You are on page 1of 5

ISA Transactions 47 (2008) 80–84

www.elsevier.com/locate/isatrans

Problem and its solution for actuator saturation of integrating


process with dead time
Maorui Zhang ∗ , Canghua Jiang
Center for Control Theory and Guidance Technology, No. 92, West Da-Zhi Street, 150001 Harbin, Heilongjiang, China

Received 6 October 2006; accepted 7 May 2007


Available online 20 August 2007

Abstract

The modified Smith predictor (MSP) for integrating process with dead time (IPDT) has at least one pole at the origin of s-plane. The integral
effect leads to large overshoot and slow settling time when the control signal exceeds the saturation limits of the system actuator. The windup
problem in MSP is presented and studied. Based on the two degrees of freedom (2DOF) control schemes presented by Normey-Rico and Camacho,
an anti-windup controller (AWC) is proposed. By simulation, significant performance improvement demonstrates the effectiveness of the proposed
scheme.
c 2007, ISA. Published by Elsevier Ltd. All rights reserved.

Keywords: Actuator saturation; Modified Smith predictor; Integrated processes; Anti-windup

1. Introduction and Ito [1] pointed out that if the process has poles near
the origin on the left half s-plane, the response may be
All processes are subjected to constraints in practice. The sluggish enough to be unacceptable, and they modified the
actuators have a limited range of action. Furthermore, in Smith predictor by replacing the exact process deliberately
practice, the operating points of plants are determined to by a mismatched process. Watanabe’s controller could yield
satisfy the overall economic goals of the process and lie at the zero-steady-state error and the desired transient response to
intersection of certain constraints. The control system normally step disturbance. Since then, a number of different approaches
operates close to the limits and constraint violations are likely have been developed to control IPDT [1,3–15,18]. Normey-
to occur. Therefore, in order to be successful, any controller Rico and Camacho modified Watanabe and Ito’s dead-time
must deal with constraint violations and correct them. Model compensator structure by adding a filter F(s) [10,13,14] for
Predictive Control (MPC) gives a systematic method to cope set-point signal. The controller [14] is compared to the other
with constraints. However, an MPC application is more costly, algorithms using robustness and performance indices; and for
time consuming and requires personnel with better training in the cases studied it is demonstrated to have better or at least the
control [2]. same performance–robustness as the others, and could maintain
If the closed-loop set-point response must guarantee a the simple tuning characteristics of the original SP [14].
desired performance with disturbance rejection, two degrees of However, even if omitting the nonlinear system leads to a
freedom (2DOF) controller is a general solution [14]. Because controller, an important problem in practice is still left behind.
2DOF controller structure decouples the set-point response It is known that the actuator saturation in practical system
from the disturbance response, both the set-point response is common, and windup phenomenon is observed in the PI
and the disturbance response can be tuned separately. Most or PID control systems. Pagano et al. [16] mentioned the
of the MSPs for IPDT are 2DOF controller structures. IPDT effects of saturation element in the system with Smith predictor.
is a representative model in the industry. In 1981, Watanabe They studied the effects of the dead-time uncertainties and the
actuator saturation based on the control scheme in paper [10].
∗ Corresponding author. Tel.: +86 451 86402214. In order to avoid the actuator saturation, the idea of shaping the
E-mail address: zhangmaorui@hit.edu.cn (M. Zhang). control signal was proposed in [17] to offer a two-level control

c 2007, ISA. Published by Elsevier Ltd. All rights reserved.


0019-0578/$ - see front matter
doi:10.1016/j.isatra.2007.05.004
M. Zhang, C. Jiang / ISA Transactions 47 (2008) 80–84 81

Fig. 1. Actuator saturation process with dead time and integrator.

Fig. 3. Original Pagano’s saturated Smith predictor [16].

problem. The IMC principal for MSP is attempted by Pagano


et al. [16].
K
G m (s) = sV n (1 − s · L n ) was suggested in this scheme.
Fig. 2. Equivalent system of MSPs for IPDT. When IMC principal is applied for these MSPs as in Fig. 3,
whether or not the system is internal stable, and whether or not
for processes with dead time and input constraints. However, the performances can be improved need to be verified carefully.
this technique is restricted to the first-order plus dead-time Because there are various MSPs for IPDT, the following
process (FOPDT). analysis will be based on the scheme proposed by Normey-
This paper studies the windup problem in MSP when there Rico and Camacho. The actuator saturation and our proposed
is a saturation element in the system. The organization of the AWC are shown in Fig. 4. If the control signal is within the
paper is as follows. In Section 1, the windup problem in the saturation bounds, u 3 − u 1 = 0, and the AWC does not affect
MSP for IPDT is presented, which is followed by the proposed the system. A PI controller, C(s) = K C · (1 + 1/s · Ti ) with
anti-windup controller. Then, the stability of the system with Ti = 2T0 + L n and K c = 2T0 + L n /(K vn (T0 + L n )2 ),
anti-windup and MSP is analyzed in Section 3. In Section 4, and F(s) = (1 + sT0 )2 /(1 + sTi )(1 + sT1 ) are suggested
comparisons between the scheme with and without anti-windup in this scheme. The proposed controller has four parameters
are carried out by simulation. Finally, conclusions end the (K vn , L n , T1 and T0 ). When a model of the process is obtained
paper. experimentally, only two parameters (T1 and T0 ) need to be
tuned.
2. Windup problem of MSP for IPDT and anti-windup The internal feedback channel can be rewritten as
controller
H (s) = G m (s) − Pn (s) = K vn (1 − L n s − e−L n s )/s
The saturation function f SAT is defined as: 1 − e−L n s
= −K vn L n + K vn , (2)
s
UupL , U I ≥ UupL

UO = UI , ULoL < U I < UupL (1) where the first part is a proportional gain, and the second part is
ULoL , U I ≤ ULoL . a zero-order holder with coefficient K vn . The internal feedback

channel is internal stable.
When the input signal is within the range specified by The feedback controller is:
the Lower limit (ULoL ) and Upper limit (UupL ), it passes
u c (s) C(s)
through the saturation element without changes. Without loss C1 (s) = =
of generality, we let ABS(ULoL ) = ABS(UupL ) = U L . When e1 (s) 1 + C(s)H (s)
the input signal is outside these bounds, the signal is clipped to K c s(1 + Ti s)
= . (3)
the upper or lower bound. Ti s + K c K vn (1 + Ti s)(1 − L n s − e−s L n )
2
sL
The transfer function of IPDT is P(s) = KsV e− . Let the If the system has a zero-steady-state error to step
K
model of the process be Pn (s) = sV n · e−s L n . IPDT with the disturbance, the controller must have at least one zero pole.
actuator saturation element is shown in Fig. 1. From Eq. (3), it follows that
Theoretically, for disturbance rejection, the control schemes
K K 1
of MSPs for IPDT in existence can be simplified to Fig. 2. The C1 (s) = ∞ + · ∞ . (4)
sL Ti · s Q
controller G c (s) may be the function of s and e− n . It is much (τi s + 1) (τi s + 1)
Q
more complicated than classical PI or PID controller. i=1 i=1
As each MSP needs to reject step disturbances, the controller As C1 (s) is stable, Re(1/τi ) > 0, 1 ≤ i < ∞ should
G c (s) must have at least one pole at the origin of s-plane. be satisfied. By Eq. (4), the controller C1 (s) is divided into
So, the existing schemes of MSP for IPDT will have windup two parts, one has proportional characteristic, and another
problem when the control signal u 1 exceeds the saturation has integral characteristic. Theoretically, the windup problem
bounds of system actuator. should be caused mainly by the integration part.
Windup problem may be partially solved by Internal Model Windup exists in the system with saturation if the controller
Control (IMC), especially for FOPDT. If the process is modeled has integrating function. The results of windup effect include
as IPDT, because the traditional SP cannot be used [4], MSP excessive process overshoot and long process settling time.
must be introduced. The authors think that whether or not the Based on Normey-Rico and Camacho’s controller, the saturated
IMC principal is valid for the existing MSPs is still an open SP was proposed in [16]. Unfortunately, it is internal unstable.
82 M. Zhang, C. Jiang / ISA Transactions 47 (2008) 80–84

Fig. 4. Normey-Rico and Camacho’s structure with AWC.

If the actuator saturation is applied only to the model Pn (s)


as suggested in [16], even if the real implementation is done
as in Fig. 2, the system is not internal stable. Supposing the
disturbance, d(t) ≡ 0, and u c (t) = Am · u(t) (Am > 0, u(t)
is a unit step function) is the step signal with amplitude Am ,
the large amplitude Am will make the control signal exceed the
actuator bounds. If the system is stable, the control signal Am Fig. 5. Simplified control structure.
can be represented as

Am = ∆Am + U L , ∆Am > 0. (5)

Then,
Fig. 6. The simplified system.
u 2 (t) = L −1 (U2 (s))
= L −1 (G m (s)Am /s − Pn (s)U L /s) where
= L −1
((∆Am · (1 − s L n ) K C (1 + Ti · s)
F1 (s) =
+ U L (1 − s L n − e −s L n
))K vn /s )
2 K i + Ti · s + H (s) · K C · (1 + Ti · s)
Ki
= K vn (∆Am · (t − 1(t)L n ) F2 (s) = .
K i + Ti · s + H (s) · K C · (1 + Ti · s)

((−δ(t))(i−2) L in /i!)),
X
− UL (6) From Eq. (7), it follows that
i=2 
∞, Ki = 0
which shows that the inner feedback has the effect of lim u 1 (t) = KC (8)
integration. t→∞ ±U L − m 1 · , K i > 0.
Ki
If the actuator saturation is applied to both Pn (s) and G m (s),
According to Eq. (8), when K i > 0, the larger K i is,
simulation study shows that the close-loop performance cannot
the more accurate anti-windup effect may be obtained. When
be improved, although the system is internal stable.
K i = 0, limt→∞ u 1 (t) = ∞, which shows the controller has
Because the integral part can be separated from the MSP, integration effect. However, as the controller C1 (s) is not a PI
the similar anti-windup structure, which is shown in Fig. 4, for controller, some values of the gain K i , for the system might not
normal PI controller can be applied. However, it is much more be taken as suggested in classical anti-windup scheme for PI or
complicated, since from Eqs. (2) and (3), the controller C1 (s) PID controller. The stability of the system with the introduced
has as many as infinite of poles on the left half s-plane as well anti-windup scheme needs to be analyzed.
as one pole at the origin.
For studying the characteristic of the controller, we assume 3. Stability
that the control signal exceeds the saturation bounds. Because
u 3 is the output of the actuator, the absolute value of u 3 is not Because the amplitudes of harmonics are relatively small
larger than U L when windup happens. Assume that u 3 is a step in comparison with that of the fundamental, and the linear
signal with amplitude U L , i.e. u 3 (s) = ±U L /s, which may be part of the system can attenuate the harmonics sufficiently,
the worst case. For simplicity, the error e1 (t) is assumed to be the describing function method can be applied. If the model
a step signal with amplitude m 1 , though other forms of e1 (t) is perfectly matching, the stability of the system with the
may reveal more information. A qualitative analysis can show introduced anti-windup scheme may be analyzed by the
the performance improvement by introducing the anti-windup describing functions in the simplified system shown in Fig. 6.
technique. If the disturbance d(t) ≡ 0, Fig. 4 can be simplified G(s) is the equivalent transfer function. It is given by:
to be as in Fig. 5. NG K V K c (1 + Ti · s)e−Ls − K i · s
The control signal u 1 (t) is given by G(s) = = . (9)
DG s[K i + Ti · s + H (s)K c (1 + Ti · s)]
 
−1 m 1 UL The proposed controller has four parameters (K vn , L n ,
u 1 (t) = L (u 1 (s)) = L
−1
· F1 (s) ± · F2 (s) , (7) T1 and T0 ), and when a model of the process is obtained
s s
M. Zhang, C. Jiang / ISA Transactions 47 (2008) 80–84 83

Fig. 7. The stability equivalence system.

experimentally, only two parameters (T1 and T0 ) need to be


tuned. T1 does not affect the system stability. To analyze the
stability, the following time scale is employed,

s1 = L · s
(10)
T0 = α · L α ∈ (0, ∞).
With this time scale, the analysis is greatly simplified and
system’s stability will not be changed at the same time. For
K
example, the model Pn (s) = sV n · e−s L n becomes Pn (s1 ) =
K V n1
s1 ·e−s1 with K V n1 = K V n · L n . The parameters of the model Fig. 8. Output responses y(t): reference input r (t) = 28 · u(t), “—” y(t) of the
are reduced from two to one. scheme with AWC, “- -” y(t) of the scheme without AWC.
Large α will lead to slow disturbance response and, at the
same time, a robust behavior of the system. With the above time The system with high gain K i anti-windup feedback
scale, it follows from Eq. (9) that performs better than that with low gain, when there is no
noise in the system. Higher value of K i makes the anti-windup
G(s1 ) = performance more sensitive to high frequency noise, though it
−K i (α + 1)2 s1 /(2α + 1) + (1 + (2α + 1)s1 )e−s1 provides the stronger feedback to avoid controller saturation.
.
1 + (K i (α + 1)2 /(2α + 1) + 2α)s1 + α 2 s12 − (1 + (2α + 1)s1 )e−s1 With extremely heavy noise, a rule of thumb is suggested,
(11)
K i = (1 + α/α + 1)2 . (15)
Unlike the anti-windup design for traditional PI or PID
controller, the number of poles or zeroes of G(s1 ) is infinite, 4. Simulation results
which makes the analysis very complicated. By using Pade
expansion of e−s1 , the stability may be estimated. Let the gain If the model is perfectly matching and the system is linear,
of the saturation be K , the absolute value of the bounds be there is no overshoot in the IPDT process with MSP. One
U L , and the describing function be Nsat (A). Assuming that the possible reason of big overshoot and long settling time is the
describing function of the dead zone is Ndead (A), Ndead (A) = integral effect when the control signal exceeds the saturation
K (1 − Nsat (A)/K ) can be derived by comparing the two bounds.
describing functions. It shows that the saturation unit can be In these simulations, the performance of using AWC is
equivalent to a linear gain connected with a dead zone in series. compared with that without AWC. If the unit step function is
If K = 1, Ndead (A) = 1 − Nsat (A). It can be proved that the denoted by u(t), the unit step signal delayed by t0 second can
stability of system in Fig. 5 is equivalent to the system shown be denoted by u(t − t0 ). A step set-point signal r (t) = 28 · u(t)
in Fig. 7. and a step disturbance d(t) = −5 · u(t − 80) are introduced.
The describing function of the dead zone is The process transfer function G(s) = KsV e−Ls = 1s e−5s with
K V = 1 and L = 5. If U L = 1, the saturation function f SAT is
Ndead (A) given by:
  s 
   2
2 U U U

= K 1 − arcsin
L
+
L
1−
L  1, UI ≥ 1
π A A A U O = U I , −1 < U I < 1 (16)
−1, U I ≤ −1.

A ≥ UL , (12)
In Fig. 8, let α = 0.8. The parameters of controller T0 = 4s
where A is the amplitude of the input sine signal. G 1 (s1 ) is and T1 = 2.4s are taken. Assuming that u 3 (t) is contaminated
K i (α + 1)2 /(2α + 1)s1 − ((2α + 1)s1 + 1)e−s1 by uniformly distributed Gaussian noise with mean zero and
G 1 (s1 ) = . variance 0.015, K i = 2.0864 is taken.
(αs1 + 1)2
Model mismatch can lead to integral windup. The example
(13) shown in Fig. 9 provides a performance comparison between
Compared with G(s), G(s1 ) is simpler. Nyquist criterion can be the system with and without anti-windup modification. Assume
used to determine the system’s stability. If the model is perfectly that the process parameters are K v = 1 and L = 5s, and
matching, it can be proved that the system is stable when K i there is 20% mismatch in the dead time, that is L n = 4s. The
takes the following formula, equivalence of H (s) in Eq. (2) is used in the simulation.
The simulations in Figs. 8 and 9 show the response
K i ≥ (1 + α/α + 1)2 . (14) comparison between the scheme with AWC and that without
84 M. Zhang, C. Jiang / ISA Transactions 47 (2008) 80–84

a process with an integrator and long dead time. IEEE Trans Automat
Control 1994;39:343–5.
[4] Palmor ZJ. Time-delay compensation—Smith predictor and its modifica-
tions. In: Levine S, editor. The control handbook. Boca Raton (FL): CRC
Press; 1996. p. 224–37.
[5] Matausek M, Micic A. A modified Smith predictor for controlling a
process with an integrator and long dead-time. IEEE Trans Automat
Control 1996;41(8):1199–203.
[6] Zhang WD, Sun YX. Modified Smith predictor for controlling
integrator/time delay processes. Ind Eng Chem Res 1996;35:2769–72.
[7] Zhang WD, Sun YX, et al. Two degree-of-freedom Smith predictor for
processes with time delay. Automatica 1998;34(10):1279–82.
[8] Majhi S, Atherton DP. A new Smith predictor and controller for unstable
and integrating progresses with time delay. In: Proceedings of the 37th
IEEE conference on decision & control. 1998. p. 1341–5.
[9] Matausek MR, Micic AD. On the modified Smith predictor for controlling
Fig. 9. Output responses y(t): reference input r (t) = 10 · u(t), disturbance a process with an integrator and long dead-time. IEEE Trans Automat
d(t) = −1.8 · u(t − 50), “—” y(t) of the scheme with AWC, “- -” y(t) of the Control 1999;44(8):1603–6.
scheme without AWC. [10] Normey-Rico JE, Camacho EF. Robust tuning of dead-time compensators
for processes with an integrator and long dead-time. IEEE Trans Automat
AWC. The overshoot δ p , settling time ts , maximum error emax Control 1999;44:1597–603.
caused by disturbance and disturbance settling time tds are all [11] Majhi S, Atherton DP. Modified Smith predictor and controller for
smaller than those of the scheme without AWC, even when processes with time delay. IEE Proc Control Theory Appl 1999;146(5):
359–64.
there exists model mismatch. [12] Majhi S, Atherton DP. Obtaining controller parameters for a new Smith
predictor using autotuning. Automatica 2000;36:1651–8.
5. Conclusion [13] Normey-Rico J, Camacho E. A unified approach to design dead-
time compensators for stable and integrative processes with dead-time.
The windup problem in MSP for IPDT is first studied. In: Proceedings of the 40th IEEE conference on decision and control.
Based on the scheme of Normey-Rico and Camacho, an 2001. p. 2267–72.
[14] Normey-Rico J, Camacho E. A unified approach to design dead-time
anti-windup control scheme is proposed. The original control
compensators for stable and integrative processes with dead-time. IEEE
performances are kept when the control signal is smaller Trans Automat Control 2002;47(2):299–305.
than the actuator saturation bounds. When the control signal [15] Zhang W, Wang W, He X, Xu X. Analytical formulas for the new
exceeds the actuator bounds, set-point response speeds up with two degree-of-freedom smith predictor. In: Proceedings of the 4th world
the maximal acceleration. By the proposed control scheme congress on intelligent control and automation. 2002. p. 569–72.
[16] Pagano DJ, Normey-Rico JE, Franco ALD. Stability analysis of
with AWC, a significant improvement on system’s major
a modified smith predictor for integrative plants with Dead-time
performances, such as settling time ts , overshoot σ p , maximum uncertainties and saturations. In: Proceedings of the 40th IEEE conference
error emax caused by disturbance, and disturbance settling time on decision & control. 2001. p. 1855–60.
tds are achieved, which is demonstrated by the simulations. [17] Visioli A. Time-optimal plug & control for integrating and FOPDT
The method may be possibly applied to other modified Smith processes. J Process Control 2003;13:195–202.
[18] Morari M, Zafiriou E. Robust process control. Prentice Hall; 1989.
predictors for IPDT.
Robustness analysis for IPDT with AWC might be work left
for the future. Dr. Maorui Zhang was born on 1964 in Sichuan at
P.R. China. He is an Associate Professor in Center for
Control Theory and Guidance Technology at Harbin
Acknowledgements Institute of Technology. His research fields include:
Control of systems with dead times and uncertainties,
The authors are grateful to the anonymous reviewers who Application of control theory to process control, Auto-
contributed their valuable suggestions and comments to the tuning and self-tuning controllers, Magnetic levitation,
etc.
improvement of the content and organization of this paper. This
research is funded by China Scholarship Council (CSC).
Dr. Canghua Jiang received his B.Eng. and M.Eng.
References degrees in Automation from Dalian University of
Technology in 1998 and 2001, respectively, and his
[1] Watanabe K, Ito M. A process-model control for linear systems with Ph.D. degree in Automation from Tsinghua University
in 2006. Now he is a lecturer in Center for Control
delay. IEEE Trans Automat Control 1981;26:1261–9.
Theory and Guidance Technology at Harbin Institute of
[2] Glattfelder AH, Schaufelbegcr W. Start-up performance of different Technology. His research interests are fault diagnosis,
proportional-integral-anti-windup-regulators. Int J Control 1986;44(2): adaptive estimation and observer design for time-delay
493–505. systems.
[3] Astrom KJ, Hang CC, Lim BC. A new Smith predictor for controlling

You might also like