Content uploaded by Ivan Carlos Franco
Author content
All content in this area was uploaded by Ivan Carlos Franco on Dec 05, 2016
Content may be subject to copyright.
BIOREACTOR TEMPERATURE CONTROL USING A GENERIC FUZZY
FEEDFORWARD CONTROLSYSTEM
Rodolpho R. Fonseca1, Ivan C. Franco1,2 and Flávio V. da Silva1,*
1School of Chemical Engineering, University of Campinas-UNICAMP, Campinas, SP, Brazil
2Department of Chemical Engineering, University of FEI, São Bernardo do Campo, SP, Brazil
*Corresponding author, e-mail: flavio@feq.unicamp.br
ABSTRACT
Fermentation process is widely used in industries to
produce biofuels, beverages, foods and medicines. As
example, ethanol is obtained via corn syrup fermentation
by Saccharomyces cerevisae, a bioprocess that is thermal
sensitive and requires temperature control to maintain
microorganism metabolic rate. In a CSTR, the feed stream
temperature is normally a load source, affecting process
variable control that makes a feedforward control system
highly recommended to this case. However, bioprocesses
are intrinsically nonlinear processes, becoming very
difficult to obtain satisfactory control performance using
classic feedforward control. Nevertheless, Fuzzy logic
control is indicated to this type of process because results
in nonlinear control law. For this reason, this work
presents a generic Fuzzy feedforward control strategy
developed to be used with a feedback control system in
nonlinear processes, aiming load rejection improvement.
The results shown that the Fuzzy feedforward feedback
control system (4FB) was better than only feedback
control to reject feed stream temperature loads in
Saccharomyces cerevisae fermentation in a CSTR, being
an interesting control strategy for load rejection in
nonlinear processes.
KEYWORDS
Fuzzy Logic, Feedforward, Process Control, Bioprocess.
1. Introduction
In nonlinear processes, to establish a model that
represents the process dynamic in time is a complex task,
sometimes being necessary advanced mathematical and
statistical methods to determine the parameters [1-3].
Bioprocesses are widely used in fuel, pharmaceutical and
food industry and are classic nonlinear processes, with
large amount of variables that can be determined in
laboratory experiments [4]. Fermentation of
monosaccharide to produce ethanol from sugar cane or
corn syrup is an alternative to fossil fuels and, for this
reason, has been widely employed by countries interested
in more sustainable energetic matrix using renewable
sources. The glucose fermentation using Saccharomyces
cerevisae to produce ethanol can be performed in a
continuous stirred tank reactor (CSTR) with feed stream
of glucose solution and, often, feedback control loops are
used to regulate process variables as fermentation
temperature, pH and feed rate, as examples [4-6]. This
stream is a typical source of loads as glucose
concentration variations around the steady state
operational condition.
In case of load on feed stream parameters, a
feedforward control strategy may be applied for disturb
rejection as used in other similar cases [7-9]. The classic
feedforward control law bases on process and disturbance
model and, for nonlinear processes, it can become a hard
task applying process control. Though, this setback
caused by nonlinearities can easily be transposed using
artificial intelligence techniques as Fuzzy logic, that has
been widely used to process control in industry [6,10,11].
Fuzzy logic controllers ensures nonlinear input-
output static map that can be defined for each specific
application modifying its knowledge basement,
membership functions and input/output discourse
universes [11]. Further, Fuzzy controllers are able to be
designed with multiple inputs and outputs, what is
advantageous for process control. Thus, a Fuzzy
feedforward structure could allow load rejection without
being necessary the model identification and validation,
transposing so the typical limitation of classic
feedforward control law for nonlinear processes.
Therefore, this work reports the application of a
generic Fuzzy feedforward controller in association with
temperature feedback control of a CSTR fermentation for
ethanol production, in attempt to improve load rejection
caused by feed stream temperature variations. To support
control performance analysis, criteria indexes IAE and
IFF/FB were calculated and discussed for each simulation
experiment.
2. Materials and Methods
2.1 Bioreactor model with Fuzzy - Split range control
In this work, it was applied a modelling of a CSTR
bioreactor with temperature control to produce ethanol
Proceedings of the 15th IASTED International Conference
August 16 - 18, 2016 Campinas, Brazil
Intelligent Systems and Control (ISC 2016)
DOI: 10.2316/P.2016.841-024
275
using glucose inlet stream and Saccharomyces cerevisae,
as presented by [4]. The modelling and bioprocess
parameters are shown elsewhere [4,5]. For the bioreactor
temperature control, it was used a Fuzzy-Split range
strategy presented by [5], which uses a Fuzzy-PID with
output signal split range to control bioreactor temperature,
as shown in Figure 1. The split range was used to allow
cooling and heating control action with the utilities
streams at 15°C and 60°C, respectively. As proposed by
[5], the Fuzzy-PID output was split in two ranges, one for
cold utility valve opening and the other for the hot utility
valve.
Figure 1. Fuzzy - Split range control system for the
CSTR bioreactor.
The Fuzzy - Split range control was developed with
the same bioreactor model used in this work. Therefore,
the Fuzzy-PID tuning parameters used in the simulations
were the same used by [5].
2.2 Generic Fuzzy Feedforward
As a classic feedforward, the Fuzzy feedforward output
signal is added to the feedback control output signal in
attempt to reject the load, as presented in Figure 2. As
discussed before, there are different feedforward control
laws, being the most indicated the lead-lag with delay
(Gff) shown in Equation 1, that has dynamic response
considering the process model, although a simple static
feedforward can drastically improve the control
performance [7]. Nevertheless, in both cases is necessary
to determine the process gain (Kp) and disturbance gain
(Kd) by process modelling, what is complex and time
consuming for nonlinear processes.
(1)
In order to promote load rejection in process with
nonlinearities, a generic Fuzzy feedforward control was
proposed in this work to be used with feedback control,
resulting in a Fuzzy feedforward feedback system (4FB).
The Fuzzy feedforward controller in 4FB uses the
disturbance values (D(t)) and the process variable error
(E(t)) as inputs, and the output set to increment the
feedback controller output, as a classic feedforward
control. Also a time delay was attributed to Fuzzy
feedforward signal like in feedforward lead-lag with time
delay control. The Figure 2 presents the control diagram
for the 4FB control strategy.
The load input of Fuzzy feedforward was set with 5
membership functions (MF) with the antecedents “Too
Low” (TL), “Low” (L), “Normal” (N), “High” (H) and
“Too High” (TH), and for error input were set 3 MFs
designed as “Negative” (NEG), “Zero” (ZR) and
“Positive” (POS). For the Fuzzy output was set 5 MFs
with the consequents “Too Negative” (TNEG),
“Negative” (NEG), “Zero” (ZR), “Positive” (POS) and
“Too Positive” (TPOS). For both inputs and the output,
the MFs were set as triangular form with equidistant
distribution on discourse universe. The Figure 3
represents the inputs and output MFs of the Fuzzy
feedforward.
Figure 3. MFs and discourse universes of Fuzzy
feedforward a) Load input (D(t)), b) Error input (E(t)) and
c) Fuzzy feedforward output.
Figure 2. Fuzzy feedforward feedback control strategy
(4FB) using Fuzzy – Split range as feedback control.
276
The Fuzzy feedforward controller was designed to be
only necessary to adjust the time delay and the discourse
universe of inputs and outputs, depending on the
minimum and maximum load values, and also error and
output ranges, which are typical for each process.
The time delay and discourse universes of inputs and
output are the tuning parameters of Fuzzy feedforward,
being necessary to determine them via experiments to
improve load rejection. As a rule of thumb, to avoid
feedforward control action in absence of load, set the
center of discourse universes of error input and output to
zero is needed, and for load input, set to disturbance
variable value without load step. The discourse universes
for inputs and output used in this work are presented in
Figure 3.
The base rule was set to be the same for every
application, not being designed concerning process
characteristics, but the typical response of a classic lead-
lag with time delay feedforward control in presence of a
disturbance. The Fuzzy feedforward base rule is presented
in Table 1, and was designed based on specialist
knowledge about feedforward control responses.
Table 1. Fuzzy feedforward generic base rule.
E(t)\D(t)
TL
L
N
H
TH
NEG
TPOS
TPOS
ZR
ZR
NEG
ZR
TPOS
POS
ZR
NEG
TNEG
POS
POS
ZR
ZR
TNEG
TNEG
The base rule does not have to be modified in case of
direct control or inverse control, being necessary only to
change the sum signal with the feedback control output.
2.3 – 4FB control for nonlinear process
To test the 4FB control strategy, simulations were done
with different scenarios of load in feed stream
temperature (Tin). At the first scenario, loads were
simulated with steps as shown in Table 2. For the second
scenario, a more aggressive load set was simulated as
uniform random distribution steps between 19°C and
30°C, with a sample time of 1 h and seed equal to zero. In
each scenario, 4FB control strategy was compared with
the Fuzzy – Split range system without feedforward
control action in attempt to demonstrate the advantages of
4FB.
Table 2. Feed stream temperature loads at first scenario.
Time [h]
0
50
100
150
200
250
Tin [°C]
29
24
19
28
22
30
At both scenarios, the simulation started with CSTR
at steady state operation condition and fermentation
temperature at set-point value of 30.3°C. The time of each
simulation was 300 h.
To allow control performance analysis in
experiments, indexes were calculated using Integral of
Absolute Error (IAE) criteria (Equation 2).
(2)
An index, named IFF/FB (Equation 3), especially
designed to analyse feedforward control performance was
described by [8]. This criteria permit to compare process
control performance between a simple feedback control
and a feedforward/feedback system, with values between
0 and 1.
(3)
A criteria value close to zero means that feedforward
does not reject the load satisfactorily in comparison to
simple feedback control, and a negative value means that
feedforward deteriorated the process control, what is
undesirable. However, an IFF/FB value close to 1 means
that feedforward rejects almost all effects caused by the
load and has advantages to simple feedback control [8].
3. Results and Discussion
The bioprocess was simulated with different feed stream
temperature loads, with steps of 50 h of duration or steps
during only 1 h, as mentioned before. The process
variable response for each case is shown in Figure 4.
In both cases, it is observed that 4FB rejected much
more load effects on process variable than only feedback
control. For load steps at time 0 h, 50 h, 150 h and 250 h
shown in Figure 4a, the process variable overshoots and
response times were drastically reduced with 4FB,
indicating its advantage over pure feedback system.
However, process variable response to load steps at time
100 h and 200 h indicated less overshoot and response
time, but not a response dynamic as good as for the others
steps. As a generic Fuzzy feedforward structure, it is
expected that control performance will vary case to case,
depending on operational conditions and process
nonlinearities, once the discourse universe of inputs and
output were initially set and do not change throughout the
simulation.
The process variable dynamic response observed in
Figure 4b, represents the scenario with load steps during 1
h and shows that 4FB had better control performance than
only feedback. It permits to observe less variation of
process variable in presence of more frequently load
steps. This result possibly indicates an interesting
characteristic of 4FB system, the capacity to better control
the process in case of noisy load variations, in reason of
Fuzzy inference machine that can deal with noise input
signals [5].
The analysis of control performance of 4FB and pure
feedback system can also be made using performance
criteria index IAE final value or even the evolution in
time of this index during the simulations. The Figure 5
represents the IAE evolution for both simulated scenarios
for load inputs, and the Tables 3 and 4 show the final
277
Figure4. Process variable responses with simple feedback and 4FB control strategies at different scenarios of feed stream
temperature steps a) 50 h and b) 1 h.
values of IAE and IFF/FB indexes for scenario with load
steps during 50 h and 1 h, respectively.
Table 3. Performance criteria values for simulation with
load steps during 50 h.
Feedback
4FB
IAE [°Ch]
7.8839
3.0963
IFF/FB
-
0.6073
Table 4. Performance criteria values for simulation with
load steps during 1 h.
Feedback
4FB
IAE [°Ch]
24.4145
15.0219
IFF/FB
-
0.3847
As observed in Figure 5a, for each step, IAE criteria was
more penalized with pure feedback control, resulting in a
Figure 5. Performance criteria IAE evolution at different scenarios of feed stream temperature steps of a) 50 h and b) 1 h.
278
final value higher than the obtained with 4FB, as
presented in Table 3. Calculating IFF/FB criteria, it is
possible to determinate that 4FB rejected almost 60%
more the effects in process variable control caused by
load steps in the first scenario, than pure feedback control
was able to reject. At the second scenario, with each load
step during only 1 h, the load rejection with 4FB was also
higher than simple feedback system, as shown in Figure
5b. However, the IAE difference was not as high as in the
first scenario. The Table 4 shows that IFF/FB value was
equal to 0.3847, indicating that 4FB in this case rejected
almost 38% of effects in process variable caused by load
steps. As Fuzzy feedforward in 4FB system is tuned by
discourse universe, the load rejection can be improved
depending on the range set for this parameter.
Interesting information about process control is the
actuators dynamic during the simulation, which can
indicate if some equipment is getting overloaded by the
control system. This information can be obtained
analysing, as example, the control signal that commands
actuator dynamic. For this work, the Figure 6 shows the
utilities valves opening during the simulation with loads
of 50 h.
Fermentation is naturally an exothermal process
because microorganism metabolism [12], and according
to this statement, utility valves dynamic had the expected
behaviour, as shown in Figure 6, with cold utility valve
being more required during the simulation. In comparison
between 4FB and feedback control, in the Figure 6a is
shown that for loads input at 0 h and 250 h of simulation,
cold utility valve had higher valve travel with 4FB, what
can be explained by feedforward control action in
response to load step introduction to CSTR feed stream
temperature. Obviously, it represents that actuator is being
more requested with 4FB control system and, probably,
will require sooner maintenance service than feedback
control system.
In the Figure 6b, the hot utility valve was required at
100 h for feedback control and 4FB, and also at 200 h
only for 4FB. At 100 h, cold valve opened early with
4FB, resulting in a better load rejection comparing to
feedback. In spite of hot utility valve been required by
4FB at 200 h and not by feedback control, hot valve
openings were small during all the simulation in either
cases, not reaching even 6% of opening at 100 h. Thus,
hot utility valve dynamic becomes irrelevant to be
considered for control analyses in this work.
However, in both cases of 4FB or feedback control
systems, valves dynamic does not presented significant
difference in actuator requirement for the others loads. By
this point of view, it is possible to assure that 4FB is more
indicated to regulate the process, because permits a high
load rejection, almost 60.73% for load steps of 50 h, and
does not penalize the final control elements to obtain this
result.
4. Conclusion
In this work, a generic Fuzzy feedforward control for
nonlinear process regulation was proposed to support a
feedback control system, in attempt to reject load effects
on process variable. A CSTR bioreactor with temperature
control using a Fuzzy – Split range system (feedback
control law) was used as nonlinear process model to test
the Fuzzy feedforward feedback control strategy (4FB)
for load rejection.
Figure 6. Opening dynamics of a) cold utility valve and b) hot utility valve, at scenario of feed stream temperature loads steps
of 50 h.
279
Performance criteria IAE and IFF/FB were used to
support the analyses of 4FB system, and also two
different scenarios of load steps were simulated, with 50 h
and 1 h of step duration. In either scenarios, 4FB rejected
much more load effects on process variable than only
feedback, being necessary only the discourse universes of
inputs and output adjustment for generic Fuzzy
feedforward as tuning procedure.
This characteristic gives an advantage for 4FB
system against classic feedforward feedback control law,
which requires process modelling to determine control
parameters like process and disturbance gains (Kp and
Kd), time constants (τp and τd) and time delays (tp and td).
For nonlinear processes, this parameters changes with
operational conditions, becoming difficult to implement a
classic feedforward control.
The IAE values for feedback control were higher than
for 4FB, demonstrating that Fuzzy feedforward is capable
to load rejection. The IFF/FB criterion values also
demonstrated that 4FB were able to reject almost 60% of
load effects for a scenario with steps during 50 h, and
38% of effects with steps during only 1 h. Thus, results
indicate that generic Fuzzy feedforward is an interesting
alternative to classic feedforward control for load
rejection in nonlinear process control.
References
[1] M. Lawrynczuk, Modelling and nonlinear predictive
control of a yeast fermentation biochemical reactor using
neural networks, Chemical Engineering Journal, 145(2),
2008, 290-307.
[2] A. Della Bona, G. Ferretti, E. Ficara, F. Malpei, LFT
modelling and identification of anaerobic digestion,
Control Engineering Practice, 36, 2015, 1-11.
[3] I. M. Santos-Dueñas, J. E. J. Hornero, A. M. C.
Rodriguez, I. G. Garci, Modeling and optimization of
acetic acid fermentation: A polynomial-based approach,
Biochemical Engineering Journal, 99, 2015, 35-43.
[4] Z. K. Nagy, Model based control of a yeast
fermentation bioreactor using optimally designed artificial
neural networks, Chemical Engineering Journal, 127(1-
3), 2007, 95-109.
[5] R. R. Fonseca, J. E. Schmitz, A. M. F. Fileti, F. V.
Silva, A fuzzy-split range control system applied to a
fermentation process, Bioresource Technology, 142, 2013,
475-482.
[6] J. I. Horiuchi, M. Kishimoto, Application of fuzzy
control to industrial bioprocesses in Japan, Fuzzy Sets and
Systems, 128(1), 2002, 117-124.
[7] J. L. Guzmán, T. Hägglund, Simple tuning rules of
feedforward compensators, Journal of Process Control,
21(1), 2011, 92-102.
[8] J. L. Guzmán, T. Hägglund, M. Veronesi, A. Visioli,
Performance indices for feedforward control, Journal of
Process Control, 26, 2015, 26-34.
[9] M. Veronesi, A. Visioli, Automatic tuning of
feedforward controllers for disturbance rejection,
Industrial & Engineering Chemistry Research, 53(7),
2014, 2764-2770.
[10] M. J. Fuente, C. Robles, O. Casado, S. Syafiie, F.
Tadeo, Fuzzy control of a neutralization process,
Engineering Applications of Artificial Intelligence, 19(8),
2006, 905-914.
[11] R. E. Precup, H. Hellendoorn, A survey on industrial
applications of fuzzy control, Computers in Industry,
62(3), 2011, 213-226.
[12] D. Voet, J. G. Voet, Biochemistry (Hoboken, NJ:
John Wiley & Sons, 2004).
280
