Proceedings of the 1996 IEEE InternationalSpPosiUm on Computer-AidedControl System Design Dearborn, MI September 15-18,1998

lMl7 3:iO

The Guider system for classical controller design.

A. L. Stevens arid B. Wigdorowitz Department of Electrical Engineering The University of the Witwatersrand, Johannesburg Email: [email protected]
Abstract
in terms of its specifications. The majority of controllers in the world Loday are of the PID type, usually because the process models are either unavailable or only vaguely known. For this reason, even the most advanced design tools are not much use unless the problem is posed in a way that matches the particular design technique. For example, if transfer function models with their associated polynomial coefficient variation ranges are unavailable, QFT gain-phase shaping cannot be used, regardless of how good the chosen method may be. The performance of ia control system depends on the shape in both magnitude and phase of the loop transmission L(s) and, ideally, if arbitrary loop tranmission shapes could be obtained, controller design would be quite easy. Unfortuna.tely this is not the case. However, there do exist powerful loop shaping tools for detailed shaping of L ( s ) for cases where the specifications are demanding. Classical control system design only requires simple loop transmission sha.ping in a single degree-of-freedom configuration and Guider provides a computer guided frequency domain based design methodology for doing this. More advanced loop shaping, such as is required in demanding aerospace control systems, can be accomplished using more powerful techniques such as Optimal QFT loop shaping [5]. In this paper, classical design will be discussed. (category 1 in [l]) Much of modern control systems education has lead to a de-emphasis of some important classical control concepts such as gain crossover slopes and actuator saturation due to sensor noise. Modern optimal controller design tries to pose the design problem as a specification consisting of weighting matrices and the links to classical specifications, such as overshoot, are not obvious to most practical designers. Also, design is by its very nature an iterative sequence of steps consisting of specification, analysis, synthesis and tradeoff assessment with a major component being that provided by human cognition. It is for reasons such as this that we believe that classical control system design is here to stay [9].

A classical control system design program for Bode plot loop shaping is presented. The design philosophy attempts to enforce the integration of a consistent and feasible specification with a classical design methodology. The design is divided into different categories for which suitable design procedures are provided by comparison of the specified and actual performance of the system. A number of tools for design evaluation, compensator tradeoff assessment, saturation and cost of feedback measures are included.
Keywords: Classical control, Computer guided design, Controller specification.

1. Introduction

Current CACSD programs address the problem of controller design, regarding the specifications as given. In this paper, a computer program to tackle the specifications as well as the design phases in classical controller design is presented. The program, called Guider, is designed to encourage the user to formulate consistent and feasible specifications before going on to the actual design tasks, so that time is not wasted trying to reach an unreachable design goal. The intention is deliberately to avoid a box-and-crank approach to design, in order that the designer be at all times aware of the design tradeoffs that are present during the design process. This approach, we believe, is particularly valuable when it comes to nonlinear control system design [6]. Once a set of specifications has been chosen and a performance analysis of the system has been completed, the program analyses the scenario and suggests a design strategy for the user to follow. We believe that an important factor in the selection of a controller design technique, is the amount of information available about the process to be controlled and the form it is in. With this philosophy in mind, a design hierarchy can be developed which tries to match the design technique with the way the problem is posed

0-7803-3032-3/96/$5.00 8 1996 IEEE

48 3

2. The performance specification module

The specification guidance module, discussed in detail in [l],attempts to provide assistance to the user (especially novice ones) in the formulation of a realistic and consistent classical specification set consisting of well known specification attributes. The specification attributes are phase margin ( a p ~ ) , gain margin (GM), closed loop magnitude peak ( A I m ) , closed loop magnitude peak frequency (wm), closed loop bandwidth, crossover frequency (wc), average gain gain crossover slope (V), crossover sensitivity (s,,,,), percent overshoot, time to first peak and settling time. These specification attributes are divided into amplitude and speed class variables and the user is required to choose a value for a variable in each class. Using a simple specification constraint network, an approximately consistent and feasible specification set is obtained as a guideline for the user. Certain warnings are issued, such as for example, when the user inadvertently specifies a zero ramp tracking error when it is not really needed for the application. Usually, this happens when the designer is not completely aware of the full implications of a particular action. In this case, the inclusion of two integrators, will result in the phase margin being reduced to zero, making the subsequent compensator design unnecessarily difficult. An important specification for a practical system is the maximum closed loop bandwidth that the system can support without input saturation of the process. This is allied to the performance measure called the cost of feedback [7] which is a measure of the likelyhood of sensor noise to cause saturation in the frequency region beyond we. A suitable transfer function representing this performance measure is the noise transmission from the sensor in the feedback path, to the process input U :

3. Analysis and design facilities

The program has the following analysis and design facilities:

0

A performance specification module tryspecs for generating an approximately consistent and feasible set of specifications. A systematic data gathering module for gathering detailed time and frequency-domain performance information about the system in open and closed loop, including sensitivity. Routines for plotting Bode plots of loop transmission, closed loop and sensitivity transfer functions; step responses and an algorithm for determining roots of polynomials. Design charts (charts1 and charts2) for first and second order lead-lag compensators and notch filters. These charts display phase shift, magnitude contribution and efficiency for various pole-zero separations. Compensator synthesis algorithms (ledlagl , ledlag2, notch) for determining the transfer functions of first and second order lead-lag compensators and notch filters. Useful calculation and comparison aids such as tabulated values of magnitude and phase for various compensator orders and types (compdat , t a b l e , pzldat, pz2dat) for use in loop shaping and compensator tradeoff evaluation. These are used in conjunction with the design charts to decide, for example, whether to use two first order lags or a single underdamped second order lag; to determine the compensator efficiency in terms of the amount of phase shift for a given magnitude contribution; or simply to quickly evaluate the effect in magnitude and phase of placing a pole or zero at a particular position in the s-plane. Data storage and retrieval options for system and specification data. A handy calculator for determining DC gains, magnitude and phase values at a given frequency and K , of transfer functions; conversions between arithmetic values and dBs, octaves and decades, crossover sensitivity and phase margin, pole/zero magnitude and phase contributions and polynomial multiplication. A design evaluation and guidance module which compares the actual and desired performance attributes of the system being designed and provides design guidance advice so that the user can perform a design in an iterative fashion.

where G(s) is the compensator transfer function, P ( s ) is the process transfer function and L(s) = G ( s ) P ( s ) . In order to achieve a meaningful practical closed loop bandwidth specification therefore, the process input saturation level must be taken into consideration. Based on the saturation level, the specification module uses a simple frequency domain technique to determine 3 a realistic closed loop bandwidth [3, 4, 2 which is then used in determining other speed class specification values.

484

4. Design categories

where &PM is the specified phase margin and @ p is the actual phase margin. ~ the compensator design func3. Consult tions compdat , charts, l e d l a g i and ledlag2 to choose a phase lead compensator which has the smallest magnitude contribution at the frequency of the phase peak, Mpp in order to keep the cost of feedback[7] small.

A simple classical controller loop transmission shape

is characterised by three independent parameters (DC loop gain (La,), gain crossover frequency (w,) and gain margin (GM)) and two dependent ones, namely average gain crossover slope (=) and phase margin ( @ p ~ ) which are related according to

- 180 ~ dB/oct 15 The objective of classical design is to obtain a feasible shape for the loop transmission L ( s ) so that it passes through the specified points defined by the above parameters.
M=
@ p

4.2. Design category 2 Specified: Phase margin @ P M , zero step sse, w, to be preserved, IC, or ramp sse.
The design procedure is as follows:

Two distinct design flavours can be considered, namely crossover frequency priority or steady state error priority. In the former, w, is held fixed while Ldc and @ P M are adjusted. In SSE priority, Ld, is held fixed while w, is adjusted. Crossover frequency priority occurs when the original value of w, needs to be maintained, but @ P M is too low. The design procedure is to determine how much extra phase lead is required and then using the tables and compensator design functions, choose a compensator.

1. Correct the system type as advised by the function f i x s s e . 2. The DC boost required to meet the Is', specification as calculated by f i x s s e . 3. Consult the compensator design functions compdat , charts, l e d l a g i and ledlag2 to choose a compensator which has a magnitude contribution Mpp closest to the value of ICBdc at the frequency of the phase peak. 4.3. Design category 3 Specified: Phase margin @ p ~ zero step sse, w e , to be increased to a given value. Not specified: K , or ramp sse.
There are two design options in this case. The first design procedure is as follows:

For simple loop shaping, either La, or w, can be modified without the need for complicated compensators. Roughly speaking, a three step design procedure can be followed:
Step 1: Correct the system type if required. This must necessarily be done first as the addition of an integrator will result in a 'structural change' to the process. Step 2: Adjust Ldc and we. There are seven possibilities. Six where one parameter is either too high or too low and one where they are both correct. Step 3: Adjust the gain and phase margins. 4.1. Design category 1
0

1. If the magnitude slope below OdB is constant down to -D dB, then increase the DC gain by D dB so that the desired value of gain crossover 6, is approached. 2. Add a unity gain compensator to meet the phase margin specification.
This approach will only be satisfactory in cases where the cost of feedback is not important. The second design procedure is as follows:
1. The magnitude LN and phase @N of L ( j w ) at the new crossover frequency G, is determined.

Specified: Phase margin @ P M (too low), zero step sse specified, w, to be preserved (i.e. satisfactory). Not specified: IC,, or ramp sse.

The design procedure is as follows:

1. Correct the system type (if necessary) as advised by the function f i x s s e . This must be done first as the system type may change. 2. From the values of the actual phase margin and the desired phase margin specified, the program determines the amount of phase lead boost that must be applied at the gain crossover frequency
We

2. The amount of phase boost @ B requir_ed to meet ~ the desired phase margin specification @ p is calculated according to the equation

@peak

= @PM - @PM
48 5

3. Consult the compensator design functions and charts to choose a compensator which gives the required phase boost @ E , but which also has a value of Mppwhich is closest to L N .

5. Choosing a compensator

6. C o m p e n s a t o r design c h a r t s

For a given phase shift, a number of compensator choices are possible. For example, a first order lead compensator with transfer function

where R is given by

P R=z

The maximum phase contribution Q , occurs at the centre frequency which is given by the geometric mean of the pole and zero frequencies
Wcf

In order to select the most appropriate compensator, the Guider system offers the user the facility to graphically view the magnitude and phase contributions of first and second order compensators in the form of design charts. These are useful in that they enable the user to consider design tradeoffs between different compensators. Each chart shows two plots; one of the phase shift versus R and the other, the ratio of the phase to magnitude contribution versus R. This can be regarded as a measure of compensator efficiency. In the context of phase compensation, a high efficiency compensator is regarded as one which gives a large phase shift for a given magnitude contribution for a given complexity. Compensators with underdamped zeros, for example, give large sudden phase shifts over a narrow frequency range, so there is a tradeoff between efficiency and damping factor.

=*
.

The phase and magnitude contributions at w e f are

Qmax

= arcsin [ R + 11
n = -1 if

R-1

where n = 1 if R

> 1 and

R < 1. and
7 C o m p e n s a t o r synthesis options .
There are three facilities for the synthesis of first and second order lag and lead compensators and second order notch filters. These options are l e d l a g i and ledlag2 for first and second order lead or lag compensators or notch for a second order notch filter. All of these options return the numerator and denominator polynomial coefficients of the transfer function. The first order compensators have a transfer function given by

Mcf = 10 log R
For a second order compensator, the transfer function is

and R is defined as

R=

2
Wz

the corresponding magnitude and phase relationships are

2CJZ 180 - 2 arctan IR-111

I

IJ

Rfl

where n = 1 for R

> 1 and n = -1
M c j = 20logR

for R

< 1 and

and the second order compensators have a transfer function given by

The functions compdat, t a b l e , pzldat, charti, chart2 and pz2dat are useful tools for deciding on a suitable compensator for a design problem as well as in general loopshaping tasks using poles and zeros. The function compdat enables the user to do a comparison of some first and second order compensator options for a given phase value. The function t a b l e gives a quick guide to the magnitude and phase effect of a pole or zero at frequencies above and below the pole or zero frequency position. The functions pzldat and pz2dat give the actual numerical value of the magnitude and phase effect of a pole or zero at a requested frequency for first and second order compensators.

The notch filter has a transfer function given by

All of these program options are interactive in that the user is prompted for the desired compensator parameters. In the case of the lead/lag compensators, the maximum phase contribution GmaZand the corresponding frequency wcf where this phase contribution is to occur, is requested. In the second order case, the damping factor is also required. In addition to returning the transfer function details, the magnitude contribution M c f in dB at wcf is displayed. The compensator efficiency can be evaluated and compared by calling compdat .

486

To calculate the pole-zero separation ratio R from the

specified phase value, the formula

the user to make a value judgement about the quality of the design.

9. Conclusions

is used for first order compensators and

R=
with

b+dm
2

a = t a n [ 180 -

and

A simple classical controller design program containing a specification guidance facility, compensator selection charts and transfer function synthesis tools has been discussed. While the program is still in the early stages of development, we believe that it represents a useful practical controller design tool for teaching classical control system design ais well as for use by practicing engineers.
References

for second order compensators. The notch filter option requests the user to specify the required depth of the notch M N in dB and the centre frequency wcf and returns the transfer function with ( w 1/MN.

8. Controller design using Guider The performance specification module tryspecs is first called and a specification set is generated as a design goal. At this point, the user is able to specify a saturation level for the process, in order to calculate a practically feasible closed loop bandwidth. Next, the program gathers the following system performance data from the user: Plant transfer function, poles and zeros, step response overshoot, rise and settling times, gain and phase margins, step and ramp steady state errors, gain and phase crossover frequencies, closed loop bandwidth, closed loop peaking amplitude and frequency, gain crossover slope and crossover sensitivity. In addition to this numerical data, a number of simple questions regarding the shape of the open, closed loop, sensitivity and step response curves is posed, so that narrowing of the search for a suitable design recipe can take place when the design evaluation and guidance modules are called. After the actual and specified performance characteristics are compared by the performance evaluation module, the design guidance phase takes place. The first step is for the system to identify which design category matches the given scenario. Next the system type is adjusted by fixsse if required and the text for the identified design recipe is displayed. Finally, the design is re-evaluated against the specificatlons and the process is repeated. During the performance evaluation stage, the Bode plot of the cost of feedback transfer function is plotted for

[l] Stevens A. L . , Towards a more Holistic Approach to Computer Aided Control System design. Proceedings of the 2nd European Control Conference, Groningen, The Netherlands, pp 1560-1563, June 1993. [2] Biernson G. A., HOW bandwidth of a servo the affects its saturated reslponse. IEEE Transactions on Automatic Control, pp 3-14, March 1958. [3] Klopper A. R. and Stevens A. L., Performance specification issues for a class of saturating systems. Transactions of the SAIEE, vol. 87, no. 1, pp 15-25, 1996. [4] Lozier J . C., On feedback control systems with saturable drives and noisy measurements. IEEE Transactions on ,4utomatic Control, vol. AC-26, pp 321-331, April 1981. [5] Ridley J . N. and Stevens A. L., Optimal Compensator Design in QFr. EOARD Special Contr<actReport SPC-95-4032, April 1996. [6] Wigdorowitz B. and Stevens A. L., A General Methodology for the Analysis and Design of Controllers for Nonlinear Processes. Transactions of the SAIEE, vol. 86, no. 3, pp 125-136, 1995. [7] Horowitz I. M., Optimum loop transfer functioq in single-loop minimumi-phase feedback systems. International Journal of Control, vol. 18, no. 1, pp 97113, 1973. [8] Tiplady K. and Stevens A. L., A Computerbased tutoring system for Loop Shaping on a Bode plot7, SE4 Third Biennial Conference, Natal, June 1993. [9] Stevens A. L., Is classical control theory obsolete? SAIEE Elektron, May 1995.

48 7