International Conference on Control, Automation and Systems 2007

Oct. 17-20, 2007 in COEX, Seoul, Korea

Enhanced Performance for Two-Degree-Of-Freedom Control Scheme for Second Order

Unstable Processes with Time Delay

M. Shamsuzzoha1, Moonyong Lee2*

School of Chemical Engineering and Technology, Yeungnam University, Kyongsan, 712-749, Korea
1
(Tel.:+82-53-810-3241; E-mail: smzoha2002@hotmail.com)
*2
(Tel.:+82-53-810-2512; E-mail: mynlee@yu.ac.kr)

Abstract: The present study suggested a modified disturbance estimator in a recently published two-degree-of-freedom
control scheme for open-loop unstable process with time delay. A simple PID controller cascaded with lead lag filter
replaces the high order disturbance estimator for enhanced performance. A new analytical method, which has only one
tuning parameter, is developed for the design of the modified disturbance estimator. Several illustrative examples taken
from previous works are included to demonstrate the superiority of the modified disturbance estimator. The results
show the proposed disturbance estimator provides superior performance both in nominal and robust cases. The
proposed method also has several important advantages for the process engineers in industries: it covers several class of
unstable process with time delay in a unified manner; it is simple and easy to design and tune.

Keywords: PID cascaded with lead lag filter, Unstable process, Two-degree-of-freedom control, Disturbance rejection

Due to the internal instability, IMC design principle

1. INTRODUCTION
Time-delayed unstable processes are commonly
encountered in process industries and are difficult to
control due to right half plane poles. The delay in the
process industries appears due to transport phenomena,
computation of the control input, time consuming
information processing in measurement devices etc. The
design area of the control of unstable delay system has
attracted the attention of many researchers from the last
decade [1-6]. Proportional-Integrative-Derivative (PID)
controllers are definitely the most extensive option that
can be found on industrial control applications. Their
success is mainly due to their simple structure and
meaning of the corresponding three parameters. This
fact makes the PID control easier to understand by the
control engineer than most other advanced control
techniques. In addition, the performance of a PID
controller is usually satisfactory in many situations. By
reason of the widespread use of PID controllers, it
remains an interesting issue to design a simple but
efficient method for tuning the controller.
The internal model control (IMC) is a distinguishable
framework for control system design and
implementation [3-6], which has led to much effort
being made to exploit the IMC principle to design
equivalent feedback controllers for unstable processes.
It has attracted the attention of industrial users because
there is only one user-defined tuning parameter, which
is directly related to the closed-loop time constant.
cannot be directly apply for the design of the unstable
processes. For this reason, the several modified IMC
based control strategies have been developed for
controlling unstable processes with time delay and are
available in the open literatures [4-8]. Furthermore, two-
degree-of-freedom (2DOF) control methods based on
the Smith-Predictor (SP) were proposed [9-11] to
achieve a smooth nominal setpoint response without
overshoot for first order unstable processes with time
delay. Majhi and Atherton [9] have modified the
original SP for unstable time delay process and
proposed a PD controller for stabilizing the unstable
process and obtained improved performance. It also
decouples the disturbance rejection response from the
setpoint tracking. Kwak et al. [10] have proposed a
modified SP for unstable processes, which predicts the
dynamics of the actual process from the process model.
However, the method is sensitive to uncertainties in
process parameters. Zhang et al. [11] also suggested a
modified structure of a SP for unstable first order time
delay process. They have modified the structure
proposed by Kwak et al. [10] and designed a PID
controller to stabilize the unstable process and reject the
disturbance based on the desired closed-loop response.
Recently Liu et al. [8] proposed a novel two-degree-of-
freedom control scheme with an analytical design
method as shown in Fig. 1, where a process model Gm
is divided in two parts, a delay-free part Gmo and a dead
time e −θm s , i.e., Gm = Gmo e −θm s , and a controller C is
responsible for the setpoint tracking. A disturbance

978-89-950038-6-2-98560/07/$15 ⓒICROS
240
estimator F is used for load disturbances rejection and Consequently, the setpoint transfer function
a controller Gc is employed for stabilizing the setpoint CG p 1 + FGmo e −θ m s
Hr = . (2)
response. 1 + Gc Gmo 1 + FG p
The design of Gc is such that it contributes as a P or PD
In the nominal case (i.e., Gm = G p ), the setpoint
controller and makes the system an open-loop for the
setpoint tracking. It is significant to note that both the transfer function are simplified as
nominal setpoint and load disturbance response is CG p
Hr = (3)
decoupled from each other and can be optimized 1 + Gc Gmo
separately. The developed analytical design procedure
Since the dead-time term is discarded in the
for both of C and F is on the bases of the H 2 optimal characteristic equation of the nominal setpoint transfer
performance objective [3], which is equivalent to the function, it certainly contributes to achieve the smooth
integral-squared-error (ISE) performance specification. servo response. The summarized tuning formula for the
The advantage of the Liu et al.’s [8] method is clear Gc and C for several processes are listed in Table 1 and
because it provides remarkable improvement in the the detail design procedure is available in literature [8].
reference tracking and both the nominal setpoint and The past research experience demonstrates that in many
load disturbance response is decoupled from each other. process control applications, the disturbance rejection is
The tuning for the controller C and F is easy because much more important than setpoint tracking. Therefore,
the adjustable parameter can be monotonously increased we emphasizes on the design of the disturbance
or decreased to cope with the process uncertainty in estimator for improved disturbance rejection. The
practice to make the best compromise between the closed-loop transfer function for disturbances is given
nominal system performance and its robust stability. by
Although Liu et al.’s [8] method provides the Gp
y
satisfactory setpoint tracking performance, the H di = di = (4)
disturbance estimator is somewhat complicated to di 1 + FG p
design and has a limited performance while it is ydo 1
common experience in the process industries that the H do = = (5)
do 1 + FG p
disturbance rejection has been much more important
than setpoint tracking for many control applications. For the load disturbance rejection, the complementary
Their proposed high order controller for the disturbance sensitivity function of the closed-loop between the
estimator is not much popular in process industries and process input and output is given as
difficult to implement as well. f FG p
Td = = (6)
di 1 + FG p
Despite the fact that the desired complementary
sensitivity function for the ideal condition should be in
the form of Td ( s ) = e −θ s as mentioned by Liu et al. [8],
it actually occur in the asymptotic tracking form and its
constraints is as below
e -θ m s lim H do ( s ) = 0, i = 1, 2,....m (7)
s → pi

Fig. 1. Two-degree-of-freedom control structure.
The objective of the present effort is to provide the
unified and simple design method of the disturbance
estimator to obtain the robust and enhanced disturbance
rejection performance. The proposed method provide
the disturbance estimator in the form of PID with lead
lag filter, F = KC 1 + 1 + τ D s  1 + α s , which is easy to
 τ Is 1+ β s
implement in the modern control hardware to
accomplish the improved control performance.
2. CONTROLLER DESIGN PROCEDURE
As seen in Fig. 1, the process model is related as
G e −θ m s
G 'm = mo
1 + Gc Gmo
lim H do ( s ) = 0 (8)
s →0
where pi is the process model RHP pole and m is the
number of these RHP poles. From the IMC theory [3],
on the basis of the H 2 optimal performance objective
the desired closed-loop complementary sensitivity
function is given as
∑ m a si + 1
Td ( s ) = i =1 i l + m e−θ s (9)
(λ f s + 1)
where λ f is adjustable parameter and hold the trade–off
between the performance and robustness of the
disturbance estimator, and subscript l is the model
relative degree.
The value of ai is determined by the asymptotic
constraints to cancel the unstable poles and the resulting
(1) disturbance estimator obtained as

241
 Td 1 a1 a2
Fim = . (10) KC = ; τ I = a1 ; τ D = (16)
1 − Td G p (
k 4λ f + θ - a1 ) a1

Table 1 Setpoint controller design summarization, Liu et al. [8]

Processes Gc C (s)
ke −θ s
τ 1τ 2 s 2 + ( kd k − τ 1 − τ 2 ) s + 1
(τ1 s − 1)(τ 2 s − 1) kd s
k ( λc s + 1)
2

k c , if τ 1 > τ 2 τ1τ 2 s 2 + (τ1 − τ 2 ) s + kc k − 1

−θ s
ke k ( λc s + 1)
2

(τ1s − 1)(τ 2 s + 1)
k c + k d s , if τ 1 ≤ τ 2 τ 1τ 2 s 2 + ( kd k + τ 1 − τ 2 ) s + kc k − 1
k ( λc s + 1)
2

ke −θ s
=
kφ e −θ s kc + kd s τ s 2 + ( kd k − 1) s + kc k
s (τ s − 1 ) (φ s − 1 ) (τ s − 1 )
k ( λc s + 1)
2

Let’s consider for the design of the unstable process Furthermore, it is obvious from the asymptotic tracking
with two right half plane poles as a representative model. constraints in Eq. (8) that the remaining part of the
ke −θ s denominator in Eq. (15) contains the factor of process
Gp = (11) poles (τ1s − 1)(τ 2 s − 1) . Therefore, the parameter β can
(τ 1 s − 1 ) (τ 2 s − 1 )
be obtained by taking the first derivative of Eq. (17)
where k is the gain, τ 1 and τ 2 are the time constant, below
and θ the time delay and the desired closed-loop  (
a θ 2 − a2 + 2λ f θ + 6λ f 2
1 + 1 s+
) (
a2θ 2 2 + 3λ f 2θ + 4λ f 3
s2 +
) (
2λ f 3θ + λ f 4 ) 
s3 
(17)
 (
θ + 4λ f − a1 ) (
θ + 4λ f − a1 ) (
θ + 4λ f − a1 ) 
complementary sensitivity function is given as ( )
γ s2 + β s + 1 = 
(τ1s − 1)(τ 2 s − 1)


a s 2 + a1s + 1 −θ s and substituting s = 0 as

Td ( s ) = 2 e (12)
(λ f s + 1)4
β=
(a θ
1 2 − a2 + 2λ f θ + 6λ f 2 ) + (τ +τ2 )
(18)
The ideal desired disturbance estimator obtained by (θ + 4λ f − a1 ) 1

substituting Eq. (12) into Eq. (10) The filter parameter α can be easily obtained from Eq.

Fim ( s ) =
(τ1s −1)(τ2 s −1) a2s2 + a1s +1
(13)
( ) (15)
α = 0.5θ (19)
k ( λ f s +1) − e−θ s a2 s2 + a1s +1  ( )
4
2
 Since the high order γ s term has little impact on the
The ideal disturbance estimator above can be overall control performance in the control relevant
approximated to a simple PID cascaded with a lead lag frequency range, the remaining part of the fraction in Eq.
filter: The dead time e−θ s in Eq. (13) is approximated (17) can be successfully approximated to a simple first
by 1/1 Pade expansion order lead/lag filter as (1 + α s ) /(1 + β s ) .
e −θ s =
(1 − θ s 2 ) (14) The value of a1 and a2 are determined by the
(1 + θ s 2 ) asymptotic tracking constraints
which results in Fim ( s ) lim H do ( s ) = 0 and lim H do ( s ) = 0
s →1 τ 1 s →1 τ 2
(15)
which is given as
( )
(τ 1s − 1)(τ 2 s − 1)  a2 s + a1s + 1 (1 + θ s / 2)
( )
2
Fim ( s ) =  a2 s 2 + a1s + 1 
( ) s + ( a θ 2 + 3λ ) s + ( 2λ ) s 
lim 1 − −θ s 
 a1θ 2 − a2 + 2λ f θ + 6λ f 2 2 2
θ + 4λ f 3 3
θ + λf 4
( )
k θ + 4λ f − a1 s 1 +
( )
2

(
f

)
2
(
f

)
3
e =0 (20)
s →1 τ1  
 

( )
θ + 4λ f − a1 θ + 4λ f − a1 θ + 4λ f − a1 4
 
 λf s +1 
It seems from Eq. (15), the resulting disturbance
estimator has the form of the PID controller cascaded
with a high order filter. The analytical PID formula can

lim 1 −
(
a2 s 2 + a1 s + 1
e

−θ s 
=0
) (21)
s →1 τ 2  
( )
4
be obtained by rearranging Eq. (15) as below λ f s +1
 

242
The value of a1 and a2 is obtained after simplification PID controller by Liu et al.’s [8] method. The PID
and given below. controller by Liu et al. [8] shows the longest settling
4 4 time with oscillation.
 λf   λf 
τ 12 
 τ1
+ 1 eθ τ1 − τ 2 2 
  τ2 
(
+ 1 eθ τ 2 + τ 2 2 − τ12 ) (22)
Figure 3 shows the responses for a perturbation of 5% in
a1 = each process parameters simultaneously towards the
(τ1 − τ 2 ) worst case model mismatch and given as
 λ 4 
f 
(23) G p = 2.1e −0.315 s ( 2.85s − 1)( 0.95s − 1) . Since Liu et
a2 = τ12  + 1  eθ τ 1 − 1 − a1τ1
 τ 1   al.’s [8] disturbance estimator in the PID form provides
 
Several other types of second order unstable process completely unstable response, it is not shown in the
should be transformed in the standard form of Eq. (11) figure. As seen from the figure, the proposed method
for the disturbance estimator design by adjusting their gives a better performance over Liu et al.’s [8] third
sign. The integrating with first order unstable process is order controller, particularly in the disturbance rejection.
approximated with a second order unstable process as
follows: 1

ke −θ s kφ e −θ s
Gp = = (24)
s (τ s − 1) (φ s − 1)(τ s − 1)
0.8

Remarks 1: For the second order unstable process

without any zero, it is observed that the designed value 0.6

of β is too large to obtain robust performances of the y

closed–loop system when the parametric uncertainties 0.4

are large. On the basis of extensive simulation study
conducted on different unstable processes, it is observed
that using a value of 0.1β instead of β gives robust 0.2

control performances [12]. Proposed method

Liu et al. Third order
Liu et al. PID
0
3. SIMULATION RESULTS 0 4 8
Time (sec)
12 16 20

Example 1. Process with two unstable poles Fig. 2. Response for Example 1 (Nominal).

Consider the process with two unstable poles studied by

Liu et al. [8] and Tan et al. [7] Example 2. Unstable process with large time delay
2e −0.3s
Gp = (25)
( 3s − 1)(1s − 1) The following unstable process with comparatively
large time delay is studied by Liu et al. [8].
Liu et al. [8] have already explained the advantage of
their method over those of Tan et al. [7]. In this study, e−1.2 s
Gp = (27)
the setpoint tracking controller is designed as (1s − 1)( 0.5s + 1)
kd = 3, λc = 1.7θ =0.51 and the resulting The above unstable process is transformed into the form
C ( s ) = (1.5 s + s + 0.5 ) ( 0.51s + 1) according to the Liu et of Eq. (11) and the parameters for controller design are
2 2

al.’s [8] method. The disturbance estimator F by the
Liu et al.’s [8] method is provided as
(
K C = 1.7638, τ I = 1.8679, τ D = 2.3042 i.e. λ f = 1.7θ = 0.51
in the form of PID, and their third order controller is
below
32.82 s3 + 439.41s 2 + 232.64s + 129.79
F3/ 3 ( s ) =
0.56s3 + 0.8s 2 + 100s
The disturbance estimator by the proposed method is
given by
KC = 3.5671, τ I = 1.491, τ D = 1.3364, a = 0.15 and b = 0.0058 for λf = 0.35.
For the performance comparison a unit step change to
the setpoint input at t = 0 and an inverse unit step
change of load disturbance to the process input at t = 10
is added and the simulation results shown in Fig. 2. The
proposed tuning method has a faster settling time and
smaller peak compared to both the high order as well as
obtained as: k = -1, τ1 = 1, τ 2 = -0.5 , and θ = 1.2 .
Earlier several authors [7, 9] emphasis that it is
) extremely difficult to control the large time delay
process mentioned above by using a PID type controller.
As Liu et al. [8] mentioned Tan et al. [7] and Majhi and
Atherton [9] methods are unable to damp down a step
load disturbance with magnitude of 0.05. A modified
(26)
IMC-PID approach by Yang et al.’s [6] method shows
a severe oscillation. Therefore, a high order controller
numerically derived by using the RLS algorithm was
proposed by Yang et al. [6] to obtain the smooth and
fast response. Recently, Liu et al. [8] have already
explained the advantage of their method over several
other well–known tuning methods.

243
 response and their PID controller was fully unstable
response (although not shown in the Fig. 5). The
1
performance of the proposed method for both the servo
and regulatory response is stable and satisfactory which
obviously demonstrates that the proposed method has
0.8
more robust performance.

0.6 1
y

0.4 0.8

0.2 0.6

Proposed method

y
Liu et al. Third order
0 0.4
0 4 8 12 16 20
Time (sec)

Fig. 3. Response for Example 1 (Perturbed). 0.2

0 Proposed method
Liu et al. Fifth order
As mentioned earlier, the proposed study is mainly Liu et al. PID

concerned to enhance disturbance rejection performance. 0 20 40

Time (sec)
60 80 100

The setpoint tracking is identical to Liu et al.’s [8]

approach because it really provides the sufficiently
Fig. 4. Response for Example 2 (Nominal).
improved servo response over other existing methods.
Therefore, the setpoint tracking controller is designed as
kc = 2 and λc = 3.6 and C ( s ) = ( 0.5s 2 + s + 0.5) ( 3.6 s + 1)2 .
The PID parameters of the disturbance estimator by Liu 1.6

et al.’s [8] method are 1.4

KC = 1.05, τ I = 245.8901, τ D = 0.881 for λ f = 3.2 and

1.2
their fifth order controller is below for λ f = 2.4 .
36.12s 5 + 337.86 s4 + 1168.12s3 + 1331.04s 2 + 120.05s + 1 (28) 1

F5/ 5 ( s ) =
( 4 3 2
s 2.08s + 122.32s + 9.84s + 1131.54s + 100 )
y

0.8

The tuning parameters by the proposed method are

KC =1.1165, τI = 61.3412, τD = 0.4983, a = 0.6 and b = 0.0145
0.6

for λ f = 1.3 . A unit step change to the setpoint input at 0.4

t = 0 and an inverse step change of load disturbance 0.2

Proposed method

with magnitude of 0.05 to the process input at t = 50 . 0

Liu et al. Fifth order

0 20 40 60 80 100 120 140 160

The simulation results for the comparison of the Time (sec)

proposed method and Liu et al.’s [8] PID and fifth order
controller are provided in Fig. 4.
The proposed method clearly shows the significant Fig. 5. Response for Example 2 (Perturbed).
advantage of load disturbance performance while the
setpoint response is identical. The disturbance estimator
of Liu et al. [8] method in the form of PID gives the Remarks 2: λf guidelines
long peak as well as significant oscillation, while their In the proposed disturbance estimator, λf is the only
fifth order controller has smooth response.
The robust stability of the controller has been one user–defined parameter and is directly related to the
investigated for the model mismatch by inserting a performance and robustness. It is important to have
perturbation uncertainty of 5% in each parameters some λ f guidelines in order to provide both a fast and
simultaneously towards the worst direction and robust performance. Small peak and fast speed of
assuming the actual process as response of disturbance rejection are favored by
−1.26 s . Fig. 5 shows the
G p = 1.05e ( 0.95s − 1)( 0.475s + 1) selecting a small value of λ f ; however, robustness and
perturbed response of the proposed method and Liu et
stability of the system requires sufficiently large value
al.’s [8] fifth order controller. Liu et al.’s [8] fifth order
disturbance estimator gave an unstable oscillatory of λ f . On the bases of extensive simulation studies, the

244
recommended starting value of λf can be equal as
process time delay, which can give robust control
performance. If the selected λ f doesn’t satisfy the
performance and robust stability criteria, then
monotonously increase or decrease the λ f on-line until
the desirable trade–off between the close–loop nominal
performance and its robust stability is achieved.
4. CONCLUSIONS
A new design method of the disturbance estimator has
been proposed for enhanced disturbance rejection in the
two-degree-of-freedom control scheme developed by
Liu et al. [8]. The proposed disturbance estimator uses a
simple PID cascaded with a lead lag filter structure.
Furthermore, the developed tuning formula for the
proposed disturbance estimator is very simple, easy to
memorize, and also applicable for several class of
unstable process with time delay in a unique manner. It
contains only one user defined parameter λ f , which has
a straightforward relationship to the disturbance
rejection characteristics.
The proposed method has been compared with the latest
published paper [8] and shown a clear advantage both
in the nominal and robust performance in disturbance
rejection. For the model mismatch, in particular, the
proposed method provides the clear advantage in the
servo performance as well as in the regulatory
performance.
[7] W. Tan, H.J Marquez, T. Chen, IMC design for
unstable processes with time delays, J. Process
Control 13 (2003) 203–213.
[8] T. Liu, W. Zhang, D. Gu, Analytical design of
two–degree–of–freedom control scheme for open–
loop unstable process with time delay, J. Process
Control 15 (2005) 559–572.
[9] S. Majhi, D.P Atherton, Obtaining controller
parameters for a new Smith predictor using
autotuning, Automatica 36 (2000) 1651–1658.
[10] H.J Kwak, S.W Sung, I.B Lee, J.Y Park, Modified
Smith predictor with a new structure for unstable
processes, Ind. Eng. Chem. Res. 38 (1999) 405–
411.
[11] W.D Zhang, D. Gu, W. Wang, X. Xu, Quantitative
performance design of a modified Smith predictor
for unstable processes with time delay, Ind. Eng.
Chem. Res. 43 (2004) 56–62.
[12] M. Shamsuzzoha, J. Jeon, M. Lee, Improved
analytical PID controller design for the second
order unstable process with time delay, 17th
European Symposium on Computer Aided Process
Engineering (ESCAPE–17), Bucharest, Romania,
27–30 May 2007 (Editors Valentin Plesu and Paul
Serban Agachi), 901–906.

Acknowledgement

This research was supported by Yeungnam University

research grants in 2007.

Reference

[1] J.H. Park, S.W. Sung, I.B. Lee, An enhanced PID

control strategy for unstable processes, Automatica
34 (1998) 751-756.
[2] Y.G. Wang, W.J. Cai, Advanced proportional–
integral–derivative tuning for integrating and
unstable processes with gain and phase margin
specifications, Ind. Eng. Chem. Res. 41 (2002)
2910–2914.
[3] M. Morari, E. Zafiriou, Robust Process Control,
Prentice Hall Englewood Cliffs, New York, 1989.
[4] M. Shamsuzzoha, M. Lee, IMC–PID controller
design for improved disturbance rejection of time–
delayed processes, Ind. Eng. Chem. Res. 46 (2007)
2077–2091.
[5] Y. Lee, J. Lee, S. Park, PID controller tuning for
integrating and unstable processes with time delay,
Chem. Eng. Sci. 55 (2000) 3481–3493.
[6] X.P. Yang, Q.G. Wang, C.C Hang, C. Lin, IMC–
based control system design for unstable processes,
Ind. Eng. Chem. Res. 41 (2002) 4288–4294.

245