ISA

TRANSACTIONS
ISA Transactions 44 2005 8191

Design of controller using variable transformations for a nonlinear

process with dead time
R. Anandanatarajan,a,* M. Chidambaram,b T. Jayasinghc
a

Pondicherry Engineering College, Pondicherry, India

b
IIT, Chennai, India
c
St. Xaviers College of Engineering, Nagercoil, India

Received 30 October 2003; accepted 5 June 2004

Abstract
In this work, a globally linearized controller GLC for a first-order nonlinear system with dead time is proposed.
This is similar to the GLC proposed by Ogunnalke Ind. Eng. Chem. Process. Des. Dev. 25, 241248 1986 for
nonlinear systems without dead time. Two methods are proposed. One is based on the Smith prediction from the model
in the transformed domain and the other is based on Newtons extrapolation method. The simulation study is made on
the conical tank level process and the results are compared with those obtained using a conventional PI controller and
the Smith PI controller based on the transfer function model about the operating point 39%. Finally, experimental
results on the laboratory conical tank level process are also given. 2005 ISAThe Instrumentation, Systems, and
Automation Society.
Keywords: Nonlinear control; GLC; Conical tank

1. Introduction
Time delays occur in many control applications,
typically as a result of transporting material or energy. When the time delay is significantly greater
than the process dominant time constant, serious
degradation of the closed-loop performance can be
expected with traditional controllers. For instance,
the traditional PI controller results in very sluggish
closed-loop response 1. Faster reference tracking
and disturbance rejection can be achieved by the
Smith predictor. However, its main weakness,
compared to the PI controller, is a lack of robustness to variations in process parameters 2,3.
The traditional and easiest approach to the controller design problem for nonlinear systems in*Corresponding
author.
annandanatarajan@yahoo.com

E-mail

address:

volves linearizing the modeling equation around a

steady state and applying linear control theory results 4. It is obvious that the controller performance in this case will deteriorate as the process
moves further away from the steady state around
which the model was linearized.
Apart from the local linearization approach,
there are a few other special purpose design
procedures 5 which may be applied directly to
nonlinear systems. However, as noted by Ray 5,
these usually have limited applicability and are often based on accumulated experience with a special type of nonlinear system. A new modified
Smith predictor is proposed by Vrecko 6 to improve the robustness but again it is a linear controller.
The design of a controller based on the variable
transformation for a nonlinear system without
dead time is presented by Ogunnalke 7. In this

0019-0578/2005/$ - see front matter 2005 ISAThe Instrumentation, Systems, and Automation Society.

82

Anandanatarajan, Chidambaram, Jayasingh / ISA Transactions 44 (2005) 8191

present paper, it is shown that how transformations may be found which transform the nonlinear
system with dead time into one that is ideally linear. The approach is based on the hypothesis that a
system which is nonlinear with dead time in its
original variables is linear in some transformation
of the predicted future original variables. Constructive methods for finding the appropriate transformations are presented. The design of a controller for the linear transformed system may then be
carried out with a great deal of facility. The resulting controller will of course be nonlinear when
recast in terms of the original system variables. In
the present work, the specific problem of a conical
tank level process system having dead time is used
to illustrate the potentials of this approach. An example simulation and real time implementation of
the controller on a conical tank laboratory level
process are presented.

Fig. 1. Conceptual configuration of the GLC controller.

where a and b are constants, v is a state variable,

and g ( ) is a function to be determined. Differentiating Eq. 3 with respect to t,

dz dg dx

.
dt dx dt
From Eqs. 1 and 2

2. Design of globally linearized controller

GLC

dx
c 1 f 1 x c 2 f 2 x,u tT d .
dt

2.1. Variable transformation

where F ( ) is an arbitrary nonlinear function of x,

the system state variable, and u, the control variable, and T d is the dead time in the process. The
function Fx,u ( tT d) is split up as

Fx,u tT d c 1 f 1 x c 2 f 2 x,u tT d ,
2
where f 1 ( x ) is a function of x alone and
f 2 x,u ( tT d) is a function of both x and
u ( tT d) . Both f 1 and f 2 are taken to be nonlinear. c 1 and c 2 are constants.

zg x

is a transformation for mapping the nonlinear system F ( ) to a linear system

dz
ab v ,
dt

Substituting Eq. 6 in Eq. 5,

A single-input, single-output nonlinear control

system with dead time can be represented in general by

dx
Fx,u tT d ,
dt

dz
dg
dg
c 1 f 1
c 2 f 2
.
dt
dx
dx

Let

c 1 f 1 x

dg
a
dx

and

c 2 f 2 x,u tT d

dg
b v .
dx

The nonlinear system is thus mapped to

dz
ab v .
dt

10

A linear system may be effectively controlled with

a PI controller as in Fig. 1. From Eq. 8, the transformation g ( x ) which transforms the given nonlinear system with dead time into a linear system
is

g x

c1

f dx x .
1

11

Anandanatarajan, Chidambaram, Jayasingh / ISA Transactions 44 (2005) 8191

83

2.2. Prediction of process variable and controller

output
A conventional PI controller gives an output
v ( t ) which effectively controls z. The future value
of f 2 is predicted either by using Newtons method
or by using the transformed model of the system.
Since f 2 is a function of v , once f 2 is predicted, v
can be easily obtained. The nonlinear control law
can be obtained using Eq. 12.
Fig. 2. Block diagram representation of the nonlinear controller with Newtons predictor method NEM. The PI controller is designed for the linear system Eq. 10.

2.3. Newtons extrapolation method

Newtons extrapolation formula, used to predict
the future value, is

Substituting the value of dg/dx from Eq. 8 into

Eq. 9 yields

u t

c 1 f 1 b v tT d
.
c 2 a f 2 x tT d

f ph f p h f p higher-order terms,
13

12

Let

u t q ,
where v ( tT d) is the predicted manipulated variable and f 2 x ( tT d) is the predicted function of
state variable. In deriving Eq. 12, the specific
case of f 2 ( x ) u ( tt d) is considered than
f 2 x,u ( tt d) . In general case u has to be solved
numerically.
Thus a nonlinear controller u ( t ) is designed
based on a variable transformation for the firstorder nonlinear process with dead time. The controller performance is tested by simulation of the
conical tank level process. The proposed controller is expected to be highly robust when the operating point of the process is shifted over the entire
span of the tank. q ( ) is the transformation which
transforms the linear controller output in the transformed domain into the nonlinear controller output in the original domain.
A PI controller is designed and interfaced with
the pseudolinear system shown in Fig. 2. This is
made possible by the use of the transformations
g ( x ) and u ( t ) . x * is the set point and z * is the set
point in the transformed domain. Thus the entire
design procedure boils down to the determination
of g ( x ) to compute the corresponding u ( t ) by using Eq. 12.

where f ( ph ) is the future value to be predicted

after an interval of h sec and p is the sampling
time at which the latest value is available. is the
difference operator. f ( p ) is the difference between the latest sampled value and the previous
sampled value. The number of terms that has to be
considered in the above formula depends on the
number of past data available and the designers
interest. In their work, only the first two terms are
considered 8. This method may be called Newtons extrapolation method NEM.
2.4. Variable transformation predictor
The nonlinearity in the process and hence in the
model limits the prediction. The model which is
used to map the nonlinear process by variable
transformation is linear and this model is used
here to predict the future variable as shown in Fig.
3. The configuration is a combination of Smith
predictor and variable transformation and hence
may be called the variable transformation predictor VTP. The transfer function model of the
transformed pseudolinear system 10 is
z ( s ) / v ( s ) b/s. The transfer function model b/s
will give the predicted transformed process variable z . The model prediction error e m is the difference between the predicted output and the
transformed process variable z. e m is added in the
feedback path as an additional error.

84

Anandanatarajan, Chidambaram, Jayasingh / ISA Transactions 44 (2005) 8191

Fig. 3. Block diagram representation of the variable transformation predictor. The PI controller is designed for the
linear system Eq. 10.

3. Application of GLC to conical tank level

process
The mathematical model of the conical tank liquid level system considered for the study is expressed as

dh u tT d u d tT d c h

,
dt
R2 2
2h
H

14

where H, R, c are total height, top radius, and outflow valve coefficient of the conical tank, respectively. h, u, and u d are liquid level, inflow rate, and
disturbance of the conical tank level process, respectively. The transformation g ( h ) transforms
the nonlinear system represented by Eq. 14 into
a linear system. The transformed system is a
pseudolinear system, whose mathematical model
is

dz
ab v .
dt

Fig. 4. Block diagram of the Smith PI controller.

but practically it is found that selection of large

values of a and b end up with overflow error in the
transformed variable.
The transfer function of Eq. 15 is

G s b/s.

16

The forward path transfer function of the closedloop system with a PI controller is

G f s bK c 1T is /T is 2 .

17

T i and K c can be obtained by assuming a closedloop time constant and damping factor 9. Interestingly, the tuning parameters are independent of
local time constant and local gain of the process.
The transformation which transforms the present
process in to a linear system is
g h

2aR 2
5H 2 c

h 5/2.

18

4. Tuning of controllers

15

The values of a and b are assumed as 0.7657

and 0.0177, respectively. Mathematically there is
no restriction on the selection of values of a and b

The responses are compared with a conventional

PI controller tuned about nominal operating point
of 39%. The time constant and gain of linearized
model are 76 sec and 1.2, respectively. The process dead time is 32 sec. The Ziegler and Nichols

Table 1
ISE of Regulatory responses for 15% decrease in load.

Table 2
ISE of Regulatory responses for 15% increase in load.

Operating
point
39%
24%
54%

ZN
PI

Smith
PI

GLC

Operating
point

1285
2324
1224

3269
2762
3740

928
1492
518

39%
24%
54%

ZN
PI

Smith
PI

GLC

1236
1780
1221

3367
2991
3779

807
2162
448

Anandanatarajan, Chidambaram, Jayasingh / ISA Transactions 44 (2005) 8191

85

Table 3
ISE of servo responses for 20% increase in set point.
Operating
point
39%
24%
54%

ZN
PI

Smith
PI

GLC

23670
19260
32030

35480
27530
47660

16090
15780
16260

ZN tuning parameters for the linearized model

are K c0.95 and T i0.020. The tuning parameters for the Smith PI controller as shown in Fig. 4
are based on Haalmans tuning rule. They are K c
0.5747 and T i0.022 10. Similarly for both
GLCs are K c2.87 and T i0.025.
The simulation and experimentation are carried
out by taking 39% as nominal level. 24% and 54%
are other operating points used to test the robustness of the controller tuned at 39% nominal level.
The integral square error ISE values are presented in the Tables 1 4. Regulatory responses for
a 15% decrease in load at nominal operating point
39% refer to Fig. 5 show that the proposed controller gives an improved response 28% lesser
ISE while the conventional Smith predictor
150% higher ISE gives a poorer performance
than the ZN-PI controller.
Regulatory responses for a 15% increase in load
at nominal operating point 39% refer to Fig. 6
show that the proposed controller improves the response 35% lesser ISE while the conventional
Smith predictor 172% higher ISE gives a poorer
performance than the ZN-PI controller.
Regulatory responses for a 15% decrease in load
at 24% operating point but tuned at 39% refer to
Fig. 7 show that the proposed controller provides
a better response 36% lesser ISE than the conventional Smith predictor 19% higher ISE. The
ZN-PI controller gives an oscillatory response.
Regulatory responses for a 15% increase in load
at 24% operating point but tuned at 39% refer to
Fig. 8 show that the proposed controller provides

Fig. 5. Regulatory responses for a 15% decrease in load at

39% nominal operating point.

a better response 26% lesser ISE than the conventional Smith predictor. The ZN-PI controller
gives an oscillatory response. Regulatory responses for a 15% decrease in load at 54% operating point but tuned at 39% refer to Fig. 9 show
that the proposed controller gives an improved
performance 58% lesser ISE while the conventional Smith predictor 205% higher ISE provides
a poorer performance than the ZN-PI controller.

Table 4
ISE of servo responses for 20% decrease in set point.
Operating
point
39%
24%
54%

ZN
PI

Smith
PI

GLC

46920
18918
28980

28900
14260
36270

20030
10930
18420

Fig. 6. Regulatory responses for a 15% increase in load at

39% nominal operating point.

86

Anandanatarajan, Chidambaram, Jayasingh / ISA Transactions 44 (2005) 8191

Fig. 7. Regulatory responses for a 15% decrease in load at

24% operating point.

Fig. 9. Regulatory responses for a 15% decrease in load at

54% operating point.

Regulatory responses for a 15% increase in load

at 54% operating point but tuned at 39% refer to
Fig. 10 show that the proposed controller improves the performance 63% lesser ISE while
the conventional Smith predictor 209% higher
ISE gives a poorer response than the ZN-PI controller.
Servo responses for a 20% decrease in the set
point at nominal operating point 39% refer to Fig.
11 show that the proposed controller gives an im-

proved response 31% lesser ISE than the conventional Smith predictor. The ZN-PI controller
provides an oscillatory response. Servo responses
for a 20% increase in the set point at nominal operating point 39% refer to Fig. 12 show that the
proposed controller improves the response 32%
lesser ISE while the conventional Smith predictor
50% higher ISE gives a poorer performance than
the ZN-PI controller.

Fig. 8. Regulatory responses for a 15% increase in load at

24% operating point.

Fig. 10. Regulatory responses for a 15% increase in load at

54% operating point.

Anandanatarajan, Chidambaram, Jayasingh / ISA Transactions 44 (2005) 8191

Fig. 11. Servo responses for a 20% decrease in set point at

39% nominal operating point.

87

Fig. 13. Servo responses for a 20% decrease in set point at

24% operating point.

Servo responses for a 14% decrease in the set

point at 24% operating point but tuned at 39%
refer to Fig. 13 show that the proposed controller
improves the response 33% lesser ISE more than
the conventional Smith predictor. The ZN-PI controller gives an oscillatory response. At this operating point it is not possible to decrease the set
point greater than 14% because of numerical
round-off error. Some values in the transformed

domain are lesser than the tolerance that can be

handled by the computer.
Servo responses for a 20% increase in set point
at 24% operating point but tuned at 39% refer to
Fig. 14 show that the proposed controller improves the response 18% lesser ISE while the
conventional Smith predictor 43% higher ISE
gives a poorer performance than the ZN-PI controller.

Fig. 12. Servo responses for a 20% increase in set point at

39% nominal operating point.

Fig. 14. Servo responses for a 20% increase in set point at

24% operating point.

88

Anandanatarajan, Chidambaram, Jayasingh / ISA Transactions 44 (2005) 8191

Fig. 15. Servo responses for a 20% decrease in set point at

54% operating point.

Servo responses for a 20% decrease in set point

at 54% operating point but tuned at 39% refer to
Fig. 15 show that the proposed controller improves the response 36% lesser ISE while the
conventional Smith predictor 25% higher ISE
gives a poorer response than the ZN-PI controller.
Servo responses for a 20% increase in set point at
54% operating point but tuned at 39% as in Fig.
16 show that the proposed controller improves the
response 51% lesser ISE than the ZN-PI control-

Fig. 17. Servo responses of conical tank level process for

8% step change at nominal operating point 39% using
GLC VTP and GLC NEM with the same PI settings.

ler and the conventional Smith predictor 49%

higher ISE provides a very poor response.
The servo responses of the conical tank level
process for a 8% decrease in set point at the nominal operating point of 39% using GLC
controllersone with NEM prediction and the
other with proposed VTP prediction but both with
same tuning parametersare shown in Fig. 17.
The response given by NEM has slight oscillations. The magnitude of oscillations increases as
the step magnitude increases. This indicates that
the prediction by NEM is very poor compared
with the proposed GLC with VTP.
Figure 18 shows the experimental servo responses for various set point changes at 39%
nominal level. The responses are slightly oscillatory. Figures 19 and 20 show the experimental
regulatory responses for various magnitudes of
load changes at the nominal operating point. The
regulatory responses are found to have small oscillations as compared to the servo responses. Figure 21 shows a block diagram of the experimental
setup.
5. Conclusion

Fig. 16. Servo responses for a 20% increase in set point at

54% operating point.

A nonlinear controller is designed based on the

variable transformation for the first-order nonlin-

Anandanatarajan, Chidambaram, Jayasingh / ISA Transactions 44 (2005) 8191

89

Fig. 18. Experimental servo responses of conical tank level process for set point changes at nominal operating point 39%
using GLC VTP.

ear process with dead time. The performances of

proposed GLC VTP are tested by simulation and
compared with ZN PI and Smith PI controllers.
The ZN PI controller gives an oscillatory response
for decrease in set point even at a nominal operating point. The situation becomes worse when the
operating point is shifted to 24%. For any increase
in the set point the Smith PI controller gives a
very poor response due to a very small dead time
to time constant ratio. This situation becomes

worse when the operating point is shifted to 54%.

On the other hand, the proposed controller gives
better dynamics.
Consider the regulatory responses at 24% operating point for decrease in load as shown in Fig. 7.
The ZN-PI controller gives a highly oscillatory response. On the other hand, Smith PI and GLC give
an oscillation free response. Similarly, consider
the regulatory responses at 24% operating point
for an increase in load as shown in Fig. 8. Here

Fig. 19. Experimental regulatory responses of conical tank level process for increase in load changes at nominal operating
point 39% using GLC VTP.

90

Anandanatarajan, Chidambaram, Jayasingh / ISA Transactions 44 (2005) 8191

Fig. 20. Experimental regulatory responses of conical tank level process for decrease in load changes at nominal operating
point 39% using GLC VTP.

also the ZN-PI controller gives a highly oscillatory

response but the Smith PI gives an oscillation-free
response. GLC gives an oscillatory response but
quickly settles unlike that of the ZN-PI controller.
In this case the drawback with the GLC is the
large undershoot. But it has less ISE than that of
Smith PI. Consider the regulatory responses at
54% operating point refer to Figs. 9 and 10. The
Smith PI controller gives a large undershoot and a
large overshoot compared to that of the ZN-PI
controller. But GLC gives a very small overshoot
and undershoot compared to that of the ZN-PI
controller. Consider the servo responses at 39%
and 24% refer to Figs. 11 and 13. The ZN-PI

controller gives oscillatory responses and the

Smith PI gives sluggish responses, GLC gives oscillation free responses with quick rise time. The
simulation results show the robustness of the
GLC.
The proposed controller outperforms the Smith
PI and the ZN-PI controllers when the operating
point of the process is shifted over the entire span
of the tank. While comparing VTP and NEM, VTP
gives better dynamics than NEM. As a future
work, the GLC can be tested on other nonlinear
systems; the work in this direction is ongoing.
Acknowledgments
We are grateful to Professor K. Ethirajulu for his
constant encouragement and providing the facilities. We thank Professor P. Dhananjayan for his
encouragement.
References

Fig. 21. Block diagram of experimental setup.

1 Meyer, C., Seborg, D. E., and Wood, R. K., A comparison of the Smith predictor and conventional feedback control. Chem. Eng. Sci. 31, 775778 1976.
2 Hagglund, T., A predictive PI controller for processes
with long dead times. IEEE Control Syst. Mag. 12 1,
57 60 1992.
3 Tan, K. K., Lee, T. H., and Leu, F. M., Predictive PI
versus Smith control for dead-time compensation. ISA
Trans. 40, 1729 2001.
4 Schneider, D. M., Control of process with time delays.
IEEE Trans. Ind. Appl. 24 2, 186 191 1988.
5 Ray, W. H., Advanced Process Control. McGraw-Hill,
New York, 1981.

Anandanatarajan, Chidambaram, Jayasingh / ISA Transactions 44 (2005) 8191

6 Vrecko, D., Vrancic, D., Juricic, D., and Strmcnik, S.,

A new modified Smith predictor: The concept, design
and tuning. ISA Trans. 40, 111121 2001.
7 Ogunnalke, B. A., Controller design for nonlinear process systems via variable transformations. Ind. Eng.
Chem. Process Des. Dev. 25, 241248 1986.
8 Chidambaram, M., Anandanatarajan, R., and Jayasingh, T., Controller design for nonlinear process with
dead time via variable transformations. Proceedings of
the International Symposium on Process Systems Engineering and Control ISPSEC03, IIT, Mumbai,
2003, pp. 223228.
9 Chidambaram, M., Applied Process Control. Allied
Publishers, New Delhi, India, 1998.
10 Tian, Y. C. and Gao, F., Double-controller scheme for
control of processes with dominant delay. IEE Proc.:
Control Theory Appl. 145 5, 479 484 1998.

R.
Anandanatarajan obtained his B.Sc. Mathematics
from Madras University in
1984, Bachelor degree in Electrical from the Institution of
Engineers India in 1989,
M.Sc. Mathematics and M.E.
Process control and Instrumentation from Annamalai
University in 1994 and 1998,
respectively, Ph.D. from Anna
University in 2003. Presently
he is assistant professor of the
Department of Instrumentation
Engineering, Pondicherry Engineering College. He has authored two
books titled Signals & Systems and Computer Peripherals and Interfacing.

91

M. Chidambaram obtained
his B.E. Chemical from Annamalai University, M.E.
Chemical and Ph.D. from Indian Institute of Science in
1977 and 1984, respectively.
He was faculty member in IITBombay during 1984 to 1991.
Since September 1991 he has
been a faculty member in IITMadras. He had been head of
the Department of Chemical
Engineering during the period
October 2000 to October 2003.
He has authored three books: Nonlinear Process Control. John Wiley,
1996; Applied Process Control. Allied Publishers, 1998; Computer
Control of Processes. Narosa Publishers, 2002. He has published 135
research publications in journals and 45 papers in conferences proceedings. His areas of interest in process control are PID control, relay
tuning, and nonlinear control.

T. Jayasingh obtained his

Ph.D. in Control and Instrumentation from IIT-Delhi. He
has more than two decades of
experience in teaching and research at Anna University,
Chennai, India. Presently he is
a visiting professor at St.
Xaviers College of Engineering, Nagercoil, India. His areas
of interest include process control, computer based instrumentation, and sensor modeling.