Professional Documents
Culture Documents
Sliding Mode Control An Approach To Regulate Nonlinear
Sliding Mode Control An Approach To Regulate Nonlinear
www.elsevier.com/locate/isatrans
Abstract
A new approach for the design of sliding mode controllers based on a ®rst-order-plus-deadtime model of the process,
is developed. This approach results in a ®xed structure controller with a set of tuning equations as a function of the
characteristic parameters of the model. The controller performance is judged by simulations on two nonlinear chemical
processes # 2000 Elsevier Science Ltd. All rights reserved.
Keywords: Sliding mode control; Variable structure control; Nonlinear chemical processes
general SMCr, which can be used for self-regulat- is simple to perform, and provides adequate models
ing chemical processes. The parameters of the for many applications. The curve is obtained by
model, process gain, K, process time constant, , introducing a step change in the output from the
and process deadtime, t0 , are used to obtain the controller and recording the transmitter output.
initial estimates of the tuning terms in the SMCr. From the process curve shown in the ®gure, and
This article is organized as follows. Section 2 the procedure presented in the reference, the
brie¯y presents the process model. Section 3, pre- numerical values of the terms in the FOPDT
sents some basic concepts of the SMC method. model given in Eq. 1 are obtained
Section 4 shows the procedure to design a SMCr
using the FOPDT model. Tuning equations for X
s Keÿt0 s
1
the controller are also given in this section. In U
s s 1
Section 5 the simulation of the SMCr for two
nonlinear chemical processes is presented. Section where X
s is the Laplace transform of the con-
6 concludes the paper. trolled variable, the transmitter output, and U
s
is the Laplace transform of the manipulated vari-
able, the controller output. Both X
s and U
s are
2. Process model deviation variables. In this paper we use the unit
of X
s as fraction of the transmitter output, frac-
The process reaction curve, Fig. 1, is an often- tion TO; the unit of U
s is fraction of the con-
used method for identifying dynamic models 1. It troller output, fraction CO. K, t0 and were
previously de®ned.
S
t
UD
t KD
6
S
t
The above approximation can also be written as a Substituting Eq. (7) into Eq. (1) produces
®rst-order Pade approximation
X
s K
1 ÿ 0:5t0 s
9
eÿt0 s
8 U
s
s 1
t0 s 1
1 0:5t0 s
Fig. 3 shows a comparison among the deadtime In dierential equation form
term and the ®rst-order Taylor series and PadeÂ
approximations. The ®gure shows that the Pade d2 X
t dX
t
t0
t0 X
t KU
t
10
approximation works very well between 0 and 1 but dt2 dt
beyond the approximation brakes down. On other
hand, the Taylor series approximation improves and since this is a second-order dierential equa-
as t0 increases. tion, n 2, from Eq. (2) S
t becomes
In [10] is shown that the ®rst-order Taylor
approximation or the Pade approximation can be
t
de
t
considered as good approximations for the dead- S
t l1 e
t l0 e
tdt
11
d
t 0
time term for chemical processes.
The next section shows the development of a
SMCr using both approximations. Where l1 2l and l0 l2
From Eq. (3)
4.1. SMCr development based on a ®rst-order
Taylor series approximation dS
t d2 e
t de
t
2
l1 l0 e
t 0
12
dt dt dt
In this section a SMCr is developed based on the
®rst-order Taylor series expansion. Additionally, a Substituting the de®nition of the error,
rule to choose the tuning parameters will also be e
t R
t ÿ X
t, into the ®rst two terms of the
presented. above equation gives
Fig. 3. Comparison among eÿx (1), Taylor (2) and Pade (3) approximations.
O. Camacho, C.A. Smith / ISA Transactions 39 (2000) 205±218 209
d2 R
t d2 X
t dR
t dX
t with
ÿ l 1 ÿ
dt2 dt2 dt dt
S
t sign
K
l0 e
t 0
13
t
18b
dX
t
ÿ l1 e
t l0 e
tdt
Solving for the highest derivative from Eq. (10), dt 0
substituting it into the Eq. (13), and solving for
U
t provides the continuous part of the controller Eqs. (18a) and (18b) constitute the controller
" equations to be used. These equations present
t t dX
t X
t advantages from process control point of view,
0 0
UC
t l1
K t0 dt t0 ®rst they have a ®xed structure depending on the
# l's parameters and the characteristic parameters
d2 R
t dR
t of the FOPDT model, and second the action of
l0 e
t l1
14
dt2 dt the controller is considered in the sliding surface
equation, by including the term sign
K, in Eq.
This procedure, involving Eqs. (11) and (13), to 18b. Note, that sign
K only depends on the static
obtain the expression for the continuous part of gain, therefore it never switches. From an indus-
the controller is known in the SMC theory as the trial application perspective, Eq. (18b) represents a
equivalent control procedure [2]. PID algorithm [12].
In [11] is shown that the derivatives of the To complete the SMCr, it is necessary to have a
reference value can be discarded, without any set of tuning equations. For the tuning equations
eect on the control performance, resulting in a as ®rst estimates, using the Nelder±Mead search-
simpler controller. Thus, ing algorithm [13], the following equations were
" obtained [11].
t t
dX
t
0 0
UC
t ÿ l1 . For the continuous part of the controller and
K t0 dt
# the sliding surface
X
t
l0 e
t
15 t0
t0 l1 timeÿ1
19a
t0
4.2. SMCr development based on the Pade Thus, from the previous results, the controller
approximation equation to be used is that obtained from the
Taylor series approximation. The next part illus-
This section contains the development of the trates the controller performance.
control law when the deadtime term of the FOPDT
process model is approximated by the PadeÂ
approximation, Eq. (8). The procedure followed in 5. Simulation results
this section is similar to that one presented in the
previous part. Substituting Eq. (8) into Eq. (1), This section simulates the control performance
gives of the SMCr designed and given in Eqs. (18a) and
(18b). The ®rst process, a mixing tank, compares
XC
s K
2 ÿ t0 s the performance of the SMCr with respect to a
20 PID controller. The second process, a chemical
U
s
s 1
2 t0 s
reactor, presents further performance character-
istics.
Using a similar procedure as shown above, the
continuous part of the controller, UC
s, is 5.1. Mixing tank
Table 1 where
Design parameters and steady-state values
Variable Value Variable Value W1
t = mass ¯ow of hot stream, lb/min
W2
t = mass ¯ow of cold stream, lb/min
W1 250.00 lb/min V 15 ft3 Cp = liquid heat capacity at constant pressure,
W2 191.17 lb/min TO 0.5
Cp1 0.8 Btu/lb- F Vp 0.478
Btu/lb- F
Cp2 1.0 Btu/lb- F CVL 12 gpm/psi1/2 Cv = liquid heat capacity at constant volume,
Cp3 ; Cv3 0.9 Btu/lb- F Pv 16 psi Btu/lb- F
Set point 150 F T 0.5 min T1
t = hot ¯ow temperature, F
T1 250 F vp 0.4 min T2
t = cold ¯ow temperature, F
T2 50 F A 0.2006 ft2
T3 150 F L 125 ft
T3
t = liquid temperature in the mixing tank, F
62.4 lb/ft3 m 0.478 CO T4
t = equal to T3
t delayed by t0 , F
t0 = deadtime or transportation lag, min
= density of the mixing tank contents,
lbm/ft3
. Energy balance around mixing tank
V = liquid volume, ft3
W1
tCp1
tT1
t W2
tCp2
tT2
t TO
t = transmitter output signal on a scale from
0 to 1
ÿ
W1
t W2
tCp3
tT3
t VP
t = valve position, from 0 (valve closed) to
1 (valve open)
dT3
t m
t = fraction of controller output, from 0 to 1
VCv3
22 CVL = valve ¯ow coecient, gpm/psi1/2
dt
Gf = speci®c gravity, dimensionless
. Pipe delay between the tank and the sensor Pv = pressure drop across the valve, psi
location T = time constant of the temperature sensor,
min
T4
t T3
t ÿ t0
23
Vp = time constant of the actuator, min
. Transportation lag or delay time A = pipe cross section, ft2
L = pipe length, ft
LA
t0
24
W1
t W2
t Following the procedure, presented in Section 2,
to obtain the parameters of the FOPDT model
. Temperature transmitter yields: K ÿ0:78 fraction TO/fraction CO,
2:32 min., and t0 2:97 min. Using these values
dTO
t 1 T4
t ÿ 100
ÿ TO
t
25 the tuning parameters for the SMCr are
dt T 100
SMCr initial tuning equations, but they can be mance of the PID controller. To recover stability,
adjusted, ®ne tuning, until acceptable control per- new tunings are required for the PI controller
formance be obtained. while none are required for the SMCr.
Please note that the controller equations, Eqs. In spite of the controller being synthesized using
(18a) and (18b), were developed using deviation a Taylor approximation and the tuning equations,
variables. The following changes the ``deviation Eqs. 19a±19d, are empirical, the proposed method
variables'' in the controller to ``actual variables'' can be successfully used in processes with a dead-
time to time constant ratio larger than one. In our
U
t m
t ÿ m experience, they can be applied for t0 = around 3.
and
X t TO t ÿ TO
with
t
dTO
t
S
t ÿ l1 e
t ÿ l0 e
tdt
18d
dt 0
Table 2
Design parameters and steady-state values
Also , from Eqs. (19c) and (19d) never perfect. Martin [14] considers that modeling
error of 25% in its parameters is a ``reasonable
KD 0:96 fraction CO error''. Let us consider two cases. The ®rst case is
for ÿ10% model error and the second one is for
0:76 fraction TO=min 100% in model error. The second case could be
considered an ``unreasonable error,'' but our
Fig. 8 shows the system response when a +10% intent is to judge the controller. The error used is
change in inlet ¯ow occurs. The ®gure shows that, the same in every parameter, that is, the same
because the temperature of the inlet ¯ow is cooler ÿ10% error in K, and t0 .
than the temperature in the reactor, the reactor Fig. 9 shows the open loop responses for the
temperature ®rst decreases somewhat. However, actual process and for the model with a ÿ10 and
after a short while the temperature in the reactor 100% error.
increases since more reactant is added to the reactor. Figs. 10 and 11 show the process response when
Fig. 8 shows the control performance when the the inlet ¯ow changes by 10% and the modeling
modeling error between the real process and the error used is ÿ10 and 100%, respectively. A com-
FOPDT model is small. However, the model is parison of Figs. 8 and 6, when no error in the
Fig. 10. System responses for 10% change in inlet ¯ow for ÿ10% error in modeling.
O. Camacho, C.A. Smith / ISA Transactions 39 (2000) 205±218 217
Fig. 11. System responses for 10% change in inlet ¯ow for 100% error in modeling.
model is present shows little dierence in the pro- The controller law, Eqs. (18a) and (18b) should
cess response. Fig. 9 shows that with 100% error be rather easy to implement in any computer sys-
in the model, the control performance degrades tem (DCS) [12].
somewhat. The most signi®cant dierence is that it
takes longer to return the process to the set point. References
However, even with such a large error in the
model, the control is still stable. [1] C.A. Smith, A.B. Corripio, Principles and Practice of
Automatic Process Control, John Wiley & Sons, New
York, 1997.
[2] V.I., Utkin, Variable structure systems with sliding modes,
6. Conclusions Transactions of IEEE on Automatic Control, AC ± 22
(1997) 212±222.
This paper has shown the synthesis of a sliding [3] H. Sira-Ramirez, O. Llanes-Santiago, Dynamical dis-
mode controller based on an FOPDT model of the continuous feedback strategies in the regulation of non-
linear chemical processes, IEEE Transactions on Control
actual process. The controller obtained is of ®xed Systems Technology 2 (1) (1994) 11±21.
structure. A set of equations obtains the ®rst esti- [4] J.J. Slotine, W. Li, Applied Nonlinear Control, Prentice-
mates for the tuning parameters. The examples Hall, New Jersey, 1991.
presented indicate that the SMCr performance is [5] M.C. Colantino, A.C. Desages, J.A. Romagnoli, A. Pala-
zoglu, Nonlinear control of a CSTR: disturbance rejec-
stable and quite satisfactory in spite of non-
tion using sliding mode control, Industrial & Engineering
linearities over a wide range of operating condi- Chemistry Research 34 (1995) 2383±2392.
tions. The relations given in Eq. (19) provided a [6] A.S.I. Zinober, Variable Structure and Liapunov Control,
good starting set of tunings. Springer±Verlag, London, 1994.
218 O. Camacho, C.A. Smith / ISA Transactions 39 (2000) 205±218
[7] K.D. Young, V.I. Utkin, UÈ. OÈzgumer, A control engi- [11] O.E. Camacho, A new approach to design and tune slid-
neer's guide to sliding mode control, IEEE Transactions ing mode controllers for chemical processes, Ph.D. Dis-
on Control Systems Technology 7 (3) (1999) 328±342. sertation, 1996, University of South Florida, Tampa,
[8] J.Y. Hung, W. Gao, J.C. Hung, Variable structure con- Florida.
trol: a survey, IEEE Transactions on Industrial Electro- [12] O. Camacho, C. Smith, E. ChacoÂn, (1997). Toward an
nics 40 (1) (1993) 2±21. implementation of sliding mode control to chemical pro-
[9] G.E. Young, S. Rao, Robust sliding-mode of a nonlinear cesses, in: Proceedings of ISIE'97, Guimaraes-Portugal,
process with uncertainty and delay, Journal of Dynamical 1997, pp. 1101±1105.
Systems, Measurement, and Control 109 (1987) 202±208. [13] D.M. Himmelblau, Applied Nonlinear Programming,
[10] O. Camacho, R. Rojas, W. Garcia, Variable structure McGraw-Hill, New York, 1972.
control applied to chemical processes with inverse [14] T.E. Marlin, Process Control, McGraw-Hill, New York,
response, ISA Transactions 38 (1999) 55±72. 1995.
Oscar Camacho received the Electrical Engi- Carlos A. Smith, is Professor of Chemical Engi-
neering, and M.S. in Control Engineering degrees neering and Associate Dean for Academic Aairs
from Universidad de Los Andes (ULA)m MeÂrida, at the University of South Florida. Before joining
Venezuela, in 1984 and 1992, respectively, and the USF he worked for Dow Chemical USA from
M.E. and Ph.D. in Chemical Engineering at Uni- 1967 through 1972. Professor Smith is a co-author
versity of South Florida (USF), Tampa, FL, in of two editions of Principles and Practice of Auto-
1994 and 1996, respectively. matic Process Control, 1986 and 1975, published
He has held teaching and research positions at by John Wiley & Sons. Professor Smith has been
ULA, CIED-PDVSA since 1985. He has been the in consultancy work, as well as lecturing on short
Chairman of the Electrical Engineering School at courses, for many companies in the US, Canada,
ULA since 1998. His current research interest Latin America, and Europe. Professor Smith loves
includes sliding mode control, hybrid systems and teaching, working with graduate students, and
passivity-based control. He is the author of more dealing with industry.
than 30 publications in journals and conference
proceedings.