Savaresi,
rig was built, Fig. 1. and Panarotto (2006), and Garge, Wetzel, and Ogunnaike
(2012). In Previdi, Savaresi, and Panarotto (2006) the
temperature in each zone is controlled individually by means
of identical SISO PID controllers. In the presented work we
use a similar structure, but with the SISO controllers
interlinked via the Space Synthesis block, Hulkó et al (2009),
so including into each individual control action computation
the effect of all other zones as well.
As stated, the barrel temperature field is created by heater
bands and monitored using temperature probes. The barrel-
plastic-screw system is governed by a nonlinear partial
differential equation (PDE). Modelling of this process was
Fig. 1. Extruder assembly: 1 - Hopper, 2 - Barrel, 3 - Die, done by finite element method (FEM) using ANSYS
4 - Screw, 5 - Heated part, 6 - Heater band, 7 - Temperature Polyflow and the model validated on the laboratory rig.
probe, 8 - Material feed, 9 - Extruded profile, Control synthesis was performed on a linearized interval
10 - Insulation, 11 - Cooling channel, 12 - Screw drive motor. around the operating point and the loop set up using DPS
Blockset by Hulkó et al (2003–2012).
The barrel’s heated part (Fig. 1. item 5) is subdivided into 6
zones, each having a heater band and temperature probe in The paper is organized as follows. Section 2 describes the
order to control the temperature profile, which depends on mathematical model of the barrel temperature control
the particular material being processed as well as the desired process, section 3 introduces the time-space decoupling based
extruded profile and/or subsequent operations, Harold, on the lumped-input and distributed-parameter-output
Wagner and Mount (2005). Before starting screw rotation, systems (LDS) approach by Hulkó et al. (1998) and Hulkó et
al (2009), and finally section 4 displays the results.

978-3-902823-54-0/2013 © IFAC 191 10.3182/20130925-3-FR-4043.00078

2013 IFAC CPDE
September 25-27, 2013. Paris, France

2. MATHEMATICAL MODEL
We developed a simplified model of the extrusion process
focused on the relation between the heater inputs and barrel
temperature field. Since the model is used to estimate only
temperatures and heat flux, the screw rotation and rheology
of the extruded plastic are not considered.
The PDE governing this system is
ρ c ( ∂T ∂t + ∇T ⋅ v ) − ∇ ⋅ ( λ∇T ) = q
, (1)
where T (K) is the temperature, q
(W) heat sources/sinks,

(kg·m-3) density, c (J·kg-1·K-1) specific heat,  (W·m-1·K-1)
thermal conductivity, v (m·s-1) flow velocity.

Fig. 3. Parameters of the extruded plastic.

viscous fluid throughout the whole temperature range

(including phase change in the real material) entering the
barrel at the hopper side and leaving it through the die.
Another simplification proving to be acceptable as the
comparison of measured and simulated data shows, Fig. 4.
Heater bands are six nonlinear domains serving the heat input
generation. This is generated in the heater bands by electric
resistance heating, which is controlled by thyristors for each
channel separately. Joule’s law states
Qel = R (T ) I 2 t = Pel t , (2)
where R (
) is the heater wire resistance, Table 1., I (A)
Fig. 2. Computational model and extruder geometry. current, Pel (W) electric power. The input heat generated in
the heater band (per unit volume) is
The computational model, Fig. 2, comprises the domains
B
– the barrel,
P – the plastic,
S – the screw,
I – thermal q
Hi = Qel,i ( t VHi ) i =16 = Pel,i VHi i =16
, (3)
insulation, and
H1…6 – the heater bands; further the 3
boundaries: 1 – the outer surface, 2 – the drive protection where VHi (m ) is the i-th heater band volume. This heat is
cooling channel, 3 – cold plastic inlet, and 4 – the hot being transferred via domain interfaces predominantly to the
plastic outlet. barrel, just a minor part dissipates to the surroundings thanks
to the insulation.
2.1 Computational domains Table 1. Heater properties
The barrel and the screw are considered temperature Heater # 1 2 3 4 5 6
invariant, thus linear in parameters, with
=7750 kg·m-3, R (
) 81.5 102.0 94.0 94.0 95.5 99.0
c=450 J·kg-1·K-1, and =20 W·m-1·K-1. This is a valid @15°C
assumption considering the fact that the operating P (W) 649.1 518.6 562.8 562.8 553.9 534.3
temperature range (25 – 300°C) is quite narrow for the steel V (cm3) 84.82
these parts are made of and the mentioned properties do not Temperature rise to 180°C causes resistance increase by 1%.
change significantly. The same is assumed for the insulation,
=0.2 W·m-1·K-1,
=18 kg·m-3, and c=1000 J·kg-1·K-1.
2.2 Boundary conditions
The plastic being extruded is a nonlinear domain with all
three parameters varying with temperature, the main The extruder outer surface is subject to free air cooling.
contribution being due to the phase change the plastic is
undergoing, Fig. 3. Latent heat of solidification is included in q
( x, t ) = h (T ( x, t ) − TA ) , x ∈ Γ1 , (4)
the heat capacity. Additionally, for computational simplicity, where x=(x,y,z) is the spatial coordinate, h=20 W·m ·K the -2 -1
the plastic is considered an isotropic compressible non- convection heat transfer coefficient, TA=25°C temperature of

192
2013 IFAC CPDE
September 25-27, 2013. Paris, France

the surrounding air. For the operating temperature range the

effect of radiation may be neglected, or better, can be easily
linearized by including it in the convection heat transfer
coefficient, Zhang, Maijer and Cockroft (2007).
Barrel cold-end cooling is incorporated in form of a Dirichlet
boundary condition with the surface temperature TW=15°C.
T ( x, t ) = TW , x ∈ Γ 2 . (5) Fig. 5. Lumped-input and distributed-parameter-output
-1 system.
Plastic inlet 3 has a defined mass flow rate Qm (kg·s ) and
inflow temperature Tin=25°C.
interest, the locally linearized system output can be expressed
T ( x, t ) = Tin , v ( x, t ) = Qm ( ρ ( x, T ) S ) ,
Γ3 x ∈ Γ3 , (6) as the sum of effects of all control variables. Considering the
discrete (in time) case, we have
2
where S3 (m ) is the inlet cross-section surface. The mass
6 6 k
conservation law requires the in- and outflow rate being Y ( x, k ) =
i =1 Yi ( x, k ) =
i =1
q = 0
i ( x, q ) U i ( k − q ), (8)
equal, therefore the flow velocity is changing with fluid
density throughout the domain. So the plastic outlet boundary where Y(x,k) is the system output, Yi(x,k) i-th output
condition is as follows component (effect of the i-th actuator),
i(x,k) corresponding
distributed impulse response of the system, Ui i-th control
v (x, t ) = Qm ( ρ ( x, T ) S ) ,
Γ4 x ∈ Γ4 , (7) variable, k time step. Impulse responses are either measured
or estimated by a computational model as in our case.
where S4 (m2) is the outlet cross-section surface. Amount of
heat extracted from the system is driven by the plastic flow {
( x, k ) =  ( x, k ) −  ( x, k − 1)} . (9)
i i i i =16
rate and outlet temperature at the die.
Further in the text we will work with a length coordinate only
The above described model was discretized by FEM with a (replacing x by x in the equations) since our interest lies in
tetrahedral mesh using ANSYS Polyflow and the FEM model the temperature profile along the barrel which is rotationally
then validated experimentally on the rig, Fig. 4. symmetrical.

Fig. 4. Response of the system to a step change in power of

heater #4, simulated (dashed) and measured (solid).
Temperature probes are located in the barrel 10mm from the Fig. 6. Extruder barrel temperature field step responses.
inner surface.
A closer look at the step responses, Fig. 6., reveals the crucial
fact, that for each one we find a point where the response is
3. CONTROLLER DESIGN
most intensive. This is the highest gain response i(xi,k),
The extruder barrel temperature field, a spatially distributed where xi is the maximum gain point for the i-th input, Fig. 7.
quantity, is controlled by 6 individual heater bands, whose Measurements performed on the real system are preferably
inputs are electric currents, lumped quantities. Such a system taken from the proximity of these locations, which are
in general is referred to as a lumped-input and distributed- generally closest to the actuator. The highest gain impulse
parameter-output system (LDS), Fig. 5. responses
i(xi,k) follow from (9) and transfer functions
{Si(z)} i=1…6. can be assigned. For each channel we get
3.1 The LDS approach Yi ( xi , k ) = Si ( z ) U i ( k ) , (10)

Being in neighbourhood of the operating point of our where {Yi(xi,k)}i,k are partial outputs in highest-gain points.

193
2013 IFAC CPDE
September 25-27, 2013. Paris, France

Once the system’s dynamics is decomposed into time-domain

characteristics {
i(xi,k)}i=1…6, {Si(z)}i=1…6 and space-domain
characteristics {
Ri(x,k)}i=1…6, {Ri(x)}i=1…6, the control
problem can be split into space-domain and time-domain
problems. Fig. 9. shows the control loop arrangement.

Fig. 7. Highest gain point responses.

These will represent time components of the system’s

dynamics. Furthermore, we introduce the normalized impulse Fig. 9. Feedback control loop. CS – control system, TS/SS –
responses in space domain, which represent the system’s time/space domain synthesis, K – time/space sampling,
dynamics in space. W(x,k) – desired temperature profile, E(x,k) – distributed
control error, (k) vector of transformed control errors, V(x,t)
{
R ( x, k )}
i i ,k
= {
i ( x, k )
i ( xi , k )}i , k , (11) – disturbance, (k)={Ui(k)}i=1…6 – vector of lumped control
quantities, Y(x,t) DPS output.
for
i(xi,k)  0. Then the system output can be expressed as
6 k
Y ( x, k ) =
i =1
q = 0
i ( xi , q )
Ri ( x, q )U i ( k − q ). (12) 3.2 Space-domain control synthesis
Analogously, for the steady states i(x, ) we introduce
Based on (14), the output/state of the controlled system is
normalized steady-state values expressed as a weighted sum of partial distributed-parameter
responses. In fact, these weights express the contribution of
{R ( x )}i i,k
= { i ( x, ∞ )  i ( xi , ∞ )}i , k , (13)
each actuating member to the overall output/state.
where {i(xi, )  0}i=1…6 are steady state values in the
The Space Synthesis block solves optimal approximation of
highest-gain points, Fig. 8.
the system’s output, reference or control error, depending on
the control loop configuration, on the system’s distributed
parameter output space (a strictly normed function space)
with the base {Ri(x)}i=1…6.
 6 

{E ( k )}
i i
= arg min

{Ei ( k )}i
E ( x, k ) −
Ei ( k ) Ri ( x ) ,
i =1
(15)

thus finding the optimal approximation coefficients, which is

in fact a transformation of the distributed control error to a
vector of lumped quantities. These transformed control errors
are subsequently processed by the time domain controllers.

3.3 Time-domain control problem

Fig. 8. Normalized steady states. The control problem is solved in time domain by means of a
group of SISO regulators, which are tuned to the time
The system’s steady-state output becomes components of the system’s dynamics {
i(xi,k)}i=1…6 (or the
6 respective transfer functions {Si(z)}i=1…6).
Y ( x, ∞ ) =
i =1 Yi ( xi , ∞ )  Ri ( x ). (14)
Using this approach the control problem is reduced to a group
of SISO control problems, since the individual controllers are

194
2013 IFAC CPDE
September 25-27, 2013. Paris, France

tracking the transformed quantities, instead of solving a full- The (reduced domain) control loop for the simulation study
scale DPS optimization problem or a MIMO controller was set up in MATLAB-Simulink using DPS Blockset, Fig.
synthesis. This enables to use any SISO control approach 10. In time domain PI1 controllers, Huba (2006) and Huba
without the need of special adaptation to DPS which makes (2007), were used. The control loop for the experiment was
the controller implementation very simple and done using Simulink Real-Time Windows Target and I/O
straightforward. cards.

3.4 Controllability in space 4. RESULTS AND DISCUSSION

A set-point change was considered. The nonlinear nature of
Number and location of actuators and measurement points, if the plastic, especially the changing density and heat capacity
not fixed in advance by the technology, is to be carefully cause shifts in the heat extraction by the output material not
chosen depending on the nature of the process. In order to only in the output zone, but having effect on the whole
make the control problem solution feasible in space-domain, temperature field. The challenging part is that the extruded
we must ensure profile is expected to be at a temperature, where it is holding
6 its shape. This means being close to the solid/liquid interface
W ( x ) −
i =1 Yi ( xi , ∞ )  Ri ( x ) ≤ , (16) where nonlinearities are the strongest.
where  is the tolerance given by the technology. This In the simulation study, feedback control was applied to the
condition becomes more stringent with increasing number of FEM model described in section 2, which serves as the
measuring points. In case of a square plant (as our case) this “plant” during controller tuning and testing. These controllers
depends solely on the values in highest gain points. were then put in operation on the rig with some fine tuning to
In the experimental rig (let us recall Fig. 1.), for each zone we be done remaining. The control process is shown in Fig. 11.
have one temperature probe for monitoring located in the 12. and 13.
barrel walls 10mm from the inner surface. It is obvious that
their location in the highest gain points is the optimal one (we
observe the barrel temperature along the line where these are
located). In general, number of probes should be at least the
number of actuators for the control problem (15) to be
determinate, as it is in our case. In this way, the control
synthesis (15) is performed on a reduced domain taking the
system output at the measuring points {x16 } ⊂ Ω , i.e.
solving the problem on a subset of the full definition domain

. If the reference variable WnFull is given on the full domain,
the equivalent reference values WnRed for the reduced domain
are calculated by
+
WnRed = RnRed  RnFull WnFull , (17)
where RnRed are reduced domain steady-states and RnRed
are full domain steady states (all quantities already
discretized and in matrix form). This ensures that the solution
of control problem (15) will be the same in the full domain
case as well as in the reduced domain case, which enables to
significantly reduce computational load and memory need.
The full scale effect however is still preserved being
determined by the plant’s DPS nature.

Fig. 11. Simulation of the barrel temperature field control

process (dashed lines indicate setpoints).
As can be seen from the control actions, Fig. 11, a significant
load disturbance is not present in the simulation. This can be
explained by the extruder construction and the resulting
relatively high inertia of the system. Another factor is the
ideal extruded material considered, which is assumed to have
very low variation in composition, consequently also in
thermal properties, thus not causing fluctuations in heat
fluxes.
Fig. 10. Control loop using DPS Blockset.

195
2013 IFAC CPDE
September 25-27, 2013. Paris, France

The proposed barrel temperature field control system is based

on the LDS approach using local linearization in the set-point
neighbourhood and PI1 controllers in time domain. Using
this approach the control system set-up kept very simple with
low requirements on controller hardware and software as well
as operator knowledge making it well suited to industrial
applications.
Acknowledgements: Research funded by grants VEGA
1/0138/11 “Control of dynamic systems represented by
numerical structures as distributed parameter systems” and
APVV-0131-10 „High-tech solutions for technological
processes and mechatronic components as controlled
distributed parameter systems” and by the European Union
with the co-financing of the European Social Fund, grant
TAMOP-4.2.1.B-11/2/KMR-2011-0001.

REFERENCES
Rauvendaal, C. (2001). Polymer extrusion, 4th edition, Hanser
Verlag, München.
Harold, F.G., Wagner, J.R., and Mount, E.M. (2005).
Extrusion: The definitive processing guide and
handbook, Wiliam Andrew Inc., Norwich.
Fig. 12. Barrel temperature field control process on the Previdi, F., Savaresi, S.M., Panarotto, A. (2006). Design of a
experimental rig (dashed lines indicate setpoints). feedback control system for real-time control of flow in a
single-screw extruder. Control engineering practice, vol.
The experimental data, Fig. 12., shows a somewhat different 14, pp. 1111–1121.
situation. Main disturbances are caused by the uncontrolled Garge, S.C., Wetzel, M.D., Ogunnaike, B.A. (2012). Control-
water flow in the cooling channels and the standalone poor relevant model identification of reactive extrusion
proprietary screw drive speed control. In addition to that a processes. Journal of process control, vol. 22, pp. 1457–
fluctuation in the processed material quality was observed as 1467.
well. The controller handles these disturbances well and the Hulkó, G. et al. (1998). Modeling, control and design of
resulting temperature fluctuations measured by the probes are distributed parameter systems with demonstrations in
within ± 0.5°C. MATLAB. Publishing house of STU, Bratislava.
Hulkó, G. et al. (2009). Engineering methods and software
support for modelling and design of discrete-time control
of distributed parameter systems. European journal of
control, vol. 15, (no. 3–4), pp. 55–73.
Hulkó G. et al. (2003-2012) DPS Blockset for MATLAB &
Simulink – Third-Party Product of The MathWorks.
http://www.mathworks.com/products/connections
Zhang, B., Maijer, D.M., and Cockroft, S.L. (2007).
Development of a 3-D thermal model of the low-pressure
die-cast (LPDC) process of A356 aluminum alloy
wheels. Materials science and engineering A, vol. 464,
pp. 295–305.
Huba, M. (2006) Constrained pole assignment control. In
Fig. 13. Quadratic norm of the control error. Menini, L.; Zaccarian, L.; Abdallah, C. T. (ed.), Current
trends in nonlinear systems., pp. 163-183., Birkhäuser,
5. CONCLUSIONS Boston,
Huba, M. (2007) Nonlinear systems. Controller design
We presented the design of a numerical model of an extruder respecting constraints (in Slovak Nelineárne systémy.
barrel including plastic flow and the related control system. Návrh riadenia s rešpektovaním obmedzení), Slovenská
The model was created using ANSYS Polyflow and validated e-akadémia, Bratislava
experimentally. This model serves as a generator of extruder Hulkó, G. et al. (2012). Control of energy systems as
characteristics and as the “plant” for controller tuning and distributed parameter systems with software support by
testing at various set-points. These features and capabilities virtual software environments. Proceedings of the 51st
of FEM software products are still not exploited well enough IEEE Conference on Decision and Control, pp. 2382-
for the support of DPS control problems. The authors’ works 2387. Maui, Hawaii.
also aim to facilitate this matter, as in Hulkó et al. (2012).

196