Energy Conversion and Management 47 (2006) 30453058

www.elsevier.com/locate/enconman

Computation of stabilizing PI and PID controllers

using the stability boundary locus
Nusret Tan
a

a,*

, Ibrahim Kaya a, Celaleddin Yeroglu a, Derek P. Atherton

Inonu University, Engineering Faculty, Department of Electrical and Electronics Engineering, 44280 Malatya, Turkey
b
University of Sussex, Department of Engineering and Design, Falmer, Brighton BN1 9QT, UK
Received 19 August 2005; accepted 6 March 2006
Available online 19 April 2006

Abstract
In this paper, a new method for the calculation of all stabilizing PI controllers is given. The proposed method is based
on plotting the stability boundary locus in the (kp, ki)-plane and then computing the stabilizing values of the parameters of
a PI controller. The technique presented does not require sweeping over the parameters and also does not need linear programming to solve a set of inequalities. Thus it oers several important advantages over existing results obtained in this
direction. Beyond stabilization, the method is used to shift all poles to a shifted half plane that guarantees a specied settling time of response. Computation of stabilizing PI controllers which achieve user specied gain and phase margins is
studied. It is shown via an example that the stabilizing region in the (kp, ki)-plane is not always a convex set. The proposed
method is also used to design PID controllers. The limiting values of a PID controller which stabilize a given system are
obtained in the (kp, ki)-plane, (kp, kd)-plane and (ki, kd)-plane. Furthermore, the proposed method is used to compute all the
parameters of a PI controller which stabilize a control system with an interval plant family. Examples are given to show the
benets of the method presented.
 2006 Elsevier Ltd. All rights reserved.
Keywords: Stabilization; PI control; PID control; Gain and phase margins; Relative stabilization; Uncertain systems

1. Introduction
There has been a great amount of research work on the tuning of PI (proportional integral), PID (proportional integral derivative) and lag/lead controllers since these types of controllers have been widely used in
industries for several decades (see Refs. [1,2] and references therein). However, many important results have
been recently reported on computation of all stabilizing P (proportional), PI and PID controllers after the
publication of work by Ho et al. [36]. A new and complete analytical solution based on the generalized version of the HermiteBiehler theorem has been provided in Ref. [3] for computation of all stabilizing constant
gain controllers for a given plant. A linear programming solution for characterizing all stabilizing PI and PID
*

Corresponding author. Tel.: +90 4223410010x4429; fax: +90 4223410046.

E-mail address: ntan@inonu.edu.tr (N. Tan).

0196-8904/$ - see front matter  2006 Elsevier Ltd. All rights reserved.
doi:10.1016/j.enconman.2006.03.022

3046

N. Tan et al. / Energy Conversion and Management 47 (2006) 30453058

controllers for a given plant has been obtained in Refs. [4,6]. This approach, besides being computationally
ecient, has revealed important structural properties of PI and PID controllers. For example, it was shown
that for a xed proportional gain, the set of stabilizing integral and derivative gains lie in a convex set. This
method is very important since it can cope with systems that are open loop stable or unstable, minimum or
non-minimum phase. However, the computation time for this approach increases in an exponential manner
with the order of the system being considered. It also needs sweeping over the proportional gain to nd all
stabilizing PI and PID controllers, which is a disadvantage of the method. An alternative fast approach to this
problem based on the use of the Nyquist plot has been given in Refs. [7,8]. A stability boundary locus
approach for the design of PI and PID controllers has been given in Ref. [9]. A parameter space approach
using the singular frequency concept has been given in Ref. [10] for design of robust PID controllers. More
direct graphical approaches to this problem based on frequency response plots have been given in Refs.
[11,12]. However, the requirement for frequency gridding has become the major problem for this approach.
Compensator design in classical control engineering is based on a plant with xed parameters. In the real
world, however, most practical system models are not known exactly, meaning that the system contains uncertainties. Much recent work on systems with uncertain parameters has been based on Kharitonovs result [13]
on the stability of interval polynomials. Using the Kharitonov theorem, there have been many developments
in the eld of parametric robust control related to the stability and performance analysis of uncertain control
systems represented as interval plants [14].
In this paper, a new approach is given for computation of stabilizing PI controllers in the parameter plane,
(kp, ki) plane. The result of Ref. [8] is used to obtain the stability boundary locus over a possible smaller range
of frequency. Thus, a very fast way of calculating the stabilizing values of PI controllers for a given SISO (single input, single output) control system is given. The proposed method is also used for computation of PI controllers for relative stabilization and for achieving user specied gain and phase margins. An extension of the
method to nd all stabilizing values of the parameters of a PID controller, namely kp, ki and kd in the (kp, ki)
plane, (kp, kd) plane and (ki, kd) plane, is also given. It is shown that the stability boundary of the convex polygon in the (ki, kd) plane for a xed value of kp can be generated from four straight lines. The equations of these
straight lines can be easily derived using the stability boundary of the stabilizing regions obtained in the (kp, ki)
plane and (kp, kd) plane. The proposed method is nally used for computation of PI controllers for the stabilization of interval systems.
The paper is organized as follows: The proposed method is described in Section 2. In Section 3, the computation of PI controllers for relative stabilization is given. The design of PI controllers that achieve user specied gain and phase margins is given in Section 4. Extension of the method to PID controllers is given in
Section 5. In Section 6, the computation of PI controllers for interval plant stabilization is given. Concluding
remarks are given in Section 7.
2. Stabilization using a PI controller
Consider the single input, single output (SISO) control system of Fig. 1 where
Gs

N s
Ds

is the plant to be controlled and C(s) is a PI controller of the form

Cs k p

ki kps ki

s
s

+_

C(s)

G(s)

Fig. 1. A SISO control system.

N. Tan et al. / Energy Conversion and Management 47 (2006) 30453058

3047

The problem is to compute the parameters of the PI controller of Eq. (2) that stabilize the system of Fig. 1.
Decomposing the numerator and the denominator polynomials of Eq. (1) into their even and odd parts and
substituting s = jx, gives
Gjx

N e x2 jxN o x2

De x2 jxDo x2

The closed loop characteristic polynomial of the system can be written as

Ds k i N e x2  k p x2 N o x2  x2 Do x2  jk p xN e x2 k i xN o x2 xDe x2 
RD jI D 0

Then, equating the real and imaginary parts of D(s) to zero, one obtains
k p x2 N o x2 k i N e x2 x2 Do x2

k p xN e x2 k i xN o x2 xDe x2

and

Let
Qx x2 N o x2 ;
Sx xN e x2 ;
2

X x x Do x ;

Rx N e x2

U x xN o x2

7
2

Y x xDe x

Then, Eqs. (5) and (6) can be written as

k p Qx k i Rx X x
k p Sx k i U x Y x

From these equations

kp

X xU x  Y xRx
QxU x  RxSx

ki

Y xQx  X xSx
QxU x  RxSx

10

and

Solving these two equations simultaneously, the stability boundary locus, l(kp, ki, x), in the (kp, ki) plane can be
obtained. The stability boundary locus, l(kp, ki, x), and the line ki = 0 (since a real root of D(s) of Eq. (4) can
cross the imaginary axis at s = 0, thus, for x = 0, ID = 0, and from RD = 0, it can be found that ki = 0) divide
the parameter plane ((kp, ki) plane) into stable and unstable regions. Choosing a test point within each region,
the stable region that contains the values of stabilizing kp and ki parameters can be determined.
Example 1. Consider the control system of Fig. 1 with transfer function
Gs

N s
s3 4s2  s 1
5
Ds s 2s4 32s3 14s2  4s 50

11

which has two right half plane poles and two right half plane zeros. From Eqs. (9) and (10)
kp

x8 23x6 94x4  210x2 50

x6  18x4 7x2  1

and

ki

2x8  117x6  20x4  46x2

x6  18x4 7x2  1

12

The aim is to compute all the stabilizing values of kp and ki that make the characteristic polynomial of Eq. (4)
Hurwitz stable. For a range of frequency, the stability boundary locus can be easily computed. For example,
for x 2 [0.45, 7.8], l(kp, ki, x) is shown in Fig. 2. From this gure, it can be seen that there are a few regions,
namely R1, R2, R3, R4 and R5 in which one needs to choose a test point in order to nd the stability region.

3048

N. Tan et al. / Energy Conversion and Management 47 (2006) 30453058

45

= 4.4

40

= 0.48

R4

R3

= 0.5326 and 5.547

35

= 0.45
30

R1
R5

= 0.505 and 3.25

Stability
Region

25
R2

ki 20

= 0.58

=2

15

=7

10

= 0.698 and 6.664

= 7.65

= 7.8
-5
-40

-30

-20

-10

10

20

30

kp
Fig. 2. Stability boundary locus.

For example, choosing a test point within region R5, such as kp = 13 and ki = 25, it can be calculated that the
characteristic polynomial has two right half plane complex roots, and therefore, the system is unstable for
these values of parameters. Thus, the region R5 is not a stable region. It is found that the only stabilizing region is the region denoted by R1. For example, for kp = 5 and ki = 20 within region R1, the characteristic
polynomial is D(s) = s6 + 2s5 + 37s4 + 54s3 + 71s2 + 35s + 20, which is a stable polynomial. Fig. 3 shows
more clearly R1 for all stabilizing values of kp and ki.
This example shows that the method is very fast and eective, however, frequency gridding becomes important. An ecient approach to reduce the range of frequency that needs to be gridded can be obtained by using
the Nyquist plot based approach of Ref. [8]. In this case, it is only necessary to nd real values of x that satisfy
Im[G(s)] = 0 where s = jx. Thus, the frequency axis can be divided into a nite number of intervals, and
then, by testing each interval, the stability region can be computed. For example, consider a second order
system G(s) = (s  1)/(s2 + 0.8s  0.2). From Eqs. (9) and (10), kp = (1.8x2 + 0.2)/(x2  1) and

40

35

30
Stability Region

ki 25

20

15

10
-10

-5

kp

Fig. 3. All stabilizing PI controllers.

10

15

N. Tan et al. / Energy Conversion and Management 47 (2006) 30453058

0.1

3049

=0.9

0.08
0.06
0.04
0.02

=0

=0.77

ki
0
-0.02

Stability Region

-0.04
-0.06
-0.08
-1

-0.9

-0.8

-0.7

-0.6

-0.5

-0.4

-0.3

-0.2

kp
Fig. 4. Stability boundary locus.

ki = (x4 + 0.6x2)/(x2  1). It can be calculated that the only real frequency value that satises
Im[G(jx)] = 0 is 0.77 rad/s. Thus, the frequency axis can be divided into two intervals, such as x 2 (0, 0.77)
and x 2 (0.77, 1). For 100 points within x 2 [0, 0.9], l(kp, ki, x) is shown in Fig. 4 where it can be seen that
there are stabilizing values of kp and ki when x 2 (0, 0.77). For the G(s) of Eq. (11), Im[G(jx)] = 0 for
x = 7.65. After testing that the stable region is for x 6 7.65, the stability region shown in Example 1 is then
easily computed.
To show that the region that contains all stabilizing PI controllers is not always a convex set, the control of
the transfer function for a exible arm [15] is considered in the following example.
Example 2. Here,
Gs

s 500
ss 0:0325s2 2:57s 6667

13

10

x 10

=65

=50

=68

R4
R2

ki 4
=41.8 and 70.1

=35
=30

=20

=72

R1

Stability region

=0
-2

=73.398
0.5

kp

R3
1.5

Fig. 5. Stability boundary locus for exible arm control.

2.5
4
x 10

3050

N. Tan et al. / Energy Conversion and Management 47 (2006) 30453058

From Eqs. (7), (9) and (10)

497:4x4  3333333:3x2
x6  5367:1x4  108350x2
and
k
14

i
x2  250000
x2  250000
Since Im[G(jx)] = 0 is satised for x = 73.398 rad/s, it is necessary to obtain the stability boundary locus for
x 2 (0, 73.398). The stability boundary locus when x varies within x 2 (0, 73.398) is shown in Fig. 5 from
which it has been computed that the stabilizing values of kp and ki only exist in the region R1. It can be clearly
seen that the stabilizing region that is shown in Fig. 5 is not a convex set.
kp

3. Relative stabilization
In this section, the approach is further developed for relative stabilization. In the analysis and design of
control systems, it is important to shift all poles of the characteristic equation of a control system to a desired
region in the complex plane, for example, to a shifted half plane that guarantees a specied settling time of the
response. The aim of this section is to nd all values of kp and ki that put all the closed loop poles to the left of
s = q (q = constant). Using s + q instead of s in Eqs. (1) and (2)
~ s
N s q N
~

Gs
Gs q
15
~
Ds q Ds
and
kps kpq ki
~
Cs
Cs q
sq
Eq. (15) can be written as
~ e x2 jxN
~ o x2
N
~
Gjx

~ e x2 jxD
~ o x2
D

16

17

Then, Eq. (7) for this case becomes

~ o x2 ; Rx N
~ e x2
~ e x2  x2 N
Qx qN
~ o x2 ; U x xN
~ o x2
~ e x2 qN
Sx xN
~ o x2  qD
~ e x2 ; Y x xD
~ e x2 qD
~ o x2
X x x2 D

18

Using Eqs. (9) and (10), the stability boundary locus, l(kp, ki, x), can be obtained in the (kp, ki) plane for relative stabilization, and for x = 0, the stability boundary line is
ki

~ e x2  qD
~ e x2
k p qN
2
~ e x
N

19

Thus, the stability region can be computed using the stability boundary locus and the stability boundary line
of Eq. (19).
Example 3. Consider the transfer function of the Manutec robot [15]
Gs

10
ss 2s 40s 45

20

All stabilizing values of kp and ki for q = 0 are calculated as indicated previously and are shown in Fig. 6.
Now, let us nd all the stabilizing values of kp and ki for q = 0.2. From Eqs. (15) and (16)
k p s  0:2k p k i
~
Cs

s  0:2

21

and
~
Gs
Gs q

10
s4 86:2s3 1918:04s2 2822:4s  641:89

22

N. Tan et al. / Energy Conversion and Management 47 (2006) 30453058

3051

4000

=5

=4

3500
3000

Stability region for = 0

=3

2500

= 4.5

= 3.5

2000

ki

=6
= 2.5
Stability region for = -0.2

1500
1000

= 5.326

500
0
-500

= 6.4327

=0 =0
0

1000

2000

3000

4000

5000

6000

7000

8000

9000

kp
Fig. 6. Stability regions for q = 0 and q = 0.2.

Using Eqs. (18), (9) and (10)

k p 0:1x4 190:08x2 120:64

and

k i 8:62x4 281:9x2 11:29

23

and from Eq. (19), the stability boundary line is

k i 0:2k p  12:8378

24

It can be calculated that the stability boundary locus and the stability boundary line intersect at x = 0 and
x = 5.326. Thus, all the stabilizing values of kp and ki that put all the closed loop poles to the left of
s = 0.2 are shown in Fig. 6.
4. Stabilization for specied gain and phase margins
Consider Fig. 1 with a gain-phase margin tester, Gc (s) = Aej/, which is connected in the feed forward
path. Then, the characteristic equation can be written in the form of Eq. (4) as
Ds k p AxN e x2 sin/  x2 N o x2 cos/ k i AN e x2 cos/ xN o x2 sin/
 x2 Do x2  jk p AxN e x2 cos/ x2 N o x2 sin/ k i AxN o x2 cos/
 N e x2 sin/ xDe x2 
0

25

Thus, Eq. (7) can be written as

Qx AxN e x2 sin/  x2 N o x2 cos/
Rx AN e x2 cos/ xN o x2 sin/
Sx AxN e x2 cos/ x2 N o x2 sin/
2

26

U x AxN o x cos/  N e x sin/

X x x2 Do x2 ; Y x xDe x2
To obtain the stability boundary locus for a given value of gain margin A, one needs to set / = 0 in Eq. (26).
On the other hand, setting A = 1 in Eq. (26), one can obtain the stability boundary locus for a given phase
margin /.

3052

N. Tan et al. / Energy Conversion and Management 47 (2006) 30453058

1.2

=0.7

Stability region

0.8

0.6

=0.9

=0.4

ki

Stability boundary
for = 45o

=0.4

0.4

=0.7
=0.2

0.2

=0.6

Stability region
for 45o and
A3

=0 = 0 = 0
-0.2
-1

-0.5

=1
0.5

1.5

Stability boundary
for A = 3

=1

= 0.668
2

2.5

3.5

kp
Fig. 7. Stability region for / P 45 and A P 3.

Example 4. Consider
1
Gs
s 14

27

The aim is to nd all stabilizing PI controllers that satisfy the conditions that the phase margin of the system is
greater than 45 and the gain margin is greater than 3(9.54 db). First, let us compute the stability boundary
locus. Using Eqs. (7), (9) and (10)
k p x4 6x2  1

and

k i 4x4 4x2

28

Since Im[G(jx)] = 0 for x = 1 rad/s, it is only necessary to grid x within x 2 (0, 1). The stabilizing region is
shown in Fig. 7. To nd all stabilizing PI controllers for which the phase margin of the system is greater than
45, it is required to set A = 1 and / = 45 in Eq. (26). Thus, using Eqs. (9), (10) and (26) for these values of A
and / gives
k p 0:707x4  2:82x3 4:24x2 2:82x  0:707
k i 0:707x5  2:82x4  4:24x3 2:82x2 0:707x

29

The range of x needed for stabilization for this case can be found from arg[G(jx)] = 135. It can be seen that
arg[G(jx)] = 135 for x = 0.668 rad/s. The stability boundary locus for / = 45 with x varying within
x 2 (0, 0.668) is shown in Fig. 7. Similarly, setting / = 0 and A = 3 in Eq. (26)
k p 0:33x4 2x2  0:33 and

k i 1:33x4 1:33x2

30

The stability boundary locus for A = 3 can be obtained for x 2 (0, 1) as shown in Fig. 7. The region obtained
from the intersections of these regions includes the stabilizing values of kp and ki that make the phase margin
of the system greater than 45 and the gain margin greater than 3. For example, choosing a point in the region,
such as kp = 0.35 and ki = 0.3, the closed loop poles are 1.4 0.5i, 0.16 0.36i and 0.87. The gain margin is equal to 3.04 and the phase margin is equal to 45.7.
5. Design of PID controllers
Consider that C(s) of Fig. 1 is an ideal PID controller of the form
Cs k p

ki
kds
s

31

N. Tan et al. / Energy Conversion and Management 47 (2006) 30453058

3053

Using the procedure given in Section 2, the stability boundary locus in the (kp, ki) plane can be obtained for a
xed value of kd or in the (kp, kd) plane for a xed value of ki. However, it is not possible to obtain the stability
boundary locus in the (ki, kd) plane for a xed value of kp since Q(x)U(x)  R(x)S(x) will be equal to zero for
this case. Although the stability region in the (ki, kd) plane for a xed value of kp is a convex polygon [4], the
stability region in the (kp, ki) plane for a xed value of kd or in the (kp, kd) plane for a xed value of ki is not a
convex polygon, and it may not even be a convex set as shown in Example 2. However, it is possible to obtain
the stability region in the (ki, kd) plane for a xed value of kp using the stability region obtained in the (kp, ki)
plane and (kp, kd) plane as follows:
Example 5. Consider a system with the transfer function
Gs

N s
s3  4s2 s 2
5
4
Ds s 8s 32s3 46s2 46s 17

32

which has also been studied in Ref. [4]. The aim is to nd the limiting values of the parameters of a PID controller such that the resulting closed loop system is stable. To obtain the stability region in the (kp, ki) plane in
terms of kd, it can be found using Eqs. (9) and (10) that
kp

x8  65x6 246x4  22x2  34

x6 14x4 17x2 4

33

ki

a1 x8  a2 x6 a3 x4 a4 x2
x6 14x4 17x2 4

34

and

where a1 = 12 + kd, a2 = 180  14kd, a3 = 183 + 17kd and a4 = 75 + 4kd. The stability regions for kd = 0 and
kd = 1 are shown in Fig. 8. From this gure, it can be seen that the values of kp for which stabilizing ki and kd
values exist change within the range kp 2 (8.5, 4.21), which is the same result as that obtained in Ref. [4]
where the root locus method was used. For the computation of the stabilizing regions in the (kp, kd) plane
for xed ki, it has been calculated that kp is the same as Eq. (33), and solving Eq. (34) for kd in terms of
ki, one can obtain
kd

12x8 b1 x6 b2 x4 b3 x2 b4
x8 14x6 17x4 4x2

35

kp = 0

4.21
4

kd = 1

3.76
3

kd = 0

ki 2
Stability Region
1

-1
-10

-8

k p = -8.5

-6

-4

-2

kp

k p = 4.21

Fig. 8. Stability region in the (kp, ki) plane for kd = 0 and kd = 1.

3054

N. Tan et al. / Energy Conversion and Management 47 (2006) 30453058

where b1 = 180 + ki, b2 = 183 + 14ki, b3 = 75 + 17ki and b4 = 4ki. The stability regions for ki = 0 and
ki = 1 are shown in Fig. 9. Since it is known that the stability region in the (ki, kd) plane for a xed value
of kp is a convex polygon, it is possible to obtain this polygon using Figs. 8 and 9. From Fig. 8, one can obtain
the equations of the two straight lines that contribute to the boundary of the stability region for each value of
kp. For example, for kp = 0, one straight line passes through the points (ki, kd) = (3.76, 0) and
(ki, kd) = (4.21, 1) and the other straight line passes through (ki, kd) = (0, 0) and (ki, kd) = (0, 1). Thus, the equations of these two straight lines are:
l1 : k d 2:17k i  8:17

and

l2 : k i 0

Similarly, from Fig. 9, one straight line goes through the points (ki, kd) = (0, 3.61) and (ki, kd) = (1, 3.87) and the
other one goes through (ki, kd) = (0, 7.37) and (ki, kd) = (1, 6.15). The equations of these two straight lines are:
l3 : k d 0:26k i 3:61

and

l4 : k d 1:22k i  7:37

kp = 0
4

3.87
3.61

kd

Stability Region
-2

-4

ki= 1
ki= 0

-6

-6.15
-7.37
-8
-12

-10

-8

-6

-4

-2

kp
Fig. 9. Stability region in the (kp, kd) plane for ki = 0 and ki = 1.

10

l3

l2
5

l1

Stability Region

kd
-5

l4

-10

-15
-2

-1

ki
Fig. 10. Stability region for kp = 0.

N. Tan et al. / Energy Conversion and Management 47 (2006) 30453058

3055

kp
-5
0

-10
-8

5
-6

-4

-2

kd

ki

10

Fig. 11. Stabilizing kp, ki and kd values.

These four lines are shown in Fig. 10 where the shaded region is the stability region in the (ki, kd) plane for
kp = 0. Repeating the procedure for different values of the proportional gain, kp, the stabilizing kp, ki and
kd values can be shown in the 3-D plot of Fig. 11.
6. Interval plant stabilization
There are some important results in the literature about stabilization of interval systems. For example, in
Ref. [16], it was shown that a constant gain controller stabilizes an interval plant family if and only if it stabilizes a set of eight of the extreme plants that can be obtained from the upper and lower values of uncertain
parameters. In Ref. [17], it was shown that a rst order controller stabilizes an interval plant if it stabilizes the
set of extreme plants. In Ref. [18], it was proved that a rst order controller stabilizes an interval plant if and
only if it simultaneously stabilizes the sixteen Kharitonov plants family. In Ref. [19], the generalized version of
the HermiteBiehler theorem has been used for the stabilization of interval systems. In this section, instead of
using Routh tables, which were used in Ref. [18], in order to characterize all the parameters of a rst order
controller that stabilize an interval plant, the stability boundary locus is used to nd all the values of the
parameters of a PI controller for which the given interval plant is Hurwitz stable.
Consider a unity feedback system with a PI controller of Eq. (2) and an interval plant
Gs

N s qm sm qm1 sm1    q0

Ds
pn sn pn1 sn1    p0

36

where qi 2 qi ; qi ; i 0; 1; . . . ; m and pi 2 pi ; pi ; j 0; 1; . . . ; n. Let the Kharitonov polynomials associated

with N(s) and D(s) be, respectively:
N 1 s q0 q1 s q2 s2 q3 s3   
N 2 s q0 q1 s q2 s2 q3 s3   
N 3 s q0 q1 s q2 s2 q3 s3   
N 4 s q0 q1 s q2 s2 q3 s3   

D1 s p0 p1 s p2 s2 p3 s3   
and

D2 s p0 p1 s p2 s2 p3 s3   
D3 s p0 p1 s p2 s2 p3 s3   

37

D4 s p0 p1 s p2 s2 p3 s3   

By taking all combinations of the Ni(s) and Dj(s) for i,j = 1, 2, 3, 4, the following sixteen Kharitonov plants
family can be obtained
GK s Gij s

N i s
Dj s

38

3056

N. Tan et al. / Energy Conversion and Management 47 (2006) 30453058

d(t)
r(t)

e(t)

++

C(s)

+_

y(t)

G(s)
motor

controller

Fig. 12. Position control system.

where i, j = 1, 2, 3, 4. Dene the set S(C(s)G(s)) that contains all the values of the parameters of the controller
C(s) that stabilize G(s), and then, the set of all the stabilizing values of the parameters of a PI controller that
stabilize the interval plant of Eq. (36) can be written as
SCsGs SCsGK s SCsG11 s \ SCsG12 s    SCsG44 s

39

where GK(s) represents the sixteen Kharitonov plants family given in Eq. (38).
Example 6. Consider the position control system with the block diagram of Fig. 12, where the motor is
assumed to have the transfer function
Gs

Km
Km

sJs bLf s Rf Lf Js3 bLf JRf s2 bRf s

40

Rf and Lf are the resistance and inductance of the eld, Km is the motor constant, J is the inertia, b is the viscous friction and the controller is a PI controller of the form of Eq. (2). Here, an integrator is needed for elimination of a steady state error to a ramp input or a steady state error to a torque disturbance. The major
uncertainty is assumed to be in the parameters Km, b, Lf and J. Initially, assume that all these parameters
are xed and have the nominal values Km = 60 103, Rf = 1.2, b = 2.5 103, Lf = 1.3 102 and
J = 2 103. For these nominal values, the nominal transfer function of the motor can be written as
Gno s

60
s0:026s2 2:43s 3

41

Using Eqs. (7), (9) and (10), kp = 0.0405x2 and ki = 0.00043x4 + 0.05x2. All the stabilizing values of kp and
ki for the nominal transfer function are enclosed by Gno in Fig. 13.

4.5
4

G12

3.5

G22
G14

3
2.5

G24

ki 2
1.5

Gno
G11

1
0.5
0
-0.5

G21

G13

G23
0

kp

10

12

14

16

Fig. 13. Stability regions of eight Kharitonov plants.

18

N. Tan et al. / Energy Conversion and Management 47 (2006) 30453058

3057

0.6

0.5

0.4

0.3

ki
0.2
Stability region for interval plant
0.1

-0.1
0

0.2

0.4

0.6

kp

0.8

1.2

1.4

Fig. 14. Stability region for interval plant of motor.

Assume now that the parameters Km, b, Lf and J may vary by 10%, 15%, 20% and 40% around their nominal values, respectively, then,
K m 2 K m ; K m  54  103 ; 66  103 ;

b 2 b; b 2:1  103 ; 2:9  103 ;

Lf 2 Lf ; Lf  1:04  102 ; 1:56  102 ;

J 2 J ; J  1:2  103 ; 2:8  103 

42

Thus, by over-bounding the coecients of Eq. (40), since the uncertain parameters of Eq. (40) are multi-linearly dependent on the above parameters, the transfer function of the motor can be written in the form of the
interval transfer function as follows:
q0
43
Gs
p3 s3 p2 s2 p1 s p0
where q0 2 K m ; K m  54  103 ; 66  103 , p0 0; p1 2 bRf ; bRf  2:52  103 ; 3:48  103 , p2 2 bRf ;
bRf  2:52  103 ; 3:48  103  and p3 2 Lf J ; Lf J  1:25  105 ; 4:37  105 . The stabilizing parameters
can be calculated for all eight Kharitonov plants (i = 12 and j = 14) and their bounds are also shown in
Fig. 13. Obviously, in this simple case, all G1j are outside G2j so it is only necessary to plot four loci. The interior of all these, which is bounded by parts of G21 and G23, gives the region of all stabilizing parameters for
stable control of the uncertain system and is shown in Fig. 14.
7. Conclusions
In this paper, a new approach has been presented for computation of the boundaries of the limiting values
of PI and PID controller parameters that guarantee stability. The approach is based on the stability boundary
locus, which can be easily obtained by equating the real and the imaginary parts of the characteristic equation
to zero. The computation of PI compensator parameters for relative stabilization and for achieving user specied gain and phase margins have also been studied. The proposed method has further been used to nd the
stabilizing region of PI parameters for the control of a plant with uncertain parameters. The method presented
does not require sweeping over the parameters. Also, it does not need linear programming to solve a set of
inequalities. Therefore, the method has advantages over existing results. The examples given clearly show
the value of the method presented.
References
[1] Zhuang M, Atherton DP. Automatic tuning of optimum PID controllers. IEE Proc Part D 1993;140:21624.
[2] Astrom KJ, Hagglund T. PID controllers: Theory, design, and tuning. Instrument Society of America; 1995.

3058

N. Tan et al. / Energy Conversion and Management 47 (2006) 30453058

[3] Ho MT, Datta A, Bhattacharyya SP. A new approach to feedback stabilization. In: Proc of the 35th CDC, 1996. p. 46438.
[4] Ho MT, Datta A, Bhattacharyya SP. A linear programming characterization of all stabilizing PID controllers. In: Proc of Amer contr
conf, 1997.
[5] Ho MT, Datta A, Bhattacharyya SP. A new approach to feedback design Part I: Generalized interlacing and proportional control,
Department of Electrical Engineering, Texas A&M University, College Station, TX, Tech. Report TAMU-ECE97-001-A, 1997.
[6] Ho MT, Datta A, Bhattacharyya SP. A new approach to feedback design Part II: PI and PID controllers, Dept. of Electrical Eng.,
Texas A& M Univ., College Station, TX, Tech. Report TAMU-ECE97-001-B, 1997.
[7] Munro N, Soylemez MT. Fast calculation of stabilizing PID controllers for uncertain parameter systems. In: Proc symposium on
robust control, Prague, 2000.
[8] Soylemez MT, Munro N, Baki H. Fast calculation of stabilizing PID controllers. Automatica 2003;39:1216.
[9] Tan N, Kaya I, Atherton DP. Computation of stabilizing PI and PID controllers. In: Proc of the 2003 IEEE int conf on the control
application (CCA2003), 2003. p. 87681.
[10] Ackermann J, Kaesbauer D. Design of robust PID controllers. In: European control conference, 2001. p. 5227.
[11] Shaei Z, Shenton AT. Frequency domain design of PID controllers for stable and unstable systems with time delay. Automatica
1997;33:222332.
[12] Huang YJ, Wang YJ. Robust PID tuning strategy for uncertain plants based on the Kharitonov theorem. ISA Trans 2000;39:41931.
[13] Kharitonov VL. Asymptotic stability of an equilibrium position of a family of systems of linear dierential equations. Di Equat
1979;14:14835.
[14] Bhattacharyya SP, Chapellat H, Keel LH. Robust control: the parametric approach. Prentice Hall; 1995.
[15] Dorf RC, Bishop RH. Modern control systems. New Jersey: Prentice Hall; 2001.
[16] Ghosh BK. Some new results on the simultaneous stabilization of a family of single input, single output systems. Syst Contr Lett
1985;6:3945.
[17] Hollot CV, Yang F. Robust stabilization of interval plants using lead or lag compensators. Syst Contr Lett 1990;14:912.
[18] Barmish BR, Holot CV, Kraus FJ, Tempo R. Extreme points results for robust stabilization of interval plants with rst order
compensators. IEEE Trans Automat Contr 1993;38:17345.
[19] Ho MT, Datta A, Bhattacharyya SP. Design of P, PI and PID controllers for interval plants. In: Proc of Amer contr conf,
Philadelphia, June 1998.