Professional Documents
Culture Documents
00
# Institution of Chemical Engineers
Trans IChemE, Vol 79, Part C, December 2001
I
n this study, nonparametric and parametric model based control methods were applied in
the control line in order to control the growth medium temperature of aerobic baker’s yeast
production in a batch bioreactor, and the performance was compared experimentally and
theoretically. For a non-parametric model, an experimental reaction curve was found by
performing the open-loop step test. The reaction curve was presented by the rst order dead
time model to represent the dynamic behaviour of the reactor. An Internal Model Control
(IMC) system based on a non-parametric model, has been used to track the temperature of the
reactor mixture. The simulation program is used to calculate the parameters of the parametric
Controlled Auto Regressive Moving Average (CARMA) model for Self-Tuning PID (STPID)
control. The parameters of the CARMA model are calculated using Bierman algorithm by
applying Pseudo Random Binary Sequence (PRBS).
In the rst part of the control work, a simulation program was used to observe the
performance of IMC and STPID control systems by calculating the Integral Square Error
(ISE) values. Two different kinds of operating conditions, such as load and set point effects
were applied to test the control performance for servo and regulatory behaviours. The lter
time tf for IMC and the rst parameter of Tailoring polynomial for STPID were used as tuning
parameters. In addition, PID control system performance was compared with both parametric
and non-parametric control systems.
In the second part of the control work, the growth medium temperature of aerobic baker’s
yeast production was controlled experimentally by using IMC and STPID systems in an on-line
computer controlled batch bioreactor. Non-parametric and parametric models for both control
systems were tested by considering the reaction heat as a load effect. The heat input given from
the immersed heater was chosen as the manipulated variable. It was observed that IMC was
more effective than STPID and PID methods.
Keywords: Internal Model Control; Self Tuning PID control; pole placement technique;
Integral Square Error criteria; Baker’s yeast; bioprocess control.
242
EFFECT OF TUNING PARAMETERS ON MODEL CONTROL METHODS 243
signal is generally applied to obtain input–output data from were separately sterilized in an autoclave. The bioreactor
process. Different kinds of identi cation algorithm such as and other auxiliary equipment were sterilized with ethanol.
Bierman9, Levenberg-Marquard10, Genetic algorithm11 can Air was continuously supplied to the bioreactor by a sparger
be used in the calculation of the model parameters. For non- after passing through an air rotameter and microbiological
parametric models, the parameters of these models can be lter. Cell concentration was determined turbidimetrically at
calculated by using related curves such as the reaction curve 580 nm with a Shimadzu (Tokyo, Japan) model UV-160A
and Bode diagrams. The reaction curve is one of the non- spectrophotometer.
parametric model and it is obtained from the step change A 2 l bioreactor with a cooling jacket was used. Sensors
given to the manipulated variable. It is very important to were placed in the reactor to measure pH, temperature and
de ne the time constant and dead time. The slope of rising dissolved oxygen in the culture medium. Air was continu-
curve and the second steady-state condition must be chosen ously fed into the bioreactor by passing through a rotameter
very carefully in de ning the parameter values. The system and microbiological lter at a rate of 0.83 vvm. Agitation
model parameters can be changed using several methods. was supplied using turbine impeller and studied at 300 rpm.
However, in some cases, instead of changing parameters An immersed heater for heating the culture medium to
continuously, the tuning parameters can compensate the the desired operating temperature was also placed in the
changes in the dynamic conditions of the system. bioreactor.
In this work, a well known microorganism, particularly On-line computer control was used to control the growth
used in the food industry, S.cerevisiae (Baker’s yeast), was medium temperature. By using a thermocouple, the culture
used. The temperature of the growth medium in a batch temperature was measured. An ampli er was placed
bioreactor was controlled using IMC and STPID algorithm. between the A=D converter and thermocouple to convert
An experimental step test was used to de ne the time constant the 0–4 mV pulses obtained from the thermocouple to 0–10
and dead time of rst order transfer function for the nonpara- volt pulses. For both Internal Model Control and Self-
metric model. These parameters and models were used in the Tuning PID controller related control algorithms were
IMC control system. As well as constant values of para- developed for on-line control purposes. Heat given from
meters, lter time tf was chosen as a tuning parameter of the the immersed heater was chosen as the manipulated vari-
IMC control system. The best value of lter time was able. The value of the manipulated variable was computed
calculated according to time variation of the reactor tempera- with the control algorithm. This value was converted into a
ture. In the parametric model, parameters were calculated signal and then it was passed through a D=A converter and
using the PRBS signal and Bierman algorithm. According to converted into 0–4 mV pulses before they reached the triac
the system dynamics, a second order CARMA model was module (see Figure 1).
used for STPID control. Proper values of magnitude and the In the experimental work, a computer controlled system
time step of the PRBS signal were chosen. The parameters of was used to observe and control the culture temperature.
the CARMA model were used in the STPID control algo- The experimental procedure is summarized below.
rithm. The rst parameter of the Tailoring polynomial was Initially, the bioreactor was lled with the culture medium
used as a tuning parameter for STPID control. The best value consisting of glucose and mineral salts. The medium was
of the t1 parameter was found by calculating the performance heated to the desired temperature of 30° C. The experimental
index. Theoretical and experimental control work were operating conditions are shown in Table 1. Finally, the
realized to observe controller performance. ISE values were medium was inoculated to start the cell growth. During
computed for the closed-loop process output. In the theore-
tical work, the simulation program which had energy and
mass balance equations together with control algorithms, was
used to calculate the control performance. Load and set point
effects were examined to observe the control performance for
servo and regulatory behaviours.
Table 1. Optimal operating conditions used in experimental · calculate the predicted controlled variable, i.e. the output
studies and simulations. from the model Gm (s): The dynamic model can be
Temperature 30° C simulated in the discrete form as:
pH 4
Mixing rate 300 rpm ym (k) = Km (1 - e- t=tm )u(k - N - 1) + e- t=tm y(k - 1)
Culture medium volume 1.6 l
Cooling water ow rate 1.4 ml s- 1 (8)
Qo 1.7 cal s- 1
Air ow rate 0.83 vvm where t is the sampling time;
· calculate the difference between the measured and model-
predicted controlled variables, then nd the error signal
as:
the cell growth, the increase of the culture temperature and
cell concentration were observed. Controlled and uncon- e(k) = ys (k) - [ y(k) - ym (k)] (9)
trolled conditions were realized experimentally. Reaction
heat was assumed as a load effect. · calculate the IMC controller output as:
Here r(t) represents the set point, and: By equating the denominator of equation (24) with
-1 -2
equation (25), the following relationships are obtained:
S = s0 + s1 z + s2 z (16)
t1 - a1 + 1
s0 =
Dt tD b0
s0 = K C 1 + + (17) t2 - a2 + a1
2tI Dt s1 = (26)
b0
t3 + a2
Dt 2tD s2 =
s1 = K c - 1 + - (18) b0
2tI Dt
The discrete form of the necessary incremental PID control
tD -1 law15 may be written in terms of the change in the control
s2 = K C and R = (1 - z ) (19) signal as:
Dt
Here Dt is the sampling interval. The PID constants can be DU = s0 e(t) + s1 e(t - 1) + s2 e(t - 2) (27)
found from the values of s1, s2 and s3. Substituting the
control equation into CARMA, the process model yields the The steps in the operation of the self-tuning used in this
following closed-loop response equation: study can be given as:
(1) apply a perturbation to the system as a forcing function
z- 1 BS RC and attain the plant output;
y(t) = r(t) + e(t) (20)
AR + z- 1 BS AR + z- 1 BS (2) estimate A and B from the CARMA model using Bier-
man UDUT update algorithm;
The characteristic equation is called a Tailoring poly-
(3) calculate s0, s1 and s2 from the equations;
nomial T and it is given by:
(4) nd Kc, tI and tD from the equations;
T (z- 1 ) = A(z- 1 )R + z- 1 B(z- 1 )S(z- 1 ) (21) (5) obtain the incremental control signal from equation
(27);
The properties of this closed-loop can be varied by placing (6) output the updated control signal to the process;
the poles of the characteristic equation within the unit-circle (7) return to 1.
in the z plane. The coef cients of the A and B polynomials
In this work, the form of the system model to be
are estimated from the Bierman UDUT algorithm9, and the
controlled is preserved to ensure that only one set of PID
coef cients of the T polynomial are de ned by the user. s0,
controller coef cients is produced from the design and the
s1, s2 can be found by solving the set of simultaneous
integral action in the PID controller provides steady-state
equations obtained from the closed-loop equation.
even if the parameter values of the system or controller
The degrees of the polynomials in the characteristic
change.
equation are:
na + nr = nb + ns + 1 = nt (22)
RESULTS
Where ns is the degrees of the S and it is taken as 2, and nr is Some theoretical and experimental studies have been
the degree of R polynomials, the value must be 1 because of carried out to observe the performance of IMC and
the polynomial representation of velocity form of the PID STPID control systems to track the reactor temperature at
algorithm. This means that na = nb + 2 and nt = nb + 3 = the desired set point. In the dynamic work, the culture
na + 1: If a second order A polynomial (na = 2; nb = 0 and temperature was not controlled and the inhibitory effect of
nt = 3) is selected, a unique set of PID controller coef - the increasing culture temperature on the speci c growth
cients can be obtained from the design. If the order of the A rate of cells was observed. Hence the necessity of control-
polynomial is three, i.e. na = 3; nb = 1 and nt = 4; all the ling the culture temperature was realized.
coef cients of T polynomial should be user de ned to place In the rst part of the study, the theoretical simulation
the poles of the characteristic equation easily. In this case, work was carried out to calculate the ef ciency of the related
the system transfer function chosen is a third order T control systems. This simulation program consists of mate-
polynomial (na = 2; nb = 1) and has the form: rial and energy balance developed for this batch bioreactor.
b0 z- 1 Necessary equations for the energy and material balance
y(t) = u(t) (23) were taken from literature16. Dynamic behaviour of the
1 + a 1 z- 1 + a 2 z- 2 system under various load effects were investigated by
The closed loop relationship is obtained by combining the using this simulation program. Parameters of parametric
system model equation (23) and the controller equation (15) models were also calculated with the data obtained from
as: this simulation program.
b0 z- 1 S
y(t) = r(t) (24) Table 2. Comparison of speci c growth rate at controlled and uncontrolled
R(1 + a1 z- 1 + a 2 z- 2 ) + b 0 z- 1 S temperature.
The equivalent closed loop T polynomial is of the form: Controlled temperature Uncontrolled temperature
- 1
m (h ) 0.57 0.23
T = 1 + t1 z- 1 + t2 z- 2 + t3 z- 3 (25)
A comparison of the speci c growth rate between was found by performing the open-loop step test. When the
controlled and uncontrolled cases was given in Table 2. system is at the steady state conditions as given in Table 1, a
Temperature control is very effective in keeping the yeast step change is given to the heat input (DQ = 21.7 cal s- 1).
concentration at the desired value. Very little uctuation at The time variation of temperature to this effect is presented
the desired temperature value results in high difference in in Figure 4. The representation of the transfer function of the
the yeast concentration. The increase of the culture tempera- reactor in equation (1) was obtained from this gure. Heat
ture and cell concentration during the cell growth are shown input from the immersed heater was selected as a manipu-
in Figures 2 and 3. It can be seen that maximum cell lated variable.
concentration was obtained at the end of the exponential For the IMC system, the nonparametric rst order dead
growth phase, 4.5 h. For this reason, a comparison of the time model presented in equation (1) was obtained to
two different controller performances was realized between represent the dynamic behaviour of the reactor, which is
0 and 4.5 h time interval. In the second part of the work, the given as follows:
experimental study was realized to observe the dynamic
behaviour of the batch reactor, and parameters of the 0:56e- 3:5s
Gp = (28)
nonparametric model were determined. Reaction media 37:5s + 1
was placed into the bioreactor for this purpose. The reactor
was heated by means of an immersed heater, then cooling For this transfer function of the system, the unit of the
water was passed through the reactor cooling jacket. In this time constant and dead time are minute. Sample time t was
case, the heat given by the heater was absorbed by the taken as 37.5 sec. The inverse of the Laplace transform for
cooling water and, this reactor lled with reaction media, this transfer function was evaluated and then time variation
was considered as a heat exchanger. This system can reach a of the reactor temperature was given, as shown in Figure 4.
steady state condition with de ned values of heat input and In addition, the simulation program, having only the energy
cooling ow rate (see Table 1). Therefore, IMC and STPID balance of the batch reactor, was used to represent the
systems can be designed for processes operating in the dynamic behaviour of the system to obtain the theoretical
vicinity of a steady state. An experimental reaction curve open loop step test. A similar step change was given to the
system when the reactor was in a steady state condition as
given in Table 1. The time response of the theoretical reactor
temperature is calculated and given in Figure 4. As can be
35 easily observed from Figure 4, the agreement among the
responses of the reactor temperature obtained from the
34
computer simulation program, experimental data and inverse
33 of the transfer function for the batch bioreactor is suf -
Temperature ( C)
ciently good.
o
32
The simulation program with energy balance is used to
31 calculate the parameters of the parametric CARMA model
30
for STPID control. Theoretical estimation algorithm is given
as follows. When the batch bioreactor is in the steady state
29 condition given in Table 1, the Pseudo Random Binary
28
Sequence (PRBS) signal is given to the heat input to disturb
0 1 2 3 4 5 6 7 8 the system. The Bierman algorithm is used to calculate the
Time (h) CARMA model parameters with input output data. The
magnitude of the PRBS and rate of sampling were chosen
by knowledge of the process dynamics and the behaviour of
Figure 2. The increase of growth medium temperature with time. the nal control elements. A second order polynomial is
8 42
Yeast concentration (g/L)
40
6
38
Temperature ( C)
o
36
4
34
experimental
2 32 theoretical
desired theoretically Laplace
uncontrolled experimental 30
CARMA
0 28
0 1 2 3 4 5 6 7 8 0 10 20 30 40 50 60 70 80
Time (h) Time (min)
Figure 3. Theoretically desired and uncontrolled yeast concentration
pro les carried out under optimal conditions. Figure 4. Process reaction curve.
suf cient to represent the denominator plant dynamics. set point. Tuning control parameter tf is varied to obtain the
From estimation procedure, the coef cients found are: desired and ef cient control performance in IMC system. So
the response of the IMC system to a change in the set point
0:00876
y(t) = u(t - 1) (29) is examined with different lter time constants and the
1 - 0:5387z- 1 - 0:4603z- 2 related theoretical results are given in Figure 5. As can be
The time response of the CARMA model with heat input seen from Figure 5, when larger tf values are used, system
change (Du = 21.7 cal s- 1) is given in Figure 4. Suf cient response shows sluggish behaviour, smaller values give fast
agreement is obtained. It can be concluded that the esti- and smooth response.
mated CARMA model can be used to design the STPID The theoretical temperature response of the STPID
control. system to a similar step change in the set point were
Both system transfer function and CARMA model para- calculated (see Figure 6). The tuning control parameter t1
meters are not changed during the fermentation processes. of T polynomial is varied to obtain suf cient control
These approaches are suf ciently acceptable for this study. performance. Three different t1 values were tested and
Only the tuning parameters for both control systems were system responses examined. It was seen that when the t1
changed to obtain suf cient control performance by elim- value is placed near the edge of the unit circle, the response
inating the model error from real system. is unstable, as expected. A comparison of the performance
In the third part of the study, the theoretical simulation of the controller for the set point change based on ISE
work was carried out prior to experiment. The total simula- values are given in Table 3. In addition, the ISE values of
tion program using material and energy balance, were used IMC control system gives the best performance according to
to calculate the time response of the batch bioreactor with ISE values, so nonparametric model based control gives
IMC and STPID control systems. The best values of control more satisfactory result than parametric model based
tuning parameters were determined using this simulation control.
program. As a means of comparing the performance of the In another part of the study, step change in the cooling
IMC and the STPID controllers, the integral square of the water ow rate is chosen as a load effect and the properties
error (ISE) is computed for the closed loop process output, of IMC and STPID control systems have been tested for
where:
tf
ISE = [ y(t) - r(t)]2 (30) 33
t= 0
32
to be de ned by the user viz, NP, NC and tf. NP and NC can
o
t1= -0.1
ture17. PID control parameters were adjusted; Kc = 60:25;
o
t1= -0.6
tI = 1.219 min and tD = 0.18 min. By using Cohen Coon 31 t = -0.9
1
method under all conditions, the performance of the error first set point
(ISE) is computed for the closed loop process output. second set point
disturbance rejection. Simulation programming is used to latory behaviour. System response under STPID control for
calculate the system output. load effect is shown in Figure 8. Three different t1 values are
For the IMC control system, when the system is in steady- again chosen to test the performance of the relevant control
state condition as given in Table 1, a step change from system. The ISE values are computed for both control
1.4 m s- 1 to 0.7 m s- 1 is given to the cooling water ow rate systems and the results are shown in Table 4. The IMC
and time response of the system is examined using simula- performance is marginally better than the STPID perfor-
tion program. Relevant results are given in Figure 7. Again mance. However, both of these control strategies are much
two lter time constants are chosen for examining the better than conventional PID control results in the same
controller performance. Similar calculations are made for table.
tuning the lter time constant. Small tf value gives fast and Experimental work was carried out in light of the theore-
smooth responses, greater values shows sluggish and oscil- tical studies. The tuning parameters which give the best ISE
values were used. Necessary experimental conditions are
given in Table 1. The experimental results of temperature
33 control by IMC algorithm are given in Figure 9. As can be
t = 0.1 seen, even if the culture temperature is at the set point at the
f
t = 10 beginning of the experiment, the IMC system brings this
f
temperature quickly to the desired value and then it follows
Temperature ( oC)
32
set point
set point smoothly. Second control was made with STPID
control algorithm and the relevant results are shown in
31 Figure 10. Initially the culture temperature was again at
the set point, but the response was more sluggish and
oscillatory than the IMC algorithm. The ISE values, both
30 experimental and theoretical, were computed for each
control system and are shown in Table 5. Table 5 and
Figures 9 and 10, indicate that the performance of the
29
0 1 2 3 4 5 6 7 IMC system gives better and more ef cient response in
Time (h) following the desired set point than the performance of other
algorithms.
Figure 7. Response of IMC system to disturbance rejection with varying
lter time constants (m1 = 1.4 mls- 1, m2 = 0.7 mls- 1).
DISCUSSION
35 In this study, the performance and quality of the IMC and
t1= -0.1 STPID controllers have been examined using ISE criterion.
34 t1= -0.6 Experimental and theoretical work have been undertaken to
Temperature ( o C)
t1= -0.9
set point
observe the performance of these control systems in the face
33 of load disturbance and set point change. Heat input to the
bioreactor was chosen as the manipulated variable, and the
32
primary control objective is to maintain the bioreactor
31
temperature at the desired value. In all cases, the manipu-
lated variable which is heat input produced good sensitivity
30 and rapid response. In the experimental work, when the
process is under the reaction heat effect, IMC shows better
29 performance than the STPID controller with little ISE
0 1 2 3 4 5 6 7 values, and gives more ef cient control than STPID. Similar
results were also observed in theoretical work. In all
Time (h)
theoretical and experimental studies, the PID control
Figure 8. Response of the STPID system to load rejection with varying t1 system did not show desired control performance depending
values (m1 = 1.4 mls- 1, m2 = 0.7 mls- 1). on ISE criteria.
Table 4. Subsequent closed-loop ISE values for closed-loop response method under load effect.
y process output
ysp output set point
31
Greek letters
tD derivative time constant, minutes
30 tf lter time constant
tI integral time constant, minutes
tm model time constant
Dt sampling time, minutes
29 m speci c growth rate, h- 1
0 1 2 3 4 5 6 7 8
Time (h)
REFERENCES
Figure 9. Experimental temperature control (IMC) in the face of reaction 1. Bailey, J. E. and Ollis, D. F., 1986, Biochemical Engineering Funda-
heat, tf = 0.1. mentals, 2nd ed. (Mc Graw-Hill, Singapore), pp 689–725.
2. Cardello, R. J. and San, K. Y., 1988, The design of controllers for batch
bioreactors, Biotechnol Bioeng, 32: 519.
33 3. San, K. Y. and Stephanopoulus, G., 1984, Studies on on-line bioreactor
identi cation. IV utilization of pH measurement for product estimation,
experimental
theoretical
Biotechnol Bioeng, 26: 1208.
32 4. Douglas, A. and Ramirez, W. F., 1988, Optimal temperature control for
Temperature ( C)
set point
batch beer fermentation, Biotechnol Bioeng, 31: 224.
o
5. Suzuki, T., Yamane, T. and Shimizu, S., 1986, Control of carbon source
supply and dissolved oxygen by use of carbon dioxide concentration of
31 exhaust gas in fed-batch culture, J Ferment Technol, 64(4): 317.
6. KahveciogÏ lu, A., Ertunç, S., Akay, B., HapogÏ lu, H. and Alpbaz, M.,
2000, Adaptive Control of S.cerevisiae Production, Research 2000
30 (Institution of Chemical Engineers, UK), pp 38.
7. ErdogÏ an, S., Özkan, G. and Alpbaz, M., 1998, Self-tuning control of
batch polymerization reactor, J Chem Eng Japan, 31(4): 499–505.
29 8. Yüce, S., Hasaltun, A., ErdogÏ an, S. and Alpbaz, M., 1999, Temperature
0 1 2 3 4 5 6 7 8 control of a batch polymerization reactor, Trans IChemE, Part A, Chem
Eng Res Des, 77(A5): 413–420.
Time (h) 9. Bierman, G. J., 1976, Measurement updating using the U-D factoriza-
tion, Automatica, 12: 375.
Figure 10. Experimental temperature control (STPID) in the face of reaction 10. Levenberg, K., 1994, A method for the solution of certain non-linear
heat, tf = 7 0.1. problems in least squares, Quarterly Appl Math, 2: 164–168.
11. Wang, P., Kwok, D. P., 1994, Optimal design of PID process controllers
based on genetic algorithms, Cont Eng Prac, 2(4): 641–648.
Table 5. ISE values for closed loop response method. 12. Wittenmark, B. and Aström, K. J., 1984, Practical issues in the
implementation of self-tuning control, Automatica, 20(5): 595.
ISE ´ 100 13. Seborg, D. E., Edgar, T. F. and Shah, S. L., 1986, Adaptive control
strategies for process control: a survey, AIChe J, 32(6): 881–909.
Experimental Theoretical 14. Newell, R. B. and Lee, P. L., 1987, Applied Process Control (Prentice
Hall, USA), pp 75–79.
IMC (tf = 1) 1.35 24 15. Stephanopoulos, G., 1984, Chemical Process Control: An Introduction
STPID (t1 = 7 0.1) 11.52 24 to Theory and Practice (Prentice-Hall, Inc., USA), pp 310–315.
PID 18.22 25.48 16. Bursali, N., 1997, Application of Optimal Adaptive Generalized
Predictive Control Method to a Batch Bioreactor, Ph.D. Thesis
(Ankara University, Turkey).
17. KahveciogÏ lu, A., Application of Self Tuning PID Control to a Biore-
NOMENCLATURE actor in which S. cerevisae is Produced and Determination of Optimal
Control Parameters, Master Thesis (Ankara University, Turkey).
A=D Analog=Digital converter
d(z) disturbance
d(z) feed-back signal
dm dead-time, minutes ACKNOWLEDGEMENT
e(z) IMC controller input
The authors gratefully acknowledge Ankara University Research Fund
F(z) lter transfer function
for providing nancial support.
GC(z) Internal Model Controller transfer function
GP(z) real process transfer function
Gm(z) process model transfer function
IMC Internal Model Control ADDRESS
ISE Integral Square Error
KC controller gain Correspondence concerning this paper should be addressed to Pro-
Km model gain fessor M. Alpbaz, Department of Chemical Engineering, Engineering
na degree of A polynomial Faculty, Ankara University, TandogÏ an, Ankara, 06100 Turkey. E-mail:
nb degree of B polynomial alpbaz@science.ankara.edu.tr
nr degree of R polynomial
nS degree of S polynomial The manuscript was received 24 July 2000 and accepted for publication
nt degree of T polynomial after revision 16 August 2001.