Editorial - IEEE Control Systems Society

Editorial ~~~~~

-

Charles J. Herget Lawrence Livermore National Laboratory

Alan J. Laub University of Southern California Computer-Aided Control System Design (CACSD) has begun to emerge as anindispensible tool for the control system engineer. A CACSD capability, not only frees the engineer from routine and mundane tasks but also provides a vehicle whereby complex algorithms or control methodologies are made available to and usable by those unfamiliar with the myriad of details that make the CACSD software efficient. A good CACSD system draws on expertise from many disciplines including aspects of computerscience, computer engineering, applied mathematics (for example, nunerical analysis and optimization). as well as control system engineering. The need for such breadth is partially responsible for the paucity of high quality CACSD software today. One wayof fostering a more mature CACSD discipline is to hold a number of workshops focused on some of the more pertinent topics. Such a workshop was held at the University of California, Berkeley. in April 1982, and various aspects of that meeting are being highlighted in this special issue of the CS Magazine. The Berkeley Workshop was preceded by one held in May I98 I in Schenectady and Troy. New York. sponsored by General Electric and Rensselaer Polytechnic Institute: see the article by H. A. Spang 111 and H. Emrick in the March 1982 issue of CSM for further details. Out of that first workshop emergedtwo ad hoc working groups: a working group on algorithms with A. J. Laub as Chairman, and a working group on design with C.J.Heget as Chairman.

Formal recognition of thesead hocworking groups was sought through the IEFE.A Technical Committee on Computer Aided Control Designwas proposed to theAdministrative Committee of the Control Systems Society in June 198 1. The Technical Committee was recommended by the Information Dissemination Committee of the Society and accepted by the Administrative Committee; H. A. Spang was appointed Chairman, andthe two working groups became subcommittees of the Technical Committee. A Program Committee consisting of C. J.Herget (Chairman). A. J. Laub, "

, !

and E. Polak (and later D. Q. Mayne) ;hen proposed the Berkeley Workshop atthe December 1981 Administrative Committee meeting. The workshop was approved and the program for that workshop appears elsewhere in this issue. A motion was subsequently passed atthe June 1982 Ad Com meeting which established a Steering Committee consisting of C. J. Heget. A.J. Laub. and H. A. Spang to administer and direct further "Symposia on CACSD" on a continuing basis under the auspices of the CS Society. The next such symposium is tentatively scheduled be held to September 28-30, 1983, at MIT. As part of the Berkeley Workshop it was deemed important and desirable to make some of the results of the meeting available to the CS membership through a special issue of the CS Magazine. No formal proceedings have been issued. Rather, we have of most of the presented copies abstracts,and a number of representative papersbasedontheworkshop presentations. We encourage the readers to contactindividualauthors directly should particular items in the issue be of special interest.

One of the things attempted at the Berkeley Workshop was to inject a strong computer science flavor into some of the sessions. We feel this aspect was quite well received, andan interchange of ideas resulted that might not otherwise have taken place. Indeed, there are many exciting new developments in computers and computer science that bear directly on CACSD. We could list, for example, work by committees developing E E E standards for things like floating-point arithmetic and graphics. new developments in languages (Fortran 77, Fortran 8X, Ada, ...), a veritable explosion in the use of mini- and micro- computers. and the possibilities opened up bynew architectures such as parallel processing through VLSI technology or multiprocessor environments. Not only are there many more relevant topics in computer science and engineering. but also there is much to be learned from CACSD methodoiogies in other engineering disciplines such as chem-

ical, civil, andmechanical engineering. The possibilities for future workshops are obviously enormous. Another feature of the Berkeley Workshop were the live presentations of design packages and computer graphics. Some of the demonstrations were run on computers at Berkeley, while most were linked to a computer at the speaker's home institution via telephone lines and modems. The audience was able to watch the presentations by using a television projection system to project the terminal's video output onto the auditorium screen. We now wish to give a brief overview of this special issue and take this opportunity to express our views on the current status of CACSD and make some recommendations for future actions. There are seven papers, four describing design packages (MAT R E x , LCAP2, DELIGHT, AND CLADP), one on computer graphics, one on solving ordinary differential equations, and one ona viewofthe future of CACSD. These papers are based on presentations at the workshop; summaries of most of the other presentations are also included. One section of this issue contains software summaries of hventythree design packages, most of which were not presented at the Berkeley Workshop. The publication of thesesummaries is the initial effort in a service the CSM intends to provide on a continuing basis. AII of the papers in this issue were reviewed through the usual process for regular papers in the CSM. That is, the reviewers had only the papers, not the sofhvare or the packages themselves. In these early stages of CACSD development, we feel it is important to make the community aware of what is going on even though standards for evaluating packages have not yet been established. At this time, CACSD packages have not achieved the kind of reliability enhanced by use as, for example, the EISPACK collection. But it would be a mistake to ever imagine that a "good" CACSD package will have the characteristics of an EISPACK or LINPACK. The scope of the problem is simply too immense. Certainly submodules of a CACSD package can have at least as good or better quality than these prototypical "good" mathematical software collections. But the notion of appropriate standards for a "good" CACSD package only now is evolving. All current CACSD packages have both good and bad features, and any author of an algorithm or package would like to assure us that his or hers is best. One has only to read, patiently. enough source code to discover

control systems magazine

what a longway we really have to go before suchclaimshavemore than a shred of reality. We urge our readers simply to maintain a healthy skepticism for anything and everything in this young and rapidly evolving field. We must all learn to become intelligent and discriminating users of software. And some of us will also learn how to set standards whereby future CACSD pack-

ages will be developed which can truly be used easily, efficiently, and confidently by a “non-expert“. To that end, we firmly believe that the only thing that will elevate us from an era of piecemeal, patchwork, sometimes half baked attempts will be an opportunity for long term funding at an appropriate level. A disciplined, broadly based, well coordinated team can produce tools for

CACSD of lasting quality. But exceptionally high quality software is exceptionally expensive-at least in the short run-in both timeand money. However, an economic argument for a major CACSD commitment can certainly be made, and we hope that a major effort can be undertaken which will involve both coordinated funding andcoordinated research.

CLADP: The Cambridge Linear Analysis and Design Programs J. M. Maciejowski and A. G. J. MacFarlane Control and Management Systems Division, Cambridge University Engineering Department, Mill Lane, Cambridge CB2 IRX, England Abstract An interactivesoftware facility for designing multivariablecontrol systems isdescribed.Thepaperdiscusses the desirable characteristics of such a facility, the particular capabilities of CLADP andthenumericalalgorithms whichlie behind them, and the probable course of future development.

1. Introduction The problem of cmting a feedback controller for a plant described in terms of a given dynamical model has three aspects, conventionally called analysis, synfhesis and design. In developingasynthesis technique, the aim is to formulate a desired objective as sharply-defined a mathematical problemhaving wella founded solution which is expressible in terms of a workable, efficient and robust computer algorithm. In principlethen, one loads the synthesis problem descriptioninto the computerand theanswer dulyemerges. The disadvantages of a purely synthetic approach to designare obvious in an engineering contextsince the role of the designer, particularly the exercise ofhisintuitive judgment and skill, severely is reduced. An even greater drawback is that, at the beginning of his investigations, the designer simply may be unable to specify what he wants because he lacks information on what he willhavetopay, in engineeringterms, for the various aspects of desiredfinal system performance. In developing a design technique, one

seekstogive a practicing and experienced design engineer a set of manipulativeand interpretativetools which will enable himto build up, modify and assess a design put together on the basis of the physical reasoning within the guidelines laid down by his engineering experience. Thus,design inevitablyinvolves both analysis and synthesis and hence, in the development of design techniques, consideration of the way in which a designer interacts with the computer is vitally important. It isimperative to share the burden of workbetween computer and designer in such a way that each makes an appropriate contribution to the overall solution. In developing the Cambridge Linear Analysis and Design Programs (CLADP) the aims have been to: (i) allow the designer to fully deploy his intuition, skill and experience while still making an effective useofpowerfultheoretical tools; and (ii) to harnessthemanipulative power of the computer to minimize the level of detail with which the designer has to contend. Thedesignercommunicates with the computer through an interface. This allows him to interpret what the computer has done and to specify what he wishes it to do next. In general terms we will call anything which is presented to thedesigner by thecomputer, and which is relevantto thedesign process, an indicator. The designer must operate within

an appropriate conceptual framework, and any powerful interactive design package mustpresentthe designer with the fullsetofindicatorsrequired to specify his needs and interpret his results in thecontext of his conceptual framework. The computer is used for calculation, manipulationandoptimization. In any fully-developed interactive design packagethe“tuning” of controller parameters is best done by a systematic use of appropriate optimization techniques. Generally speaking, in the design process the designer will be doinganalysis and the computer will be doingsynthesis. That is to say, the computer will be used to solve a series of changing and restrictively-specified synthesis problems put to it by the designer as he works his way through a range of alternatives, among which he choosesonthegrounds of engineeringjudgment,as he travels towards his final design. Since the designerwill usually want to think in the mostphysical way possible about the complex issuesfacing him, a high premium is placed on developing a conceptualframework which makes the maximum use of his sparial intuition, and which is formulated as much as possible in geometric and topological terms. For this reason, heavy emphasis is placed in CLADP on generalized frequency-response methods. Generalized Nyquist diagrams and multivariable root-locus diagramsare used asindicatorsofstability. These are derived from frequencydependentcharacteristic decompositions

0272-1708/82/1200-0003$00.7501982 IEEE December 1982

3