Journal of Process Control 19 (2009) 349352

Contents lists available at ScienceDirect

Journal of Process Control

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

Short communication

Guaranteed dominant pole placement with PID controllers

Qing-Guo Wang a,*, Zhiping Zhang a, Karl Johan Astrom b, Lee See Chek a
a b

Department of Electrical and Computer Engineering, National University of Singapore, Singapore 119260, Singapore Department of Automatic Control, Lund Institute of Technology, Sweden

a r t i c l e

i n f o

a b s t r a c t
Pole placement is a well-established design method for linear control systems. Note however that with an output feedback controller of low-order such as the PID controller one cannot achieve arbitrary pole placement for a high-order or delay system, and then partially or hopefully, dominant pole placement becomes the only choice. To the best of the authors knowledge, no method is available in the literature to guarantee dominance of the assigned poles in the above case. This paper proposes two simple and easy methods which can guarantee the dominance of the two assigned poles for PID control systems. They are based on root locus and Nyquist plot respectively. If a solution exists, the parametrization of all the solutions is explicitly given. Examples are provided for illustration. 2008 Elsevier Ltd. All rights reserved.

Article history: Received 27 February 2008 Received in revised form 18 April 2008 Accepted 25 April 2008

Keywords: PID controllers Dominant poles Pole placement Root locus Nyquist plot

1. Introduction Pole placement in the state space and polynomial settings is very popular. For SISO plants, the equivalent output feedback control should be at least of the plant order minus one to achieve arbitrary pole placement. Arbitrary pole placement is otherwise difcult to achieve if one has to use a low-order output feedback controller for a high-order or time-delay plant. One typical example is that in process control, the PID controller is used to regulate a plant with delay. To overcome this difculty, the dominant pole design was proposed. It is to choose and position a pair conjugate poles which represent the requirements on the closed-loop response, such as overshoot and settling time. The dominant pole design was rst introduced by Persson [1] and further explained in [2]. The methods they proposed are based on a simplied model of plants and thus cannot guarantee the chosen poles are indeed dominant in reality. In the case of high-order plants or plants with time-delay, this conventional dominant pole design, if not well handled, could result in sluggish response or even instability of the closed-loop system. To the best of the authors knowledge, no method is available in the literature to guarantee the dominance of the assigned poles in the case above. Thus, it is desirable to nd out ways to ensure the dominance of chosen poles and the closed-loop stability. This paper aims to solve the problem. The idea behind our methods is that the chosen pair of poles give rise to two real equations which are solved for I and D terms via the proportional gain and the locations of all other
* Corresponding author. Tel.: +65 874 2282; fax: +65 67791103. E-mail address: elewqg@nus.edu.sg (Q.-G. Wang). 0959-1524/$ - see front matter 2008 Elsevier Ltd. All rights reserved. doi:10.1016/j.jprocont.2008.04.012

closed-loop poles can then be studied with respect to this single variable gain by means of root locus or Nyquist plot techniques. Hence, two methods for guaranteed dominant pole placement with PID controller are naturally developed. The root locus method is used for systems without time-delay and the Nyquist plot method is developed to handle time-delay systems. The rest of the paper is organized as follows. Section 2 states the problem and preliminary. Sections 3 and 4 each present one method along with illustrating examples. Section 5 is the conclusion. 2. Problem statement and preliminary Consider a plant described by its transfer function,

Gs

Ns sL e ; Ds

s where N is a proper and co-prime rational function. A PID controlDs ler in the form of

C s K P

KI KDs s

is used to control the plant in the conventional unity output feedback conguration as depicted in Fig. 1. The closed-loop characteristic equation is

1 C sGs 0:
And the closed-loop transfer function is

H s

Ns K D s2 K P s K I eLs : Dss NseLs K D s2 K P s K I

350

Q.-G. Wang et al. / Journal of Process Control 19 (2009) 349352

R(s)

Fig. 1. Unity output feedback control system.

Suppose that the requirements of the closed-loop control performance in frequency or time domain are converted into a pair of conjugate poles [2]: q1,2 = a bj. Their dominance requires that the ratio of the real part of any of other poles to a exceeds m (m is usually 35) and there are no zeros nearby. Thus, we want all other poles to be located at the left of the line of s = ma, that is, the desired region as hatched in Fig. 2. The problem of the guaranteed dominant pole placement is to nd the PID parameters such that all the closed-loop poles lie in the desired region except the dominant poles, q1,2. Substitute q1 = a + bj into (2):

KI 1 KP K D a bj ; Gq1 a bj
which is a complex equation. The complex equations can be decomposed into two sub-equations, one from the real part and the other from the imaginary part. Solving the two equations for KI and KD in terms of KP yields

KI KD

a2 b2 K p a2 2a 1 K X2 ; 2a p

b X 1 ;

1 1 1 1 1 1 1 1 where X 1 2 ImG 2 ReG , X 2 2 ImG 2 ReG . This b q a q b q a q

1 1 1 1

simplies the original problem to a one-parameter problem for which well known methods are applicable now. The root locus method can be used for systems without time-delay, but it cannot handle time-delay systems. Hence the Nyquist plot method is proposed for time-delay systems. 3. Root locus method The root locus method is used to show movement of the roots of the characteristic equation for all values of a system parameter. Here we plot the roots of the closed-loop characteristic equation for all the positive values of KP and determine the range of KP such

Im A

-ma

1 2

A'

Fig. 2. Desired region (hatched) of other poles.

C (s

G (s

Y(s)

that the roots except the chosen dominant pair are in the desired region as hatched in Fig. 2. Substituting (3) into (2) yields

1 X2

  N seLs N seLs 2 s a2 b X 1 Ds Dss NseLs s2 2as a2 b 0: 2as Ds

2

KP

Dividing both sides by the terms without KP, after some manipulation, gives

1 K P Gs 0;
where

h i 2 Ns s2 2as a2 b eLs   Gs : 2 2aDss 2aX 2 Nss2 eLs 2a a2 b X 1 NseLs

It can be easily veried that the manipulation does not change the roots. If G(s) has no time-delay term, Gs is a proper rational transfer function since the degrees of its numerator and denominator of Gs equal those of the closed-loop transfer functions numerator and denominator respectively. The design procedure is simply: Step 1. Draw the root locus of (5) as KP varies; Step 2. Determine the interval of KP for guaranteed dominant pole placement from the root locus; Step 3. Choose KP, KI and KD accordingly. Example 1 shows the design procedure in detail. Example 1 Consider a fourth-order process,

Gs

1 s 12 s 52

If the overshoot is to be no larger than 8% and the rise time less than 3 s, the corresponding dominant poles are q1,2 = 0.4849 0.6031j. Eq. (3) becomes

K I 0:6175K P 2:6037; K D 1:0312K P 15:3910:

And it follows from (6) that

Gs

s2 0:9698s 0:5989 : 0:9698s5 11:64s4 44:61s3 43:26s2 24:24s 2:525

Re

The root locus of Gs is exhibited in Fig. 3 with the solid lines while the edge of the desired region with m = 3 is indicated with dotted lines. Note that Gs is of 5th order and has ve branches of root loci, of which two are xed at the dominant poles while the other three move with the gain. From the root locus, two intersection points corresponding to root locus entering into and departing from the desired region are located and give the gain range of KP 2 (25,83), which ensures all other three poles in the desired region. Besides, the positiveness of KD and KI requires KP > 14.9253. Taking the joint solution of these two, we still have KP 2 (25,83). If KP = 45 is chosen, the PID controller is

C s 45

30:3931 31:0110s: s

The zeros of the closed-loop system are at s = 0.7255 0.6735j. Fig. 4 shows the step response of the closed-loop system. The specications are met, whereas if the approximate dominant pole placement method in [2] is used the step response has an overshoot more than 8%. As the dominance of the chosen poles are guaranteed, one

Q.-G. Wang et al. / Journal of Process Control 19 (2009) 349352

351

Kp = 83
2

Kp = 25
2

6 12

10

Fig. 3. Root locus for Example 1.

no more pole outside the desired region to ensure dominant pole placement. Equivalently, we want the modied Nyquist plot of Gs to have the net number of clockwise encirclements with respect to K1P ; 0 equal to 2 minus the number of poles of Gs outside the desired region. This condition determines the interval of KP such that roots of (7) other than two dominant poles are in the desired region, which can guarantee the two dominant poles are indeed dominant. Note the condition comes from the criterion that the net number of clockwise encirclement equals the number of closed-loop poles outside the desired region (in this case 2) minus the number of open-loop poles (poles of Gs) outside the desired region. It is analog to the conventional Nyquist stability criterion; the only difference is that here we are concerned about poles outside the desired region, not poles in RHP. To nd the number of poles of Gs located outside the desired region is not easy at rst glance. The poles of Gs are the roots of its denominator. The denominator has time-delay terms whose dimension is innite. To solve the problem we construct another characteristic equation from the denominator of Gs in (6) as follows:

1.4

1 Go s 0;
X 2 N ss2 a2 b2 X 1 N s Ls e . Dss

1.2

Overshoot: 6%
1

Output

0.8

Rise Time: 1.46s

0.6

0.4

where Go s Go s has its rational part with the degrees of its numerator and denominator being equal to those of the original open-loop transfer functions nominator and denominator respectively. The number of the roots of (8), i.e., poles of Gs, lying outside the desired region, equals the number of clockwise encirclements of the modied Nyquist plot of Go s with respect to (1,0), plus the number of poles of Go s located outside the desired region. We note that the poles of Go s are easy to nd from the denominator of Go s, which is just D(s)s. The theoretical basis for applying the Nyquist stability criterion in this special case can be found in [36]. The design procedure is summarized as follows: Step 1. Find the poles of Go s (the roots of D(s)s) outside the ; desired region and name its total number as P Go Step 2. Draw the modied Nyquist plot of Go s, count the number of clockwise encirclements with respect to the , and obtain the number of poles of 1 + j0 point as N Go Gs outside the desired region as P N P ; G Go Go Step 3. Draw the modied Nyquist plot of Gs and nd the range of KP during which the clockwise encirclements with 1 ; 0 is 2 P . respect to the K G P We provide Example 2 here to illustrate the design procedure in detail. Example 2 Consider a highly oscillatory process,

0.2

10

15

Time (sec)
Fig. 4. Closed-loop step response for Example 1.

would expect the proposed method always works better than the approximate dominant pole placement method.

4. Nyquist plot method If G(s) has time-delay, so will be Gs. Then, drawing the root locus for it could be difcult and checking locations of innite poles is a forbidden task. Hence we propose the Nyquist plot based method since the Nyquist plot works well for time-delay systems. The Nyquist stability criterion determines the number of unstable closed-loop poles based on the Nyquist plot and the open-loop unstable poles. We use the same idea but have to modify the conventional Nyquist contour. The Modied Nyquist contour is obtained by shifting the conventional Nyquist contour to the left by ma, as Fig. 2 shows. The image of G(s) when s traverses the modied Nyquist contour is called the modied Nyquist plot. Hence the number of poles located outside the desired region plays the same role as that of unstable poles in the standard Nyquist criterion. Rewrite (5) as

Gs

1 e0:1s : s2 s 5

If the overshoot is to be not more than 10% and the settling time to be less than 15 s, the dominant poles are q1,2 = 0.2751 0.3754j. Eq. (3) becomes

K I 0:3937K P 1:8773; K D 1:8173K P 7:7760:

We have

1 Gs 0: KP

Go s 7

7:776s2 1:877 0:1s : e ss2 s 5

Eq. (7) always has q1,2 as its two roots by our construction. These two poles lie outside the desired region (shown in Fig. 2). We want

Take m = 3. We have ma = 0.8253 and all three poles of Go s outside 3. Fig. 5 is the modied Nyquist plot of the desired region and P Go Go s and there is one anti-clockwise encirclement of the point

352

Q.-G. Wang et al. / Journal of Process Control 19 (2009) 349352

8 6 4

1.4

1.2
Overshoot: 10%

1
Settling Time: 13s

Imaginary axis

2 0 2 0.4 4 6 8 14 0.2

Output

0.8

0.6

12

10

0 0

10

15

20

25

30

Real axis
Fig. 5. Modied Nyquist plot of Go for Example 2.

Time (sec)
Fig. 7. Closed-loop step response for Example 2.

0.4 0.3

ications are met. The approximate dominant pole placement method in [2] gives an unstable closed-loop system in this case.

5. Conclusion
0.2

Imaginary axis

0.1 0 0.1 0.2 0.3 0.4 0.4

0.3

0.2

0.1

0.1

0.2

0.3

Real axis
Fig. 6. Modied Nyquist plot of G for Example 2.

Two simple yet effective methods have been presented for guaranteed dominant pole placement by PID controllers, based on root locus and Nyquist plot, respectively. Each method is demonstrated with examples. It should be noted that due to the closed-loop zeros and other effects the specications cannot always be met exactly. Obviously it is the limitation of the system, not the problem of the proposed method. In that case one can try changing m or loosing the specications and re-do the design. Obviously, the methods are not limited to PID controllers. They can be extended to other controllers where one controller parameter is used as the variable gain and all other parameters are solved in terms of this gain to meet the xed pole requirements. Extension of the method to multi-loop PID controllers is possible, but it would create an essentially different problem. In terms of controller parameters, the MIMO case is a nonlinear problem whereas the SISO case discussed in the paper is a linear problem. References
[1] P. Persson, K.J. Astrom, Dominant pole design a unied view of PID controller tuning, in: L. Dugard, M. MSaad, I.D. Landau (Eds.), Adaptive Systems in Control and Signal Processing 1992: Selected Papers from the Fourth IFAC Symposium, Grenoble, France, 13 July 1992, Pergamon Press, Oxford, 1993, pp. 377382. [2] K.J. Astrom, T. Hagglund, PID Controllers Theory Design and Tuning, second ed., Instrument Society of America, Research Triangle Park, North Caorlina, 1995. [3] G.F. Franklin, J.D. Powell, A. Emami-Naeini, Feedback Control of Dynamic Systems, Pearson Prentice Hall, Upper Saddle River, New Jersey, 1995. [4] S. Thompson, Control System Engineering and Design, Harlow, Essex, England, 1989. [5] Charles L. Phillips, Royce D. Harbor, Feedback Control Systems, Englewood Cliffs, New Jersey, 1994. [6] P.K. Stevens, A Nyquist criterion for distributed systems subject to a oneparameter family of feedback compensators, IEEE Transaction on Automatic Control 27 (3) (1982) 700702.

(1,0), that is, N 1. Therefore, Gs has two poles located in the Go N P 2. It means the modied desired region since P G Go Go Nyquist plot of Gs should have its clockwise encirclement with 0, that is zero net respect to the point (1/KP,0), equal to 2 P G encirclement, for two assigned poles to dominate all others. Fig. 6 shows the modied Nyquist plot of Gs, from which 1/KP 2 (1, 0.2851) is determined to have zero clockwise encirclement. A positive KP could always make KD and KI positive. Therefore, we have the joint solution as KP 2 (0,3.5075). If KP = 1 is chosen, the PID controller is

C s 1

2:2709 9:5933s: s

The zeros of the closed-loop system are at s = 0.0521 0.4837j. Fig. 7 shows the step response of the closed-loop system. The spec-