236 IEEE TRANSACTIONS ON CONTROL SYSTEMS TECHNOLOGY, VOL. 8, NO.
2, MARCH 2000
Control of Neutralization Processes by Robust
Loopshaping
Fernando Tadeo, Omar Pérez López, and Teresa Alvarez
Abstract—This paper studies the control of a pH neutralization
process by using a robust loopshaping approach. The loop-
shaping method is applied to calculate an optimal controller. The
problem of choosing the desired shape of the open-loop transfer
function needed by this method is addressed by considering
the available uncertainty information and applying graphical
loopshaping ideas. Thus, this methodology considers not only the
robustness properties of the shaped plant, but also those of the real
plant. The designed controller was tested in real-time on a bench
plant. On-line results show that the designed control system allows
the plant to operate in a range of pH values, despite variations of
the plant parameters, obtaining good performance at the desired
working points. The methodology presented in this paper can
be applied to other chemical processes: It is only necessary to
consider the possible uncertainty in the nominal model and using
available software to design the controller.
Index Terms—Frequency domain synthesis, loopshaping,
process control, robustness.
NOMENCLATURE
Total ion concentration in the influent stream,
mol/L.
Normalized uncertainty.
Flow rate, L/s.
Ion concentration, mol/L.
Imaginary unit, . seems to be the wide range of operating conditions over which
Dissociation constant of acid at room temperature.
Dissociation constant of salt at room temperature.
Dissociation constant of water at room temperature.
Set of possible plants.
Logarithmic dissociation constant of salt,
. transitions between different regimes. The control system must
Volume of the reactor, .
Total ion concentration in the effluent stream,
mol/L.
Subscripts
Acid (hydrochloric acid).
Salt (sodium acetate).
Graphical loopshaping.
Manuscript received July 31, 1997; revised December 28, 1998. Recom-
mended by Associate Editor, S. B. Engell. This work was supported by the
CYTED (Proyecto Precompetitivo VII-5) and CICYT (Proyecto TAP97-1144).
F. Tadeo and T. Alvarez are with the Departamento de Ingeniería de Sis-
temas y Automática, Facultad de Ciencias, Universidad de Valladolid, 47011
Valladolid, Spain (e-mail: {fernando; teresa}@autom.uva.es).
O. Pérez Lópezis with the Departamento de Procesos y Sistemas, Universidad
Simón Bolívar, Sartenejas, Baruta, Estado Miranda, Caracas 1081-A, Venezuela
(e-mail: operez@usb.ve).
Publisher Item Identifier S 1063-6536(00)01783-8.
1063–6536/00$10.00 © 2000 IEEE
loopshaping.
Linearized model index.
Nominal value.
I. INTRODUCTION
C ONTROL of pH neutralization processes plays an
important role in chemical plants, such as biological,
wastewater treatment, electrochemistry, and precipitation
plants. However, it is difficult to control a pH process with
adequate performance due to its nonlinearities, time-varying
properties, and sensibility to small perturbations when working
near the equivalence point [13]. Many different approaches
to pH control have been applied previously, such as linear
adaptive, model-based, nonlinear adaptive, and neural-network
controllers ([7], [9]–[11], [13], [22] and the references therein).
Unfortunately, as noted by these authors, there are some
weaknesses in these solutions: the control structures are quite
complex, so they could be difficult to implement in existing
distributed control systems. Also it is difficult to ensure the
robustness of the controlled system. Even under ideal condi-
tions the transient performance of nonlinear controllers based
on the “strong acid equivalent technique” [22] is not adequate,
because it is a nonlinear function of the pH [9]. A key difficulty
good control is required, and the extent to which the dynamics
vary as a consequence.
Moreover, chemical plants where pH neutralization processes
are important usually work with more than one type of product,
with frequent product and grade changeovers, giving as a result
regulate the plant not only at these varying working points, but
also in transition regions. This topic is of considerable economic
importance due to strict environmental regulations that limit the
amount of allowed off-specification materials. Obtaining a pre-
cise nonlinear model that accurately matches the plant at all
working points is a difficult problem. So, most of the advanced
controllers are usually designed using a linear model of the
process based on fixed information of the plant that is imper-
fect and incomplete. Thus, control quality may deteriorate when
working conditions change.
In view of these difficulties, robust control design methods
seem especially appropriate, since they give linear controllers
with good stability margins [22]. The essence of robust control
is to model the uncertainties themselves and to incorporate them
in the design procedure of the control system, with the aim of en-
suring stability and performance at all working points. Usually
it is possible to identify multiple local linear models at different
operating regions, which can be used to evaluate the expected
TADEO et al.: CONTROL OF NEUTRALIZAIOTN PROCESSES BY ROBUST LOOPSHAPING
uncertainty of the nominal model. Then this uncertainty infor-
mation is used to design a controller that ensures robust stability
and performance.
Among all the available robust control techniques the
loopshaping procedure has been chosen, because it
has been proved to be efficient in process control applications
[6]. The approach involves the robust stabilization to additive
perturbations of normalized coprime factors of a shaped plant.
Prior to robust stabilization, the open-loop singular values are
shaped using pre- and post-compensators. Then, the resulting
shaped plant is robustly stabilized with respect to coprime factor
uncertainty using optimization.
As pointed out in [14], one difficulty of the design
method is that it does not directly address the robustness prop-
erties of the real plant, but rather it is concerned with the shaped
plant, and, unfortunately, there is no direct connection between
the robustness of the shaped and unshaped plant. This paper
shows a methodology that solves this problem by considering
the robustness properties of the real plant in the selection of the
weights of the shaped plant. Constraints are added to the se-
lection of the shaped plant by considering the uncertainty in
the real plant, as in [4]. Once selected a robust shaped plant
the controller is designed by application of the design
method. This method is simpler and more intuitive than other
robust control techniques previously applied to the pH neutral-
ization problem, such as the presented in [18].
This paper is organized as follows. First, the loop-
shaping and graphical loopshaping methods are presented.
Next, the pH neutralization process is presented and its model
is derived. Then, the design of the controller is shown. Finally,
results of the on-line implementation of the designed controller
are given, which are then discussed.
II. LOOPSHAPING
, as introduced and solved in [5], considers the stabi-
lization of a plant which has a normalized left coprime factoriza-
tion: . That is, and are stable transfer function
matrices ( ) such that there exists
which fulfills the identities and
[where denotes ].
In this technique two uncertainty blocks are used, as depicted
in Fig. 1, one on each of the factors in the coprime factorization:
, where .
The objective of robust stabilization is to stabilize the nominal
plant and the family of perturbed plants defined by
(1)
where is the stability margin. Objectives of LS are the
maximization of this stability margin and achieving good
input–output performance. It can be shown that this is equiva-
lent to find a stabilizing that minimizes
which can be calculated by solving an algebraic Riccatti equa-
tion, as shown by [12].
237
Fig. 1. Coprime perturbed plant.
Compared with other design methods, the main advan-
tage of the LS method is that it does not require the so-called
-iteration to calculate the optimal controller. Also there are
available relatively simple formulas to calculate the controller.
On the other hand it does not (directly) include any closed-loop
specification, which must be included by considering, instead
of the nominal plant, a shaped plant. In practical designs, the
loopshaping design procedure (LSDP) can be applied [12]. The
complete design procedure is the following:
1) Using pre- and post-compensators ( and ) the sin-
gular values of the nominal plant are modified to give
a desired loopshape: , which should not
contain unstable hidden modes.
2) is considered to be perturbed by normalized coprime
uncertainties, and an optimal feedback controller is
then synthesized using the LS approach.
3) The combination of the LS controller and the com-
pensators gives the final controller: .
Different methods to select the compensators have been
studied: [21] proposes the use of the inequalities method,
[14] the use of the phase crossover frequency, and [19] the
use of genetic algorithms. In order to consider the robustness
properties of the real plant in the design, this paper shows
a methodology that solves this problem by considering the
robustness properties of the real plant in the selection of the
weights of the shaped plant. This technique is based on a
graphical approach to loopshaping proposed in [4].
III. GRAPHICAL LOOPSHAPING
Prior to coprime robust stabilization, the open-loop singular
values must be shaped in order to obtain an adequate (and al-
lowable) open-loop transfer function. A graphical loopshaping
(GLS) method, based on the ideas of [4], which will be used to
select the desired open-loop transfer function is now presented.
Suppose that the plant can be represented by a transfer func-
tion , which belongs to a set of possible plants . A con-
trol provides robust stability if the feedback system is in-
ternally stable for every plant in . Let contain plants with
the same number of unstable poles. Consider perturbed plant
transfer functions of the form , where is
a variable stable transfer function such that (output
multiplicative direct uncertainty). It is possible to prove [4] that
238 IEEE TRANSACTIONS ON CONTROL SYSTEMS TECHNOLOGY, VOL. 8, NO. 2, MARCH 2000

provides robust stability if and only if , where
is the complementary sensitivity. Simi-
larly, if an output inverse multiplicative uncertainty description
is used the controller
provides robust stability if and only if , where
is the sensitivity.
The robust performance problem can be stated as calculating
a controller that provide robust stability and certain perfor-
mance specifications are met for all the plants in the uncertainty
set. The following robust performance problem will be consid-
ered: design a controller such that the feedback system for
the nominal plant is internally stable and
(2)
In [4] the GLS technique is proposed to obtain an approxi-
mate solution of this problem, which can be readily extended to
different weighted transfer functions. The steps are as follows.
1) From the available information of the plant select ade-
quate weighting functions and , such that at any
frequency either or . the uncertainty comes from nonlinearities in the system, as is
2) Construct a loop transfer function fulfilling
the condition given in (2), such that contains the plant
unstable poles and zeros, and gives internal stability of
the feedback system.
3) Then, the controller K can be calculated as
, canceling common poles and zeros.
For simplicity reasons the method is shown for single
input–single output (SISO) systems, but can be extended to
multiple input–multiple output (MIMO) systems applying the
structured singular value theory [15]. The robustness condition
given in (2) can be made equivalent to the following one:
(3)
near the desired crossover frequency (about 20 dB/dec
for step changes in references and disturbances [16].
It must be pointed out that the design procedure is valid for
plants with poles or zeros in the right half plane (RHP), as long
as they are included in the selected open-loop transfer function,
and the cross-over frequency is selected to fulfill the achiev-
able bandwidth limitations. A complete study of these limita-
tions can be found in [16]. For instance, it can be shown that in
the presence of a single RHP zero the maximum crossover
frequency is , and in the presence of a single RHP pole
the minimum crossover frequency is .
This is the GLS method, as proposed in [4]. Unfortunately
the conditions given in (4) are only sufficient conditions, so in
many practical problems they could be too strict to obtain an
adequate open-loop shape in all the frequency range, so usually
the designed fulfills (4) only at low and high frequencies.
In fact, as will be shown later, in our pH control problem, with
the weights obtained from the uncertainty information of the
plant, there is no feasible solution to (4). Also, these bounds
are sufficient only if the uncertainty in the model is linear. If
usual in practical problems, the conditions in (4) are no longer
sufficient. To overcome these problems we propose the appli-
cation of necessary bounds for solving the robust performance
problem in (2). The deficiency of stability robustness due to the
necessary conditions will be remedied by the further robust sta-
bilization design for the normalized coprime factor description.
These necessary bounds can be derived multiplying both terms
in (3) by and applying the fact that
to derive the following necessary conditions for robust perfor-
mance:
in the frequency range where
and (5a)

A necessary condition is that (Step in the frequency range where

1), so it is convenient to study the shape of in the re-
gions where (usually at low frequencies) or and (5b)
(at high frequencies). Sufficient conditions to
Then, in the modified GLS method, conditions given in (5)
fulfill the robust performance condition in these regions are
will substitute the conditions given in (4). By selecting adequate
weights it is possible to find an appropriate , although, even
(4a) with the necessary conditions, it is not guaranteed that an appro-
priate exists. The modified GLS method can be applied to
obtain an open-loop transfer function , which then is robustly
stabilized by application of the LS approach, obtaining a ro-
(4b) bust open-loop transfer function . It is important to notice
that the available information about uncertainties in the model
the design can be carried out graphically. and performance specifications is considered when applying the
1) Plot where . GLS method to design . When applying the LS method
2) Plot where . the uncertainty is considered unknown and coprime (1).
3) Find an open-loop transfer function such that
lies above the first curve (4a) and below the IV. DESCRIPTION OF THE EXPERIMENTAL SETUP
second (4b). To make the controller proper, roll-off at The process under study is the neutralization of an aqueous
high frequencies at least as fast as . To obtain solution of sodium acetate (CH COONa) with hydrochloric
internal stability the slope of should be small acid (HCl) in a continuous stirred tank reactor (CSTR). The
TADEO et al.: CONTROL OF NEUTRALIZAIOTN PROCESSES BY ROBUST LOOPSHAPING 239

the streams, to check the shape of the titration curve. The exper-
imental results were consistent with the nonlinear model.
The reactions taking place in an aqueous mixture of sodium
acetate with hydrochloric acid are as follows:

CH COONa Na CH COO
HCl Cl H
CH COO H O CH COOH OH
H O OH H

Because of the small reaction enthalpy of this neutralization

process and the small concentration of the reactants the tem-
perature within the reactor can be considered constant (25 C).
Supposing that the input liquid is a pure dissolution of sodium
Fig. 2. Laboratory plant.
acetate, that the hydrochloric acid has constant concentration,
and that there are perfect dissolution, mixing, and no additional
experimental setup, shown in Fig. 2, consists of a CSTR buffering effects, the following model can be obtained:
where a liquid of variable pH is mixed with a solution of high
concentration of HCl. pH variations can be realized within the
setup by adding varying amounts of sodium acetate to the tank
liquid. This liquid is fed from the tank using a pump, which
produces a variable flow depending on the level of liquid in
the tank. The liquid in the mixing tank overflows (outlet not
shown), so the volume of liquid in the tank can be considered
constant. The control variable is the flowrate of the titrating
stream (normalized to the maximum value), which is applied
using a peristaltic pump (ISMATEC MS-1 REGLO/6-160).
The output variable is the hydrogen ion concentration in the
effluent stream. The mixture pH is measured using a Ag-AgCl
electrode (Kent 1180/700) and transmitted using a pH-meter (6)
(Kent EIL9143). The electrode dynamic response presents
where the ion concentrations within the tank (and in the
appreciable and asymmetric inertia. The pH measured and
effluent stream) are Cl , Na
the control signal are transmitted through an A/D interface
CH COOH CH COO and the dissociation con-
(ComputerBoards CIO-AD16, 0-5 V). The plant is controlled
and monitorzed from a PC-486 computer, using the software stants are H OH mol L
CH COOH OH
package REGULA. This is a real-time control software package mol L . is the
developed at the University of Valladolid. It runs on personal CH COO
control acid stream flowrate, the inlet stream flowrate,
computers and is able to control plants up to ten control loops
is the concentration of sodium aceate in the inlet stream and
providing different advanced control techniques (adaptive,
is the acid concentration in the control acid stream. The
optimal, robust, etc.) [1]. After some open-loop experiments,
measured pH is pH , which is affected by a time constant .
at different working points, the sampling time was selected to
To design the robust controller it is necessary to obtain a
be 1.6 s.
set of linear models of the plant at different working points,
Due to the nonlinear dependence of the pH value on the
and a nominal model which represents the plant at the nominal
amount of titrated agent the process will be inherently non-
working point. The most important disturbance is due to varia-
linear. Moreover, variations of the buffering effects could
tions in the concentration of salt in the inlet stream. The mea-
make the process time-varying. Both effects make the process
sured concentration varied between mol L
difficult to control with classical process control techniques
and mol L . The working point could also
[13].
vary between pH = 6 and 8. Also, depending on the level in
the liquid tank, the inlet stream flowrate could vary between
V. PLANT MODEL cm s and cm s .
Although the modeling of pH-control processes has been well These variations must be taken into account when designing the
studied [8], in this case it is only necessary to have a simplified control system.
model, because when designing the controller the available in- The nominal model was calculated by linearization at
formation on plant uncertainty can be considered. This model the bench plant steady-state nominal conditions, which
was obtained based on first principles, and then validated in are cm , cm s ,
the real plant, by carrying out experiments at different working cm s , mol L , ,
points. Also experiments were carried out by eliminating one of mol L , mol L
240 IEEE TRANSACTIONS ON CONTROL SYSTEMS TECHNOLOGY, VOL. 8, NO. 2, MARCH 2000

and mol L . The obtained nominal

model is

(7)

VI. CONTROLLER DESIGN

A. Weight Selection
First the effect of parameter variation on the transfer func-
tion was studied. The uncertain nonlinear model in (6) was ap-
proximated by a set of local linear approximations. Assuming
the parameter variations uncorrelated, the nonlinear model was Fig. 3. Weight on multiplicative inverse uncertainties.
linearized considering the extreme values of each uncertain pa-
rameter. This approximation is suitable for slowly varying pa-
rameters. For example, the model with the fastest poles and the
highest gain was

and the one with the slowest poles and the lowest gain was

The uncertainty due to the nonlinearity is bounded with a linear

uncertainty, calculated from the (weighted) difference between Fig. 4. Weight on multiplicative direct uncertainties.
the linearized plant at different working points and the nominal
plant. It must be pointed out that although the stability con-
ditions are not sufficient in this situation, this is a standard tech- • The high frequencies uncertainty is due to variations in
nique by now, to design linear controllers for nonlinear plants, the measurement time constant and in the inlet stream
that have been shown to be adequate for systems with smooth flowrate ( ). They can be modeled as direct multiplica-
nonlinearities [17], as in the neutralization process studied in tive uncertainty, and bounded using the weight
this paper. Observe that the controller calculated will be conser-
vative, as part of the information on the uncertainty (its nonlin-
(8b)
earity) is lost.
The uncertainty which affects each linearized models
was considered either as a multiplicative inverse uncertainty The inverse of this weight bounds the complementary sen-
( ), if the variation of the frequency response sitivity. Fig. 4 shows the uncertainties and the selected
could be noticed mainly in low-frequencies, or as a multi- weight.
plicative direct uncertainty ( ), if it was a
high-frequencies effect. Then, the weights that bounded the B. Graphical-Loopshaping of the pH Plant
multiplicative inverse uncertainties and the multiplicative
Using the weights from the previous section, the GLS
direct uncertainty were calculated. This calculus was
method was applied to design an adequate shape of the
performed keeping in mind the specifications of no steady-state
open-loop transfer function. The open-loop transfer function
error and noise attenuation, as required by the GLS method.
was selected to be
• The low frequencies uncertainty is due to variations in
salt concentration in inlet stream ( ) and working point
variations (pH). The corresponding uncertainty model is a (9)
multiplicative inverse uncertainty and the selected weight

The magnitude of this open-loop transfer function is shown

(8a)
The inverse of this transfer function bounds the sensitivity
transfer function. This weight and the corresponding set
of uncertainties are shown in Fig. 3.
in Fig. 5, together with the obtained robustness bounds. There
is no feasible solution of the sufficient robustness conditions
(4). Thus, in this case, the conditions given in (4) provides very
little information for designing the control system. On the other
hand the necessary robustness conditions (5) are fulfilled by the
TADEO et al.: CONTROL OF NEUTRALIZAIOTN PROCESSES BY ROBUST LOOPSHAPING
Fig. 5. Graphical loopshaping.
designed open-loop shape. The controller corresponding to the
nominal loop-shape given in (9) is
(10)
The order of the controller in (10) was reduced by means
of the balanced truncation method [6]. The reduced-order con-
troller is
(11)
This controller was discretized and implemented in the dis-
tributed control system. Different experiments were carried to
show that although the nominal performance was adequate, the
robustness was poor, due to the reduced phase and gain margins.
To improve the robustness of this controller the proposed
loopshaping technique was applied, as shown in the next sec-
tion.
C. Loopshaping for the pH Plant
Once an adequate open-loop transfer function (s) have
been selected, the design can be made more robust by consid-
ering additional coprime uncertainties of . These
uncertainties include additional unmodeled dynamics, giving
additional robustness. The resulting compensator can be
included in the feedback controller block by augmenting the
controller designed using the GLS method. is a stabilizing
241
compensator of and of the set of uncertain plants such that
. There-
fore, the robust controller can be calculated as follows:
(12)
This method was applied to design a controller for the pH
control plant. The obtained algebraic Riccatti equation was
solved using the Matlab robust control toolbox [3]. Considering
the desired loop-shape the one given in (9), the LS method
was applied. The calculated compensator was
(13)
The open-loop shape when this compensator is included
( ) is shown in Fig. 6. It can be seen that
the overall effect of the compensator is a gain reduction,
especially at high frequencies. The open-loop cutoff frequency
is decreased, so a reduction on the speed of response of the
closed-loop system is achieved. The feedback system with the
designed open-loop transfer function presents good robustness
characteristics: the phase margin is 74 , and the gain margin is
21 dB.
Following (12) the optimal controller is given by (14) at the
bottom of the page.
The order of the controller in (14) was further reduced using
the balanced truncation method [6], obtaining the following
controller:
(15)
(14)
242
Fig. 6. Final loopshaping.
Fig. 7. Characteristic transfer functions magnitude with the H controller.
The magnitude of the characteristic transfer functions of the
feedback system with the reduced-order controller are shown
in Fig. 7, together with the plant and controller frequency
responses. The steady-state gain is 0.99 (observe that the
controller has a pole near ), which is adequate for this
problem.
It is important to notice that the robust performance condi-
tions corresponding to (5), obtained in the first step of the de-
sign, are fulfilled with the compensated controller in (15), as
shown in Fig. 6. If this were not true, it would be necessary to
find a new desired open-loop shape by application of the GLS
method, but with better robustness characteristics (i.e., greater
magnitude and phase margins).
IEEE TRANSACTIONS ON CONTROL SYSTEMS TECHNOLOGY, VOL. 8, NO. 2, MARCH 2000
VII. REAL-TIME CONTROL
Different experiments were carried out with the designed
feedback controller working in real-time. The robust controller
was discretized and implemented in the distributed
control system, which controlled and monitored the plant.
Fig. 8 depicts a run where nominal conditions were main-
tained during the experiment: the inlet stream flowrate and con-
centration were maintained at their nominal values. During the
experiment step changes to the command signal were intro-
duced: first the desired pH was increased from 6.0 to 6.5, then
to 7.0 and finally to 7.5. Afterwards the pH command was de-
creased from 7.5 to 6.5, and finally to 5.5. By using the robust
TADEO et al.: CONTROL OF NEUTRALIZAIOTN PROCESSES BY ROBUST LOOPSHAPING
Fig. 8. Robust controller test with nominal conditions.
Fig. 9. Robust controller test with nominal conditions.
controller in (15) good performance was achieved. The results
demonstrate that the plant can operate in the desired range of
pH variations, with neither oscillations nor steady-state error.
Also, the speed of response to command changes and the shape
of the control effort are adequate. Moreover, the system remains
stable even outside the pH range considered in the design. (The
final command of pH 5.5 is outside the design range.) It can
be noticed that the response to increasing commands is slightly
243
slower than the response to decreasing commands when the con-
trol signal saturates. This is due to the fact that the hydrochloric
acid is a strong acid, and the sodium acetate acts as a weak base.
Observe that both plant and controller are stable transfer func-
tions, so in this situation the stability is ensured, and no anti-
windup correction is necessary.
An experiment that reproduces more realistic working condi-
tions is shown in Fig. 9. In this experiment we wanted to check
244
Fig. 10. Robust controller test with real conditions.
Fig. 11 Disturbance rejection of the robust controller test with real condition.
the robustness of the designed control system when the working
conditions were different from the nominal ones. During the
experiment the inlet stream concentration was three times its
nominal value, and the inlet stream flowrate was uncontrolled, so
it varied in the range given previously. To study the performance
of the feedback system step changes were introduced in the de-
sired pH. It can be seen in Fig. 10 that the performance, compared
with the previous experiment does not deteriorate: The response
is very similar to the one obtained with nominal conditions. Also
the disturbances in inlet concentration and flowrate are rejected.
IEEE TRANSACTIONS ON CONTROL SYSTEMS TECHNOLOGY, VOL. 8, NO. 2, MARCH 2000
Besides, the control signal varies smoothly during the experi-
ment, without saturating. The disturbance rejection properties
of the designed robust controller were tested by introducing step
changes in the inlet flowrate. The result is shown in Fig. 11. It can
be seen the controller rejects these disturbances regulating the pH
within a range of 0.1. These results confirm the robustness of
the controller when controlling the real plant.
These results can be compared with the controller perfor-
mance of a PI controller designed for nominal conditions using
the Ziegler–Nichols ultimate-sensitivity tuning rules [2]. Al-
TADEO et al.: CONTROL OF NEUTRALIZAIOTN PROCESSES BY ROBUST LOOPSHAPING
Fig. 12. PID controller with real conditions.
though the PI controller tuned using this method (Kp , Ti
0.8) gives adequate response for nominal conditions, its perfor-
mance worsens in the presence of parameter variations: Fig. 12
depicts a typical run when the inlet concentration is one third
of the nominal value, and the input flowrate is maximum. It can
be seen that the control signal (and therefore the output signal)
oscillates in steady-state, which is unacceptable.
VIII. CONCLUSIONS
This paper has presented the improvement of the robustness
of a pH-control process by using a robust loopshaping approach,
considering coprime uncertainties. loopshaping is an ap-
pealing approach for controller design, as it addresses explicitly
the problem of model uncertainty. However, this design method
does not directly address the robustness properties of the real
plant, but rather it is concerned with the shaped plant. This paper
has discussed a methodology that solves this problem by con-
sidering the robustness properties of the real plant in the selec-
tion of the weights of the shaped plant. Then a shaped plant is
selected following the graphical loopshaping ideas, where in-
stead of using sufficient conditions (which are only valid for
linear uncertainty), necessary conditions for robust performance
have been applied. Once selected a robust shaped plant the con-
troller is designed by application of the loopshaping de-
sign method, following McFarlane–Glover ideas. This two-step
design methodology makes possible to take advantage of the
positive properties of the loopshaping design method, but
considering at the same time the robustness properties of the
real plant. It must be pointed out that it may be necessary to
reexamine the robust stability and robust performance condi-
tions with the final controller, as there is not guarantee on robust
245
performance for the final controller. If these conditions are too
strict, it may be necessary to relax them and repeat the design.
The resulting controllers were tested in real time by using a
control software package. With the controller designed using
the robust loopshaping approach the plant operated in a wide
range of pH values, in spite of variations of the parameters. The
performance at the desired working points is good. The system
is stable in the desired range of parameter variations, with a good
response to command changes.
The idea shown in this paper of combining graphical and ro-
bust loopshaping has been shown to be promising. The theory
presented in this paper can be applied to other chemical process,
being only necessary to consider the possible uncertainty in
the nominal model and using available software to design the
controller. Compared with other robust control approaches this
technique is more intuitive to the control engineer, thanks to the
fact that the design parameter is the open-loop transfer function
itself. Further work must be done to extend the proposed tech-
nique to multivariable systems and to ensure that the final con-
troller fulfills the robustness conditions stated in the first step of
the design.
ACKNOWLEDGMENT
The authors are thankful to W. Colmenares, M. Uría, P. Vega,
L. Alonso and C. de Prada, for their helpful support during this
work. They would like to thank the anonymous reviewers for
the comments on the preliminary version of this paper, which
contributed greatly to improving it.
REFERENCES
[1] L. F. Acebes, J. Achirica, M. Garcia, and C. Prada, “A simulator to train
plant operators of a beet-sugar factory,” Syst. Anal. Modeling Simula-
tion, vol. 18, pp. 659–662, 1995.
246 IEEE TRANSACTIONS ON CONTROL SYSTEMS TECHNOLOGY, VOL. 8, NO. 2, MARCH 2000
[2] K. J. Astrom and B. Wittenmark, Computer-Controlled Systems: Theory
and Design, 3rd ed. Englewood Cliffs, NJ: Prentice-Hall, 1997.
[3] R. Y. Chiang and M. G. Safonov, Robust Control Toolbox: The Math-
works Inc.
[4] J. C. Doyle, B. A. Francis, and A. R. Tannenbaum, Feedback Control
Theory. New York: Macmillan, 1992.
[5] K. Glover and D. McFarlane, “Robust stabilization of normalized co-
prime factor plant descriptions with H bounded uncertainty,” IEEE
Trans. Automat. Contr., vol. 34, pp. 821–830, 1989.
[6] M. Green and D. J. N. Limebeer, Linear Robust Control. NJ: Prentice-
Hall, 1995.
[7] T. K. Gustafsson and K. V. Waller, “Nonlinear and adaptive control of
pH,” Ind. Eng. Chem. Res., vol. 31, pp. 2681–2693, 1992.
[8] T. K. Gustafsson et al., “Modeling of pH for control,” Ind. Eng. Chem.
Res., vol. 34, pp. 820–827, 1995.
[9] M. A. Henson and D. E. Seborg, “Adaptive input-output linearization
of a pH neutralization process,” Int. J. Adapt. Contr. Signal Processing,
vol. 11, pp. 171–200, 1997.
[10] K.-U. Klatt and S. Engell, “Nonlinear control of neutralization processes
by gain-scheduling trajectory control,” Ind. Eng. Chem. Res., vol. 35, pp.
3511–3518, 1996.
[11] A. P. Loh, K. O. Looi, and K. F. Fong, “Neural network modeling and
control strategies for a pH process,” J. Proc. Contr., vol. 6, pp. 355–362,
1995.
[12] D. C. McFarlane and K. Glover, “Robust controller design using normal-
ized coprime factor plant descriptions,” in Lecture Notes Contr. Inform.
Sci.: Springer Verlag, 1990.
[13] M. C. Palancar, J. M. Aragon, J. A. Miguens, and J. S. Torrecilla,
“Application of a model reference adaptive control system to pH
control: Effects of lag and delay time,” Ind. Eng. Chem. Res., vol. 35,
pp. 4100–4110, 1996.
[14] A. Pantas and S. Walsh, “Evaluation of H loop shaping controller
design on a process control problem,” in UKACC Contr. 96’, Exeter,
U.K., 1996.
[15] S. Skogestad and M. Morari, “Some new properties of the structured
singular value,” IEEE Trans. Automat. Contr., vol. 33, pp. 1151–1154,
1992.
[16] S. Skogestad and I. Postlethwaite, Multivariable Feedback Control,
Analysis and Design. New York: Wiley, 1996.
[17] F. Tadeo, P. Vega, and R. G. Cameron, “The robust control of a milk of
lime blending tank,” Applicat. Multivariable Syst. Techniques Mechan-
ical Eng. Public, pp. 109–117, 1994.
[18] F. Tadeo, A. Holohan, and P. Vega, “ -Optimal regulation of a pH control
plant,” Comput. Chem. Eng., vol. 22, pp. S459–S466, 1998.
[19] K. S. Tang, K. F. Man, and D. W. Gu, “Structured genetic algorithm for
robust H control systems design,” IEEE Trans. Ind. Electron., vol. 43,
pp. 575–582, 1996.
[20] J. F. Whidborne, I. Postlethwaite, and D. W. Gu, “Robust controller
design using H loop-shaping and the method of inequalities,” IEEE
Trans. Contr. Syst. Technol., vol. 2, pp. 455–461, 1994.
[21] R. A. Wright and C. Kravaris, “Nonlinear control of pH processes
using the strong acid equivalent,” Ind. Eng. Chem. Res., vol. 30, pp.
1561–1572, 1991.
[22] K. Zhou, J. C. Doyle, and K. Glover, Robust and Optimal Con-
trol. Englewood Cliffs, NJ: Prentice-Hall, 1996.
Fernando Tadeo was born in Medina del Campo, Spain, in 1969. He received
the bachelor’s degree in physics in 1992, and in electronic engineering in 1994,
both from the University of Valladolid, Spain. After he received the M.Sc. de-
gree in control engineering in the University of Bradford, U.K., he received the
Ph.D. degree from the University of Valladalid in 1996.
Currently he is a Lecturer at the “Departamento de Ingeniería de Sistemas
y Automática” of the University of Valladolid. His main interests include
robust control, process control, and analog filters. During his studies, he visited
Dublin City University, Ireland, Strathclyde University, Glasgow, U.K., and
Simón Bolívar University, Caracas, Venezuela.
Omar Pérez López was born in Venezuela. He
received the bachelor’s degeree in electronic
engineering in 1992 and the M.Sc. degree in system
engineering in 1995, both from the Universidad
Simon Bolivar, Caracas, Venezuela.
He is currently Assistant Professor in the Depart-
ment of Processes and Systems and a member of the
Industrial Automation Center (CAI) at the same uni-
versity.
Teresa Alvarez received the M.Eng. degree with distinction in computer science
from the University of Valladolid, Spain, in 1992, the M.Sc. degree in control
engineering from the University of Bradford, U.K., in 1994, and the Ph.D. de-
gree from the University of Valladolid in 1997.
She has been a Lecturer at the University of Valladolid, Spain, since 1995. Her
current interests include multivariable predictive control, constraint handling,
optimization, and the use of expert systems in control and supervision, which
she has applied to solve several problems in different European industries.