2017 Indian Control Conference (ICC)

January 4-6, 2017. Indian Institute of Technology, Guwahati, India

Modified Smith predictor and controller for

unstable first order processes

Moina Ajmeri Ahmad Ali

Electronics and Comm. Engineering Dept. Electrical Engineering Dept.
National Institute of Technology Patna Indian Institute of Technology Patna
Patna, India Patna, India
[email protected] [email protected]

Abstract—A number of tuning rules have been reported in integrating and unstable processes. Later, W. Tan [12] has
literature for unstable first order plus time delay processes. designed the controllers of the double two degree of
However, at least to the author knowledge most of the reported freedom structure using IMC (Internal Model Control)
methods fail to give robust performance for process models approach to achieve improved robustness. It is to be noted
with time delay equal to or greater than the process time that no guidelines are provided to select suitable values of
constant. In this work, a robust tuning method based on the tuning parameters in [7] and [12]. Furthermore, suitable
modified Smith predictor is proposed for unstable first order ranges of design parameters are provided in [8] and [9] and
plus time delay process models. Set point tracking controller is thereby making it difficult to select the suitable values of the
designed using direct synthesis approach. The PD controller
tuning parameters. As in practice, an approximate model of
which is used for load disturbance rejection is designed so as to
satisfy the Routh Hurwitz’s stability criterion. Suitable value
the plant is used for controller design and also, the
of the tuning parameter is recommended after studying its parameters of the physical systems vary with operating
effect on the system performance and robustness. This is also conditions and time, tuning method must be robust to
an advantage of the present work over the other recently process parameters’ variations. Main drawback of the above
reported strategies where suitable ranges of the tuning cited works is that they fail to give robust performance for
parameters’ values are provided. process models with time delay equal to or greater than the
process time constant. In the present paper, a modified
Keywords—unstable process, modified Smith predictor, Smith predictor is proposed and new tuning rules for
robustness, direct synthesis unstable first order plus time delay (UFOPTD) process
models are reported. After analyzing the effect of tuning
I. INTRODUCTION parameter on the closed loop performance and robustness,
A tremendous attention has been devoted to the design suitable value of the design parameter is recommended. This
of PID or its variants for unstable processes with time delay paper is organized as follows: Section II describes the
in the last decade [1-5]. For process models with large time proposed modified Smith predictor. Controller design based
delay, controllers based on unity feedback configuration on the modified Smith predictor is presented in the Section
gives poor control performances. Several modified Smith III followed by the stability and robustness analysis in
predictors have therefore been reported in literature [6-10] Section IV. Next section gives guidelines for selecting
for unstable processes with large time delay. Rao and suitable value of the tuning parameter λ. Simulation studies
Chidambaram’s [7] structure has three controllers namely are done in the Section VI and lastly conclusions are given
Gc1, Gc2 and Gc3 which were designed for the following in the Section VII.
unstable first order plus time delay process model:
II. PROPOSED MODIFIED SMITH PREDICTOR
 s
Ke The proposed modified Smith predictor is shown in Fig. 1,
Gm  (1)
Ts  1 where Gm = Gmo e-θs represents the nominal model of the
actual process Gp which is to be controlled. Gc1 and Gc2 are
where K is the gain, θ is the time delay and T is the time
constant. The set point tracking and load disturbance the two controllers used for reference input tracking and
rejection controllers are tuned based on H2 optimal control load disturbance rejection. Gc is a proportional controller
specification in [8]. A proportional controller is also used in which stabilizes the delay free part of the process model
the above cited work to stabilize the delay free part of the Gmo. A low pass filter with transfer function, G f  1  f s  1 
process model. In [9], set-point tracking controller was is used to filter the predicted signal (y*) as suggested in [7]
designed based on direct synthesis approach whereas the to achieve improved robustness. The closed loop response
load disturbance rejection controller was tuned using Tan et (y) under nominal conditions (Gp = Gm) is given by
al.’s [10] method. Authors in [9] report that they achieved
improved performance and robustness as compared to Liu et
al.’s [8] method. A double two degree of freedom control
scheme with four controllers has been proposed in [11] for

978-1-5090-1795-9/16/$31.00 ©2017 IEEE 397

Authorized licensed use limited to: NATIONAL INSTITUTE OF TECHNOLOGY WARANGAL. Downloaded on July 04,2023 at 06:11:26 UTC from IEEE Xplore. Restrictions apply.
 controller and therefore the characteristic equation,
1  Gm  s  Gc 2  s   0 , takes the form:

Ts  1  KK p 2 1  Td 2 s  e
 s
0 (6)
substituting  1  0.5 s  1  0.5 s  Pade’s
 s
e
approximation) and assuming Td 2  2 , we get

T KK p 2   2  s  1  1 KK p 2  0 (7)

The constraint 1 K  K p 2  2T  K must be satisfied for

Fig. 1: Proposed modified Smith predictor
stability as per Routh Hurwitz’s criterion. Assuming
K p 2    2T  K  where  is a factor whose values are
G c 1G m o e
 s
1  G G  G G  G G G
c mo c1 mo c1 f mo
e
 s
G mo
e
 s

y r d obtained based on extensive simulation studies. In this

1   G c  G c1  G m o 1   G  G  G  1  G
c c1 mo c2
G mo e
 s
 work,  f   is used to get improved robustness [7]. The
(2)
tuning formula of Gc2 is summarized in Box II.
The symbols r, y, and d represent the set-point, controlled
variable, and the load disturbance at the plant input,  3 KK c
3
3 
2

respectively. Ti 1  3    2  2
   K K c  1
T T T 
III. CONTROLLER DESIGN 3 T   
3
T  KK c
3
T  3
2
  KK  1
c
K p1 
Substituting Gc = Kc and G m o  K  Ts  1  , the inner K 
3
T  KK c
3
T  3 
2

loop characteristic equation, 1  G c G m o  0 becomes 

3

Tf1 
 T  KK c T  3
3 3 2
Ts  1  K K c  0 (3)
Box I
Root of the above equation lies in the left half of s-plane
only if K c  1 K . In the present work, K c  4 K is  0.55;  / T < 1

considered as it results in stable response for a wide range of Td 2   / 2 ; K p2  2  T /  K  where    0.60;  / T  1
UFOPTD process models.  0.66;  / T  1


Substituting  1  1 and K in Box II

G c1  K p 1  1   G mo 
 T s
i1 
T s 1 Ts  1
f1
IV. STABILITY AND ROBUSTNESS
(2) we get
The plant model which is used for designing a controller
y Ti 1 s  1  s is only an approximation of the true dynamics of the actual
 e
r TTi1T f 1 3
 TTi1  Ti1T f 1 K c Ti1T f 1  2 K T T  system. Hence, it is necessary to analyze the robust stability
s   s 
c i1
 i 1  Ti 1  s  1 of the closed loop system for uncertainties in the process
K K p1  K K p1 K p1   K K K p1 
   p1 
and for the load disturbances. The closed loop system is
robustly stable if and only if [13]
(4)
 m  j  C  j    1      ,   (8)
Assuming the desired transfer function,
 y Ti 1 s  1   s
   e , and comparing its coefficients with where C  j    G c1 G mo e
 s
is the complementary
  s  1
3
 r  desired
1  G c  G c1 G mo
(4), we get the parameters of Gc1 as summarized in Box I.
sensitivity function and   j    G p  j    G m  j   is the
Here, λ is the desired closed loop time constant. The closed m
G m  j 
loop transfer function between the system output and input
load disturbance is bound on the process multiplicative uncertainty. If
uncertainty exists in both the process gain and time delay, λ
y 1  G G c mo
 G c 1 G mo  G c 1 G mo G f e
 s
  s
should be selected such that
 G e (5)
1  G c G mo  G c1G mo  1  G c 2 G mo e 
mo
d  s
C ( j ) 
1 (9)

 K K  1e
  s
1
It can be observed from the above equation that the
characteristic equation has two factors: (1+GcGmo+ Gc1Gmo) Furthermore, the sensitivity and complementary
and (1+ Gc2Gmo e–θs). The purpose of Gc2 is to stabilize the sensitivity function should satisfy the following constraint to
unstable process model Gmoe–θs and and improve the load ensure that the closed loop performance is robust [13].
disturbance rejection performances. Assuming Gc2 as a PD
m  j  C  j    w m  1  C  j    1 (10)

398
Authorized licensed use limited to: NATIONAL INSTITUTE OF TECHNOLOGY WARANGAL. Downloaded on July 04,2023 at 06:11:26 UTC from IEEE Xplore. Restrictions apply.
 where w m is the uncertainty bound on the sensitivity 10
1

function S  j   1  C  j  .

V. SELECTION OF λ
0
To illustrate the approach used for selecting suitable value 10

Magnitude
of λ, a process model with K = T =4 and θ =2 is considered.
The effect of λ on the robust stability and robust
performance of the system is studied by introducing a -1
perturbation of +15% in the process time delay. Only delay 10
for  = 0.3
uncertainty has been considered because the delay
for  = 0.5
compensators are more sensitive to time delay mismatches
[14]. Fig. 2 shows the magnitude of 1  e  0.3 s  1  and that of for  = 0.7
-2
10
complementary sensitivity function for various values of λ. -2 -1 0 1 2 3
10 10 10 10 10 10
It can be observed from Fig. 2 that stability margin increases
Frequency, 
as λ is increased from 0.3θ to 0.7θ. The robust performances
Fig. 3: Magnitude plots of   s  C  s   w  s  S  s  for +15%
are compared by assuming wm  1  3 [14] andthe m m

corresponding plots are shown in Fig. 3. It is observed that perturbation in the process time delay
from Figs. 2 and 3 that λ = 0.5θ satisfies both the robust
stability and robust performance conditions given by (8) and
1
(10) for the considered process model. The nominal and
perturbed system outputs for a unit step change in the set
0.8
point are shown in Figs. 4 and 5. λ = 0.5θ gives satisfactory Process variable
nominal and perturbed responses as is evident from the
0.6
above said figures. Similar trend is also observed for
balanced and delay dominated process models. Hence, λ =
0.4
0.5θ is recommended in the present work.  = 0.3
 = 0.5
VI. SIMULATION STUDY 0.2
 = 0.7
ISE (integral of squared error), IAE (integral of absolute 0
error) and TV (total variation) of the manipulated variable 0 10 20 30 40 50 60 70
Time (sec)
are calculated using following formulas for quantitative Fig. 4: Effect of λ on the system response under nominal conditions
comparison of the closed loop performances of various
methods. 1.5

(11)
IAE   e t  dt
0

(12) 1
Process variable

 e t  dt
2
ISE 
0
4
10

0.5
2  = 0.3
10
 = 0.5
 = 0.7
Magnitude

10
0 0
0 10 20 30 40 50 60 70
Time (sec)
15% perturbation in  Fig. 5: Effect of λ on system response under perturbed conditions
-2
10 C(s) for  = 0.3
C(s) for  = 0.5 If u(t) is discretized as a sequence [u1, u2, u3,…ui, …], then
C(s) for  = 0.7 TV is given by
-4
10 
-2 -1 0 1 2
TV  u  ui (13)
10 10 10 10 10 i 1
i0
Frequency, 
Fig. 2: Magnitude plots of 
1 e
 0 .3 s
1  and C(s)
Example (i) Assuming K = 4, T = 4 and θ = 4 and
applying a unit step change in the set point at t = 0 and
inverse step load disturbance of magnitude 0.05 at t = 130.

399
Authorized licensed use limited to: NATIONAL INSTITUTE OF TECHNOLOGY WARANGAL. Downloaded on July 04,2023 at 06:11:26 UTC from IEEE Xplore. Restrictions apply.
 System outputs and control signals are shown in Figs. 6 (a)
1.2
and 6 (b). Performance measures corresponding to various
tuning methods are given in Table 1. This table shows that 1
the methods reported in [7] and [9] give smaller values of

Process variable
IAE and/or ISE for set-point tracking. Furthermore, Rao and 0.8
Chid.’s [7] work results in smaller IAE and ISE for load
disturbance rejection response as compared to that of 0.6
proposed one. To check the system robustness,
0.4 Proposed
uncertainties of +5% in K and θ and –5% in T are
considered and the corresponding plots are shown in Fig. 6 Rao et al.
0.2
(c). It is observed from Fig. 6 (c) that the methods reported Rao and Chid.
in [7] and [9] fail to give robust performance. However, the 0
robustness of the proposed scheme is satisfactory. System 0 50 100 150 200 250 300
T ime (sec)
output of the proposed control scheme in presence of a Fig. 6 (c): step responses (perturbed) for example (i)
random measurement noise (zero mean and variance equal
to 0.001) at the process output is shown in Fig. 6 (d). It is 1.4
observed that the proposed method tracks the reference
1.2
input signal and rejects the load disturbance even in a noisy
environment. The high frequency noise is filtered out by a 1
low pass filter (filter time constant is assumed as 0.01 times

Process variable
0.8
the derivative time constant) used in cascade with the
derivative term of Gc2. 0.6

0.4

0.2
1 0

-0.2
0.8 0 50 100 150 200 250 300
Time (sec)
Process variable

Fig. 6 (d): step response of proposed scheme in presence of noise for

0.6 example 1

0.4 Table I: Performance measures for example (i)

Proposed

0.2 Rao et al. Tuning methods ISE IAE TV ISE IAE TV

Rao and Chid.
Set-point tracking Disturbance rejection
0
0 50 100 150 200 250 300 Proposed 6.8 8.5 0.47 5.9 12.5 0.54
T ime (sec) Rao and Chid. [7] 6.3 12.8 1.0 4.0 9.5 0.47
Fig. 6 (a): step responses (nominal) for example (i)
Rao et al. [9] 5.6 7.2 0.55 2.7 7.2 0.64

0.8 Example (ii) W. Tan [12] has considered a delay dominated

Proposed process model with K = 1, T = 1 and θ = 1.2. A unit step
0.6 Rao et al. change in the set point at t = 0 and an inverse step load
Rao and Chid. disturbance of magnitude 0.1 at t = 40 are assumed. The
variable variable

0.4 resulting process and control variables are shown in Figs 7

(a) and 7 (b). ISE, IAE and TV are summarized in Table II
0.2 which shows that proposed method gives smaller ISE and
ControlControl

IAE as compared to Tan’s method [12] for set-point

0 tracking. For load disturbance rejection response, the
methods reported in [9] and [12] outperform the proposed
-0.2 work under nominal (perfectly matched) conditions. Process
time delay is increased by 7% to check the system
-0.4 robustness and resulting responses are shown in Fig. 7 (c). It
0 50 100 150 200 250 300
Time (sec) is to be noted that Tan’s [12] method gives unstable
Time (sec)
Fig. 6 (b): control signals (nominal) for example (i) response and hence, is not included in Fig. 7 (c). Fig. 7(d)
shows the magnitude of 1 e  0 .084 s  1 and that of
complementary sensitivity function for the considered

400
Authorized licensed use limited to: NATIONAL INSTITUTE OF TECHNOLOGY WARANGAL. Downloaded on July 04,2023 at 06:11:26 UTC from IEEE Xplore. Restrictions apply.
 process model. The validity of robust performance condition
is checked by assuming wm 1  3 and the corresponding plot 2
10
is shown in Fig. 7 (e). Figs. 7 (c), (d) and (e) confirm that
the proposed controller settings give robust closed loop
performance.

Magnitude
0
10
1

0.8 -2 7% perturbation in 
10
Process variable

C(s)
0.6
Proposed -1 0 1 2 3
10 10 10 10 10
W. T an Frequency,
0.4
Rao et al. Fig. 7 (d): Magnitude plots of 1 e  0 .084 s  1 and C(s) for example (ii)

0.2 0
10

0
0 10 20 30 40 50 60 70 80
T ime (sec)
Fig. 7 (a): step responses (nominal) for example (ii)

Magnitude
-1
10
1
Proposed
W. T an
0.5
variable

Rao et al.
variable

0 -2
10 -2 -1 0 1 2 3
Control

10 10 10 10 10 10
Control

-0.5 Frequency,

Fig. 7 (e): Magnitude plot of  m  s  C  s   w m  s  Sfor

 s example (ii)
-1

Example (iii) Dynamics of an isothermal chemical reactor

-1.5 is described by the non-linear differential equation [9]:
0 20 40 60 80 100
Time (sec)
Time (sec)
Fig. 7 (b): control signals for example (ii)
dC

Q
C  C 
k 1C (14)
k 2 C  1
f 2
dt V
1.4
where Q (the inlet flow rate) = 0.03333 L/s , V = 1 L, k1 =
1.2
10 L/s, and k2 = 10 L/mol and Cf is the inlet concentration
1 [9]. An unstable steady state at C = 1.316 is obtained for
Cf = 3.288 mol/L. The transfer function obtained by
Process variable

0.8 linearizing the manipulated variable (feed concentration)

around C = 1.316 is 3 .433 103 .1s  1 . Assuming
0.6
measurement delay of 20s, we get
0.4 G m  s   3 . 433 e
 20 s
103 .1s  1 . Fig. 8 (a) shows system
Proposed outputs for a step change from the steady state value of
0.2
Rao et al 1.316 to 5 and an inverse step load disturbance of
0 magnitude 4 at t = 600 s. Fig. 8 (a) shows that Rao et al’s
0 20 40 60 80 100 method [9] fails to reject the disturbance. +10% change is
T ime (sec) assumed in the process time delay to check the system
Fig. 7 (c): step responses (perturbed) for example (ii)
robustness, and the corresponding plot is shown in Fig. 8
(b) from which we observe that the proposed settings give
satisfactory robustness.

401
Authorized licensed use limited to: NATIONAL INSTITUTE OF TECHNOLOGY WARANGAL. Downloaded on July 04,2023 at 06:11:26 UTC from IEEE Xplore. Restrictions apply.
 6 class of delay unstable processes,” ISA Trans., vol. 49, pp. 318–325,
2010.
[3] O. Arrieta, R. Vilanova, and A. Visioli, “Proportional-Integral-
5 Derivative tuning for servo/regulation control operation for unstable
and integrating processes,” Ind. Eng. Chem. Res., vol. 50, pp. 3327–
Concentration (C)

3334, 2011.
4
[4] F. Padula and A. Visioli, “Optimal tuning rules for proportional-
integral-derivative and fractional-order proportional-integral-
3 derivative controllers for integral and unstable processes,” IET
Control Theory Appl., vol. 6, pp. 776–786, 2012.
[5] B. Vanavil, K. K. Chaitanya, and A. S. Rao, “Improved PID
2 Proposed controller design for unstable time delay processes based on direct
Rao et al. synthesis and maximum sensitivity”, Int. Journal of Systems Science,
vol. 46 pp. 1349–1366, 2013.
1
0 200 400 600 800 1000 1200 [6] S. Majhi and D. P. Atherton, “Modified Smith predictor and
Time (sec) controller for processes with time delay,” IEE Proc. Control Theory
Fig. 8 (a): step responses (nominal) for example (iii) Appl., vol. 146, pp. 359–366, 1999.
5.5 [7] A. S. Rao and M. Chidambaram, “Enhanced Smith predictor for
unstable processes with time delay,” Ind. Eng. Chem. Res., vol. 44,
5
pp. 8291–8299, 2005.
4.5 [8] T. Liu, Y. Z. Cai, D. Y. Gu, and W. D. Zhang, “New modified Smith
predictor scheme for integrating and unstable processes with time
Concentration (C)

4 delay,” IEE Proc Control Theory Appl., vol. 152, pp. 238–246, 2005.
3.5 [9] A. S. Rao, V. S. R. Rao, and M. Chidambaram, “Simple analytical
design of modified Smith predictor with improved performance for
3 unstable first order plus time delay (FOPTD) processes,” Ind. Eng.
2.5 Chem. Res., vol. 46, pp. 4561–4571, 2007.
[10] W. Tan, H. J. Marquez, and T. Chen, “IMC design for unstable
2 processes with time delays,” Journal of Process Control, vol. 13, pp.
1.5
203–213, 2003.
[11] X. Lu, Y.-S. Yang, Q.-G. Wang, and W.-X. Zheng, “A double two-
1 degree-of-freedom control scheme for improved control of unstable
0 200 400 600 800 1000 1200
Time (sec) delay processes,” Journal of Process Control, vol. 15, pp. 605–614,
Fig. 8 (b): step response of the proposed method for example (iii) 2005.
under perturbed condition [12] W. Tan, “Analysis and design of a double two-degree-of-freedom
control scheme,” ISA Trans., vol. 49, pp. 311–317, 2010.
Table II: Performance measures for example (ii) [13] M. Morari, E. Zafiriou, “Robust Process Control,” Englewood Cliffs:
Tuning methods ISE IAE TV ISE IAE TV Prentice Hall, 1989.
[14] A. S. Rao, V. S. R. Rao, and M. Chidambaram, “Set point weighted
Set-point tracking Disturbance rejection modified Smith predictor for integrating and double integrating
processes with time delay,” ISA Trans., vol. 46, pp. 59–71, 2007.
Proposed 2.0 2.5 1.8 1.4 3.7 1.6
W. Tan [12] 2.2 3.2 1.5 0.7 2.4 1.1
Rao et al. [9] 1.7 2.1 2.1 0.6 1.9 1.8

VII. CONCLUSION
A new modified Smith predictor structure is proposed for
unstable plus time delay process models. Direct synthesis
method is used to design the set point tracking controller
whereas the load disturbance rejection controller is tuned such
that Routh Hurwitz’s stability criterion is satisfied. After
studying the effect of tuning parameter on the system
performance and robustness, suitable value of the tuning
parameter is recommended. This is an advantage of the
proposed method over other published works where suitable
ranges of the tuning parameters are provided. As compared to
the recently reported strategies, improved robustness is
achieved with the proposed scheme.
REFERENCES
[1] M. Shamsuzzoha and M. Lee, “Design of robust PID controllers for
unstable processes,” SICE-ICASE International Joint Conference, pp.
3324–3329, 2006 .
[2] S. C. Lee, Q.-G. Wang, and L. B. Nguyen, “Stabilizing control for a

402
Authorized licensed use limited to: NATIONAL INSTITUTE OF TECHNOLOGY WARANGAL. Downloaded on July 04,2023 at 06:11:26 UTC from IEEE Xplore. Restrictions apply.