(AP2) PFC Control Autoclave

You might also like

Download as pdf or txt
Download as pdf or txt
You are on page 1of 10

Predictive Functional Control of an Autoclave

Aleksander Preglej1, Igor Steiner1, Igor Škrjanc2


1
INEA d.o.o., Stegne 11, 1000 Ljubljana, Slovenia
2
Faculty of Electrical Engineering, University of Ljubljana, Tržaška 25, 1000 Ljubljana, Slovenia
E-mail: aleksander.preglej@inea.si

Abstract. This paper represents mathematical model and further predictive functional control of an autoclave.

Key words: autoclave, mathematical model, predictive functional control, temperature, pressure

Matematični model in prediktivno funkcijsko vodenje avtoklava

Povzetek. V članku je predstavljen matematični model in  we can control very easy or very complex
nadalje prediktivno funkcijsko vodenje avtoklava.
processes, processes with delays, non-minimum
Ključne besede: avtoklav, matematični model, prediktivno phase systems and non-stable systems,
funkcijsko vodenje, temperature, tlak  such control can be implemented in multi-variable
systems,
 obtained control laws are linear and easy to
1 Introduction implement.
****************not yet finished**************** The main disadvantages are the lack of theoretical
The mathematical modelling is presented in [1]. knowledge as stability and robustness in key points, and
the need of knowing the process model.
The paper is organized in the following way: in
Section 2 the predictive control is introduced, in Section The methodology of MPC controllers can be
3 predictive functional control law is described, in collapsed in three points [3]:
Section 4 the linearization and reduction of the non- 1. At some time point with help of the model we
linear model and determination of controller parameters predict process outputs for Nth horizon, which base
of the inner model is presented, in Section 5 on previous inputs and outputs, and future control
proportional control of the pressure changes is presented signal that will be estimated and sent to the
and in Section 6 PFC control of the heating and cooling system.
process is presented. 2. The future control signal is estimated with the
****************not yet finished**************** optimization of the criterion function, which
minimizes the sum of squared error between
2 Introduction in the predictive control reference signal and model output.
3. The control signal is sent to the system and in the
The name Model Based Predictive Control (MPC) next time point is this procedure repeated.
covers more algorithms [2], which base on use of the
inner model for prediction of the process output, The easiest and in industry most used model is step
estimation of the control signal with minimization response model presented in transfer function form.
criterion function and the strategy of the moving
horizon. These algorithms differ in used process model,
criterion function and noise they try to minimize. 3 Predictive functional control – PFC
The main advantages [2] are: Predictive functional control approach is analytically
 attractive for operators because for work with solvable. The main idea [3] is that in the some point
these controllers relatively little knowledge is the given reference process output coincides with
needed and estimation of the parameters is easy, real process output, which means that we estimate and
minimize criterion function in mentioned point. Another
requirement, when we estimate prediction, is the same
Received xx.xxxx, 2011
Accepted xx.xxxx, 2011
control signal:
2 Preglej, Steiner, Škrjanc

. (1)
In equation (8) stands for trajectory transfer
function. The parameter is chosen by us
In equation (1) stands for control signal, stands
, where stands for desired trajectory time
for time point and stands for number of all time
constant. From transfer function (8) we can with similar
samples. Above, stands for time point of criterion
derivation as for equation (4) get:
minimization, which is also PFC controller parameter.
. (9)
Most of the real processes can be good enough
described with first order model with delay or with In equation (9) stands for reference output and
second order model with delay. Our process of stands for reference signal.
autoclave heating and cooling will be approximated
with first order transfer function: Suppose that at time appears the reference change
. (2) and so in time the process output must change
for:

In equation (2) stands for model transfer (10)


function, stands for model gain, stands for model ,
time constant and stands for model delay.
and process model changes for:
3.1 Control law for the first order model
First we transform continuous transfer function into (11)
discrete transfer function with sample time : .

, (3)
In equation (10) stands for process output.

where , and Next we equalize both differences and , and


. we get:

For start, we will take the model without delay. . (12)


Process output prediction can be presented as following:
(4)
If we presume and , we
(5)
can write equation (12) in more transparent form:
. (13)

(6)
. The above control law (13) can be presented as seen
in Figure 1.
In equations (4) to (6) stands for model output
prediction.
Taking requirement (1) and
into account
we get:

(7)
.
Figure 1. Block scheme for the process without delay.
Reference trajectory is also given with first order
transfer function: Slika 1. Blokovna shema za proces brez zakasnitve.

. (8)
Predictive Functional Control of an Autoclave 3

On the principle of Smith predictor [3] we add delay method to cover also the influence of the disturbance
in the control law (14) and the scheme presented in input transfer function:
Figure 2, and so we get:
, (15)

(14) . (16)
.

In equations (15) and (16) stands for heating


In equation (14) stands for delayed model
process transfer function and stands for cooling
output prediction.
process transfer function.

4.2 Reduction of the higher order model


For the reduction of the third order to the first order
models we used the method Hankel minimum degree
approximation (MDA) without balancing. Basis of this
method is that for a stable system Hankel singular
values indicate the respective state energy of the system
and so the reduced order can be directly determined by
examining the system Hankel singular values. This
method guarantees an error bound, which is computed
based on Hankel singular values of transfer function, on
the infinity norm of the additive error for well-
conditioned model reduced problems [4]:
Figure 2. Block scheme for the process with delay on the
principle of the Smith predictor. . (17)
Slika 2. Blokovna shema za proces z zakasnitvijo na osnovi
Smithovega prediktorja. In equation (17) stands for original transfer
function, stands for reduced transfer function,
4 Linearization and reduction of the non- stands for desired reduced order, stands for total
singular values number and stands for single singular
linear model value.
For desired control with PFC controller we must set
inner model parameters and parameters and , which The algorithm of the mentioned method, which
can be for stable processes easily estimated from model truncates the original state space model to the order
transfer function. reduced model, when we have state space model with
When dealing with first order model with delay, we matrixes , , , and the desired reduced order , is
have to set three parameters, process gain, process time presented in continuation of this paper.
constant and delay.
First it finds the controllability and observability
4.1 Linearization in the working point grammians and .
Next, forms the descriptor
Our mathematical model is non-linear and so we have to
, (18)
fix the value of that is included in the heat transfer
coefficient equations .
Without material placed in the autoclave we have where , and descriptors state space.
the system of third order, which we linearized in Then it takes singular values of the descriptor and
working point . Next we can transform partitions the result into order truncation form:
linearized differential equations into the state space,
from where it is easy to get the model transfer function. , (19)
Because the process is affected by one input (heaters
or cooler) and one disturbance input (environment
temperature), naturally we get two transfer functions, . (20)
from which has the input transfer function major
influence. For our linearization we used before
mentioned input transfer function with slightly Afterwards, it applies the transformation to the
increased gain with already mentioned optimization descriptors state space equation (20):
4 Preglej, Steiner, Škrjanc

(21)
,

, (22)

, (23)
Figure 3. Original (solid line) and reduced (dashed line)
heating process model in frequency domain.
. (24)
Slika 3. Originalen (polna črta) in reduciran (prekinjena črta)
proces segrevanja v frekvenčnem prostoru.
And finally, it forms the equivalent state space
model:

(25)
.

The final order Hankel MDA is stable part of the


equation (25).

The proof of the Hankel MDA algorithm can be Figure 4. Original (solid line) and reduced (dashed line)
found in [5]. The error system between the original cooling process model in frequency domain.
system and the Zeroth Order Hankel MDA is an Slika 4. Originalen (polna črta) in reduciran (prekinjena črta)
all-pass function [4]. proces hlajenja v frekvenčnem prostoru.

Our transfer functions were well presented with We can see that in the reduced transfer functions the
reduced transfer functions, but the steady state error, gain is preserved and that in the beginning of the
which must be minimized for the system control, was frequency area reduced transfer functions follows the
too big. Therefore we used acquired pole and preserved original transfer function very good, but then the
steady state gain from the original third order transfer following is worse and worse.
function.
4.3 Determined parameters for the heating process
With described MDA method and preserved steady
state gain we got process reduced transfer functions: We approximated heating third order model with first
order model response very good as presented in Figure
, (26)
5.

(27)

In equations (26) and (27) stands for heating


process reduced transfer function and stands for
cooling process reduced transfer function.

The original and the reduced transfer functions can


be best compared in frequency response in Bode
diagram. In Figure 3 and Figure 4 are presented
comparisons of heating and cooling process transfer
functions in Bode diagrams. Figure 5. Third (solid line) and first (dashed line) order
heating model response comparison.
Predictive Functional Control of an Autoclave 5

Slika 5. Primerjava odzivov procesa segrevanja tretjega (polna


črta) in prvega (prekinjena črta) reda.

From equation (26) we can notice


and , and delay of the real process is
approximately .

4.4 Determined parameters for the cooling process


Next we approximated cooling third order model with
first order model response also relatively good as
presented in Figure 6. Figure 7. ON/OFF (solid line) and P (dashed) controlled
pressure comparison, both following the reference signal
(dotted line).
Slika 7. Primerjava ON/OFF (polna črta) in P (prekinjena črta)
vodenega tlaka, kjer oba sledita referenčnemu signalu (črtkana
črta).

Figure 6. Third (solid line) and first (dashed line) order


cooling model response comparison.
Slika 6. Primerjava odzivov procesa ohlajanja tretjega (polna
črta) in prvega (prekinjena črta) reda.

From equation (27) we can notice Figure 8. P controller output: pressure increasing (solid line)
and , and delay of the real process is and decreasing (dashed line).
approximately . Slika 8. Izhod P regulatorja: naraščanje (polna črta) in padanje
(prekinjena črta) tlaka.
5 Proportional control of the pressure
changes 6 PFC control of the heating and cooling
The real process pressure changes are controlled with process
ON/OFF controller, because the entry and escape valves In section 9 we concerned with the settings of the inner
are discrete. model parameters and in this sections we deal with the
We designed simulation of the proportional settings of the parameters and , which affect the
controlled pressure changes and compared its responses system response speed.
to the real process responses. We set the controllers First we considered proposed [3], [6] settings rules:
proportional part on value 25 and limited its output for
, (28)
the valve opening from 0 to 1.
, (29)
In Figure 7 we can see that limited P controller is
basically a bit improved ON/OFF controller. While the
increasing and decreasing slope of the real process with which did not return satisfying results, because we
proportional controller would be the same as with needed faster responses and also steady state error
ON/OFF controller, the improvement would be quicker smaller than 0,5 °C and maximum ramp tracking error
nearing to the set point, which is of course dependent of smaller than 5 °C, so we appreciable decreased . For
the controllers proportional part value, and reduced the parameter settings we tried various values
steady state error. In Figure 8 are presented both P meeting requirements from equation (29) and finally
controllers outputs.
fixed it to the value .
6 Preglej, Steiner, Škrjanc

In simulation of step tracking, ramp tracking and


load disturbance rejection we used the same settings,
which were for the heating process
and , and for the
cooling process and
.

*****************not yet tested*****************


In the real system we used slightly changed
parameters settings.
*****************not yet tested*****************
Figure 11. Simulated processes step tracking errors: heating
(solid line) and cooling (dashed line) error.
6.1 Step tracking
Slika 11. Napaka simuliranega sledenja stopnici: segrevanje
6.1.1 Simulation (polna črta) in hlajenje (prekinjena črta).
In Figure 9 is presented simulated PFC controlled step
tracking of both processes heating and cooling, in From presented Figure 9 to Figure 11 it can be seen
Figure 10 are presented simulated step tracking PFC that the PFC controller tracks the step very good
controller outputs and in Figure 11 are presented both because the response is quick, there is no overshoot and
errors. the error in steady state is less than 0,5 °C.

6.1.2 Real process


*****************not yet tested*****************
In Figure 12 is presented real PFC controlled step
tracking of both processes heating and cooling, in
Figure 13 are presented real step tracking PFC
controller outputs and in Figure 14 are presented both
errors.

Figure 9. PFC controlled step tracking of the simulated


processes: the reference (solid line) and the autoclave
temperature (dashed line).
Slika 9. Simulirano sledenje stopnici s PFC regulatorjem:
referenca (polna črta) in temperatura avtoklava (prekinjena
črta).

Figure 12. NOT REAL FIGURE: PFC controlled step tracking


of the real processes: reference (solid line) and autoclave
temperature (dashed line).
Slika 12. Realno sledenje stopnici s PFC regulatorjem:
referenca (polna črta) in temperatura avtoklava (prekinjena
črta).

Figure 10. Simulated step tracking PFC controller outputs:


heating (solid line) and cooling (dashed line) controller.
Slika 10. Izhod PFC regulatorjev simuliranega sledenja
stopnici: segrevanje (polna črta) in hlajenje (prekinjena črta).
Predictive Functional Control of an Autoclave 7

Figure 13. NOT REAL FIGURE: Real step tracking PFC Figure 15. PFC controlled ramp tracking of the simulated
controller outputs: heating (solid line) and cooling (dashed processes: reference (solid line) and autoclave temperature
line). (dashed line).
Slika 13. Izhod PFC regulatorjev realnega sledenja stopnici: Slika 15. Simulirano sledenje rampi s PFC regulatorjem:
segrevanje (polna črta) in hlajenje (prekinjena črta). referenca (polna črta) in temperatura avtoklava (prekinjena
črta).

Figure 14. NOT REAL FIGURE: Real processes step tracking


errors: heating (solid line) and cooling (dashed line) error. Figure 16. Simulated ramp tracking PFC controller outputs:
heating (solid line) and cooling (dashed line) controller.
Slika 14. Napaka realnega sledenja stopnici: segrevanje (polna
črta) in hlajenje (prekinjena črta). Slika 16. Izhod PFC regulatorjev simuliranega sledenja rampi:
segrevanje (polna črta) in hlajenje (prekinjena črta).
From presented Figure 12 to Figure 147 it can be
seen that the PFC controller tracks the step very good.
*****************not yet tested*****************

6.2 Ramp tracking


6.2.1 Simulation
In Figure 15 is presented simulated PFC controlled
ramp tracking of both processes heating and cooling, in
Figure 16 are presented simulated ramp tracking PFC
controller outputs and in Figure 17 are presented both
errors.
Figure 17. Simulated processes ramp tracking errors: heating
(solid line) and cooling (dashed line) error.
Slika 17. Napaka simuliranega sledenja rampi: segrevanje
(polna črta) in hlajenje (prekinjena črta).

From presented Figure 15 to Figure 17 it can be seen


that the PFC controller tracks the ramp even better that
the step because the response is also quick, there is no
overshoot and the error in steady state is less than 1,5
°C.
8 Preglej, Steiner, Škrjanc

6.2.2 Real process Slika 20. Napaka realnega sledenja rampi: segrevanje (polna
črta) in hlajenje (prekinjena črta).
*****************not yet tested*****************
In Figure 18 is presented real PFC controlled ramp From presented Figure 18 to Figure 207 it can be
tracking of both processes heating and cooling, in seen that the PFC controller tracks the ramp very good.
Figure 19 are presented real ramp tracking PFC *****************not yet tested*****************
controller outputs and in Figure 20 are presented both
errors.
6.3 Load disturbance rejection
Although PFC tracks signals very well, it is quite bad in
the load disturbance rejection. Because the disturbance
appears before the process and it is not measurable, is
the actual process input different than the expected input
and so are different also actual process and model
responses.
In Figure 21 is presented simulated disturbance of
heaters power fall and the bad PFC’s load disturbance
rejection. It can be seen that the output signal is rising
again, but very slowly as in 12000 seconds the
disturbance is not rejected yet.
Figure 18. NOT REAL FIGURE: PFC controlled ramp In Figure 22 is presented simulated disturbance of
tracking of the real processes: reference (solid line) and coolers power rise and the bad PFC’s load disturbance
autoclave temperature (dashed line).
rejection. It can be seen that the output signal is falling
Slika 18. Realno sledenje rampi s PFC regulatorjem: referenca again, but very slowly as in 9000 seconds the
(polna črta) in temperatura avtoklava (prekinjena črta). disturbance is not totally rejected yet.
In both Figure 21 and Figure 22 the disturbance is
presented illustrative.

Figure 19. NOT REAL FIGURE: Real ramp tracking PFC


controller outputs: heating (solid line) and cooling (dashed
line) controller. Figure 21. Disturbance and the bad PFC's load disturbance
Slika 19. Izhod PFC regulatorjev realnega sledenja rampi: rejection by heating: disturbance (solid line) from the
segrevanje (polna črta) in hlajenje (prekinjena črta). reference (dotted line) and autoclave temperature (dashed
line).
Slika 21. Motnja in slabo regulacijsko delovanje PFC
regulatorja pri segrevanju: motnja (polna črta) od reference
(črtkana črta) in temperatura avtoklava (prekinjena črta).

Figure 20. NOT REAL FIGURE: Real processes ramp


tracking errors: heating (solid line) and cooling (dashed line)
error.
Predictive Functional Control of an Autoclave 9

correction joint value 20981 by heating and 0,0173 by


cooling.
Correction P joint really improves load disturbance
rejection by heating as the disturbance is rejected in less
than 800 s as seen in Figure 24. In Figure 25 is
presented PFC controllers and added correction P joints
output by heating.

Figure 22. Disturbance and the bad PFC's load disturbance


rejection by cooling: disturbance (solid line) from the
reference (dotted line) and autoclave temperature (dashed
line).
Slika 22. Motnja in slabo regulacijsko delovanje PFC
regulatorja pri hlajenju: motnja (polna črta) od reference
(črtkana črta) in temperatura avtoklava (prekinjena črta).

To improve PFC’s load disturbance rejection we Figure 24. Load disturbance rejection with PFC controller and
must add the loop presented in Figure 23, which on the added correction P joint by heating: disturbance (solid line)
base of model and process output difference from the reference (dotted line) and autoclave temperature
additionally affects control signal. (dashed line).
Slika 24. Regulacija motnje s PFC regulatorjem in dodanim
korekcijskim P členom pri segrevanju: motnja (polna črta) od
reference (črtkana črta) in temperatura avtoklava (prekinjena
črta).

Figure 23. Improved PFC scheme for the load disturbance


rejection.
Slika 23. Shema izboljšanega PFC regulatorja za odpravo
motenj.

Afterwards we must carefully set the PFC and Figure 25. PFC controllers and added correction P joints
correction joints parameters that the added correction output (solid line) by heating.
joint does not have too negative affects in tracking and
at the same time it well rejects load disturbance. Slika 25. Izhod PFC regulatorja in dodanega korekcijskega P
člena (polna črta) pri regulaciji motnje pri segrevanju.
The most intuitive correction joint is [3] proportional
(P) joint with P part set to , where
Correction P joint also perceptibly improves load
parameter varies between values 0,3 and 0,7
disturbance rejection by cooling as the disturbance is
dependent on the model and the real process difference.
rejected in less than 700 s as seen in Figure 26. In
The bigger is the model and the real process
Figure 27 is presented PFC controllers and added
difference, the greater is the correction joints affect on a
correction P joints output by cooling.
transient phenomenon and we have to be careful with
the parameter settings.

6.3.1 Simulation
Our PFC controller is because of the small tracking
errors requirements already close to instability, so we
set parameter to lowest value 0,3, which gives us P
10 Preglej, Steiner, Škrjanc

8 References
[1] Preglej, A., Karba, R., Steiner, I., Škrjanc, I.
Mathematical Model of an Autoclave. Journal of
Mechanical Engineering, vol. , no. , pp. 1-14, 2011
[2] E. F. Camacho, C. Bordons, Model Predictive Control,
Springer, London, 1999
[3] Dovžan, D., Škrjanc, I. Self-Tuning Algorithms for
Predictive Functional Controller. Electrotechnical
Review: Journal of Electrical Engineering and Computer
Science, vol. 76, no. 4, pp. 205-210, 2009
[4] K. Glover, All Optimal Hankel Norm Approximation of
Figure 26. Load disturbance rejection with PFC controller and Linear Multivariable Systems and Their -error
added correction P joint by cooling: disturbance (solid line) Bounds, Int. J. Control, vol. 39, no. 6, pp. 1145-1193,
from the reference (dotted line) and autoclave temperature 1984
(dashed line). [5] M. G. Safonov, R.Y. Chiang and D.J.N. Limebeer,
Optimal Hankel Model Reduction for Nonminimal
Slika 26. Regulacija motnje s PFC regulatorjem in dodanim
Systems, IEEE Trans. on Automat. Contr., vol. 35, no. 4,
korekcijskim P členom pri hlajenju: motnja (polna črta) od
pp. 496-502, 1990
reference (črtkana črta) in temperatura avtoklava (prekinjena
[6] I. Škrjanc, D. Matko, Predictive Functional Control
črta).
Based on Fuzzy Model for Heat-Exchanger Pilot Plant,
IEEE Trans. on Fuzzy Systems, vol. 8, no. 6, pp. 705-
712, 2000

Aleksander Preglej received his B. Sc. degree in electrical


engineering from the University of Ljubljana in 2008. He is
currently employed as young researcher in economy by the
company INEA, Ljubljana. His main research interests are in
advanced process control.

Igor Steiner received the B.Sc. and the M.Sc. degrees in


electrical engineering, in ___ and ____, respectively, from the
Faculty of Electrical and Computer Engineering, University of
Ljubljana, Slovenia. He is currently employed as automation
Figure 27. PFC controllers and added correction P joints project manager by the company INEA, Ljubljana.
output (solid line) by cooling.
Slika 27. Izhod PFC regulatorja in dodanega korekcijskega P Igor Škrjanc received the B.Sc., the M.Sc. and the Ph.D.
člena (polna črta) pri regulaciji motnje pri hlajenju. degrees in electrical engineering, in 1988, 1991 and 1996,
respectively, from the Faculty of Electrical and Computer
With these settings of PFC and P part parameters we Engineering, University of Ljubljana, Slovenia. His main
research interests are in adaptive, predictive, fuzzy and fuzzy
encounter stability problems by the step and ramp
adaptive control systems.
tracking, so we should increase PFC parameters and
, but then we would not meet small tracking errors
requirements.
Easy practical solution for meeting the small
tracking errors requirements and good load disturbance
rejection is to turn the correction joint on just in steady
state and turn it off at the reference change.

7 Conclusion
****************not yet finished****************
Designed PFC control shows great response tracking
performance, because we easily kept the errors within
the requirements by the simulation and real process. The
PFC has problems with load disturbance rejection,
which we reduced with added P correction joint, but
then it impairs response tracking performance. So we
must pay attention and turn the correction joint on just
in steady state and turn it off at the reference change.
****************not yet finished****************

You might also like