You are on page 1of 5

Journal of Process Control 20 (2010) 235–239

Contents lists available at ScienceDirect

Journal of Process Control


journal homepage: www.elsevier.com/locate/jprocont

Short communication

Stabilization of all-pole unstable delay processes by simple controllers


See Chek Lee, Qing-Guo Wang *, Cheng Xiang
Department of Electrical and Computer Engineering, National University of Singapore, 119260, Singapore

a r t i c l e i n f o a b s t r a c t

Article history: The stabilization of a class of all-pole unstable delay processes of arbitrary order with single unstable pole
Received 12 May 2009 by means of simple controllers is investigated in details. Complete stabilizability conditions are estab-
Accepted 15 May 2009 lished and the computational methods for determining stabilizing controller parameters presented. They
provide theoretical understanding of such a stabilization problem and can also serve as practical guide-
lines for actual controller design.
Keywords: Ó 2009 Elsevier Ltd. All rights reserved.
Unstable time-delay processes
Stabilization
PID controller
Stabilizer parameterization

1. Introduction whole stability domain is drawn in two-dimensional plane by


sweeping the remaining parameter(s). However, this result only
Unstable processes coupled with input time-delay have been provides sufficient condition regarding the size of the time-delay
gaining much attention from the control community, due to its for stabilization of first order unstable processes. Xiang et al. [7]
complexity and limitations in achieving stabilization or perfor- exploited frequency response technique to study stabilizability by
mance. For an open-loop unstable process, a sufficiently large of P/PI/PD/PID controller and obtained complete stabilizability results
feedback gain must be used to stabilize the process before address- in term of the upper limit of time-delay for a class of second-order
ing performance and robustness, and yet the feedback loop may processes. The stabilizability condition for higher-order unstable
become unstable again if the gain is too large. Design of single loop delay processes still remains open.
PID controllers, the most commonly used controllers in industry, In this paper, we investigate stabilization for a class of unstable
for such processes have been reported [6,2]. However, these works delay processes described by GðsÞ ¼ m QKn eLs ; m P
s ðks1Þ ðT k sþ1Þ
k¼1
do not provide a clear scenario on what kind of process could be
stabilized by PID controllers. 0; n P 0; T k > 0; L > 0, by PID controller or its special cases. Note
that when n ¼ 0, the process GðsÞ contains no stable lags. The com-
Huang and Chen [3] used root locus to study stabilizability prob-
lem of unstable delay processes by simple controllers and showed plete stabilizability conditions are established and given in Table 1.
And the procedures for computing the parameters of stabilizing
that the normalized time-delay should be less than 1 for P/PI con-
troller to stabilize the first order delay unstable process, while it controllers are also presented and illustrated with examples. The
rest of the paper is organized as follows. Some preliminaries are
should be less than 2 for PD controller to stabilize the first order de-
lay unstable process. Silva et al. [5] investigated the complete set of presented in Section 2. The stabilization by P/PI controller is ad-
stabilizing PID controllers based on the Hermite–Biehler theorem dressed in Section 3 while the case of PD/PID controller is in Sec-
for quasi-polynomials, which involves finding the zeros of a tran- tion 4. Section 5 concludes the paper.
scendental equation to determine the range of stabilizing gains.
But this approach is mathematically involved, which does not pro-
2. Problem formulation and preliminaries
vide an explicit characterization of the boundary of the stabilizing
PID parameter region, and the maximal stabilizable time-delay
To formulate the stabilization problem with fewest possible
for some typical yet simple processes still remains obscure. Hwang
parameters, some normalization is adopted throughout the paper.
and Hwang [4] applied D-partition method to characterize the sta-
This is best illustrated by an example. Let the actual process and
bility domain in the space of system and controller parameters. The
controller be GðsÞ ¼ m QKn eLs and CðsÞ ¼ K P 1 þ K D þ KsI Þ,
stability boundary is reduced to a transcendental equation, and the s ðks1Þ
k¼1
ðT k sþ1Þ

respectively, in unity output feedback configuration depicted in


* Corresponding author. Tel.: +65 6516 2282. Fig. 1. One can scale down the time-delay and all time constants
E-mail address: elewqg@nus.edu.sg (Q.-G. Wang). by k, and absorb the process gain K into the controller so that

0959-1524/$ - see front matter Ó 2009 Elsevier Ltd. All rights reserved.
doi:10.1016/j.jprocont.2009.05.005
236 S.C. Lee et al. / Journal of Process Control 20 (2010) 235–239

Table 1 Proof 1. Suppose that a process GðsÞ ¼ NðsÞ


DðsÞ
eLs is stabilizable by a
Stabilizability results. PD controller, then there exist some K P and K D such that the char-
Process P/PI PD/PID acteristic equation, 1 þ K P ð1 þ K D sÞ NðsÞ
DðsÞ
eLs ¼ 0, has stable roots
Pn
Qn1 e Ls
None L < 1  k¼1 T k only. The closed-loop characteristic equation with PID is
sðs1Þ ðT k sþ1Þ
k¼1
qffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi  
Pn P P 1 þ K P 1 þ K D s þ KsI NðsÞ eLs ¼ 0, which can be rewritten as
Qn 1
eLs
L<1 k¼1 T k L < 1 þ nk¼1 T 2k þ 1  nk¼1 T k DðsÞ
ðs1Þ k¼1
ðT k sþ1Þ
 
KI
DðsÞ þ K P ð1 þ K D sÞNðsÞeLs þ K P NðsÞeLs ¼ 0;
L ¼ L=k, T k ¼ T k =k; K D ¼ K D =k; K I ¼ K I k; K P ¼ KK P km . It follows that s
 
the open-loop transfer function is expressed as KI K P NðsÞeLs
" # 1þ ¼ 0: ð3Þ
s DðsÞ þ K P ð1 þ K D sÞNðsÞe Ls
K Ls
GðsÞCðsÞ ¼ Q e
sm ðks  1Þ nk¼1 ðT k s þ 1Þ This can be viewed as the closed-loop characteristic equation with
2 3
" !# K I =s controlling the plant:
KI s¼ks 6 1 7
 KP 1 þ KDs þ ;)6 4 eLs 7
5 K P NðsÞeLs
s Qn
; ð4Þ
sm ðs  1Þ ðT k s þ 1Þ DðsÞ þ K P ð1 þ K D sÞNðsÞeLs
k¼1
  
KI which has a non-zero static gain due to Nð0Þ – 0 and K P – 0 for sta-
 KP 1 þ KDs þ ;
s bilizing, thus asserts that no cancellation of the pole of K I =s at origin
with the normalized process of interest: with (4). The plant (4) is stable as its denominator is the same as the
closed-loop characteristic quasi-polynomial for the process and its
1 stabilzing PD controller. It follows from the Root–Locus technique
GðsÞ ¼ eLs ; m P 0; n P 0; T k > 0; L > 0;
Q
n
that there is always a non-zero positive K I such that the closed-loop
sm ðs  1Þ ðT k s þ 1Þ
k¼1 characteristic equation in (3) is stable, that is, there also exists a PID
ð1Þ controller stabilizing GðsÞ. The stabilizability proof from P to PI is
  the special case of the above by setting K D ¼ 0. This completes
and the normalized controller, CðsÞ ¼ K P 1 þ K D s þ KsI , which can be
 the proof. h
in one of the following forms: C 1 ðsÞ ¼ K P ; C 2 ðsÞ ¼ K P 1 þ KsI ;
 
C 3 ðsÞ ¼ K P ð1 þ K D sÞ; and C 4 ðsÞ ¼ K P 1 þ K D s þ KsI .
Remark 1. In our subsequent discussion, sufficient P-stabilizabilty
The corresponding open-loop transfer function, Q i ðsÞ, can be ex-
condition, by Lemma 2, automatically becomes sufficient PI-stabi-
pressed in the form of
lizabilty condition. The converse of Lemma 2, that is, PI-stabiliz-
NðsÞ Ls ability implies P-stabilizability, is not true in general [8]. Hence,
Q i ðsÞ ¼ C i ðsÞGðsÞ ¼ K v e ; i 2 f1; 2; 3; 4g; ð2Þ
s DðsÞ necessary stabilizability condition by P and PI controllers, on the
other hand, is to be proven independently. The same is true for
where K is the gain, v a non-negative integer representing type of
PD and PID controllers.
the loop, NðsÞ and DðsÞ both polynomials of s with Nð0Þ ¼ Dð0Þ ¼ 1.
Application of Nyquist stability criterion to open-loop Q i ðsÞ in
(2), leads to Lemma 1 [7]. The Nyquist contour consists of the en- 3. P/PI controller
tire jx axis from x ¼ 1 to þ1 and a semicular path of infinite
radius in the right half s plane if v ¼ 0. But for v > 0, the contour The stabilization by P or PI controller will be discussed first, fol-
near the origin is modified to be a semicular path with infinitesi- lowed by PD or PID controller in the next section. Due to the sym-
mal radius  (where   1) to the right half plane [1]. metry property of the Nyquist curve, subsequent analysis focuses
Lemma 1. Given the open-loop transfer function Q i ðsÞ in (2) with P þ on the positive frequency, x > 0, unless otherwise indicated.
unstable poles, the closed-loop system is stable Theorem 1. The all-pole unstable process, GðsÞ ¼ Qn1 eLs ;
ðs1Þ k¼1
ðT k sþ1Þ

(1) if and only if the Nyquist plot of Q i ðsÞ encircles the critical n P 0; T k > 0; L > 0, is stabilizable by P or PI controller if and only if
P
point (1, 0), Pþ times anticlockwise. L < 1  nk¼1 T k .
(2) only if limx!1 jQ i ðjxÞj < 1.
(3) only if K < 0 when Pþ ¼ 1 and v ¼ 1; 2. Proof 2. Sufficiency: For P controller, C 1 ðsÞ ¼ K P , the open-loop fre-
d v Ls mþ1 quency response is
(4) only if the polynomial, HðsÞ ¼ eLs ds mþ1 ½s DðsÞe , has all its

roots in the open left half plane (LHP), where m is the degree KP
Q 1 ðjxÞ ¼ C 1 ðjxÞGðjxÞ ¼ ejLx ;
of NðsÞ. Q
n
ðjx  1Þ ðjT k x þ 1Þ
k¼1
Lemma 2. If a process GðsÞ ¼ NðsÞ DðsÞ
eLs ; Nð0Þ – 0, is stabilizable by a PD
controller, so is it by a PID controller. Similiarly, stabilizablity by P con- þ
with P ¼ 1 and v ¼ 0. It follows from Lemma 1 that K ¼ K P < 1,
troller implies stabilizability by PI controller. or K P > 1. Then the loop has its magnitude as

1
MQ 1 ðxÞ ¼ K P sffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi ;
Qn

ð1 þ x2 Þ 1 þ T 2k x2
k¼1

which decreases monotonically from K P to 0. The phase is


X
n
UQ 1 ðxÞ ¼ Lx  p þ arctanðxÞ  arctanðT k xÞ; ð5Þ
Fig. 1. Unity output feedback. k¼1
S.C. Lee et al. / Journal of Process Control 20 (2010) 235–239 237

with its derivative as Necessity: By Lemma 1, C 1 ðsÞ ¼ K P can achieve stabilization only
d 1 X
n
Tk if HðsÞ ¼ cnþ1 snþ1 þ cn sn þ    þ c1 s þ c0 , has stable roots only,
UQ 1 ðxÞ ¼ L þ  ð6Þ Q
dx 1 þ x2 k¼1 1 þ T 2k x2
: where the highest-degree coefficient is cnþ1 ¼ L nk¼1 T k , and the
Pn
constant term is c0 ¼ 1  L  k¼1 T k .
It follows immediately that UQ 1 ð0Þ ¼ p and ddx UQ 1 ðxÞ x¼0 ¼ Note cnþ1 > 0 since L and T k are positive. Thus c0 > 0 is
P P Pn
L þ 1  nk¼1 T k > 0, using the assumed condition L < 1  nk¼1 T k , necessary for stability of HðsÞ, and
gives L < 1  k¼1 T k . Similarly,
KI
so that there is x > 0 such that UQ 1 ðx Þ > p. In fact, one sees
  if PI controller, C 2 ðsÞ ¼ K P 1 þ s , is used, one obtains HðsÞ ¼
from (5) that max ðUQ 1 ðxÞÞ never exceeds p=2. Hence, the phase cnþ2 snþ2 þ cnþ1 snþ1Q
þ    þ c1 s þ c0 , with the highest-degree coef-
will first increase from p, and later decrease back to it with fre- ficient, cnþ2 ¼ L2 nk¼1 T k , and the constant term, c0 ¼ 2ð1  L
Pn Pn
quency, and there is one intersection with the negative real axis k¼1 T k Þ, respectively. Thus, L < 1  k¼1 T k is also necessary
with UQ 1 ðxc Þ ¼ p for some positive frequency xc . In order for stabilizability condition by PI controller. h
the anticlockwise encirclement of the critical point to occur, this
In the following, we proceed to determine the set of stabilizing
intersection should lie between 1 and 0, that is, M Q 1 ðxc Þ < P
P and PI controllers for given GðsÞ under L < 1  nk¼1 T k . For P con-
1; UQ 1 ðxc Þ ¼ p, or equivalently
vffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi troller, the stabilizing gain K P should be within the range of (7),
u n
with the phase crossover frequency xc satisfying
u Y
1 < K P < t 1 þ x2c 1 þ T 2k x2c ; ð7Þ X
n
k¼1 Lxc þ arctan ðxc Þ  k ¼ 1 arctan ðT k xc Þ ¼ 0: ð8Þ
where the left inequality, K P > 1, is from the early discussion. And For stabilizing PI controller, we need K P K I > 0 by Lemma 1. By the
for x > xc ; MQ 1 ðxÞ is always less than 1 so that there is no encircle- Root–Locus argument from Lemma 2, we assert that both K P and
ment around the critical point thereafter. Consequently, there is ex- K I have to be positive. One chooses K I such that
P
actly one anticlockwise encirclement when L < 1  nk¼1 T k and (7) UQ 2 ðx Þ > p; for some x > 0: ð9Þ
are all true, which asserts the stabilization by Lemma 1. The PI case
follows from Lemma 2. Then the range of K P is given by
vffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi
vffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi

ffi u Q n

u u
Nyquist Diagram u 1 þ x2 Qn 1 þ T 2 x2 u 1 þ x2c2 1 þ T 2k x2c2
0.4 u c1 k¼1 k c1 u
u
t
2 < KP < u
t
k¼1

2 ;
KI
1 þ xc1 1 þ xKc2I
0.3
ð10Þ
0.2
with xc1 < xc2 being the first two phase crossover frequencies
0.1 solved from
Imaginary Axis

  X n
0 KI
Lx þ arctanðxÞ  arctan  arctanðT k xÞ ¼ 0: ð11Þ
x k¼1
−0.1
For illustration, consider a third order unstable dead-time process
−0.2 [2]:
1
−0.3 GðsÞ ¼ eLs : ð12Þ
ð5s  1Þð2s þ 1Þð0:5s þ 1Þ
−0.4 Let L ¼ 0:5. By the normalization technique in Section 2, one obtains
−2 −1.5 −1 −0.5 0 0.5 1
the normalized process, GðsÞ ¼ ðs1Þð0:4sþ1Þð0:1sþ1Þ e0:1s . We have the
Real Axis
(a) Pcontroller. normalized time-delay L as 0.1, and normalized time constants,
T 1 ¼ 0:4 and T 2 ¼ 0:1. Since L ¼ 0:1 < 1  T 1  T 2 ¼ 0:5, it follows
from Theorem 1 that the process is stabilizable by P/PI controller.
Nyquist Diagram One calculates xc ¼ 2:1435 from (8), and 1 < K P < 3:1864 from
0.5
(7). Let K P ¼ 2, the P controller is C 1 ðsÞ ¼ 2, which leads to
0.4 C 1 ðsÞ ¼ 2. The Nyquist plot of C 1 G is given in Fig. 2, which indicates
0.3 a stable closed-loop.
To design PI controller, let K I ¼ 0:05, where (9) is met. From
0.2 (11), we obtain xc1 ¼ 0:371, and xc2 ¼ 2:044. Then K P is bounded
Imaginary Axis

0.1 by (1.070, 2.998), from (10). Let K P ¼ 2, the PI controller is


C 2 ðsÞ ¼ 2ð1 þ 0:05=sÞ, which leads to C 2 ðsÞ ¼ 2ð1 þ 0:01=sÞ. The Ny-
0 quist plot of C 2 G is given in Fig. 2, which indicates a stable closed-
−0.1 loop.

−0.2 Corollary 1. The all-pole process, GðsÞ ¼ Q1n eLs ;


sm ðs1Þ k¼1
ðT k sþ1Þ

−0.3 m P 1; n P 0; T k > 0; L > 0, is not stabilizable by P or PI controller.

−0.4
Proof 3. Corollary 1 follows from applying Theorem 1 to the plant,
−0.5 Qm e
−2.5 −2 −1.5 −1 −0.5 0 0.5 e l¼1 T l Ls
GðsÞ ¼ hQ i Q e ;
Real Axis m e þ 1Þ ðs  1Þ nk¼1 ðT k s þ 1Þ
l¼1 ð T l s
(b) PIcontroller.
e l ! 1; l 2 f1; 2; . . . ; mg, giving L < m1.
and let T h
Fig. 2. Stabilization of GðsÞ in (12), L ¼ 0:5:
238 S.C. Lee et al. / Journal of Process Control 20 (2010) 235–239

4. PD/PID controller where the left inequality, K P > 1, is from the early discussion. And
for x > xc ; MQ 3 ðxÞ is always less than 1 due to (13), so that there
is no encirclement around the critical point thereafter. Conse-
Theorem 2. The all-pole unstable process, GðsÞ ¼ Qn1 eLs ;
ðs1Þ k¼1
ðT k sþ1Þ quently, there is exactly one anticlockwise encirclement when
n P 0; T k > 0; L > 0; is stabilizable by PD or PID controller if and (15)–(17) are true, which asserts the stabilization by Lemma 1.
qffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi
P ffi P The PID case follows from Lemma 2.
only if L < 1 þ nk¼1 T 2k þ 1  nk¼1 T k .
Necessity: By Lemma 1, PD controller, C 3 ðsÞ ¼ K P ð1 þ K D sÞ, can
achieve stabilization only if HðsÞ ¼ cnþ1 snþ1 þ cn sn þ    þ c1 s þ c0 ,
Proof 4. Sufficiency; For PD controller, C 3 ðsÞ ¼ K P ð1 þ K D sÞ, the
has stable roots only, where the highest-degree coefficient
open-loop frequency response is Q
is cnþ1 ¼ L2 nk¼1 T k , and the constant term is c0 ¼
1 þ jK D x h Pn 2 P i
Q 3 ðjxÞ ¼ C 3 ðjxÞGðjxÞ ¼ K P Q ejLx ;  L þ k¼1 T k  1  nk¼1 T 2k  1 .
ðjx  1Þ nk¼1 ðjT k x þ 1Þ
Note cnþ1 > 0 since L and T k are positive. Thus c0 > 0 is
with P þ ¼ 1 and v ¼ 0. It follows from Lemma 1 that K ¼ K P < 1, necessary for stability of HðsÞ, and gives L<
qffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi
Pn ffi Pn
or K P > 1. Then the loop has magnitude as 2
1 þ k¼1 T k þ 1  k¼1 T k . Similarly, if PID controller, C 4 ðsÞ ¼ K P
vffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi

u 1 þ K D s þ KsI , is used, one obtains HðsÞ ¼ cnþ2 snþ2 þ cnþ1


u 1 þ K 2D x2
M Q 3 ðxÞ ¼ K P t Q
:
ð1 þ x2 Þ nk¼1 1 þ T 2k x2 snþ1 þ    þ c1 s þ c0 , with the highest-degree coefficient, cnþ2 ¼
Q h P 2
L3 nk¼1 T k , and the constant term, c0 ¼ 3 L þ nk¼1 T k  1 
It follows from qffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi

Pn P P

2
k¼1 T k  1, respectively. Thus, L < 1 þ nk¼1 T 2k þ 1  nk¼1 T k is
! 
2
d M Q 3 ðxÞ 2x 1 þ K 2D x2
1 also necessary stabilizability condition by PID controller. h
¼ Q
K 2D 1 þ K 2D x2
dx K 2P n 2
ð1 þ x2 Þ k¼1 1 þ T k x2 In the following, we proceed to determine the set of stabilizing
# PD and PID controllers for given GðsÞ under L <
qffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi
 1 X n
1 P P
þ 1 þ x2 þ T 2k 1 þ T 2k x2 ; 1 þ nk¼1 T 2k  nk¼1 T k þ 1. The parameters for stabilizing PD con-
k¼1 troller, K D and K P , can be found from (16) and (17), respectively,
P

that if 1 þ nk¼1 T 2k  K 2D > 0, or equivalently with the phase crossover frequency xc satisfying
qffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi
Xn 2 X
n
KD < 1þ k¼1 k
T ; ð13Þ Lxc þ arctanðxc Þ þ arctanðK D xc Þ  arctanðT k xc Þ ¼ 0: ð18Þ
k¼1
then the magnitude, M Q 3 ðxÞ, decreases monotonically from K P . The
phase is For stabilizing PID controller, we need K P K I > 0 by Lemma 1. From
Lemma 2 together with the results obtained from PD controller, we
UQ 3 ðxÞ ¼ Lx  p þ arctanðxÞ þ arctanðK D xÞ assert that all K P , K D and K I have to be positive. One first chooses K D
X
n from (16). Then K I should be such that
 arctanðT k xÞ; ð14Þ UQ 4 ðx Þ > p; for some x > 0: ð19Þ
k¼1

with its derivative as Then the range of K P is given by


vffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi

vffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi

u u
d 1 KD X
n
Tk u 1 þ x2 Qn 1 þ T 2 x2 u 1 þ x 2  Qn 1 þ T 2 x 2
UQ ðxÞ ¼ L þ þ  : u c1 k¼1 k c1 u c2 k¼1 k c2
dx 3 1 þ x2 1 þ K 2D x2 k¼1 1 þ T 2k x2 u
2 < KP < u
2 ;
t t
1 þ K D xc1  xKc1I 1 þ K D xc2  xKc2I
Under the assumed condition,
vffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi ð20Þ
u
u X n X
n
L < t1 þ T 2k þ 1  Tk; ð15Þ with xc1 < xc2 being the first two phase crossover frequencies
k¼1 k¼1 solved from
  X
k¼1
it is possible to choose K D within the following non-empty set: KI
vffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi Lx þ arctan K D x  þ arctanðxÞ  arctanðT k xÞ ¼ 0:
u x n
X
n u X n
L1þ T k < K D < t1 þ T 2k : ð16Þ ð21Þ
k¼1 k¼1
Consider again the process (12), but with L ¼ 5:5. One obtains the
With K D in the range given by (16), it follows that UQ 3 ð0Þ ¼ p and 1
normalized process as GðsÞ ¼ ðs1Þð0:4sþ1Þð0:1sþ1Þ e1:1s . Since L ¼
P qffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi
d
U ðxÞjx¼0 ¼ L þ 1 þ K D  nk¼1 T k > 0, using the assumed con-
dx Q 3
1:1 > 1  T 1  T 2 ¼ 0:5, and L ¼ 1:1 < 1 þ T 21 þ T 22  T 1  T 2 þ 1 ¼
dition (15), so that there exists x > 0 such that UQ 3 ðx Þ > p.
1:582, it follows from Theorems 2 and 1, respectively, that the pro-
In fact, one sees from (14) that maxðUQ 3 ðxÞÞ never exceeds 0.
cess is stabilizable by PD/PID controller, but is not stabilizable by P/
Hence, the phase will first increase from p, and later decrease
PI controller. According to (16), there exists a stabilizing gain of K D
back to it with frequency, and there is one intersection with the
in the range (0.6, 1.082). Let K D ¼ 0:75. One calculates xc ¼ 0:6329
negative real axis with UQ 3 ðxc Þ ¼ p for some positive frequency
from (18), and 1 < K P < 1:1052 from (17). Take K P ¼ 1:05. The
xc . In order for the anticlockwise encirclement of the critical point
resultant PD controller is C 3 ðsÞ ¼ 1:05ð1 þ 0:75sÞ, which leads to
to occur, this intersection should lie between 1 and 0, that is,
C 3 ðsÞ ¼ 1:05ð1 þ 3:75sÞ. The Nyquist plot of C 3 G is given in Fig. 3a,
M Q 3 ðxc Þ < 1; UQ 3 ðxc Þ ¼ p, or equivalently
vffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi which indicates a stable closed-loop.
u  Qn

u 1 þ x2c 1 þ T 2 2
x To design a PID controller, choose K D in the range of (16), say
t k¼1 k c
K D ¼ 0:85. One finds that for K I ¼ 0:03 (19) holds. From (21), we
1 < KP < ; ð17Þ
1 þ K 2D x2c obtain xc1 ¼ 0:3959, and xc2 ¼ 0:7107. Then K P is bounded by
S.C. Lee et al. / Journal of Process Control 20 (2010) 235–239 239

 "b
!#
Nyquist Diagram b C b 2 ðsÞ ¼ 1 bP 1 þ KI
0.5 GðsÞ Q eLs
K
ðs  1Þ nk¼1 ðT k s þ 1Þ s
0.4  " !#
1 bI 1 þ 1 s
bPK
¼ Qn eLs K
0.3 sðs  1Þ k¼1 ðT k s þ 1Þ bI
K
0.2 ¼ GðsÞC 3 ðsÞ:
Imaginary Axis

0.1 One sees that there exists a PD controller, C 3 ðsÞ, to stabilize


GðsÞ ¼ Qn1 eLs , under the sufficient condition of L < 1
0 sðs1Þ ðT sþ1Þ
k¼1 k
Pn
−0.1 k¼1 T k . PThe PID case follows from Lemma 2. The necessity of
L < 1  nk¼1 T k , can be verified by Lemma 1, similiarly as in Theo-
−0.2 rem 1 or 2.
−0.3 For m P 2, the proof follows from applying the above result to
the plant,
−0.4 Qm1 e
e l¼1 T lh i eLs ;
−0.5 GðsÞ ¼ Qn Qm1
−1.2 −1 −0.8 −0.6 −0.4 −0.2 0 0.2 0.4 0.6 sðs  1Þ k¼1 ðT k s þ 1Þ e
l¼1 ð T l s þ 1Þ
Real Axis
(a) PD controller. e l ! 1; l 2 f1; 2; . . . ; m  1g, giving L < ðm  1Þ1.
and let T h

Nyquist Diagram Remark 2. In practice, PD/PID controller is not implemented in the


0.5 form discussed. Instead, it is always cascaded with an additional
1
0.4 first order filter, f ¼ T sþ1 . The stabilizability results can still apply
f
by absorbing the filter into the plant transfer function. Then the
0.3 stabilizability conditions for GðsÞ in (1) become, L < 1  T f 
Pn qffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi
P P
0.2 1 þ T 2f þ nk¼1 T 2k þ 1  T f  nk¼1 T k , respectively,
k¼1 T k , and L <
Imaginary Axis

0.1 for PD and PID controllers.

0
5. Conclusion
−0.1

−0.2 In this paper, the stabilization of a class of unstable all-pole de-


lay processes of arbitrary order with single unstable pole is inves-
−0.3 tigated. The stabilizability conditions by the four simple controllers
−0.4 are established explicitly in term of the maximum allowable time-
delay and they are both necessary and sufficient. Besides, we also
−0.5 provide the procedures for determining the parameter ranges of
−1.2 −1 −0.8 −0.6 −0.4 −0.2 0 0.2 0.4 0.6
Real Axis the stabilizing controllers and have demonstrated them through
(b) PID controller. examples. Both stabilizability conditions and controller design pro-
cedures are easy to apply in practical process control.
Fig. 3. Stabilization of GðsÞ in (12), L ¼ 5:5.

References
(1.034, 1.117), from (20). Take K P ¼ 1:08. The PID controller is
C 4 ðsÞ ¼ 1:08ð1 þ 0:85s þ 0:03=sÞ, which leads to C 4 ðsÞ ¼ 1:08ð1þ [1] T. Dougherty, Systems and Control: An Introduction to Linear, Sampled and
Non-linear Systems, World Scientific, 1995.
4:25s þ 0:006=sÞ. The Nyquist plot of C 4 G is given in Fig. 3b, which [2] C.T. Huang, Y.S. Lin, Tuning PID controller for open-loop unstable process with
indicates a stable closed-loop. time-delay, Chem. Eng. Commun. 133 (1995) 11–30.
[3] H.P. Huang, C.C. Chen, On stabilizing a time-delayed unstable process, IEE Proc.
Corollary 2. For the all-pole process, GðsÞ ¼ m Q1n eLs ; n Contr. Theory Appl. 28 (4) (1997) 289–299.
s ðs1Þ k¼1 ðT k sþ1Þ
P 0; T k > 0; L > 0, [4] C. Hwang, J.H. Hwang, Stabilisation of first order plus dead-time unstable
processes using PID controllers, IEE Proc. Contr. Theory Appl. 151 (1) (2004) 89–
94.
(i) in case of m ¼ 1, it is stabilizable by PD or PID controller if [5] G.J. Silva, A. Datta, S.P. Bhattacharyya, PID Controllers for Time-Delay Systems,
P Birkhauser, Boston, 2004.
and only if L < 1  nk¼1 T k . [6] R.Padma Sree, M. Chidambaram, Control of Unstable Systems, Narosa
(ii) in case of m P 2, it is not stabilizable by PD or PID controller. Publishing House, Chennai, India, 2006.
[7] C. Xiang, Q.G. Wang, X. Lu, L.A. Nguyen, T.H. Lee, Stabilization of second-order
unstable delay processes by simple controllers, J. Proc. Contr. 17 (2007) 675–
Proof 5. We discuss for m ¼ 1 first. From Theorem 1, there is PI 682.
  [8] Z. Zhang, Q.G. Wang, Y. Zhang, Relationship on stabilizability of LTI systems by P
controller, C bP 1 þ b
b 2 ðsÞ ¼ K KI b
to stabilize GðsÞ ¼ Qn1 and PI controllers, Can. J. Chem. Eng. 85 (3) (2007) 374–377.
s ðs1Þ k¼1 ðT k sþ1Þ
P b C b 2 ðsÞ as
eLs if L < 1  nk¼1 T k . Rewrite GðsÞ

You might also like