Report R E S U M E S

REPORT R E S U M E S

ED 011 862 SE 001 762 PROCEEDINGS ON THE CONFERENCE ON THE IMPACT OF COMPUTERS ON EDUCATION IN ENGINEERING DESIGN (CHICAGO, APRIL.2123,1966), B Y- FENVES STEVEN J. AND OTHERS COMMISSION ON ENGINEERING EDUC., WASHINGTON, D.C. PUB DATE 66 EORS PRICE MFS0.27 HC -$7.20 160P.

D ESCRIPTORS- *CONFERENCE REPORTS, *CURRICULUM DEVELOPMENT, *ENGINEERING EDUCATION, *TECHNICAL EDUCATION, COMPUTER ASSISTED INSTRUCTION, EDUCATIONAL TECHNOLOGY, INSTRUCTION, INFORMATION RETRIEVAL, SCIENCE EDUCATION, NATIONAL SCIENCE FOUNDATION, CHICAGO, DISTRICT OF COLUMBIA

PAPERS PRESENTED AT A 1966 CONFERENCE ON ENGINEERING EDUCATION ARE INCLUDED IN THIS CONFERENCE REPORT. THE CONFERENCE WAS PRIMARILY CONCERNED WITH THE IMPACT OF COMPUTERS ON EDUCATION IN ENGINEERING DESIGN. PARTICIPANTS WERE FACULTY MEMBERS FROM SCHOOLS OF ENGINEERING, WO HAVE THE RESPONSIBILITY FOR COURSE AND CURRICULUM DESIGN..A LONG -RANGE OBJECTIVE WAS TO SET DIRECTIONS FOR ENGINEERING EDUCATION AND RESEARCH IN DESIGN AND ANALYSIS, WHEREVER THE LATTER WAS EXPECTED TO CHANGE UNDER THE IMPACT OF COMPUTER CONCEPTS. SESSIONS OF THE CONFERENCE WERE DEVOTED TO (1) APPLICATIONS IN ENGINEERING DESIGN,(2) COMPUTERS IN ENGINEERING DESIGN AT UNIVERSITIES, (3) A MANUFACTURERS FORUM, AND (4) FUTURE PROSPECTS AND REQUIREMENTS. SUMMARIES OF DISCUSSIONS THAT FOLLOWED PRESENTATIONS OF PAPERS ARE INCLUDED. (AG) THE COMMISSION ON ENGINEERING EDUCATION NFVJ 1-1ANIPcsHIPF AVFNIIc- N Vv 1A/AL_uIr. T rl u r \ 4 , I .-c , I, / - 't k' _ 1 r

1 1 1

1 1 ' " I

' 1 1 1 , I 1 ' , 1 I 1 1 ' 1

1 1 I I,

II A PRECEDING PAGEBLANK1101E1MM Program and Table ofContents

Section A OPENING SESSION CHAIRMAN: S. J. FENVES, Prof. of Civil Eng., University of Illinois, Urbana, Ill..

WELCOME N. A. PARICER, Vice President, University of Illinois at Chicago Circle, Chicago, Ill

CHARGE TO CONFERENCE W. L. EvEarrr, Dean of Engineering, University of Illinois, Urbana, Ill.

THE IMPACT OF COMPUTERS ON EDUCATION IN ENGINEERING DESIGN J. 0. R. LICICIADER, Cotwultant to the Director of Research, IBM Corp., Yorktown Heights, N Y , 3

SOME COMMENTS ON THE FUTURE PRACTICE OF ENGINEERING C. L. MILLER, Head, Dept. of Civil Eng., Massachusetts Institute of Technology, Cambridge, Mass 5 Section B APPLICATIONS IN ENGINEERING DESIGN CHAIRMAN: F. H. BEANIN, JR., Development Engineer, IBM Corp., Poughkeepsie, N. Y.

EFFECT OF 'COMPUTERS ON ELECTRONIC DESIGN G. L. BALDWIN, Head, Machine Aids Development Dept., Bell Tele- phone Labs., Murray Hill, N J 11

STRUCTURAL ANALYSIS BY FINITE ELEMENT METHODS M. C. C. BAMPTON, Structures Research Engineer, the Boeing Co., Seattle, Wash 17

APPLICATIONS OF COMPUTERS TO POWER SYSTEM ENGINEERING G. W. STAGG and A. H. PALMER, Head, Eng. Analysis and Computer Division, American Electric Power Co., New York, N. Y. 25 APPLICATIONS OF COMPUTERS IN HIGHWAY ENGINEERING PRACTICE T. F. Mow, Deputy Chief Highway Engineer, Illinois Division of Highways, Springfield, Ill 29 COMPUTERS AND ENGINEERING DESIGN IN INDUSTRY D. Haw, Head, Computer Techsology, General Motors Research Labs., Warren, Mich. 84 Section C THE UNIVERSITY OF MICHIGAN PROJECT ON COMPUTERS IN EN- GINEERING DESIGN CHAIRMAN: B. CARNAHAN, Asst. Prof. of Chemical Eng., University of Michigan, Ann Arbor, Mich.

COMPUTERS IN ENGINEERING DESIGN: D. L. ICATZ, Prof. of Chemical Eng., University of Michigan, Ann Arbor, Mich

PREPARING THE UNDERGRADUATE FOR COMPUTER-AIDED DE- SIGN C. IliscincE, Prof. of Mechanico2ing., Iowa State University, Ames, Iowa 49

THE SIMULATION SPECTRUM N. HAUSER, Prof. of Industrial Eng. and Management, Polytechnic Institute of Brooklyn, Brooklyn, N Y 54

OPTIMIZATION METHODSA REVIEW AND SOME EXAMPLE AP- PLICATIONS B. CARNAHAN 56

THE SYSTEM APPROACH TO COMPUTER-AIDED DESIGN: USE OF PROBLEM-ORIENTED LANGUAGES R. L. MOTARD, Assoc. Prof. of Chimical Eng., University of Houston, Houston, Texas 77

Section D COMPUTERS IN ENGINEERING DESIGN AT UNIVERSITIES CHAIRMAN:. L. A. Sonars JR., Prof. of Structures, Engineering Division, Case Institute of Technology, Cleveland, Ohio

COMPUTER-AIDED DESIGN A. B. ROSENSTEIN, Prof. of Eng., UCLA, Los Angeles, Calif... . 85 FROM PERSONAL ANALOG TO MIGHTY DIGITALNOVEL COM- PUTER PROGRAMS IN DESIGN J. B. REMICK, Prof. of Eng. and Director of Eng. Design Center, Case Institute of Technology, Cleveland, Ohio

THE IMPACT OF TIME=SHARING COMPUTERS IN EDUCATIONIN ENGINEERING DESIGN P. T. SHANNON, Assoc. Prof. of Eng., Thayer School ofEngineering, Dartmouth College, Hanover, N. H 93

THE NEED FOR COMPUTERS IN TEACHING AND RESEARCH H. IL PAYNTER, Prof. of Mechanical Eng., MassachusettsInstitute of Technology, Cambridge, Mass 106 iv Stv Irr`A:Zr'1- V",

A LANGUAGE PROCESSOR FOR INTRODUCTORY ENGINEERING COMPUTATION R. W. CONWAY, Prof. of Computer Science, Cornell University, Ithaca, N. Y 111

Section E MANUFACTURERS' FORUM CHAIRMAN: S. G. CAMPBELL, Assistant Vice President, Xerox Corporation, Rochester, N. Y. ROUND-TABLE PRESENTATION BY COMPUTER MANUFACTURERS ON PRODUCTS AND CONCEPTS PERTINENT TO THE USE OF COM- PUTERS IN ENGINEERING DESIGN

AUTOMATED DESIGN WITH HONEYWELL SYSTEMS R. R. BROWN, Manager-Design Automation, EDP Division, Minne- apolis- Honeywell

AUTOMATED TECHNICAL DOCUMENTATION SYSTEMS AND EN- GINEERING DESIGN PAUL. H. ROSENTHAL, Manager, Generalized Applications, Data Proc- essing Division, Univac 119

COMPUTER CONTROL COMPANY PRODUCTS AND THE ENGINEER- ING DESIGN ENVIRONMENT R. COWAN, Marketing Manager of Systems Software, Computer Control Co 123

THE COMPUTER AS A DESIGN TOOL J. WORTHINGTON, Manager-Scientific Systems Review, Data Process- ing Division, IBM 127

ANALOG AND HYBRID COMPUTERS IN ENGINEERING DESIGN G. MOSER, Director of Training, Electronic Associates, Inc..... R. LONERGAN, Manager, Product Programming Planning, EDP Divi- sion, RCA

RECENT DEVELOPMENTS IN ON-LINE, REAL-TIME, TIME SHARING AND MULTIPROGRAMMING COMPUTER CAPABILITIES A. M. ROSENBERG, Member of Technical Staff, Product Planning, Scientific Data Systems 133

GENERAL ELECTRIC REACTIVE DISPLAY SYSTEM M. T. S. LING, Consulting Engineer, Computer Graphics, General Electric Computer Equipment Dept 137

CONTROL DATA'S ACTIVITIES W BEILFUSS, Director, Computer Dept., Meiscon Corp., Subsidiary of Control Data Corp 141

CRT DISPLAYS IN ENGINEERING DESIGN M. FORD, Marketing Manager, Graphic Arts, Digital Equipment Corp. 147 +./

Section F PROSPECTS AND REQUIREMENTS .4 CHAIRMAN: N. M. NEWMARK, Head, Dept. of Civil Eng., University of Illinois, Urbana, Ill.

COMPUTERS IN ENGINEERING DESIGN VS. DESIGN OF COM- PUTERS L. A. ZADEH, Prof. of Electrical Eng., University of California, Berkeley, Calif 158

PROLEGOMENA TO A SYSTEMS ANALYSIS OF EDUCATION R. E. MACHOL, Head, Dept. of Systems Eng., University of Illinois at Chicago Circle, Chicago, Ill 158

EMBEDDING A DIGITAL COMPUTER IN A NEW CAMPUS AND SCHOOL OF ENGINEERING R. M. SAUNDERS, Dean of School of Eng., University of California, Irvine, Calif. 163

A PLAN: IMPROVING INTERDISCIPLINARY COMMUNICATION FOR ENGINEERING DESIGN B. Mum, Assoc. Prof. of Industrial Eng., University of Michigan, Ann Arbor, iTtich 166

CONCLUDNG REMARKS N. M. NEWMARK 174

Section G ATTENDANCE: CONFERENCE ON THE IMPACT OF COMPUTERS ON EDUCATION IN ENGINEERING DESIGN 175 PROCEEDINGS OF THE CONFERENCE ON THE IMPACT OF COMPUTERS ON EDUCATION IN ENGINEERING DESIGN

FOREWORD

The Commission on Engineering Education, The conference was intended to serve the faculty through its Advisory Committee on The Use of Com- members of engineering schools who are responsible puters and Mathematical Techniques in Engineering for the development of courses and curricula in en- Design, sponsored a conference at the Univer=sity gineering design. Its immediate objective was to of Illinois at Chicago Circle, Chicago, Illinois, April 21- 28,1966, dealing with the impact of computers on explore all the facets that intend on aign, design education in engineering design.The conference education, and computer technology. The long-range steering committee was composed of : objective was to set directions for engineering PROFESSOR STEVEN J. FENVES, Conference Chairman education and research both in design and analysis, Dept. of Civil Engineering wherever the latter is expected to change under the University of Illinois impact of computer concepts. DR. FRANKLIN H. BRANIN, JR. Data Systems Division IBM Corporation These proceedings comprise most of thepapers presented by the participants, as well as the discus- SULLIVAN A. CAMPBELL Xerox Corporation sions which followed in some cases. The conference program and a list of those who attended it are in- PROFESSOR BRICE CARNAHAN Dept. of Chemical and Metallurgical Engineering cluded. Univecsity of Michigan PROFESSOR SAMUEL E. SHAPIRO The conference was partially financed bya grant Department of Systems Engineering from the National Science Foundation to the Com- University of Illinois at Congress Circle mission on Engineering Education. 5!:,7% -g,

OPENING SESSION

SECTION A

CHAIRMAN: S. J. FENVES, Prof. of Civil Engineering University of Illinois w,riprorrirrWONV..1k.INIMPRW,,

eRECEDINGPAGE BLANKMUSLIM THE IMPACT OF COMPUTERS ONEDUCATION IN ENGINEERING DESIGN J. C. R. LICKLIDER Thomas J. Watson Research Center International Business Machines Corporation Yorktown, N. Y. (Summary)

The keynote is the rate of advance of thetech- spiration followed by a phase of clerical execution, nology that provides the medium and themeans the total effort involves hundreds of thousands of and constitutes in large part thecontext and the such phases. objectof engineering design.It is a high note. Thus creative engineering design requires close Several of the basic capabilities of informational interaction between the heuristic factor and the technology, such as capacity for storage and speed algorinmic factor.Because people are so greatly of processing, have been doublingevery two years. superior to machines in respect of the former and Over all, though undoubtedly not inevery sector machines are so greatly superior to people in respect of the technology, a high rate of changeseems likely of the latter, the requirement is essentiallya re- to persist throughout the universityyears of engi- quirement for close interaction betweenmen and neers who are now freshmen and well on into their computers. And that is precisely what is afforded postgraduate careers.Engineering education must by the new informational technology, which makes do its best to prepare them to expect, to master, to it possible: employ, and, indeed, to contribute to the development 1. To bring the time scale of man-computer inter- of the more effective ways of thinking and working action down to seconds,or even milliseconds, that the advanced technology will make possible. and thus to substitute for the old,ccarse con- At the present time, we see a majorwave of tech- catenation of human operations and machine nical advance about to breakupon the field of engi- operations a fine, tight interweaving in which neering design.It may be only the first ofa succes- even the smallest area contains both the man's sion of waves, or it may be the great main force heuristicandthemachine'salgorithmic that will shape the coming generation.All that is threads. sure is that the wave is massive and that its impact 2. To rescale the computer's contribution from will be great. uninterrupted execution of an ad hocmono- The essential nature of the creatingwave is lithic program prepared to solvean entire caught up by the current phrase, "close interaction." problem or carry out an overall task torespon- The process of engineering design involves inter- sive execution of generalpurpose atomic pro- play between creative and routine activities,be- grams prepared to place a broad repertoire of tween activities that are heuristic, suchas formula- fundamental information operations at thede- tion of hypotheses, selection of approaches,assess- signer's fingertips. ment of values, and activities thatare algorithmic, 3. To free the designer from the dilemma of hav- such as determination of costs and calculation of ing either to plunge forwardone creative step stresses.In design by classical methods, those two to another with only the support of hisown kinds of activity were insofaras possible separated engineering estimates and artistic intuitions from each other, the designer attending to the or to wait after each creative step for slow heuristic part himself and assigning to draftsmen, and laborious implementation and testing. clerks, and computers the part that he could lay The close-interaction approach to engineeringde- out step-by-step in explicitly detailed procedure. sign depends upon synergic use ofnew hardware, That approach was successful in inverse proportion new software, and new techniques.In this con- to the challenge of the problem, for only when the ference, we shall examine all three. design was confined to preformed channels of solu- In the field of hardware, the mainareas are tion could anything like a clean separation of the storage, processing, transmission, and input-output. algorithmic from the heuristic be achieved.It be- In storage, the most dramatic advanceis the advance came obvious, indeed, that in truly creative engi- up the scale of capacity.Stores capable of holding neering design the two kinds of activity guide and a million analog images or a trillion binary digits support each other in intimate interplay and that, and affording access in a few secondsto any image if there is such a thing as a phase of creative in- or any bit are in advanced stages of. development. 8 The trillion-bit digital stores seem especially signif- computer and communication industries, and several icant to me because they will make it possible to governmental and government-related non-profit maintain vast libraries of engineering information organizations. Only two years ago, a large fraction in readiness for rapid access and for processing of the progress in this area could be subsumed by under program control. citing "Sketchpad" and other programs at Massa- In processing, the key advance is multipleaccess chusetts Institute of Technology in Cambridge and through time-sharing and multiprocessing, with it the Lincoln Laboratory, the DAC-I Project at memory protection, dynamic relocation, and related General Motors, and related work at Bolt Beranek features that make it possible for a teamor com- and Newman, System Development Corporation, munity of engineers to use extensive (andexpen- and very few other other places. In the interim, the sive) facilities in a coordinated and efficientway. field has burgeoned. In transmission, the main relevant movement is the Undtr "new methods and techniques" in the growth of digital transmission services and the phrase "new hardware, new software, andnew melding of digital transmission with digitalcompu- methods and techniques," I meant to includenew tationwhich will let the user work in or near ways of using on-line facilities for engineering de- his office with information stored at diverse remote sign.It is now well understood that the mainpro- locations and processed wherever the computersare. mise of the on-line approach is not merely to In input-output, the critical area of advance is that automate drafting, but no one I know thinks he of consoles for on-line man-computer interaction. yet sees clearly just how designers will use thenew Most of the basic capabilities are provided byone or facilities -or just what the process of design will another of the new-generation consoles. As consoles be within the new context. The new ways of work- move from their positions of relative obscurity in ing and thinking will evolve out of experimental and traditional computing systems to become the main pilot-operational use.One of the main responsi- foci of interaction between theusers and the facili- bilities of educators, therefore, is to incorporate such ties of on-line information systems, there is inten- use of on-line facilities into the process of education sive development aimed at improving the basic in engineering design.Students should learn the functions of display and control, integrating them new methods and techniques not so much by being more effectively for convenience and rapidity of taught them, as by contributing to their conception interaction, and making them widely available and and development. affordable. What beyond on-line interactive design? Perhaps it is not too soon to think about that. Myown best The software requirements imposed by engineer- guess is that, when many engineers are working ing design and releted multiple-access, interactive within the context of an on-line community, there applications are more basic and more difficult, in will be a marked acceleration in the understanding my assessment, than are the corresponding hard- of the heuristic aspects of design. More andmore ware requirements.The general requirement is of the exploration of partly random, partly controlled the "software base," the on-line library of widely variations on themes, for example, will be carried useful subprograms; the on-line languages andpro- out by computer programs, and designers will tend gram-processing programs through whichusers increasingly to become the strategists, rather than will call and apply the subprogram from the library; the conductors, of explorations and experiments. the on-line programming systems that will facilitate Designers will function also as evaluators, ofcourse, preparation, testing, modification, documentation, but perhaps even in that area they will become and communication of new programs and subpro- strategistsstudents of the processes through which grams, and on-line, controlled-access data banks. "time" performs its ultimate evaluations.In any Problems raised by those general software require- event, new possibilities and new opportunities will ments are being attacked vigorously now, and the arise, and (from the somewhat non-central butnone- most basic requirements are gradually being met theless interesting point of view from which Isee by research and development projects in universi- it) the future of engineering design looks both chal- ties, the automobile and aerospace industries, the lenging and exciting.

4 SOME COMMENTS ON THE FUTURE PRACTICE OF ENGINEERING C. L. MILL= Massachusetts Institute of Technology Cambridge, Mass. THE YOUNG ENGINEER-1976 THE NEXT DECADEINFORMATION The young engineers who will be 28 years old SYSTEMS in 1976 will be entering college as freshmen this As we look ahead to the next decade, there is fall.While 1976 sounds like a long way off to you ample evidence that it will be characterized as the and me, age 28 sounds quite young to all of us. Cer- decade of engineering information systems. What tainly we will want our 28-year-old engineers to be has been done in the past for engineering calcula- well prepared for their professional work.This tions will be done in the future for the whole area means that the formal education we give these of engineering communications, decision-making, entering freshmen during the next four to eight and systems design. We will be concerned with the years should reflect engineering as it will be prac- computer as a component in large - (scale engineering ticed in 1976, that far-off date. information systems which will encompass the total spectrum of information acquisition, processing, THE NEW TECHNOLOGY 1976 storage, retrieval, and communication. We will be concerned not only with computer technology, The new technology for engineering which is but also with information technology in the broad already available has been excellently described in sense of the word. the previous address by Dr. Licklider.With our concern for the 28-year-old engineer in 1976, we must project this new technology ahead for at least THE NEW CAPABILITYA NEW a decade of what is sure to be a period of accelerated PRACTICE progress. Information technology and the information sys- While no one can predict what the technology for tems of 1976 will represent new kinds of capabilities engineering will be in 1976, it will surely be much for engineering practice, not just improvements on more advanced than. that available for exploitation the old capabilities of past decades. To exploit the today.One conservative approach to projecting new capabilities will not be as simple as adding ahead to 1976 is to say that the most advanced of them to the old capabilities.It may take a rather the new technology which is just emerging in re- complete restructuring of the practice of engineer- search environments in 1966 will be commonplace ing and wholly new approaches to the execution in design and production environments in 1976. of the engineering process. It is reasonable to anticipate that the new tech- THE PAST DECADEELECTRONIC nology providing new capability will lead to a new CALCULATION practice of engineering. By a new practice is meant significant changes .in how engineering isac- The past decade might be characterized as the complished.The relative roles of men, machines, decade of electronic calculation in engineering. and organizations may be considerably different Since 1956, excellent progress has been made in the from that which we observe in engineering practice application of the computer for engineering calcu- of today. Who does what, when, where, and how lations and in the use of the computer as an aid in will change.Engineering will still be an art in engineering analysis.In engineering schools the 1976, but it will be quite a different art. focus has been on the teaching of elementary programming and an introduction to appropriate numerical methods.In addition, extensive use has THE NEW PRACTICEA NEW MAN been made of the computer in engineering schools If we are to have a new practice of engineering for research calculations.During the past decade, in 1976 based on new capabilities, it follows that interest in the potential role of the computer in we may need a new kind of man to exploit these engineering design has developed, and some progress new capabilities and practice the new art.By a has been made.Excellent examples of such prog- new man is meant one who is educated and trained ress will be presented during the balance of this for the new art and who has the talents andcompe- program. tence to use the new technology. 5 Engineering schools look ahead with respect to 5. All aspects of engineering design which can engineering science-oriented teaching and building be programmed will be programmed, includ- on the science-oriented research.However, there ing aspects of design which at this point in is a tendency to look backward with respect toen- time are not well structured or well under- gineering practice-oriented teaching and to do little stood, resulting in a significantly different de- or no practice-oriented research. The art of engi- sign process from what we know today. .neering tends to be taught on a "this is the way it is 6. Communications, in the broadestsense of the done, boys" description of current and past practice. word, will be at least as important as proces- If the practice of engineering is going to change in sing in the design system of 1976, and both the next decade, we have a responsibility to look will relegate calculation to a minor subset ahead and prepare these young men for the future, of total design activity.The conventional rather than for the soon-to-be-obsolete methods of engineering drawing per se willcease to be 1966. the prime communications medium for engineering design decisions. Dynamic communi- HEY NOTESBEY POINTS cations will replace static communications. A keynote address should presumably makea few 7. Design activity in an organization will be key notes, or key points, to stimulate discussion dur- integrated with other decision-making and ing a conference.It is not always clear just what action-initiating activities, and will becon- key points a keynote speaker is trying to make and, ducted interactively with the total environ- in order to bring mine into sharp focus, I will state ment and operations of the organization. them in summary form. My pointsare a combina- 8. The design systems for the late 70's have tion of speculation, opinion, and proposal.I trust not yet been fully designed, nor have the de- that you will not entirely agree with all those Iam signers to use them been designed. The initia- going to make because, ifyou do so, the keynote tive and leadership for accomplishing such address will not have served its basic purposeto designs should be taken by the engineering stimulate discussion. To avoid the remote possibility schools. that you might agree, I will deliberately statea 9. Engineering design education during the next somewhat extreme position which issure to be some- five years should be highly research-oriented. what controversial. You should bear in mind that I The best design education we can giveour am speculating on engineering design practiceas it students during the immediate years ahead might be in 1976 and not as it is in 1966. is to make them junior partners with the The ten key pocats I would like to makeare as faculty in conducting design research and follows : development 1. Engineering design will be accomplished by 10. Future formal education for engineering dean elite of gifted, brilliants creative, highly sign practice will require a studyprogram skilled, individual professional designers through the Ph. D., plus additional require- hereafter called "master designers." ments. The program should be highly selective and only those youngmen who are truly 2. The master designers will be concerned with gifted for engineering design should survive. generating design concepts and with deciding The design-oriented Ph. D. program for the between significant alternatives for broadly future professional designer shouldemerge defined design problems. as the highest attainment in engineering edu- 8. The master designers will work directlyand cation, involving all of the conventionalre- interactively with highly responsive design quirements for demonstrated researchcapa- systems based on powerful and large-scale bility,plus demonstrated originaldesign information systems, component of which capability. will be ultra high-speed computers, butthe THE PROTOTYPES heart of which will be information filesand While the ten key points representan extreme totally new kinds of software for working view, they stem from extrapolation to 1976 ofex- with these files. perimental systems already under development in 4. The engineering designprocess will be formal- 1966. In particular, they are heavily influenced by ized and organized intoa highly sophisticated current work on the development of the Integrated strategy for the optimumuse of man and Civil Engineering System (ICES) underway atthe machine.The total design system will in- MIT Civil Engineering Systems laboratory and volve a highly sophisticated machine,a highly the large-scale, time-sharing hardware/software sophisticated man, and a highly sophisticated developments underway in industry and theuni- theory of design and decision- making. versities, such as in MIT Project MAC, The1976

tr, ,

versions of these prototype systems,even if pro- Perhaps the most intriguing aspect of the chal- jected in a very conservativeway, provide the basis lenge we face is that there are a variety of ways to for the ten key points. pursue the challenge. Like all real design problems, The prototypes for the hardware/softwaresys- there is no one solution. There are many solutions. tems are under development.The evidence that Every school and every individual can participate in the technology will be available is quite clear. The the search for good designs. We are sure to hear evidence that a man will be available to use this and see evidence of this during the conference. All technology is not as clear.That the engineering of us have a role to play in designing the new prac- organizations will be available is even less clear. tice of engineering for the 28-year-old engineers There is very little evidence that organizations are of 1976. responding to the implications of such change by advance planning and preparation.However, it is anticipated that the prototype organizations will PAYNTER: Taking Miller's fourth and tenth points emerge in the next five years, and competition will I get the distinct impression that the person we take care of the balance in the following five years. are realty talking about bears a strong resem- blance to the person we have heard about in the RATE OF CHANGERESISTANCE past who was called the creative mathematician. Change of any kind is always resisted by a sub- I believe that creativity in basic science and set of the people and organizations which are affect- in mathematics is not as different evennow as ed. The subset includes those who are are already you seem to make it out to be, or hint that it may content but who fear that their relative position be, from creativity in engineering design. And I will be adversely affected.Engineering is no ex- definitely feel that, in the period ten and twenty ception.Many will adopt the attitude that the years from now, the man whom you will screen new capabilities are not needed.Historically, this out at this level of the Ph.D. and creative engi- has always been the case.(It wasn't too long ago neering design I la computer will be aman who that the log-table people argued that desk calculators simply has a tremendous, rational, creative gift. were not needed, and it was just a few years ago If you are talking about the strategy and the that many questioned the need for computers in development of strategy, certainly no discipline engineering calculationssome still do!) has done more incisive, introspective looking at There is no question but that the impending the strategy for creation of formal systems than change in engineering practice will be resisted in a mathematics itself.I would like to hear a variety of subtle and not so subtle ways. There is response to this. also no question but that there will be pressures for change. The only question is how the resistance MILLER: I don't know whether Ican give you a will affect the rate of change. The rate of change very eloquent response. I certainly will be willing is a major unknown at this time. It is quite possible to admit rather candidly that, as you know, we that we may indeed fall way short of the situation are trying to practice a little of what I've just described in the ten points by 1976 if the anticipated been preaching in our own effort of takinga resistance to change is sufficiently strong. responsible role and attempting to design and develop these new design systems. Some of the THE CHALLENGE most able and valuable young men workingon Our engineering schools can have considerable that project actually did their undergraduate work influence on the rate of change in engineering prac- in mathematics. The young men whoare ,making tice during the next decade. We can be an effective the most notable creative contribution to the de- source of resistance and slow down the rate of sign of the ICES system are people who havea change. Or we can accept the challenge of being great flair for mathematics, am& very abstract an effective source of leadership.The choice is mathematics as far as that goes. I'munsure of ours. This Conference on the Impact of Computers how far what you say is correct in terms of deci- on Education in Engineering Design is particularly don making. Even in 1978, witha very powerful timely as there will be an impact in the immediate design system of the type weare discussing; we years ahead. We have the opportunity now to ac- are going to be working with incomplete and cept the challenge of being the designers of that inaccurate information with problems for which impact and the instigators of change. The alterna- there are resource constraints, that is, limitson tive is to be followers and to face a future of the available amounts of money and time. Inthis responding to an impact which results from external whole area of compromise and trade-offs, I'mstill change.If we do not accept the challenge by con- betting rather heavily on theman to play a major scious decision, we decide on the alternative by de- role in making many of the judgment decisions, fault. but he will be assisted in these bysome very 7 sophisticated, very formal strategies behind the enormous amount more of design, ofa different scene represented in terms of the programming kind and a higher level therein, which will facilitate his than we did before. I do ability to make feel that the effectivedesign technician, while he what may in the final analysis bejudgment deci- will still exist and playa role,and there will be sions. What's going on inside will,in fact, be a relatively large numbers ofthem,will operate representation of a degree of mathematicalsophis- differently in terms of where tication that one would associate he fits into the pic- with the best of ture. But the basic pointis that the wholenew creative mathematics. Whatgoes on outside when technology is meaningless, he turns around and walks or at least it does not way from that con- achieve its full potential,unless we have the sole, I think, will still representcharacteristics equivalent of the that we associate with the engineer man I was labelling master de- today, at least signer to really master thetechnology that will our image of a good engineer, that is, his abilityto be available. make decisions in the face ofuncertainty and constraints and to dovery well in the major RESWICK: Itseems to me all the trouble started decisions he makes. back with the Tower ofBabel when people became unable to talk to each other. LICKLIDER: I would like to add Licklider a couple of per- talked about the problem ofdesign by teams, that haps divergent comments here.First, I largely is, by bringing various kinds agree with Paynter that creative mathematics of people together and to achieve solutions. He also hashad experience creative design, within the limitsof what I know with group dynamics. about it, have a lot in I am thinking of what common. Second, I want happens when an architect andan engineer get to suggest that if there isany thought in your together to work cooperatively, mind that designers will be rather to do creative a drug on the design or innovation.It seems tome that Lick- market because onlya few will be able to handle lider was saying that the all the jobs, I would disagree computer offers a solu- strenuously be- tion to the Tower of Babel,that everyone will cause it seems to me that the search, forsolutions have to talk computer language in problem space, and therefore will or the search for good and have to talk in thesame language, and therefore reliable synthesis, isan extremely difficult and will be able to understand each low probability pursuit. As other, and that out soon as we have the of this can come optimumcreative solutions. And apparatus, there are goingto be facilities for yet, what actually does happen when doing a lot of that kind of such a group thing. Then, I think, of people meets to doa creative job; and how does it will become importantto do very muchmore the best kind of idea than we now do, to come out of this group? carry planning whichnow, So many things happen whichare way beyond even at the international level, is about1 1/2-or the informationor the knowledge content of the 2-atep affair.Well, to carry designvery much people. What is the information farther, into working designs content of crea- and into planning, tivity in a group, and whencan a computer system will take all the intellectualresources there are. ever reach the capability of transferring all of One final thought is thatif we ever do startto the things that must saturate what one might call happen between people in modal design, that such a situation to reachsomething like an opti- is, the finding ofstraightforward solutions to mum decision? design problems, whichI don't thinkwe have come near yet, then there will bea great premium LICKLIDER: I certainly haveno adequate answer on achieving the odd-ball, but happilyvery good, or solution, but I would suggesta couple of things. designs, that you can'tget to along the standard First, I wouldn't hold thatall the people would channels, and that will readilytake a lot of intel- have to speak the computer'slanguage. I think lectual effort. the trend is for thecomputer to speak the languages of the various disciplines. MILLER: Let me follow The second up on Licklider's comment point would be that muchof the misunderstand- by noting that atno place did I say we needed ing between people from just a few of these master different disciplines is designers. By analogy, misunderstanding basedon the fact that much of in the last tenyears, with electronic computers their talk is merely naming, we didn't just chew up in and names are static. a relatively few milli- With the aid of thecomputer, one could substitute seconds all the calculations thatwe would nor- for the mere mally have done passage back and forth of static over many days.We simply messages, the demonstration of behaviors. Ifyou went ahead and dida lot more computing than see the behavior, even if it's just pictures of otherwise would have been possible. be- I think this havior on a scope face, it'sbetter than passing the is also true of design.Certainly we will doan words back and forth.

8 1

APPLICATIONS IN ENGINEERINGDESIGN

SECTION B

CHAIRMAN: F. H. BRANIN, JR., IBM Corporation, Poughkeepsie, N. Y.

ti #%-!41.,KAplif5i/P,rwt.,0,40'N,L.

PRECEDING PAGE BLANK- NOI FILMED

THE EFFECT OF COMPUTERSON ELECTRONIC DESIGN G. L. BALDWIN Bell Telephone Laboratories, Incorporated Murray Hill, N. J.

1. INTRODUCTION quired to realize these equivalent circuitpa- Over the past decade, the computer has beenplay- rameters? ing an ever-increasing part innearly all of our Characterization: What is the total performance activities at the Bell Telephone Laboratories.Since of the transformerwe have either selected or these activitiesspan a wide range of technical areas synthesized? from basic physical and mathematicalresearch Specification : What is the detailed information to the design of pushbuttons for telephonesets required for manufacture of this transformer ? it would be impossible to makea complete accounting Before the availability of the computer, these of all the ways in whichwe have found the computer basic functions were carried out manually by refer- to be a useful tool. We will taketwo examples of ence to catalogs, rough slide-rule calculations, labora- practical design problems in thecommunications tory tests and drafting. We willnow attempt to industry and givea brief history of how computer- show how, over theyears, increased use of the aided techniques have developedover the years in computer has gradually automated each of these each of these. Byso doing we will be able to give a functions. flavor of the manner in which the computerhas changed, and is continuing to change, the Analysis way we The first use of a computer in the design 0,o our work. The two examples themselvesserve process to illustrate the variety of our problems. The first is typically that of analysis: How willa given that of a communication transformer design device or system behave under various conditions of operation ? illustrates some of the problemswe face in selection The first step in performing such and design of components, while the secondthat analysis is to develop a mathematical model which of designing large systemsshows whatwe need suitably represents the physical situation.This to do in order to assemble hundreds of thousands requires a combination of intensive theoretical and of components into working systems. experimental studies.In some casesas in linear lumped-parameter filtersthe approximation isvery 2. COMMUNICATION TRANSFORMER DESIGN close, and one is able to make calculations with Quite naturally communications equipment is full confidence that they will be accurate withina small of all sorts of broadband transformers, each of fraction of a percent. In othersas in modeling the which must be designed to meet the particular cir- behavior of a typical modern central officeweare cuit requirements dictated by its environment. The unable to develop a model simple enough to handle history of the development of computer techniques on a computer which is, in turn, sufficiently realistic in this area is particularly enlightening forit shows to provide more than fairly gross approximations to how the computer has been used atmany levels, the system's real behavior.In our example of ranging from specific one-shot calculationsto a designing broadband transformers, the model is in highly complex set ofprograms which virtually the form of an equivalent circuit madeup of linear automates the entire process from design concept lumped-parameter elements. to the generation of detailed manufacturing infor- Next, one must developa method of relating the mation. values of the parameters of the mathematicalmodel The word design as we here apply it to trans- to the physical characteristics of the deviceor sys- formers is really a misnomer in that itencompasses tem being studied.Similar relationships between a good deal more than physical realization itself. the behavior of the model and that of the physical Also involved in the overall problemare the pro- system are needed in order to be able to interpret cesses of : the results of the study. Inour example the phy- Search : Can an existing transformer handle this sical characteristics of the transformerare core job? size and permeability, wire size, turns ratio,insu- Synthesis :From the given electrical circuitre- lation thickness and several others. Theparameters quirements, what are the parameters of the of the mathematical modelare the values of the equivalent circuit of the transformer? elements in the equivalent circuit,and in this Design: What are the physical quantitiesre- case the behavior of both the model and the device 11 is either the transient response or the frequency already have a transformer in stock which fills the characteristic curves, depending upon the trans- bill), the second generates the detailed manufactur- former's application.Since our emphasis here is ing information required by the factory, andthe more on the design process itself rather than on third allows his cut-and-tryprocess to take place broadbwi transformers, we will not go into detail more conveniently and far more rapidly. as to the specific equivalent circuit which was used.1 In the. first step of introducing the computer to Search the problem of design, we program it to accept phy- Once a suitable model of our transformer exists, then a specific transformercan be exactly described sical parameters of a given transformer, calculate by a set of numbers. the values of the elements in the equivalent circuit, If we have recorded all previously designed transformers in computermemory and then perform standard network calculationson that circuit. (typically on a magnetic tape "catalog"),we are This allows an engineer to selecta able to match the parameters of the desired trans- set of physical parameters, examine the resulting former against those for which designs already performance, and zero in on a satisfactory design by exist.Should any fall within suitable tolerance, repeated runs, varying the physical parameterson a cut-and-try basis. then the analysis program is brought into play and the designer cansee how well any existing Synthesis design fits his requirements.If there are any, that One would initially suppose that the problem of come close to fitting the requirements, they are synthesis would simply be a matter of reversing the so indicated and their complete characteristic curves problem of analysis, that is, sot ving the equations are presented to allow the engineer to see if one for the physical parameters in terms of the desired of them fits the bill.Otherwise, we go through characteristics. However, this is generally not the the synthesis phase (and, ofcourse, add this new case; there are only a small number of physical transformer and its characteristics to the catalog). parameters which can be altered andone usually Output has a larger number of external characteristic val- Once a design has been completed, the engineer ues to be satisfied. Thus a good deal more investiga- has (as a direct result of theprocess described tion, both experimental and theoretical, is needed above) a complete functional characterization of in order to determine what specificexternal mem"- the transformer- But the computernow "knows" urements are of greatest significance,so that the all of the physical quantities required for actual inverse solution can be made possibleat all. We manufacture.Therefore, a part of our growing are generally faced with a set of non-linear equations design system is the production of manufacturing which defies solutioneven with a computer at hand. information in a format which is directly acceptable If we are fortunate,some further analysis will pro- to the factory. duce a few gross relationshipswhich allow at Let us see what the computer has providedus least an approximate solution to theseequations. with so far.Our program system provides the Now we are faced with approximateanswers to a engineer with a significant saving in time, cost mathematical model which in itself isan approxi- and accuracy.The system, through its search mation of the actual device. The next step is utili- routine, will minimize the number ofnew designs. zation of the actual device. The next step in utiliza- If a new design isnecessary, the frequency band tion of the computer is to writea program which ac- of interest is centered, thereby providing greater cepts desired external characteristicsas input and possibility for multiple-use transformers. Thiscen- uses the rough approximations to estimate the tering procedure will tend to minimize overlap in physical parameters, providing input directly to designs; hence, a tendency toward fewer different the analysis program. designs. By filling out the transformer designre- This then allows the engineer tozero in on his quest, the engineer can have withina day's time final design with fewer computerruns.It also complete information on any transformer which shows us the start of a system ofprograms rather meets his needs and/or design information, if it is than of a single one-shot effort.The next three possible to produce such a transformer. stages of development in our transformer design Although we do indeed have a system ofpro- example are somewhat independentoneaids the grams, it was designed in such a fashion that part designer in his catalog search (after allwe may of it may be used independently.For example, both the synthesis-design and the analysis portions 1 For a treatment of the broadband transformer design have found extensive use as individualprograms. problem in greater detail, see "CommunicationTransformer Engineering byComputer,"by C. M. Bailey, F. J. Kasper Direct Interaction and D. Kati, in the Proceedings of the 1965 ElectronicsCom- About the time that our transformer select-and- ponents Conference. design system became operational,a completely new 12 facility was made availableto us at the Laboratories or so attemptseach depending heavily on the on an experimental basis. The present talk is not results of former trials. Using typical batch proc- intended as a description of that facility,for other essing procedures, this might stretch the overall talks in this session willgo more deeply into time between origination of the needs and realiza- similar developments.Briefly stated, this equip- tion of a satisfactory design to several days.Of ment, known as GRAPHIC-I, providesa graphical course, during this time period the engineer is input/output capability wherebya user can com- presumably off doing something else, but the turn- municate with a large computer viaa CRT display around time is still measured in days.Having a and a light pen.Physically it consists ofa CRT, directly interacting facility for entering data, view- a keyboard, a light pen, and a smaller computer ing results and enteringnew data can reduce this (a PDP5) withan added memory, plus hardware to a matter of minutes. which allows data transmission to and from the While we are on the subject of electroniccom- 7094. It differs from other similar devices (such ponents, we might profit by looking into another as SKETCHPAD and the General Motors DAC application of graphic consoles. A large part of System) primarily in that the local computer and what we generally refer toas design consists of its added memory providea good deal of stand- looking through catalog information fora com- alone capability. Thus communication to andfrom ponent which meets various combinations ofre- the 7094 tends to take place relatively infrequently quirements,includingelectricalandphysical but with fairly large chunks of data handled in characteristics,costs,reliability, and numerous one interaction.GRAPHIC-I is really an experi- others.Although we have not yet builtup a file, mental vehicle leading to the design of GRAPHIC- we have developed a structure for such an applica- II, which will be used to provide graphicalcom- tion and have played with it enough to be convinced munication with our GE 645 system abouta year that this will be of great value.What we are from now.Its major limitation is one basedon really doing is searchinga huge file structured in 7094 softwarewe didn't feel it worth whileon the form of a tree, and at each node asking the our particular time scale to provide capability for user which branch he wishes to take.Although interruption of the 7094on other than a between- the identical searching process is feasible usinga job basis. For this reason,one might have to wait typewriter as the input/output device, the CRT up to a few minutes to capture the 7094, but from light pen is far faster in displaying the information then until the transaction is completed, hecom- pertinent to each node, and it is far easier for the mands the 7094's full attention. user to point to something by a light pen than to So much for the equipment.How has this type his request out. changed our outlook on design in general, andon 3. SYSTEM DESIGN broadband transformer design in particular? Let's In addition to designing components review the basic method of designwe have used we are, of all along the way. Essentially it has been to course, interested in gathering these components guess into various levels of sub-assemblies whichare what the proper physical configuration would be, built up to form large systems. analyze this and then make anotherguess.The It is not at all un- computer's role has been partly clerical and partly usual today to discuss a system with nearlya million calculational. electrical components and hundreds of thousands of It has been clerical in that the pro- interconnecting wires. gram first looks in a catalog to see if a new design In this area, I am sorry to is really needed. say, we can't point to as neat and orderly a growth It has also been clerical in that of automation as in my first example, although there it "remembers" pastguesses and their consequences and in that it presents the finally agreed-on design have been a few notable approaches to complete design automation systems. in the proper form for manufacture. Thecalcula- tional roles of the computer have been When we look at system design, the first problem to perform that faces us is the sheer bulk of the data, together the analysis and to assist the engineerin making better-educated guesses. Now what the GRAPHIC with the complexity of the interrelationshipsamong facility provides is greater simplicity of subsets of the entire data.For this reason, most use (the design automation efforts aimed at large systems engineer no longer has to punch cards andassemble mysterious decks in theproper order) .* and an have first of all attacked the clerical, record-keeping order of magnitude in compression problems rather than the activities involvingsome of the time creativity. scale. A complex design might requirea dozen Automation of record-keeping is virtually mandatory here, not only from thepoint of 2 At present he has to learnother mysterious things like view of economy, but also in orderto provide the pushing "light buttons," waving the lightpens and the like. host of design engineers up-to-dateand consistent Our ultimate aim is to make the systemas simple as possible information as to the current statusof the system. for the user. Furthermore, one can ask questionsabout the ap- 18

,11111),M1111.

.,...711.11 t.,r; v.-Jr:a:

proximate cost of the entire system or of a sub- board, location of boards on a frame, or even loca- system, or find out how many times and where a tion of equipment frames in an office.Many at- particular component is used, and receive answers tempts have been made to do placement entirely economically and quickly.With manual methods automatically.One technique developed at Sandia of keeping records, either of these questions would involves establishing artificial force fields among the require so much thumbing through stock lists that components. 3Another developed at the Bell Lab- the answers would cost too much, and by the time oratories involves a rough placement of clumps of they were generated they might well be out of date. components, followed by an iterative scheme of in- It is appropriate here to point out one of the terchanging pairs of components and determining major problems we have encountered in using the the resulting effect.I know of other techniques and computer to document large systems. Especially in I am sure there are many more of which I have no engineering development efforts, the nature of the knowledge.Of one thing I am sure, however information we wish to keep is continually changing. that is, that none have been found to be entirely Furthermore, the various people involved with the satisfactory in a practical system design situation. job are interested in different subsets of the entire Let us observe that the previous paragraph dis- information.However, the data-processing soft- cussed a problem in automated design without men- ware we have available to us today renders changing tioning any criteria by which a given design can the format of a file very difficult, since it necessitates be evaluated. This is really the heart of the matter. changing each individual program which works with The most easily stated criterion is minimum total the file.Thus we are faced with a combination wire length, and even in this case no completely of two problems: a good deal of our programming automatic techniques have been found to solve prac- effort is wasted by trivial but time-consuming edit- tical problems. But the problem is far more serious ing to introduce changes in file or input/output in that most design engineers are unable to specify formats; and many potentially useful ideas are all of their criteria at all clearly.In some cases abandoned because of the cost involved in imple- crosstalk is important, so that there exists a limit menting them. on the distance over which a pair of wires can run A second area in which automated design of parallel to each other.In other cases, capacitive large systems has suffered stems from a heavy de- loading or delay are critical, and here one wishes pendence of most of them on the detailed nature to minimize the length of the longest lead in a cer- of the equipment being designed.That is to say, tain elites.Finally, esthetic considerations come things like the type of logic used and the size and into play, primarily as an aid to the maintenance terminal identification of sockets tend to become man. If we have a structure (say a shift register) so deeply ingrained in these programs that we al- which has a definite electrical pattern repeated many most have to write a complete new system of pro- times, it is desirable to have the corresponding phy- grams to handle a new line of equipment. sical pattern repeated similarly. So much for the negative side.There are a For two reasons, then, most of us have given dozen or so companies now designing and manufac- up on fully automatic placement, and have turned turing large electronic systems, and each has its own to use of graphic consoles operated by the design computer-based record-keeping system.Each of engineer himself.Here the computer is able to these provides to one degree or another a voice of keep accurate records and remember, and cause to conscience for the designer, checking his design be repealed, specific patterns; it can also evaluate a against various rules of fan-in, fan-out, impedance given layout with respect to any of a number of matching and the like.This, incidentally, has different criteria. We leave to the judgment of the proVed to be one of the most valuable of the ser- man, then, those criteria he cannot really explain vices provided by the record-keeping type of pro- explicitly. gram. I feel with a fair degree of confidence that Another closely related problem is one we refer the problems relating to software and standards to as treeing and routing of wires.Given the will soon be solved to the point where we will have electrical schematic and the position of each com- really flexible engineering data-processing systems. ponent, we know which terminals must be connected What now about the design of large systems? together.In this instance, completely automated There are some design problems which are common methods have been developed for determining the to nearly every large system, and some are unique proper order of connecting the terminals to give to certain specific technologies.In this review we minimum wire length, and also for finding a route will look at a few of the common ones. Perhaps the most worked-over problem in large "ACEL" Automated Circuit Card Etching Layout," system design is that of placement.This might C. J. Fisk, D. D. Isett, Proceedings of the. 1964 SHARE mean location of components on a printed circuit Design Automation Workshop. 14

Z^ for each wire such that no rules related to bunching HART: I would like to have your opinion about of wires are violated. knowing something about computing, but not Now let's look at the early stages of designing having any courses in it.Would you explain it a a large system. To my knowledge, each effort here little? has been of a one-shot nature.It is not unusual to BALDWIN: I think learning how to use a computer see two kinds of simulation used in the early stages is becoming easier and easier.I think that if a of development of a project.The first of these is student were to spend a lot of time learning For- a fairly gross overall representation of the system tran, Algol, Flugol, or what have you, and also to allow the designers to find out where the major the mechanism of coding, he would be wasting flows of information will occur and which are likely his time.There is a need for the high priests to be the most critical parts of the system in deter- of computing who will supply the eventual soft- mining its ultimate performance. The second type ware whereby the computer will talk people's of simulation is to take a small chunk of the system languages; but a man who is going to be a designer and simulate its behavior on a pulse-by-pulse basis. needs to know very little about computers. What As far as I know, even our fastest computers are he needs to know is a lot more basic mathematics too slow to allow a complete, detailed simulation and physics. of a sensibly large system. Indeed, even if this were CONWAY: You, of course, are aware that almost possible, it is hard to imagine what one would do all engineering students are now learning to solve with all of the numbers which would result. problems with a computer and understand how to In the area of detailed logic designwe again look use it. My observation is that if they don't learn it to the graphic console to be of great use. The day in school, once they get into industry it isvery is not far off when a designer can sit in front ofa difficult for them to learn it. console, draw chunks of circuitry, asking the com- BALDWIN: I would like to address myself to two puter to repeat given chunks when a repetitive points.First, today you need to teach some pro- circuit comes up.The machine will then check gramming because we have not reached the uto- the circuit against the design and manufacturing pian situation where it iseasy to use the com- rules appropriate to the particular technology and puter.But, to me, the reason you teach pro- tuck the information away for recall at any later gramming is so that studentscan use the com- time either by the designer, by one of his colleagues, puters in their other courses to develop a good or by programs used later on to generate the added fundamental education. information required to manufacture the system. Now, my particular job is hiring people who will become these high priests and supposedly 4. CONCLUSION make the computer easier to use, but we're mak- Computers have already become indispensable in ing it more and more difficult.The last machine the design of communications components and sys- I really understood was the IBM 650. But when tems. Today the uses are largely of a clerical na- I hire people, I do not care whetheror not they turereceiving information, massaging it gently, have had any programming experience in school, and printing it out in various forms. However, the and I run a programming department. -What I problems are rapidly being solved to the point where want is good, fundamental mathematics and we are looking to the design process itself.In some physics and a statement of interest in thispar- cases we are able to develop algorithms which solve ticular field, because peoplecan pick up pro- a design problem completely.In most cases, on gramming easily if they are going to be goodat the other hand, we look to the development of pro- it.They can pick it up certainly ina year on the grams involving direct communication between the job, and we're hiring these people for 40years. designer and the computer through on-line graphic A second point is that schools have machines consoles. which are far behind the ones thatare used in industry. So you get a fellow thatsays "Ah ha, ACKNOWLEDGMENTS I know how to program the Blubiac," and this Practically none of the work described here can doesn't do him any good at all because we'vegot I claim as my own. In particular, the use of gra- something else. phic consoles in general has been carried out by BRANDT: I'd like to makea comment on that point. several people at the Laboratories who are too I feel that once an individual has learnedto pro- numerous to mention.The use of these console gram in any machine language, or any language for transformer design and file searching has been at all, he has mastered the fundamental principles. done primarily by Messrs. P. G. Dowd and T. J. The translation to another language is nowhere O'Conner. near as difficult as learning the first process. 15 Fer,.."" "&";,_gpv-,,,,,-".

, u'AAr

BALDWIN: Maybe he learns to mistrustthe com- don't suppose that you must thencontinue with puter.This is indeed a lesson thatone must more how-to-do-it courses, but give enough train- learn. ing in programmingso that a student can ap- LICHLIDER: I'd like to make anti intermediatepro- proach a computer and solvea problem with some posal because I agree with you that there shouldn't insight into what this involves. be courses in computers which spend three weeks BALDWIN: One of my reactions isthat he can on binary notation and teach how to program learn the problem-solving procedure,which is in some particular machine language.1 also actually mostly problem formulationand not agree that people can learn pretty quickly on problem solving.Typically, you're reformulating the job, but I notice that the self-taught people the problem as you solve it, whetheryou usesc miss something very important.They haven't computer or not. But can't he do thisby standard really learned to understand aboutformalisms analytical methods? You don't thinkso. or the syntax of computer languages, and they don't have a set of the abstractions withoutwhich BRANIN: There is a strong point thatHart brings they cannot really makeany progress in the out, which is that the discipline ofprogramming modern, world of computers. So I wouldargue a problem forces you to learn it inmore detail that there is a course to be taughtabout com- and more thoroughly thanyou would otherwise, puters; maybe it isn't being taught anywhereyet, so I would agree with him on that. but there is some lore somewherebetween com- SHANNON: With the introduction ofour time- puters and mathematics that would be awfully sharing system, we have moved inexactly that good to have. direction. We moved the teachingof computer HART: It is very easy fora student to go through programming into the freshmanmathematics four years of engineering school andnever solve course, deleted a course in engineering in "In- a problem. I think one of the opportunities the troduction to Computation in NumericalMeth- computer provides is that the student, bygetting ods," and inserteda new course in "Field and involved in the sweat and tears ofwriting a Electricity and Magnetism." This is thecurrent program to solve a problem, develops the concept direction in which we're moving. of the problem-solving procedure, theidea of an BALDWIN: Well, that sounds good.I don't be- algorithm, and I thinkexposure to this early in lieve everything I say myself. I thinkcertainly his educational career is extremelyimportant. I this is a very serious problem ineducation.

16 STRUCTURAL ANALYSIS BY FINITE ELEMENT METHODS M. C. C. BAMPTON The Boeing Company Seattle, Wash.

I. INTRODUCTION they are much more capable of producing refined The intention of this paper is to discuss only one descriptions of complex structural behavior.This aspect of aeronautical design procedure which is paper will describe several typical applications and, influenced by the existence of digital computers. where possible, show the correlation between test This particular aspect is the analysis of structures. results and analysis. While limiting the technological scope of the dis- It is believed that there will continue to be pro- cussion, this topic provides an excellent illustration gress made in developing and applying finite element of the dramatic development of an analysis capa- methods of sizuctural analysis.It is also believed bility to cope with a very wide range of problems. that considerable progress will be observed in what Such progress would have been impossible without might be regarded as a lateral direction, that is, the associated development of digital computation the application of such methods to other technolo- machines. gical areas.This trend can already be evidenced During the last decade the aeronautical industry with the molecular vibration analysis by Kaufman has been mainly responsible for the introduction and and Kaufman, ,2the seepage analysis by Zienlde- establishment of finite methods.These methods wicz o and temperature analysis by Visser Is and have been denoted by various names depending Bampton. 14To lend weight to this contention, upon whether they have inherently regarded forces two applications of relevance to aircraft design or deformations as the unknowns of an analysis, are included in this paper. or whether the method has been expressed solely in Finally it can be said that, just as the finite ele- terms of matrix algebra or has implied certain comment methods are dependent upon the digital computing algorithms.These methods all possess one puter machines for their existence, so are they feature in common. That is, they depend upon the also dependent upon the appropriate education of process of idealizing the structure into an assembly engineers for their correct application, interpreta- of finite elements which are analytically connected tion and development. The overwhelming response at nodes or boundaries. to related courses being offered within the industry Examination of the literature will show that de- is a clear indication of the need for such education. velopment of the finite element method is continuing vigorously both within industry and at universities. II. FINITE ELEMENT METHODS Effort is being directed along two main avenues It is not the intention of this paper to discuss of development : improvement of the element and finite element methods with any thoroughness, since improvement of the system that accepts the element. this has been successfully accomplished elsewhere The former implies more precise representation of in the literature.14. 10.15The objective is to dwell technological behavior and the latter implies more upon applications and thus stimulate interest in the efficient computing procedures. methods.However, some statements concerning The aircraft structural problem which can be concepts should prove beneficial within this context. said to deserve the strongest claim to having Before proceeding to a finite element analysis, an prompted the development of finite element methods engineer must have at his command within the com- is the multi-spar, multi-rib swept wing. This struc- puter capability an array of mathematically defined ture possesses both complexity and a high degree of structural elements.The force-deflection relation- indeterminacy. A decade ago, such a problem was ships of each element must be explicitly expressed. a challenge to analysis; today, finite element methods An array of typical linear isothermal elements is have made its solution commonplace.It is also shown in Figure 1. The existence of such elements evident from reports that much experience has been will enable an engineer to idealize a structure into gained in applying such methods to other aircraft an assembly of such elements.An idealization structural analysis problems.These applications is shown conceptually in Figure 2 in which the plan of finite element methods reflect advantages over view of an aircraft swept wing is covered byan classical aircraft analysis procedures in that they analysis grid.The grid, in turn, implies the loca- reduce considerably the weight penalties associated tion of analysis nodes and finite elements. with making assumptions in idealization and that The forces and deflections of the elementsare 17 FLANGE BEAM

.1111111111111111 SHEAR PLATE PLATE

FIG. 1TYPICAL LINEAR ISOTHERMAL ELEMENTS

FIG. 3 BOEING MODEL 737 DESIGN STUDIES

structure, contains a formidable challenge to classical analysis procedures even when supported with empyrical data. Region 6 has similar features to those possessed by 1.Region 2, 4 and 5 are wing structure and inherently containan analysis problem due to their being swept. Fora two-spar structure =11.S 1 S2 this is the shear transfer of load from the front to the rear spar and the amountcan not be determined lC 712 it by classical methods unless supported by empyrical 03 IC==14S4 data. Figure 4, a region on the Boeing Model 727

FIG. 2 IDEALIZATION INTO FINITE ELEMENTS then expressed in a set of linear 'algebraic equations such that equilibrium and compatibility are satisfied at the nodes and boundaries of the structure. For a large complex structure, this leads to a large number of simultaneous equations.In realistic problems, many hundreds of equations can be involved. Thus it becomes clear that for an engineer to merely discuss his problem he must resort to the compact notation of matrix algebra, but that to solve it, he must rely upon the availability of a digital computer. III. APPLICATIONS OF THE FINITE ELEMENT METHODS FIG. 4 WING ROOT SPANWISE STRESSES 1. Airplane Design Studies Extensive application of finite element methods main wing, shows the excellent agreement between occurred during the design of the Boeing Model 787 stresses obtained by finite element analysis and Transport Airplane. A representative selection of stresses obtained during static tests on the airplane. the studies is shown in Figure 8 and their pertinent Regions 8, 7 and 8 of Figure 8 represent cut-out features will be described.In all cases, the com- analyses around doorways, windows andaccess putations were automatic and the engineering effort panels respectively. arose in connection with the interpretation of re- One major consideration in such analyses are the sults and preparation of data for idealization.It is boundary conditions for the regions of analysis. in the two latter phases that engineering judgment These are usually obtained from a prior grossan- must be exercised if an analysis is to be successful. alysis or by invoking Saint Venants Principle. Re- Referring to Figure 8, Region 1 represents an gion 9 involves ducting around the engines and analysis of the complex wing-fuselage intersection requires consideration of thermal effects.Region structure.This area, with its conglomeration of 10 represents an analysis that is small by compari- bulkheads, flexible frames, cut-outs, fuselage cross- son with those previously mentioned and yet con- section transitions and wing-fuselage interaction tains enough indeterminate elements to justify the_ 18 use of a finite element method. The problem con- particular values. A typical curve is shown in Fig- cerns interactions of floor beams and door sills. ure 7.The solution, which follows the approach In order to give some meaning to what might currently be considered a large analysis, one of the preliminary design studies performed for the Boeing version of the C5A Airplane is shown in Figure 5. 0(S) kflrFeE S=b.R -$. H (S)

(v)

FIG. 7 NON-LINEAR STRESS STRAIN REPRESENTATION outlined by Araris, 14 uses an initial strain con- FIG. 5 CM PRELIMINARY DESIGN ANALYSIS cept, and involves both iteration and inversion processes. This region of fuselage includes large access open- ings for loading vehicles, etc. The analysis grid is indicated in Figure 6 and the details that describe the size of the problem are included in the figure.

2b a/b=.279

ELEMENTS 1615 REDUNDANTS 928 ELEMENT FORCES____5653 FIG. 6 CRACKED PANEL ANALYSIS GRID LOAD CASES 6 MACHINE TIME----4.5 hrs An example of a simple problem is shown in analysis grid form in Figure 8.The plate is uni- FIG. 6 CM REAR FUSELAGE ANALYSIS GRID form, loaded by a uniform stress along the edges parallel to the crack and, since it is doubly 2. Non-Linear Material Behavior sym- metric, only one quarter need be analysed.The In assessing the fatigue and fail-safe characteris- distribution of grid lines reflects tics of an airplane component, a preliminary phase an anticipation of the location of the stress concentrations.The involves the determination of stresses that are dis- grid lines define the positions of flange elements and tributed around an existent crack. A typical com- the rectangular spaces imply uniform shear fields. ponent to receive such attention is a panel comprised of cover plate and stiffeners.It is usually assumed that in the middle of the pane! there exists a crack of finite length and zero width.In other words, the panel possesses a structural discontinuity additional to those caused by its attachment to stiffeners. To solve the stress distribution, finite element methods are used. It is known that for significant loadings on the panel, the concentrations will cause the stress within a certain region to exceed the elastic limit. Thus nonlinear material behaviors have to be represented. These can easily be read into the computer in the form of a table that relates stresses to strains for FIG. 9CRACKED PANEL RESULTS 19 In Figure 9 can be seen the results due to a cer- 10 and part of the analysis grid and the pressure tain magnitude of applied stress.The stresses loading is shown in Figure 11. shown are those acting normal to the crack at the The finite element method used to analyse the line co-axial with the crack (Y=0). Assuming cowling structure included a triangular plate ele- first a fully elastic material, there can be a very ment that possessed bending and membrane prop- high magnitude of stress near the crack tip.This erties.This was used to represent the honeycomb corresponds to the singular point that would occur panels in the idealization.The computing details in an analytical solution.Allowing the material to of this analysis, which produced both stress and have non-linear properties shows a reduction in deflection distributions and an influence matrix, peak magnitude, yielding and general redistribu- are as follows: tion. Number of nodes =364 3. Engine Cowling Analysis Number of structural elements =401 The engine cowling structure was analysed by a Computing time = 45 minutes finite element method and certain critical deflections (IBM 7094 Mk II) obtained. The cowling was a tube constructed from Tests were performed upon the cowling and deflection measurements obtained. A set of these titanium honeycomb panels and the tube was loaded results is shown in Figure 12 for one cross-section of the structure.It can be seen that. the finite

SUPPORT POINTS element analysis predicted the test resulth considerable accuracy. 4

HONEYCOMB PANEL

478 lay/

FIG. 12 COWLING DEFLECTION RESULTS 0 SUPPORT POINTS IV. THERMAL AND ACOUSTICAL APPLICA- FIG. 10 ENGINE COWLING TIONS by internal pressure.Analysis was made complex The concepts expressed previously with regard by virtue of the irregular shape of the tube cross- to structural analysis by finite element methods section, the presence of plates (splitters) connect- are valid also for analysis in other technologies. ing opposite sides of the tube, and the need to In the matter of analysing temperature distribu- represent the bending behavior of the honeycomb tions, it can be shown that for linear conditions panels. A view of the cowling is shown in Figure a direct analogy exists between the formulation of the structural and thermal equations.Thus one solution program could solve both classes of problems, provided that the appropriate data was supplied.In the acoustical problem, a fluid element is treated as though it were a structural element and is therefore easily incorporated into the structural solution. 1. Thermal Analysis Using a variational approach, Visser u has shown that a close analogy exists between the solution of conduction problems in thermal analysis and the solution of structural problems.Thus it is possible to conceive of finite elements representing FIG. 11 COWLING ANALYSIS GR ID AND PRESSURE LOADING assumed temperature behavior and expressed math- 20 emetically by the corresponding heat flow-tempera-

ture relationships. In matrix form this is denoted: to 5 .5.75 1.001.25 1.51.752.0 Where 6=0.T 8=Column matrix of nodal heat flows

T=Column matrix of nodal temperatures 10 6 4 2 0 0= Square matrix of influence coefficients APPLIED ABSOLUTE VALUES fora thermal element POINT TEMP I.5 .5.751.01.251.5 1.5 2.0 In the analogous structural equations, the mat- 30 rix 0 would correspond to the stiffnessmatrix of a structural element. Visser explained the 9) matrix fora flange and 30 420 the 9) matrix for a triangular plate isderived in DIFFERENCE FROM AVERAGE VALUES QM Appendix A.Using similar approaches, mathematical expressions fora. comprehensive array of FIG. 14 SAINT VENANTS PRINCIPLE - TEMPERATURE DISTRIBUTIONS idealized thermal elements.can be constructed. This array could match the types. of eleniente used in The second application is shown in Figure 15 a structural analysis. . and was performed to test correlation with another If it is possible to idealizea thermal problem method.2'The exact analysis involved a solution into a set of mathematically defined finite elements, to the heat-conduction partial- differential equatton it .einains fora solution procedure to assemble and and the matrix analysis used finite .elementet.A. 'solve 'for the' unknowns. Themethod used can be linear temperature distribution was imposiidAong 4.:jdonticatio'lliat used in solvinga: structural prob- one edge, zero temperatures maintained SiSit two ..10*'. but .Witiot bediscUssed in detail hem others and zero heat flow along the reniainlift 04gre.. e apirations of thermal conduction analysis Plots of temperature distribution show 414cores. using- the triangular plate element will be de- relation between the two methodswas1 excellent. 'scribed.The first is a demonstration of Saint I 1 VenantsPrinciple. Irregular .distributionsof temperature were imposedon the boundary of a ZERO NEAT rI, 1.0 plate that was "long" in one direction compared .2 MEE! to the other direction.The grid used is shown in .6 MEW111 II MEM:L .4 II II M Figure 13 and the boundary condition of the three .2 II IIII ME x

nATIER)mums

FINITE ELEMENT METHOD HILDEBRAND IREF.21) 1. 1. TA1 TN 0.6

APPLIED TA2 .5 .5

TEMPER- . ATURES TA3

.5 X 1. TA4 2L

FIG. 15 PLATE CONDUCTION ANALYSIS TA5 2. Cavity Analysis The cavity illustrated in Figure; 16 represents a problem in acoustics.The figure shows a two- FIG. 13 SAINT VENANT THERMAL ANALYSIS GRID dimensional view of a rectangular air cavity remaining edges is such that there iszero heat bounded on five sides by rigid surfaces and closed flow across them.Interior nodes also have zero on the sixth side by a flexible panel. The problem resultant heat flow. is to determine the response of the combinedpanel The results of imposing two sets of temperature and enclosed air volume toan impinging acoustical conditions are shown in Figure 14.Plots of tem- environment. The representation of thepanel in- perature distributions across the width of the volves standard structural idealizationprocedures, plate are given at various distances (0)from the and to represent the air system withinthe cavity application edge.It can be seen that the irregu- a fluid finite element can be used. " larities have smoothed out to give 'uniformaverage To test the quality of the representation,a finite Values within two width-distances of application. element analysis was performed foran air system 21 7. Robinson, J. and Regl, R. R., "Automated Matrix Analy- sis for General Plane Frames," Journal of the Ameri- can Helicopter Society, October 1963. 8. Clough, R. W., "The Finite Element Method in Structural Mechanics,"Chapter8,"StressAnalysis,"Editors Ziekiewicz, 0. C. and Hollister, G. S., Wiley, 1965. 9. Zienkiewicz, 0. C., "Solution of Antisotropic Seepage by Finite Elements," A. S. C. E. Journal of the Engineering Mechanics Division, February 1966. 10. Fenves, S. J., "Structural Analysis by Networks, Matrices and Computers," A. S. C. E., Journal of the Structural Division, February 1966. 11. Bampton, M. C. C., "Matrix Structural Analysis-A Course in the Theory and Application of Matrix Methods FIG. 16 CAVITY ANALYSIS USING FLUID ELEMENTS of Structural Analysis to Airplane Structures," Vol. I, Boeing Document D6-2587 TN, 1964. surrounded entirely by rigid surfaces.Boundary 12. Kaufman, S. and Kaufman, J. J., "Molecular Vibrations by a Matrix Force Method," Martin-BaltimoreReport conditions were such that motion normal toa sur- ER 13536, June 1964. face was permitted.Submitting the assembled air 13. Visser, W., "A Finite-Element Method for the Determina- system matrices to a vibration analysis yielded tion of Non-Stationary Temperature Distributions and modes and frequences that correspond to standing Thermal Deformations," Conferenceon Matrix Methods waves,., in the cavity. in Structural Mechanics, Wright-Patterson A. F. B., In" Figure 17, results for the October 1965. frectirencies of the finite -element approachare corn- 14. Bampton, M. C. C., "The Matrix Temperature Method of Thermal Analysis," Boeing Document D6-4370 TN,1964. 15. Argyris, J. H., "Modern Fuselage Analysis and the Elastic Aircraft," Butterworth, 1968. 11;11 /114 16. Argyris, J. H., "Recent Advances in Matrix Methods of Structural Analysis," Pergamon Press, 1964. MUM" 17. Bendavid, D., "Application of the Direct Stiffness Method to Static and Dynamic Analysis of Pressurized Shells," Boeing Document D6 -18118 TN, July 1965. 18. Kapur, K. K., "Buckling of Thin Plates Using Matrix MODE M1 MODE M2 MODE M3 Stiffness Method," Ph. D. Thesis, University of Wash- FREQUENCY CPS ington, 1965. ANALYSIS 92.8 140.0 168.0 MATRIX METHOD 93.0 19. Bampton, M. C. C., "The Stiffness and Mass Representa- 143.0 170.0 tion of Finite Fluid Spaces," Boeing Document D6-15163 FIG. 17 CAVITY MODES AND FREQUENCIES TN (in preparation). 20. Archer, J. S., "Consistent Mass Matrix for Distributed pared with those obtained by classical analysis." Mass Systems," A. S. C. E., Journal of the Structural The two sets agree very closely and mode shapes Division, August 1963. 21. Hildebrand, F. B., "Advanced Calculus forEngineers," were virtually identical for the two methods. Prentice-Hall, 1949. It can be seen that the ideas expressed hereare 22. Morse, P. M., "Sound and Vibration," McGraw-Hill,1948. capable of considerable extension, suchas application to more complex configurations and torepre- FENVES: I have two questions.First, how does sentations of fluids other than air. the engineer describe an indeterminatestructure 7 consisting of 5,000 nodesor whatever the example .* was?Second, how can all this data beinput REFERENCES with any reliability, considering thenumber of 1. Argyris, J. H., "Energy Theorems and Structural Analy- input cards that must be punched? sis," Butterworth, 1960. 2. Turner, M. J., et al., "Stiffness and Deflection Analysis of BAMPTON: That isa very good question and I Complex Structures," Journal of the AeronauticalSci- know one or two engineers with bloodshoteyes ences, September 1956. who would sympathize with it. 3. Denke, P. H., et al., "A Computerized Static andDynamic For one thing, Aircraft Structural Analysis System," S. A. E.SP-264, the program that resulted in this analysisinvolv- 1964. ing 5,000 nodes presumeda certain configuration 4. Gallagher, R. H., "A Correlation Study ofMethods of so that the preparation of datawas capable of Matrix Structural Analysis," A. G. A. R. D. 14thMeet- being automated. Now this ofcourse won't gen- ing, Paris, 1962. erally be the situation. But there 5. Klein, B., "A Simple Method of MatrixStructural Analy- was a certain sis," Journal of the Aeronautical Sciences,January 1957. prescribed configuration built into theprogram, 8. Irons, B. M. and Draper, K. J., "LagrangeMultiplier namely, that particular problemhappened to be a Techniques in Structural Analysis," A. I. A.A. Journal, fuselage problem and itwas known that it would June 1965. consist of grids defined by stringerlines and 22 frames, so this presented a latticeupon which lem of optimizing what may not turn out to be the engineer could hang his data.This eased a good idea, or attempting to get an efficient de- the problem which I thinkyou implied by your sign based on a preliminary design which is not question, namely, how canany engineer or team necessarily the best of alternatives?Is there of engineers handleso much data. any activity in using computers in this prelimi- FENVES: My next question isa little bit more nary design stage in your organization? fundamental and has something to do with the BAMPTON: Yes, there is, and the activity is di- discussion we had earlier concerning the theoreti- rected partially toward structural representation, cal approach vs. the algorithmic approach.The with which I'm most involved, and also toward procedure which you discussedwas a very valid what might be called integrated analysis.That application of computers, that is, you takea is, you have a structures program,a weight pro- continuous elliptic differential equation problem, gram, and aerodynamics program, etc., and you chop it up into pieces so that you have only small attempt to integrate these in order to getan pieces to concern yourself with, and then youuse efficient preliminary design.This has only been finite mathematics to combine these pieces to- possible on a limited basis because you're faced gether. Now you spend a tremendous amount of with the task of preparing data.Thus, certain research on describing these particular elements. configurations have been assumedor built into On the other hand, the technique, the algorithm, the integrated design program and they do not for interconnecting thisvery large number of reflect a structure thatcan be sophisticated as elements so as to form something thatcan be our structural program can produce for us. computed, you dismissed in one sentence.You CHASE: I'd like to answer the question simply said, "We obviously don't use matrix alge- a little bit bra." Why is there this disparity of emphasis more.Actually there is a great deal of work and of prestige? going on in this area of the basicprocess. We There appears to be a great have at Lockheed, for example, amount of prestige associated with derivingnew a bank of com- finite elements, but all the work involved in try- puter-system programs dedicated to theman- ing to put together a couple of thousand finite computer interaction that is being discussed to- elements, so that today's computerscan get a day.In particular, one of the majorareas of solution rather than waiting till the seventhgen- investigation involves the prelimiary design of eration machines are big enough and fast enough, aircraft in general, the surfaces, etc., in which is tossed away as a coding trick. structures of various sorts will be designed at some later stage.It is in its infancy.There's BAMPTON: I don't know what is the basis foryour a great deal of work being put into it.Definition assumption. I'm not tossing aside the problems of surfaces is a complex problem within itself, involved in the system which is accepting the but it is being conducted in industry. Some of elements, and I didn't say that obviouslywe the work is done at MIT under Coons in thecom- can't manipulate matrices ina tone which im- puter-aided design program. But the wholecon- plied that this was a trivial matter. I just hap- cept of design in itsmany phases and many pened to have picked up a crutch,you might say, aspects is being investigated and we're whittling for discussionnamely, the finite element. There away at little pieces at a time. I think that you is no doubt about it; there isa lot of work going will see within a matter of twoor three years on in examining the system which incorporates quite a large accomplishment in which allthese these finite elements in order to handle them in features will be put together ina somewhat more the most efficient manner. sophisticated package. SCHMIT: I would like to ask if you could tellus PAYNTER: I'd like topropose a question today anything about what effect, ifany, the computer which I'll have a chance to ask againtomorrow, is having in your organization, or in organizations but which is particularly appropriate tothe pres- you are familiar with, on arriving at the system ent speaker. So far we have not yetcome to grips you had when you set about making this idealiza- with the problem of howcan you evaluate whether tion; in other words, you describedan aft end a model is really any good or not. Wesaw here of a fuselage and you had something thatyou comparisons of a point-for-point typeresponse were ready to analyze.Certainly a great deal between some test data andsome model predic- of preliminary design or some sort of effort went tions.I'd like to raise a question in linewith into bringing you to that point. Are computers the comment just made in Fenves'second ques- having any impact on this sort of thingthe tion: whether the following problemought to preliminary design phase? What I'm reallycon- be posed, that is, how does thepresumed affect cerned with is this: are we faced with the prob- the decision process itself?In the context of 28 r

iron

many of your situations,yOu are not going to SAUNDERS: I would like to inquire aboit the design plates with variable thicknesses at differ- make-up of the teams who are doing this work. ent points on the aircraft shell, so that, if you Are they predominantly of an engineering back- are restricted to using plates of uniform thick- ground, or, mathematics, or computer people, or ness, isn't it true that as long as the model pre- just who are they? dicts within one gauge thickness, you would ac- BAMPTON: It's a combination of engineers and cept it as an acceptable representation? mathematicians with the main emphasis on engineering. At Boeing, there is a distinct dividing BAMPTON: I think the answer to that is in the line between mathematicians and engineers and affirmative, and I can only comment further that they have to communicate across that line.. Occa- we have been investigating optimisation pro - gamily, mathematicians do cross that line, but gram3 using linear programming where such con- the engineers are not allowed to do 80.They co- -atraints as ,standard plate gauges have been operate pretty closely, and it a combined effort entered into the programming problem. that produces these kinds of formulations.

U APPLICATION OF COMPUTERS TO POWER SYSTEM ENGINEERING G. W. STAGG AND A. H. PALMER American Electric Power Co., New York, N. Y.

This paper will describe the application of com- The Engineering Information Processing Section puters in power system engineering and operation. develops programming systems for collection, pro- It is essentially a case history of engineering . cessing, recording and analysis of the company's applications in the American Electric Power operation information in order to produce reports Company. necessary for engineering control and evaluation, The American Electric Power System provides and also to satisfy regulatory requirements. electric service in a seven-state area.It has a In each of these groups a distinction is made generating capacity of over 8 million kilowatts, in- between engineer and programmer-analyst.The steled in 16 plants with 38 generating units. Power programmer, or analyst, generally provides a high is distributed over 81,000 miles of transmission and level of mathematical training, whereas the en- distributing facilities.The System is connected gineer provides an extensive background in the to 21 neighboring power systems by 46 tie lines. analytical aspects of the problem area.There are In 1965 the peak load was slightly over 7 million forty baccalaureate degrees and thirteen advanced kilowatts, of which approximately 99% was gen- degrees among the forty people in the two Sections. erated by steam units requiring 16 million tons of It should be stressed here that there exist varied coal. educational backgrounds within the division.In- This particular case history demonstrates a more cluded are electrical, mechanical, civil, chemical general operation than many others in that it in- and industrial engineers, mathematicians, physicists volves electrical, civil, and mechanical engineering, and business administration personnel. as well as engineering information and project The staff has grown partially by obtaining per- control applications.The achievements made in sonnel from outside the organization, although of these areas will be described and some conclusions late some emphasis has been placed on recruiting will be drawn concerning what the future portends. from within, particularly in the Engineering Infor- Initially mention should be made of the back- mation Processing Section where experience within ground and resources which have been brought to the organization is invaluable. bear on the problem.The word problem is used Mention has been made of the resources which here in reference to provision to the practicing have been used. An important facet has been the engineer of the computer capability of the digital availability of computing at reasonable cost.Al- computer, without requiring him to perform the though the units of computer usage have increased task of program development. There exists a cen- considerably, the cost through the years has grown tralized group which is responsible for the develop- at a much slower rate. ment of computer applications. A significant step in the continuing program of For the past ten years, AEP has been requiring computer utilization for the solution of complex and training a staff of professional people. At pre- engineering problems occurred last year. On Nov. sent, the Engineering Analysis and Computer Divi- 1, 1965, an IBM/360 Model 40, specifically intended sion consists of forty engineers, analysts and for engineering, was installed in the Service Cor- programmers.In addition, a supporting group of poration Offices in 'New York.This system is ten people performs the computer operation, key- scheduled to be replaced by a Model 50 in December, punching and computer services. 1966.It is expected that these machines willpro- Two major groups operate with the Division. vide an impetus to additional expansion in the The Engineering Computer Applications Section is computer application area. concerned with the development of programs for The efforts of this staff will be the concern of the use in electrical, civil, and mechanical engineering remainder of this paper.Several broad subdivi- activities. A prime responsibility lies in the evalua- sions of application areas immediately arise.Al- tion and development of techniques suitable for though one of these is identifiedas design, all computer implementation.The applications are evidence a relationship to the designprocess to generally large-scale programs which consider the some extent.In the context which follows, design entire problem area and are intended for system is to be considered the planned change of theAEP design and simulation. system.This is a continuing, complex, multi-step 25 process which has as its aim the expansion and theory for a plate on an elastic subgrade. A third improvement of the system. program is also under development which considers The several areas which will be indicated by ex- the same foundation slab on piles. ample are: In transmission line design, the computer enters 1. Design. the problem area at an early stage.Once a line 2. Simulation. plan is selected, field survey crews obtain terrain 3. Control. data. The field survey notebooks provide the data 4. Information processing. for the terrain model.It is hoped that eventually DESIGN this data may be obtained even more directly by Among the most promising engineering applica- means of aerial photographs and subsequent auto- tions, from an economic viewpoint, are those in the matic digitization. design area. Programs of this type provide design In addition to the terrain data, it is necessary assistance in connection with power-plant steel and that constraints be specified for line crossings, con- concrete work and in location and design of trans- ductor clearances and restrictions on tower posi- mission towers. In addition to material and design tioning.The Transmission Tower Location and time-saving, the means are available for a more Selection Program uses a dynamic programming comprehensive analysis of each problem. technique to achieve an optimum economic ar- The power-plant design function is supported by rangement of tower locations and types.Several a system of programs developed to aid the design additional supporting programs available for the engineer in achieving an economical design within transmission area are sag and tension analyses of time limitations associated with this type of work. conductors, both for the design and field-stringing The Structural Steel Framing Program is used in conditions. the design of plants of braced frame construction. Two salient factors should be noted.First, an In the design of a plant of this type, the flooring integrated system of programs provides the capa- system, made up of rolled sections and welded plate bility of operating in the entire range of the prob- girders, is designed somewhat independently of the lem area.Second, this system of programs, to be bracing and column systems. effective, must be integrated into the day-to-day Data prepared for the flooring system ispro- work load. cessed with or without specification of column re- A number of other special-purpose programs are quirements. As the floor design is developed, the available for use in civil engineering functions; for reactions contributed to columnsare automatically example, analysis of backwater curve, gravity dam, retained for column design.The normal sequence reservoir capacity and concrete stack. of events on the computer provides the designer SIMULATION with all Mooring member sizings and, when required, The planning and design of a large complex power column sizes. system is virtually impossible without a computer. An interesting extension to this program is the In the simulation of such a system it is necessary production of detail fabrication drawings.The to develop a mathematical model which can be used information necessary for this step is all available for studying changes in the real system. By the within the design programs, and the technical and use of this model, any number of system conditions economic feasibility has been demonstrated of sub- can be simulated and their effects observed. These sequently processing a magnetic tape file to produce applications run the gamut from steady-state elec- drawings. The high-speed microfilm recorder dis- trical network analysis to transient analysis. plays information on a cathode-ray tube which A steady-state model of the electric power system is film-recorded to yield a drawing. is the load flow.This program evaluates station The Foundation Slab Analysis Program currently conditions and power flows and considers the volt- being used considers the mat to be a rigid body age-regulating capability of generators, synchronous developing a planar soil-bearing pressure distribu- condensersandtap-changing-under-loadtrans- tion, an assumption which is valid for the heavy formers.In addition, desired energy interchange slabs generally used in power-plant work. between systems is considered. The utility of this program lies in the possibility The system is described in terms of network of evaluating a number of alternative designs and connections, transmission line impedances, trans- considering the vast number of load combinations. former impedances and capacitor data. Fora given Although this is a fairly simple approach to the study year, loads, generation, voltage levels and general problem of mat foundations, considerable interconnection data are given.The study results work has been done in the investigation ofa pro- include a detailed report on the status of thesys- gram to design the mat based on thin-plate bending tem during the study period. 26 ,

This program permits the planning engineer to a 1,700-mile microwave system, analog and digital consider a wide variety of cases.The effect of telemetering, analog equipment and a digital con- future load, generation, transmission lines, trans- trol computer. formers, interconnections and voltage control equip- The digital control system provides, on a real -time ment may be evaluated.Furthermore, the outage basis, economic loading schedules for 38 generating of various components can be simulated. units and, in addition, performs power-system oper- This type of program can be used to analyze vast ating studies and logging functions on a time-shared systems.This particular version of the IBM/360 basis. The control system has proved to be effective can analyze systems with as many as 2,000 stations in allocating generation for optimum economy in and 4,000 lines, and, with random storage, systems AEP System operation as well as affording other many times this size.Until the general availability savings. of digital computers this was not possible, nor were The functions of the Digital Control System are : the solution techniques sufficiently well defined. 1. Economic dispatch under normal conditions. Today, the design engineer has available the capa- 2. Study mode to permit investigation of city to study in detail large interconnected systems, energy interchange. and thereby plan better future systems. 3. Logging. The Transient Stability Analysis program simu- INFORMATION PROCESSING lates behavior of the electrical network during The logging function associated with the Canton switching operations and fault conditions. The dif- control computer is a part of the AEP information ferential equations describing the machines are collection system. solved numerically to determine the system condi- The MHW (megawatt hour) Data System estab- tions, following the specified operations. Of parti- lishes a history file from the hourly collection of cular interest is the response of generators during generation, interchange and load data. Hourly and the sequence of events. The aim in using this pro- daily reports are produced automatically from this gram is to provide data for use in the design of file and are retained in conjunction with regulatory electrical facilities. requirements. The Switching and Lightning Surge program also evaluates transient currents and voltages. The time SUMMARY scale of interest here is of a much shorter duration In summary it should be indicated what plans exist than in the transient stability analysis.In effect, for continuing the efforts described.The primary a voltage wave is simulated in order to determine objective is to give to the practicing engineer caught design criteria for lightning arresters and switching up in the swirl of details a more effective path to equipment.In addition, insulation requirements computing capability. This is not to be construed as for electric facilities can be determined. a path to the raw computer, as much as a channel In the mechanical engineering area a program of communications with the programs whichare appropriate to the special problem.This entails a is available to evaluate a steam power-plant cycle. three-step process: A one-line diagram defining the cycle is the starting point forcomputer solution.The cycle is simu- 1. Development of integrated sets of computer lated by interrelating the various components of the applications associated with each problemarea. cycle. A subprogram utilizing the characteristics These are well under way. of a component as defined by input data can simulate 2. Development of a data base upon which these performance of a segment of the cycle.Subse- programs draw.Some inroads have been quently, the remaining components are evaluated made in this area. and eventually this iterative process converges. 3. Development of a program to interrelate the Results are provided for various points on the cycle. problem program, the data base, the engineer This program is used to design new cycles, evaluate and the computer. This presentsa significant current performance and to simulate modifications. obstacle which at this point remains to be passed. CONTROL (Operational) The allocation of power-plant generation to effect optimum operating economy is an important prob- PAYNTER: It looks to me as if you've outlineda lem in an electric utility system.The complexity system with a great many fragmentedprograms. of the problem for a large, integrated power system You spoke of integrating these ina system of pro- induced American Electric Power Company to in- grams but how will Miller's master designer fit stall, in 1964, an automatic real-time control system into your scheme of things? to provide optimum operation of its power facilities. PALMER: Well, I don'tsee that this is at all in con- The facilities to provide on-line control consist of tradiction to Miller's approach.I think this is 27 what we are heading toward. Perhaps the path were thinking in terms of doing something of that which we travel to get thereat somewhat dif- sort. Butalthough I don't think much work has ferent from his, but I think basically ills are been done, I'm not'really qualified to speak on coming to that same point. this matter.

LICKLIDER: Ars there any parts of,the spectrum COMMENT OFF MIKE:Would you care to elab- in power distribution that are interference-free orate upon that? 'enough to consider sending signals over them, and, if so, are there any thoughts about sending PALMER: In AEP's case I think this idea was computer signals over your power lines? And pretty. muck replaced by the decision togo to a if not, are the problem legal or technical? microwave system.But, in answer to the question, it, is now being used by a number of com- PALMER: In my experience the only time we be- panies for control systems, particularly in the came involved with something of that sort was far west where a power-line has been in use for several years ago when the Civil Defense people twenty to thirty years. APPLICATIONS OF COMPUTERS IN HIGHWAY ENGINEERING PRACTICE THEODORE F. MORF Illinois Division of Highways Springfield, Ill. There is an old saying that has come down to us and mathematical thought, the engineering profes- to the effect that "the cobbler's children go bare- sion has advanced on the break-throughs of the foot." As in the case of many other enduring ex- God-gifted geniuses of the. pastAristotle, Archi- pressions of traditional wisdom, this saying survives medes, Galileo, Napier, Newton; and more recently, because it contains a large core of truth. Einstein, Fermi, and many others. We may all In a certain sense this saying has been true of stand a little taller and hold our shoulders a little the techniques of the engineering profession. Engi- straighter when we acknowledge this heritage of neers, as a profession, are a very creative group genius, as it speaks to us today across the echoing perhaps the most creative of all the professional chasms of the centuries. groups that there are.Yet this creativityis While the bold advances in conceptualizations directed outward toward the design of products which we owe to these geniuses of the past and the and remarkably little change has been made within present have normally led to advances in technology the engineer's own shop in the techniques that he as we have come to comprehend their meanings, uses during the creative processes. we may today be seeing something of an inversion It always gives me the greatest pleasure to study of this normal process in the adoption of the elec- the exhibit cases of the antique engineering, astro- tronic computer in modern uses. nomical, and measuring intruments on display in The electronic computer, as it has been initially the Adler Planetarium and the Museum of Science used, has. Wei simply a far, far better mousetrap and Industry, here in Chicago, or in the Smithsonian than we have had before. In its initial applications Institution in Washington, D. C. Many, if not most, it has been like a $15 hamburgeran enormously of these exhibited instruments are represented in a advanced device. By its means we have done our modern form in our present-day engineering offices. conventional computations in an unconventionally If we seek differences between these antique rapid manner. instruments and their modern equivalents,we see However, as we begin to appreciate the pos- less jewellike elegance and perhaps somewhatmore sibilities of the device, we find ourselves pressing utilitarian precision in the new ones.But, by and against the limitations of our convention-bound large, a modern engineer, draftsman,or surveyor techniques. We are indeed looking for new con- could use without difficulty the instruments of his ceptual approaches to exploit, in presently unknown centuries-ago predecessor. ways, the vast potentialities that we have discovered Indeed, I often think that Copernicus, who lived in the binary nature of an electrical pulse when used about 450 years ago, would have far less difficulty in combination with the rigors of yes-no logic. We in accustoming himself to the instruments and tech- are beginning to see, in real-time applications of niques in a modern state highway department de- computer control as well as in strictly mathematical sign office, than his wifeMrs. Nicholas Kopernik processes, the intimations of a brave new world of would have in preparing a batch of Polish Christmas conceptualization made possible by technical ad- cookies in my wife's modern kitchen. vances. Until recently, the most complex mathematical In the field of highway engineering it was my devices available to engineers were the desk adding privilege to have been among the first to be exposed machines and desk calculators.Far from being to the possibilities of the electronic computer. This modern, we may be surprised to learn that completely was in December of the year 1955, and the place workable models were constructed by Blaise Pascal, was the annual meeting of the American Associa- who died at the age of 89 in 1662, about 800years tion of State Highway Officials in New Orleans. ago. To add a personal note, Philip Matthew Hahn, Until then I had been dimly aware of the work a south German Evangelical pastor (and my great- going on at Harvard and a few other places.I great-great-grandfather) designed and builta cal- had stood in the blinking, oven-like presence of culating machine which was probably the prototype Iliac at the University of Illinois.This was the of the modern CURTA computer over 200years "Tom Swift and His Magic Brain" period of popular ago. journalism. As I remember, at this time a Univac Standing as it does in the mainstream of scientific was helping out on one of the popular television 29

V.:7e , t geA. 2, 1 programs by finding, through mysterious mathe- .and accountinguses of the equipment. matical means, a magical matching of However, matrimonial since this conference is concerned withengineering mates or something else equallypreposterous. education, I would like to confine Then at Miami in January; 1956, my remarks. to another manu- engineering uses rather than accountinguses, while facturer stimulated thinking amongengineers by an always acknowledging theirpresence. exhibit at the American Road Bid lders Convention. Apart from conventional data processing,the My own first serious introduction tothis equipment uses in technical applications are quite broad,rang- was in a small discussion group, following the1966 ing from information storage Highway Research Board meeting, and retrieval through attended by only the fields of traffic assignments, traversecomputa- three or four highway engineers inthe office of the late H. A. Radzikowski, at which tions, structural computations, and soilmechanics his two assistants, if one might suppose there to besome kind of syste- S. E. Ridge and L. R.Schureman, described the matic or orderly continuum. results of their preliminaryinvestigations into electronic computers during thepast year. There is also a certain futilityin describing how electronic computers The first electronic computer are being used in highway conference, under the engineering, either technicallyor organizationally, auspices of the ElectronicsCommittee of AASHO, because changes was held in Chicago in early March, 1956, followed are taking place so rapidly. Changes in the types of problemssusceptible of by meetings in Atlanta, LosAngeles, and Boston. solution, in the scope of Within about ayear, nearly every highway depart- programs, in the software which permits better operationalscheduling of work, ment either had an electroniccomputer on the floor or on order.And within two also more efficient, so-called smallcomputers and years the various more practicable remote operationson large com- highway departments andconsulting engineers in the highway field had puters are all factors which modulatethe momen- over 1,000 programs in use tary advantages ofany particular mode of use. applicable to the solution ofhighway engineering problems.In the years which havepassed since This is why a discussion of whatany organization then, the scope of applicationshas become so broad is doing must be understoodto be a kind of amalgam that inventories of numbershave become meaning- of the outmoded methods andobsolete equipment less. that is actually inuse, as modified by the intentions Without intending to ventureinto the jungle of of use for the equipmenton order, and liberally superlatives as to which specialty,within the whole seasoned by ideas on techniquesnow only dimly engineering profession,was the earliest, or the perceived for the remote future. fastest, or the most efficientin applying thecom- Generally theuse of electronic computation in puter to engineering problems,I would say that the highway engineering today is not initself a system highway groupas a whole has made a remarkable in the sense that dataprocessors use the term, nor record in this respect. Iwould not be able to (nor in the light of present knowledgeis it likely to if I could, would I wish to)bore you witha catalog become, broadly, a system in theforeseeable future. of the highway engineeringproblems which have Some parts might be termed thus,perhaps, but in been aided by the electroniccomputer.These are the main, computations, whetherdone by pencil all well known, and, if Icould find a problem not or by electronic computer,are only a part of a yet within therange of analysis by computer, I larger system of design. would only be putting the spotlightof challenge on Generally also, in the presentstate of the art, the next problem to be solved,in the words of the highway engineering problemsare mainly of the mountain climber, Mallory,"because it's there." analytical rather than of the designtype.In the Certainly, the electroniccomputer, used asa analytical situation, the designersets certain con- calculating device, has profoundlyaffected highway ditions and the computer isprogrammed to calcu- engineering practice and allof those who are in late certain resultsor to optimize certain aspects. highway engineeringare aware of this change in The designer reviews these results and,if he is not varying degrees. Theresponse to this change has satisfied, he will change the variablesand repeat by no means been uniformamong individual en- the analysis. gineers, as they showreactions ranging from This feature ofa dialogue between the engineer- tremendous enthusiasm throughstolid acceptance ing designer and the computeroften argues for the and all theway to baleful skepticism; but not indif- closest association between thetwo and against ference.No thinkingperson can be indifferent the big centralized computer with when his mental its theoretical cage is being so vigorously shaken. efficiencies for data processing.Functionally, an How are electronic computersbeing used in high- electronic computer inan engineering analysis use way departments? Of course,as in any other large might be compared simply with business organization, a tremendously there are fiscal, statistical powerful desk calculator available, ifnot immedi- 30 ately at the designer's elbow, at least ina menage given ability as humans to Understand and where scheduling is nota formidable problem. analyze a problem or process. Another growingarea of use of computers is in 3. Instruction in a pedagogical climate which the real-time application traffic surveillance and acknowledges that yesterday's brave new world control.For so e highway engineers have is already here and takes for granted the thought ..of the trafficusage of an urban freeway ever-widening application of computer uses. as being analogous to any other continuously operat- As to specific computer instruction for practicing ing process.In a chemical plant, for instance,a engineers of our staff, we have been conducting sensing and feedback control of readings ofpres- schools of instruction and would prefer to continue sure, temperature, and composition has enabled this practice.. Our reason for this is that the in- process automation through the operation of the struction must be closely related to the specific servo values. Similarly a sensing of speed, volume, computer or computers which are being used by. our and density of traffic ona freeway will enable the engineers, and the program languagesare those optimization of vehicle throughput,or speed, or which are applicable to our programs, andwe want gross time of vehicles in the transportation corridor to continue having it this way. by controlling the traffic signals at the entranceand Such schooling has been given to more than 500 exit ramps of the freeway. of our engineers in both the central office andin We now have the hardware and the knowledge the: district field offices. A further advantage of for traffic controls of this kind andare, in fact, having this instruction done by engineers ofour doing it to some extent on the freeways inthe Chi- computer staff is that it establishes their .authority ..cago area, and. in a somewhat different manner in in the field and also aids in communications bett4enn Detroit, in river tunnels in New York, in Seattle, those providing and those using the computir-cen- and in other places. ter services. ,.., In Baltimore; New York, Toronto, and elsewhere, Beyond the bachelor's level there woulc1,;Of coOrpe, entire grids of traffic signalson surface streets are be specific instruction in oneor more of the common controlled by electronic computerson a real-time program languages, and this is taken for granted. basis. Here, however, the emphasis should beon the pos- Further uses of process-control computersare sibilities of the computer to lead the profession into seen in continuous sensing in the production of higher levels of technical capability through better concrete, asphalt, aggregate, and other materials, means of analysis and fewer approximations, or but here the problem is lessone of control than rule-of-thumb solutions. of knowing how to sense the variables thatit is As I reach the conclusion of my remarks Iam desirable to control. struck by the paradox ofsome of the positions I What do we expect from the engineering schools have taken. I have written with deepest respect of in educating the engineers whomwe will be hiring the capabilities of the equipment and withsome as graduates this summer and from now on? At the disdain as to their physical presence,or :the lan- bachelor's level, not too much is expected beyond guage in which we must communicate with them. I a general familiarization.What I am saying is am more appalled than impressed that some of them that, given a fixed number of classroom hours,we will generate seven acres of paper covered with would rather have the schools makingour prospec- printing in a few minutes.I conclude that I will tive employees better engineers thanpoor computer have to explain this paradox to myself by seeing programmers. these devices as the man-made wingswingi unim- At this level there should be several points of portant in themselves that will give added lift emphasis that are more important than instruc- to the soaring spirit of man to enable him tomove tions in how to managea problem in any particular toward the fulfillment of those ends of creativity computer language: for the benefit of mankind for which hawas made 1. First and foremost is the appreciation ofthe in the image of God. rigors of binary logicas applied to the solution of a problem. Every daywe can see an in- LICKLIDER: Two or three times therehave been crease in the evaporation of the velvetfog of comments about the likelihoodor unlikelihood "engineering judgement" andwe are applying of ever pulling together the myriadproblem- informed and discreet choices inmore and oriented programs ina field into a system.This more situations. was something that was approached this morning 2. A disenchantment with the magic ofignorance by Miller and itseems as though we ought to concerning these fabulous devices andthe get a little better agreement about theground acknowledgement that the limitationson com- rules. I think there is at leastone concept that puters are mainly the limitations ofour God- doesn't suppose thatyou will ever pull sufficient 81 number of program together to form a system technical disciplines which are integrated in the that will itself do design, if you mean a system sense that one can transfer data between these of programs and computer hardware. But there sub-systems and carry out in one continuous pro- is an alternative which says one should couple cess, using one master set of files, a rather com- with the programs enough language implementa- plete design job.I agree with you entirely that tion to make it easy for a person skilled in the this matter of the objectivity, the criteria, or what art of the particular field to shape up, with the it is you're trying to do, is still open.The first help of these already programmed resources, most operational version', of ICES will rely almost of the software he needs to solve his particular entirely on the designer himself, to provide problem. Now this coupling of the concept of these criteria. He will have a very powerful de- language to the concept of a system of programs sign tool; how he will use it, though, is still going seems to me to be extremely basic and I'm to be up to him. thinking it would be important for this group to MORF: I didn't by any means mean to say that this approach closer and closer to an understanding wasn't an eventuality. But I think that there are of what's possible there. a lot of considerations that have to be thought of MORF: I think that there are about 12 reasons, and before we have something that is available in the there might be 15 or 20, why this is a result that real world. I see as being rather distant. I don't say that it BAUMANN: We have heard various things about id ?machinable, and in fact there are a number of making languages more useful for people. .effert8 being made in this direction. One of these of the things that people do very well is useuse lan- is being sponsored in part by the Bureau of Public guage redundantly; computers de this n'okat all. Roads-. It is called TIES: Total Engineering The real thing Miller is talking about 4.0tegrat.- Integrated System. Another similar effort is being some of these special programs 4b.? h4ltinc . f ing made at MIT, called ICES. There are prob- them done under one roof. What we really need lems here and they are very deep.One of the to accomplish in terms of language is to somehow reasons is that we don't quite know how to score make it possible for redundant languages to be- the ball game.Let's say, if you try to work a come useful. An example I think we might well matrix on how to maximize the profits of a com- look to is in the field of electronics.If you take pany, you have a rather simple objective.But a new television set down to your repairman, who what are we trying to do in the highway field? may never have seen a model of this kind before, Does anybody know what the objective of high- he can locate right away the power supply, the way administration or a road system is? high voltage section, the picture tube, etc. In MILLER: We had some exposure, to TIES a year other words, there is a lot more pattern in the electronics set-up than there is in computer pro- or so ago when it was first being formulated grams as we now use them. We somehow need as an ideal. Our efforts are now devoted entirely to put this pattern back into our programming to ICES, and this is different from TIES as is language so someone else can recognize a standard shown by the absence of the word Total.The element by just looking at it.Maybe we can go ICES system, by decision, makes no pretense to be a total system.But it makes at least a start completely to a system where we can program towards the goal of integrating what in the high- computers topologically rather than by a listing. way field, as an example, have been literally hun- This seems to be one of the directions which will dreds of piecemeal programs. And it is one single be very useful in integrating some of these system, a whole set of languages or commands. separate items; that is, we'd be able to take a look The term integrated means that these sets of at a specific computer program- and see what in- commands, or languages, are so implemented that puts and 80 on are required for another routine. a bridge designer or roadway designer, a geo- MORF: If that was a question, the answer is yes. metries designer or foundation designer can all PAYNTER: I would hope that this issue, which has cross-reference a common data base; meaning been raised several times, is typical ofsome prob- that any one designer who is concerned with the lems that we might come to grips with in the location of a highway can, via commando, carry next few days.It is your feeling, going back to out a locations study which sets up files which can Licklider's charge to this group, that it is just be cross-referenced by another man doing, let's a matter of time before we progress to a situa- say, geometric design of an interchange. His files tion in which we can expect in our computer dis- in turn can be cross-referenced and utilized by course the same type of freedom that we almost the man designing the bridge, and so on.There demand of each other if we are working together is a whole set of commands for the various on a problem? Or do you sense that there may be 82

±?1- ;C4,71,* some fundamental limitations to what we will ever organization and bureaucracy of a large industrial be able to attain in the way of an integrated design organization. This again would argue that some- procedure, thing must be changed in the way of organisation MORF : Well, I have answered a similar question if you expect to put the punch card in at the front in one way. Now I will go ahead and answer it in door and have a final design come out at thoother another.This is an organisational function. end.This world involve cutting across quite a There are people who determine alignments, there large number of organizational lines.None of are others who determine the tension in lines, these are impossible but they're not the kind of or those who determine the voltage distribution, things you're going to get right away and they do and so on, and these different functions are really provide restraints, and they also pro 'vide an ex- in various compartments.Perhaps one of the planation of the reason that the solutions in the restraints on a totally integrated system is the earlier phases are undertaken one part at a time.

rt ro

d. COMPUTERS AND ENGINEERING DESIGN IN INDUSTRY D. E. HART Research Laboratories, General Motors Corporation Warren, Mich. I was asked to summarize industry's needs for Let me elaborate on the concept of the "main future. engineering graduates.Quoting from the stream" of design.In Figure 1, the main design advance abstract, "Industry needs engineers who stream is depicted as an arrow pointing to the can approach new problem situations with intelli- right. gence, imagination, and 'confidence, and who can Outside of the main stream isa function labelled arrive at satisfactory solutions in a reasonable Research and Development. It might also becalled period of time." advanced design, or some similarname. That statement was true ten years ago. The R&D function developsnew technology which It. will still be true ten. years from now. is fed into the main stream after it has been The abstract also States that the computer is one thoroughly tested,e.g., Oldsmobile's front-wheel- the.many tools which the engineer should be able drive Toronado. The function is also availableon I.. 'ork e;g: .:4":1.! Use...I am sure you would all agree that that is a consulting and troubleihooting basis. The three e statement, and, if that was all there was to bars in the main stream illustratethat it takes . story; 'there would be no need for me to say three years to complete the design ofa new car, but anything.further. that a new design cycle must be initiatedevery Of,all the tools available to the engineer, however, year. the computer undoubtedly has the greatest potential This might be referred toas a steady-state design for, increasing his productivity as a designer.Yet environment in contrast with thespace-age design only a small fraction of the engineers'now in indus- environment where the object is to inventa way tryand only a slightly larger fraction of the to do something which hasnever been done before, engineers now entering industry from the univer- e.g., get to the moon and back. sitiesare making use of computers in design. My subsequent remarks will be primarilydirected This is the problem which we all face: -How do toward this steady-state design environmentwhich we put the power of the computer into the hands I believe is reasonably typical ofmuch non-defense of the engineer so that he can make effectiveuse of industry. it as a design tool ?And, from industry's stand- Where in this environmentare computers not point, how do we get the computer into the main being usedand why? stream of industrial design activity? In the R&D area there has beena rapid increase What do I mean by "the main stream of design ?" in computer useas an analysis tool, but it is not In the ,automobile industry, it is the design activity being used as muchas it should be.There is a phenomenon here which might be referredto as the which starts with a blank sheet of paper and ends Threshold Problem: up with finished automobiles coming off an assembly an engineer will use a computer line three years later. only if it's easier thansome other way. Even if a program exists as a black box whichconverts Who works in this "main stream?"Included data sheets into printouts, the engineer among the workers are artists and clay modelers must still fill out the data sheet, get it keypunched, and then in Styling, but the largest numberare engineering wait a few hours for the results designers in hundreds of drafting to come back from rooms through- the computer. Asan extreme case, this would be out the Corporation who spend millions ofman- a ridiculous way to solve a single quadratic hours each year designing GM's principal product equation; however, an engineer mightneed to solve the automobile. several hundred quadratics ina monthone at a We pretty obvious why these designers have not time.If a program doesn't exist, theengineer can been able to use computersthe man and thema- wait in line for aprogrammer or try to learn (or chine don't speak each other's language:The recall) progra -ming ina language like FORTRAN; language of computers is numbers, while the lan- however, FORTRAN is not good forthe casual user guage of the designer is graphicslines ona since there are toomany details that can be for- drawing. This language problem will be discussed gotten. in more detail later. But something exciting is happening.I like to

r!' AggfitidtdtitrldtSISL,

call it conversational computing. We recently in- wayby sending it letters.Three or four hours stalled a GE 265 computer system with remote tele- after asking the computer a question, the answer type terminals which operates on a time-sharing comes back, and if the question wasn't asked exactly basis over telephone lines using the Dartmouth- right, the answer may be only, "WHAT?" developed BASIC system .1BASIC is easy to learn, There is another aspect of the design process : and if the engineer does forget the rules he is re- Design can be said to be about ten percent inspira- minded of them at once, not four hours later. Term- tion and ninety percent perspiration. But the plan- inals are presently installed in test cells, drafting ning and the doingor in this case, the designing rooms, and engineers' offices. and the draftingare so tightly interwoven that An important point is that the threshold has been the designer has had to be his own draftsmanhe significantly loweredthe computer can now be has to see what one line looks like before he can used quickly to solve even one quadratic. decide how to draw the next one.In a sense, he Most important, a whole new group of engineers engages in a sort of conversation with his drafting are now using computersprimarily because they board. can remain involved in the solution of their prob- The conclusion which we come to is this:"If the lems on a minute-by-minute basis.I conclude that computer is going to help a designer, he has to be it's got to be good if an engineer will put up with able to use it himself," and to do this we must a teletype terminal as a means for communicating make the computer appear to the designer to be with a computer. a kind of high-speed mechanical draftsman which he can control on a minute-by-minute basis as he Speaking of involvement, I recently received the now controls his drafting board. Proceedings of the Third Conference on Engineer- ing Design Education on "Authentic Involvement In this way, the computer can do much of the routine drafting, and the engineer will have more in Interdisciplinary Design."2 This was a report on a series of workshops in which faculty members at- time to dream up the real design which only a man, temped to guide their own students through un- and not a machine, can do. structured design projects.In this report, I was In 1960, after two years of study and experimen- unable to find reference to the use of computers tation, the GM Research Laboratories initiated a major research and development project whose pur- on any of the design projects. pose was to learn how to put the power of the My conclusion is that we indeed have not suc- digital computer directly into the hands of these ceeded in learning how to put the digital computer designers, and ultimately to move the computer into into the hands of the engineer as a design tool. the main stream of GM's automobile design. Referring back to the main stream of design, This project is known as Design Augmented by you will recall that designers now working on draft- ComputersDAC-I.3 Let me stress Design Aug- ing boards are unable to use computers because they mented by Computersnot design by computers. don't speak the same language.It has become This type of work is often referred to as computer- apparent that the solution is not to teach the de- aided design, a term originally used by MIT. signer the language of computers, but to teach the A key element in the DAC-I system is a graphic computer the language of the designer. console which is attached directly to the Research I would like to point out that two designers often 7094 computer (Figure 2).It makes possible con- engage in a "conversation" about a design problem. versation between the designer and the computer During this conversation, they make reference to in the language of the designer which is, of course, drawings, handbooks, and models which have been graphics. constructed to help visualize the problem.They This graphic console is a one-of-a-kind unit which make sketches to help express their ideas. was designed and built to our specifications by Such a conversation is essential if the job is to IBM. be completed on scheduleit couldn't be done if This computer can display drawings to the de- they had to send letters back and forth to each signer on the cathode-ray tube.He, in turn, can other.I'm emphasizing this point again because give directions to the computer by pointing at items most problems are solved on a computer just that in the display with the special pencil or by means 1P. T. Shannon, "Impact of Time-Sharing Computers on of the keyboard. Education in Engineering Design," Conference on the Impact Inside the computer, and available on a moment's of Computers on Education in Engineering Design, April notice, are stored his drawings and his drafting 21-28, 1966.Sponsored by the Commission on Engineering Education. 3E. L. Jacks, "A Laboratory for the Study of Graphical * Proceedings, Third Conference on Engineering Design Man-Machine Communication," Proceedings, 1964 Fall Joint Education, "Authentic Involvement in Interdisciplinary De- Computer Conference, The American Federation of Informa- sign," July 12-18, 1966: Carnegie Institute of Technology. tion Processing Societies. 85 toolsthe computer equivalents of the T-square, strated how to make the computer availableto a triangle, ruler, compass, french curves, etc. large group of engineers who previouslyhad not All of the many computer programsthe soft- been able to use it. warewhich make the DAC-I system workwere Perhaps there is a parallel between whatI have developed by the GM Research Computer Tech- just described and the field of engineeringeducation. nology Department. If we change only a few words, Figure The DAC-I system went into operation in early 1 pretty well characterizes the educationalprocess. 1963, two and a half years after the projectwas The started.This was a unique laboratory facility arrow now becomes the main stream of education; the products are engineers.The university, too, whose purpose, as I indicated before, was to learn could be describedas a steady-state environment; how to put the power of the computer directly into it takes four years to produce the hands of the designer. an engineer, but a new crop of students enters eachyear. Most uni- But one important elementwas missingthe versity computer use has also been designers who were supposed to be able to in the R&D use area, and so far it has had very little impacton the DAC-I. Computer technicianscan design computer main stream of education. programs, but we can't design automobiles. Both Fisher Body and Styling agreed to assign I would like to suggest that theuniversity's goal designers to the DAC-I project.These were real should be to learn how to place thepower of the designers, not computer people, and it wasn't long computer directly into the hands ofthe student and before they were making significant contributions how to start moving the computerinto the main based on their experience and knowledge of design. stream of engineering education. During their first year using DAC-I, theywere able to design a complete trunk lid, both outer panel and PA YNTER: I gather thatnow we're ready to inner panel, on an experimental basis.Figure 3 come back to Miller's master designer, because shows the results of their efforts.This drawing he is obviously not theperson Hart was speaking was recorded directly on film by the computer using of, certainly not the three-,four- or five-year the Image Processor unit of the DAC-I system. product, although perhaps thesix-, seven- or In a written paper, it is not possible to describe eight-year product. I got thedistinct impression the dynamic nature of the conversational interaction from Hart's picture oftwo men talking to each between designer and computer usinga graphic other over the drafting boardthat these people console. For those whoare interested, a 3-minute were not necessarily master designers, but simply movie on DAC-I is availableon loan. 4 competent, typical engineers. Wouldyou care to Figures 4 and 5 are examples of the type of image suggest how you think schools mightmake this which is displayed for the designer at theconsole. particular group of peopleaware of what they Figure 4 is a blow-up ofa small portion of the could do with computers? inner panel of the trunk lid ofa car.It shows the HART: I think this cutout hole and reinforcement where the courtesy comes back to the last topic, light will be mounted. which was that part of thegoal of the educational process is to impart to the student When the designer studied this display, hesaw new concepts and ideas, and the goal ofimproved education, that the subassembly was not properly located rela- by moving the computer tive to the surrounding surfaces.In only a few into the mainstream of minutes at the console he was able to supply all engineering education, isto increase the rate of concepts per unit time that you'reable to stick of the information necessary to redesign thissub- into the headbone of the assembly 11/2" to the left. The resultsare shown in student. My initialre- Figure 5. marks were thatwe are interested in engineers who have intelligence,imagination and the ability With experimental applications suchas this, we to approach a new job with have shown that it is perfectly possible to place the confidence (so far I haven't seen that the universityimparts anything power of the computer into the hands of the engi- to the student but the intelligence neer so that he can use it as a design tool without and the rest of it he learnsas he goes along). becoming a computer programmer.I should point Part of the out that this was no small jobthe DAC-I software current attempt to get studentsinvolved in actual design problems during theiruniversity training runs into well over a million computer instructions. is the wish to give them But with the DAC-I laboratorywe have demon- something more than tools to work with, and alsosome knowledge of 4"Design Augmented by Computers:DAC-I," available how to apply these toolsand what itmeans to from the General Motors Corporation, PublicRelations Staff, apply these tools. I thinkthis is very worthwhile. Film Library, General Motors Building, Detroit,Michigan, I would like tosee the computer used more in this 48202. type of design workshop. 86 r.4 Wii 1.1011111WWINIWIZINIME

LOWE: It appears to me that- todaywe have menced to use the tools, they began to get some sensed, in the minds of many people here, the insight into the real capability which this new feeling that there is a real barrier between the machine would provide for them. For example, computer designer and the computeruser. How- if you work all day on a sheet of paper which is ever, you said that the men who design the car two-dimensional, you begin to think in terms of body actually contributed to the design of the a two-dimensional world. We found that the de- system, and that they came in as guinea pigs and signers who came in and worked with us did just in a very short time turned into contributors.I thatThey were working in a two-dimensional wonder if you could amplify thisa link. environment and it took them a while to adjust HART: We could have spent tenyears doing a to the fact that, in working with the computer, systems study on how automobiles get designed. they were really dealing with a three-dimensional But we concluded that we weren't going to be- medium, and they had to think in terms of work- come smart enough fast enough to learn how to ing in three dimension. So their concepts of how design automobiles so that we could design an they ought to do design changed as the tools automobile design system. So our goal and ap- changed. These two phenomena interplayed back proach had to be entirely different. We observed and forth in developing the laboratory to the one thing about the design process: whatever a state where it is today. man was designing, whether it was automobiles, QUESTION: Would not the same system, which spark plugs, or what have you, he used the same could be the operational base upon which your tools. He used a ruler, a french curve, a compass, engineers actually conduct design within General and a straight edge. So our basic approach was Motors, be a very powerful tool for teaching a to provide the designer with a set of reasonably new generation of designers within the engineer- primitive tools and means by which he could ing school? Suppose that we could afford to give operate these tools in sequence in order to carry the young men, before they came to you, two out a design process. As I indicated, we had to yews of apprenticeship training under the guid- develop these tools in a kind ofa vacuum, be- ance of some pretty good design people. Wouldn't cause, when we approached the designers and the DAC system be a wonderful way to teach de- asked them what sort of things they would need sign in a university? if they were going to use a computer to help them with design, they just raised their eyebrows atus HART: I think it would because it's a rather pain- and suggested that we quietly retire toour lab- less way to learn descriptive geometry and the oratory and leave them alone to do their jobs. So visualisation and interactions of things in space. we had to do the best we could to build what we BRANIN: I just want to make a brief summary of thought would be an adequate set of tools to get a couple of points which seem to have emerged started. And the the designerscame in as from our session today. The question of acces- guinea pigs and attempted to use these tools to sibility to the computer has been brought out. It's solve real design problems on an experimental really not accessibility to the computer per se, basis. They rapidly discovered thatsome of the but to its facility.The other question was: how tools we provided weren't good andsome of the do we get engineers to use the computer? tools they needed we hadn't provided. Butan- think these points both have been very sharply other phenomenon also took place:as they cora- highlighted by Hare. disc:anion.

87 c Itt

r 7 41

INV

IromowAllOw/pwooto

tstiritteosotttitorteriltztzle

PRECEDING PAGE BUNK- NOT FILMED COMPUTERS IN ENGINEERING DESIGN DONALD L. KATZ The University of Michigan Ann Arbor, Mich.

At a meeting in 1962 of the Committeeon Engi- Ford Foundation project. 1It had as major objec- neering Design of the Commission on Engineering tives: (1) the training of engineering design teach- Education, one task outlined for engineering edu- ers in subjects related to computer-aided design, cators was the development of the computer as a (2) the study of the role of the computer in engi- design tool.Following these discussions, a pro- neering design education, and (8) the generation posal was written to obtain support for the project of a substantial number of completely documented which is being reviewed on this program.In the computer-oriented engineering design problems. spring of 1964, a grant was received from the In the fall of 1964, engineering design teachers National Science Foundation for support. with substantial computing experience were asked ENGINEERING DESIGN to apply for an intensive Ale-I,veek program to be In the span of years encompassed by my contact held at the University of Michigan during the wan- with engineering education, engineering design has mer of 1965. An Advisory Committee of the Com- gone through three stages. The first stage was that mission on Engineering Education sassisted in of adopting the current technology along with modi- selecting from among the applicants twenty-nine fications required by the new process under con- engineering design teachers from twenty-three sideration.In 1930, as a student, I designed a different engineering schools (list attached). petroleum .separation plant utilizing information Five weeks of the summer program were devoted available on fired heaters,distillation columns, to formal presentations on computation, numerical condensers, and treaters. Gradually there appeared mathematics, modeling and simulation, mathemati- a conflict between the developing engineering sci- cal optimization techniques, problem-oriented lan- ences and the art in form of technology. Accord- guages, man-machine interaction, economics, and ingly, design problems of the second variety were industrial design practice.During the remaining developed.These problems required the creation four weeks, each participating professor solved and of a process or system based on fundamental cal- documented one or more individua?, design problems culations for the various components using rigorous appropriate for inclusion in an ugineering design rate process equations, separation process rela- course in his own field. tionships, and other quantitative material.Such During the fall of 1965, the project staff and one designs became rather involved and tedious.It was summer-program participant from each of five necessary to make a good choice of the parameters engineering disciplines edited or revised solutions in the first place to permit completion of the design to the participants' design problems and prepared problem within the time limits permitted in the material for the final project report.This report educational environment. consists of six volumes, one summarizing project With the advent of computer usage in the engi- activities and five oriented toward design in indivi- neering sciences, it became rather clear that engi- dual engineering disciplines as follows : neering design would soon be relegated to the Volume I. Summary computerin good part because of the need for Volume II. Chemical Engineering rapid calculations. It appeared that the universities Volume III. Civil Engineering should find a format by which design could be Volume IV. Electrical Engineering upgraded to include the computer techniques being Volume V. Industrial Engineering adopted by industry.Also, the need for having Volume VI. Mechanical Engineering design teachers conducting studies at the forefront Volume I includes an historical description of the of knowledge in this area was indicated.It was project, details of the summer program, recom- this view and the discussion with the Commission mendations, review papers on the subjects covered on Engineering Education's Conirnittee on Design during the summer program, and abstracts of that prompted the preparation of the proposal for forty-three of the design problems prepared by the ouK project. The project organized in the fall of 1964 at The 2N. M. Newmark, Chairman (University of Michigan) University of Michigan was modeled after an earlier S. E. Elmaghraby (Yale University) S. J. Fenves (University of Illinois) 1Preject on Use of Computers in Engineering Education, D. E. Hart (Gener'tl Motors Corporation) -"University of Michigan (1969-63). A. Schultz (Cornell University) summer program participants.Each of the five its full potential for engineering design work disciplinary reports includes a brief statementon and wish to assign rather sophisticated prob- the role of computer-aided design in the field and lems, special purpose problem-oriented lan- several completely documented computer problems. guages should be used when available.If this These reports result from the combined efforts of is not practicable, then at the very minimum the project staff and the first authoras principal a substantial library of useful design sub- editor. The example problems insome cases were routines should be available. modified after correspondence between editorand 6. Design problems usually have an infinite problem author. number of feasible solutions, and experience Some general conclusions and recommendations and intuition play a significant role in the of the project staff are: solution process.Engineering design proce- 1. Engineering schools should continue to develop dures are often very difficult to automate introductory computingcourses to provide completely. In order to use the computer most every engineering student with an adequate creatively, the designer must be able to con- background in digital computation and instruc- verse with the computer in a virtually conver- tion in at least one procedure-orientedpro- sational way. Turnaround times of days, hours, gramming language. or even minutes are not generally acceptable. 2. Computing work should beincorporated into The interaction must be effectively instantane- the required upper-level analysis (engineering ous. science)courses;before graduation, each 7. Time-shared computers will havea great student should solvemany engineering prob- impact on the place of computers in engineer- iemzt of graded difficulty usinga computer. ing education in general. The student will 8. If r. thi introductory programmingcourse is be able to debug hisprogram and solye a taught early in the curriculum,as usually oc- problem at one sitting. He should be able to curs and is recommended, then there should solve reasonably difficult problems between be a later treatment ofmore sophisticated class meetings. In addition, it will be practical mathematical topics available, to engineering for the engineering teacher to bringan on-line undergraduates. The student will need: console, perhaps with graphical input and a. More advanced mathematics, in particular output, directly into the classroom. He will numerical mathematics and mathematical be able to display solutions for all tosee as statistics; the computation is being done.Engineering b. more careful instruction in the techniques faculty should be prepared to teach theuse of mathematical model building andsimu- of such approaches within the next twoor lation; three years. c. an understanding of the most important 8. With the introduction of time-shared hardware of the mathematical optimization tech- and the development of problem-or design- niques; oriented software, the student designer for the d. exposure to a variety of computing sub- first time will have the opportunity to investi- jects, including specialpurpose computing gate a large number of feasible solutions to languages and the manipulation of graphi- his design problem. He will be able to gain cal information. some facets of engineering experience which 4. The mathematical tools(numerical mathe- cannot now be acquired outside the industrial matics, mathematical model building, optimiza- environment.Hence, the next generation of tion techniques, etc.) needed for successful computing systems should enable the student integration of computer work into engineering to improve his intuition about and under- design courses are basically thesame for all standing of the nature of good design, which engineering disciplines, although thereare is, after 4ll, the raison d'être of engineering some differences in emphasis among theen- design courses. gineering specialties.Therefore,itis not 9. For financial reasons, itmay not be possible unreasonable to teach such material inan or practical for a small school to acquire a interdisciplinary course. At thesame time, it large time-shared computing system forits should be recognized that problemsare most exclusive use.The best approach formany meaningful when taken from the areaof schools will be to acquire remote terminals interest of the student. Hence, problemsused connected to a centralprocessor shared with in an interdisciplinarycourse would have to other university or industrialusers.Even be chosen very carefully. the larger schools will find the totalcost of 5. If we expect students touse the computer to operation, though smalleron a unit computa- 46

4. ,^-

tion basis, to be substantially larger than for Thomas J. Schriber, Assistant Professor of Math- the_ present generation of.computing equip- ematiCs,Esstern Michigan ,University ment. A very large increase in demands for Donald B. Wikoii; Astilitant-Profeeieer of Chemi-- computing services is almost certain to take cal Engineering, New Mexico State University place over the next few years.This will John M. Woods, Associate Professor of Chemical probably mean commitment of a growing Engineering, Worcester Polytechnic Inrititute share of the university's budget to support the computer and its services. University adminis- CIVIL ENGINEERING trators should be planning now for support Rodolfo J. Aguilar, Assistant Professor of Civil of these activities. Engineering, Louisiana State University 10. The key to successful integration of computers Richard T. Douty, Associate Professor of Civil into engineering design course work is the Engineering, University of Missouri at Colum- knowledgeable andinterestedteacher. A. bia considerable educational effort will be required Ti Huang, Associate Professor of Civil Engineer- to train our engineering design teachers to ing, New Mexico State University use the powerful new computing systems and Albert D. M. Lewis, Associate Professor of Civil languages. Engineering, Purdue University Copies of the summary and individual disciplinary Allan J. Malvick, Assistant Professor of Engi- reports are being distributed to appropriate depart- neering Science, University of Notre Dame ments at all accredited engineering schools.Hard- Charles H. Schilling, Professor of Military Art bound copies of the six-volume report are being andEngineering,UnitedStatesMilitary sent to engineering libraries.Copies of individual Academy report volumes will be furnished without charge to Wolsey A. Semple, Assistant Professor of Civil full time engineering faculty members. A limited Engineering, Howard University number are available for sale to industrial organi- Theodore G. Toridis, Assistant Professor of Civil zations from the Publications Distribution Service Engineering, George Washington University of The University of Michigan. During the 1966-67 school year, the project staff ELECTRICAL ENGINEERING expects to conduct several two-day "information" Ernest E. Erickson, Associate Professor of Elec- sessions to stimulate interest in introducing com- trical Engineering, Louisiana State University puter work into engineering design courses.Rep- L. Dale Harris, Acting Dean of Engineering, resentative engineering design teachers will be University of Utah invited to attend local meetings to be held in New Jens J. Jonsson, Professor of Electrical Engi- York, Boston, Atlanta, Dallas, Denver, San Fran- neering, Brigham Young University cisco, and Ann Arbor.In addition, an intensive E. Lawrence McMahon, Associate Professor of two-week workshop will be conducted in Ann Arbor Electrical Engineering, University of Michigan for about fifty engineering design teachers. Staff and Lecturers on Project INDUSTRIAL ENGINEERING Brice Carnahan, Associate Director in Charge Harry Benford, Professor of Naval Architecture Warren Sieder, Graduate Assistant and Marine Engineering, University of Michi- Dale Rudd, University of Wisconsin gan Robert Morris, Bell Laboratories, Inc. Norbert Hauser, Associate Professor of Indus- R. S. Miles, Autonetics, Inc. trial Engineering, New York University (now C. Christensen and R. H. Roth, Bell Laboratories, Professor of Industrial Engineering, Polytech- Inc. nic Institute of Brooklyn) Robert Norman, Chevron Research Gonzalo Mitre-Salazar, Professor of Industrial R. Gellatly, Bell Aero Systems Engineering, Monterrey Institute of Tech- Don Hart, General Motors Corp. nology and Advanced Studies (Mexico) Paul Shannon, Dartmouth College Frank D. Vickers, Assistant Professor of Indus- Participants on Projee trial Engineering, University of Florida CHEMICAL ENGINEERING MECHANICAL ENGINEERING Donald R. Brutvan,.Associate Professor of Chemi- Rollin C. Dix, Assistant Professor of Mechanical cal Engineering, State University of New York at Buffalo Engineering, Illinois Institute of Technology Rodolphe L. MotaTd, Associate Professor of I Also served as participants in fall term 1965 to assist Chemical Engineering, University of Houston with preparation of reports. 47

', 4 fp, - ft ,

Charles R. ht e, PrOfessor of Mechanical En- Paul F. Youngdap, Associate Professorof Meth: intieringilovwStito-Vailveridty James P. O'Leary, Aidatant Professor of En- 1 John R. Zimmerman, Assistant Professorof gineering Graphics and Design, TuftsUniver- Mechanical Engineering, PennsylvaniaState sity University Chester A. Peyrounin, Jr., Professorof Mechani- cal Engineering, Tulane University Charles A. Timko, Assistant Professor of Mech- 'Alive served as participant in fall tom, 1$65 to sigh* anical Engineering, Universityof Notre Dame with preparation of reports.

1 .ressos.ainFramusl=maasoi --4=0"47,Z=Mmrorsew-=1"--.4

PREPARING THE UNDERGRADUATE FOR COMPUTER-AIDED DESIGN CHARLES R. MISCHKE Iowa State University Ames, Iowa

The principal effort in the preparation of the of the synthesis) until the ultimate form of synthe- undergraduate for participation in computer-aided sis is possible.Although this state of affairs is a design is made before his last design course.The goal of engineering, its very attainment relegates preparation by the student is not as involved, nor such synthesis to the status of routine. as extensive, as superficial examination might indi- The frontiers of engineering include the excite- cate. However, the work to be done by the faculty ment (and demands) of design work in a domain seeking to offer computer-aided design is different wherein understanding is poor, but the need is great. to an extent that is greater than is generally ap- Hence there is a high challenge in searching for preciated. The purpose of this paper is to state the meritorious solutions by better analysis techniques problem and to offer a technique for solution. (one route) or by clever search techniques in the DEFINITIONS domain of satisfactory alternatives (another route). An engineer plans and/or controls (i. e., designs) PERSPECTIVES CONCERNING THE COMPUTER the interaction between matter, energy, men and The role of the computer is becoming more power- money in order to accomplish a specified purpose. ful as the ability to simulate systems on it is devel- Design is the essence of the engineering function, oped. The computer calculates with speed, precision although there are many engineering functions that and without fatigue.Are these abilities useful to are not design. the designer?Certainly. To design is to specify precisely how a task is to Should ability to exploit these capabilities be be accomplished. To specify requires creativeness taught to undergraduates in engineering?The and understanding. 1Clearly, design is not taught rapid growth of computer utilization in industry in school. It is learned, if, indeed, it is learned at all, over the last ten years indicates "perhaps," but the in the field, with and under a master artisan. advent of time-sharing computer hardwarecan lead Schools can, however, teach attitudes, approaches, to the development of a computer utility, likean conceptual tools and an appreciation of the mor- electrical power utility, which could place in every phology of design. engineering office a remote access to the largest The basis of understanding is usually analysis, computers available. The use of large computers will which is itself based upon mathematical and em- become a common part of design life.The advent pirical models and experience, and is a useful means of time-sharing will demand that four-year engi- of evaluating alternatives.Selection of the best neering graduates have some experience working alternative requires : fruitfully with a computer and have a broad under- (1) ability to recognize a satisfactory solution ; standing of current useful techniques. (2) ability to assess the merit of alternatives and Let us consider wherein computer-aided design order them accordingly. assists engineering education, and what skillsare Synthesis means the generation of a solution and needed.The engineering designer often proceeds it is arrived at by understanding the problem. The as follows: degree of merit of the synthesis varies directly (1) Defines his problem in engineering terms. with the ability of the student to understand. That (2) Decides how he will recognize a satisfactory is, poor understanding generates only weak solu- solution. tions, whereas complete understanding may enable (3) Determines how he will recognizea better the student to derive a solution of high merit at (or best), alternative. the first attempt. (4) Generates alternatives. Engineering advances in the field of analysis pro- (5) Through analysis, retains suitable alterna- mote understanding (and therefore the elegance tives. (6) Chooses best alternative. 1Strangely enough, although the handmaidens of design are creativeness and understanding, engineering schools do (7) Acts upon the decision in (6). not select students on creative potential, nor do they as a Since the computer can only converse ina mathe- rule make earnest constructive efforts to polish or cull on matical language, the hovering ofa computer in the creative bases. background affects these seven stepsas follows : 49 (1) The engineer must eventually prepare a straints, or whims, or prejudices of thedeci- mathematical statement of his problem for sion-maker. the computer. The pressure to state his If we concede that the computercan be helpful problem in mathematical engineeringterms is a wholesome influence. to the designer (and therefore tothe student designer), what is the price of givinga suitable (2) A description ofa satisfactory solution in introduction to computer utilization in designto the mathematical terms is required bythe com- four-year engineering student? puter;every constraint uponsolutions First, let us be practical. should be involved here. If an engineering prob- The computer re- lem had to be solved and programmedfrom scratch quirements pressure the designerto be on each and every occasion, the computer would find orderly, explicit, free of ambiguity, quantita- little use in the designroom. The overhead to plan tive and exact. and debug the computer (3) A figure of program would be large, merit.(utility function, object and only a small number ofprograms wo) be- function, etc.) is always required byengi- come economically feasible. Any notion thata prob- eers.It may be non-quantifiable, suchas lem once successfully completedwould be encoun- an experienced expert's value judgement (a tered again should be suppressed. tea-taster, for example). The computercan- If the overhead of time andmoney associated not function as a tea-taster unless the infor- with computeruse is substantial, how can previously mation and experience is placed in mathe- invested time and experience besalvaged so that matical form.Again the pressure isupon the long-term costper problem becomes reasonable? the designer to make his utilityfunction The answer to this questionis the same, surpris- mathematical. (It also prevents double-talk, ingly, whetheryou are talking about a human such as "I made itas cheaply and reliably problem-solver or a mechanicalone. A human is as possible," while offering a mediocre alter- trained to provide conditionedresponses to very native.) A computerprogram would reveal elementary situations.A designer functions be- explicitly the designer'sutilityfunction. cause he is able to bring to bearupon a problem Again, the pressure to quantifyjudgements a whole bagful of conditionedresponses, arranged is wholesome. and ordered by him through intellect,in such a (4) In this generation of alternativesthe com- way as to solve the complicated problemat hand. putDr works witha speed and accuracy with Just as the human is taughta repertoire of condi- which no man can compete. In thisstep the tioned responses,so can the computer be instructed. economic contribution of the computeris It then becomes the role of strong. the human to arrange the responses intoa sequential strategy fora solu- (5) Analysis programming hasa history as long tion. The computer capabilitycorresponding to the as that of the computer, and again the speed learning and remembering ofconditioned responses and accuracy of the computer exceeds thatof to elementary problems is calledthe subroutine manual calculation. capability of the computer. (6) Using the computer for this stepremoves A design room (oran educational department) much of the subjective prejudicein deci- requires a library of subroutines,available for sion-making ("I likea cam for this applica- quick access in any specified orderby the designer tion rather than a four-bar linkage.").Dis- (student) wishing touse the computer for design. enchantment with computer decisions tells The details on how to assemble andhow to document the engineer either that his programming of such a subroutine libraryare not germane to my step (3) is open to improvement,or that subject, but the existence of sucha library is pre- his prejudices are showing. sumed for the remainder of thispaper. (7) The computer even contributes to theimple- Given a problem in analysis, allthat remains for mentation of the decision.If the engineer the computeruser to do is to plan a strategy of needs clearance from a senior beforehe can calling library subroutines in proceed, the complete computer a specified order to program to- initiate computation which willproduce the required gether with the recommendation willallow analysis in numerical terms.Our subject here is the senior to quickly inspect the meritfunc- design by computer. Theprevious procedure is tion employed and consider the implications. necessary to computer-aided design, but itis not The latitude available to the decision-maker sufficient. is apparent. A very fiat figure ofmerit sur- What can a computer do? face in the domain of the extremum It can add, subtract, allows multiply, divide and perform otherconcerted actions considerable variation in parametersat small which we simply call calculation. cost to accommodate non-quantifiable The fundamental con- and critical activity in designis decision-making. 50 30, 4irAt'r,-'1,.4,1,1, 44i

Can the computer make decisions? Thecomputer tion). This is a procedure that can be effectedwith can take direction andmake logical decisions of the the computer and involves itsdecision-making following kinds : capabilities. Step (4) requires creativeness andthe a. It can take directionto the extent of trans- computer (a model of consistency) has no creative ferring control to a given location in thelogic talent. the flow diagram of a program, regardless of A TECHNIQUE FOR COMPUTER-AIDED DESIGN circumstances (the unconditional GO TO state- We have mentioned the elements of computer- ment). aided design. A technique for making it practical b. It can make a quantitative comparisonand, depending upon whether a mathematical ex- consists of the following steps: pression is positive, negative, or zero, it will a. The designer-student writes an EXECUTIVE branch in its logic and undertake adifferent PROGRAM to read in all necessary informa- possibility (the IF tion and write the output documentation to his course of action for each This executive program calls the statement). problem. c. It can make aquantitative determination and SEARCH subroutine. undertake a number of different coursesof b. The SEARCH subroutine is a library program action depending upon the value of amathe- that will crawl over a merit hyper-surface and matical quantity (the computed GOTO state- discover the location of the largest ordinate ment). to the accuracy specified in the executive pro- d. It can perform a calculation or aroutine of gram. The search program calls the MERIT calculation iteratively for any specified num- subroutine. ber of times (the DO statement). c. The designer-student writes the MERIT sub- e. It can read andwrite alphanumeric informa- routine which generates alternatives and, if tion. they are satisfactory, calculates their merit.. f. It can make decisions based onthe truth or In so doing this routine calls DESIGN library falsity of a statement and proceed on alogic subroutines. path dependent upon that determination(the d. The DESIGN library subroutines are analysis Boolean algebraic capability). routines that have been written, checked out, g. It will remember(store in memory and recall) found useful and are documented in a con- quantitative and alphabetic information,and venient way to facilitate their use. (These are it will seem to remember how to do aspecified largly prewritten routines occasionally aug- task in which it has had prior experience(the mented by the designer-student.) SUBROUTINE capability). In the beginning of an operation; as in a school h. It can draw, i. e., plot, two-dimensionalfigures, seeking to initiate computer-aided design experience and (with some computers) draw on a cathode- among its students, the design library will be modest ray tube tocommunicate with the designer and prewritten by the design faculty, but, as prob- (the graphical output capability) . lems are solved and new routines added to the We can roe that the computer has somedecision- library, an impressive design capability can be making capabilities. To the extent that thedesigner amassed. can reduce hisdecisions to combinations of the Now, given the design library and the search computer's decision capabilities, then thecomputer program, just what mathematical preparation must can be trained tofunction as a capable apprentice. the students have in order to use the computer for. consistency, it In some things, such as in speed and design ? will excel its master. The analysis capability of the computerwith a The composition of merit functions is technically library of subroutines serves thedesigner in step simple, but conceptually can be very complicated. (5) of the design process.Step (2)is usually This is not a computer subject; it is a design subject. suggested by the constraints unearthed in step(1). When lacking much sophistication, one can resort Step (8), which is central todesign, is often sup- to the use of the simpler merit functions such as pressed, or provided by a judgement. Thecomputer weight, cost, or geometrical attributes. demands that this step be quantitative.The de- Searching procedures need not be learned to any signer must provide a quantitativerelationship great extent because the search program is provided. whose magnitude represents the merit of asatis- The designer's programming problems will be factory alternative.As soon as this is done, the principally those of solving algebraic, transcenden- problem remaining is to search over the domainof tal, differential and simultaneous equations, evalu- the satisfactory alternatives, and seekthe one with ating integrals and derivatives, curve fitting, and the highest merit (largest value ofthe merit func- using some probabilistic ideas. 51

SC To a large extent most of the places where these In other words, cannot the basis forunderstand- techniques are useful and applicableare within the ing- this Mathematical materialbe brought into design subroutines themselves, whichare provided coincidence with the basis of computer principally by the design subroutine techniques? library.To To do so would placea double-edged sword in the some extent the methods are used by the designer hand of the student. If the mathematical in constructing his merit subroutine. sequence of some twenty-odd semestercredit hours taken by The University of Michigan Projectproduced 48 engineers is taught toa lesser extent from the tra- problems, solved and documentedby individuals ditional viewpoint and to working without the benefit of a greater extent from a design subroutine a numerical analysis vantage point, thestudent library or a rather general searchprogram. Twenty- pattern of thought would be much nine of the problems utilized more computer optimization (search) oriented, even if the computerwere never men- techniques and were of varyingcomplexity, cutting tioned! (And his mathematicswould be no less across most of the engineering disciplines.Never- useful.) theless, the list of optimization techniques and Given a new mathematicalsequence and a simple numerical methods is neither largein extent nor programming experience ina language such as impressive in complexity. FORTRAN in the firstyear, and given continued Thus, an instructorcan choose several design interaction with the computer in problems constituting the entire the mathematics computer-aided courses, more experience with it thereafteris redesign experience wherein the levelof sophistica- quired.In order to keep alive the tion and involvement with specialized student's pro- techniques is gramming ability and his numericalanalysis ideas, in keeping witha very modest preparation and every engineering course should require ability among his students. at least Not much numerical one programming problem each semester.I am not analysis preparation is easential toshow the student suggesting the use of just the potential of computer-aided any old program to keep design. a curriculum committee happy.The program Although some fine problemscan be solved with problem should be selected a modest executive routine, how in order to make the can some prepara- course better for its inclusion.Such opportunities tion in mathematical methods suitableto computer abound. implementation be given to theundergraduate Under this arrangement, student? engineering students reaching their senioryear and first comprehensive A SUBTLE PREPARATION design course (not necessarilytheir first design Where, oh, where, inan already overcrowded course. There is something to be said fora design curriculum, can this material be introducedwithout course every year) will be able touse the computer displacing other fundamental studies? as a genuine tool in their design problems. This is a perplexing question. Perhapsit is time Design is a broad word andthe meaning of the for the engineer to re-examine hisview of mathe- words design problemhere is one in which the matics and its use in the light of thecomputer's totality of the student'sprior experience is drawn role, present and future.The place to teach the upon, one that taxes their creativeskills, and one material needed for solving equations,integrals and that requires a search foroptimality (say cost)or derivatives is in the mathematicalsequences re- a trade-off optimum (say weight,cost and relia- quired of all engineering students and,to a lesser bility). extent, in every engineeringcourse in the curricu- A FRINGE BENEFIT lum. Probably the most importantreason for intro- Let's be specific.Is the time approaching when ducingcomputerprogramming it would be more useful to view the earlyinthe entire spectrum curriculum has to do withthe handmaidens of de- of mathematicsgermane in engineering from a signcreativeness and numerical analysis viewpoint? Cannot understanding. Engineering differentia- students are attracted toengineering partlyas a tion be taught from a finite differenceand numerical place to vent their creative analysis viewpoint? abilities.The usual Cannot infiniteseries be engineering curriculum stiflesthis natural enthu- viewed as partial sums with compensatingtrunca- siasra by demanding "conformnowcreate later" tion errors?Cannot integration be viewedas a until by the senioryear they are largly sheep. summing up of rectangular (or trapezoidal) (If areas? my colleagues doubt this, try the followingexperi- Can't sophomores study linear algebra,matrix ment:Ask a group of seniorsand a group of ideas, and solution to simultaneous equations(alge- freshmen for an "off the braic, differential, partial) top of the head" solution together with computer to a broad problem.You'll be surprised wherethe implementation of same? Under thesecircumstances creative answers originate.) students would not knowa Bessel function when The writing of a computer they saw one, but do they know program is a creative one now? act.Since there aremany, many ways to skina 52 ..04F7WittP,r7

cat, creativeness can flourish here, and there is little QUESTION: You keep talking about developinu a cost associated with it' Mistakes- are not-preserved Campater4rOgnitittning' Skill" ands then" keeping in cast iron, intricate machiningor non-functioning it alive until you corn use it.Why not develop equipment. The merits of alternative appioaches it just before you need it? can be demonstrated quickly. MLSCHKE: Ideally, I think, a man ought to beex- Another act of the designer which is submerged posed to design from the freshmanyear all the in engineering education is that of decision.Cour- way through.There's a lot to be said fora age in decision-making -comes from confidence in design course every year in the man's experience, tools of decision and a prior history ofsuccess. A and it can start out very trivially in terms of computer program is replete with decisions and the background required and givesome ideas con- consequences follow immediately (within the turn- cerning optimization and simple problems (they around time of the computer installation).What might even be ones to showup the math depart- better opportunity to sharpen the tools of decision ment) but problems whichare too hard to do and build confidence than in this aspect of computer by pushing a pencil. programming? Thus, a simple course inprogram- QUESTION: Are you then equating theuse of the ming can have secondary products of exercisingcre- computer and the use of the design educationas ative and decision-making skills. being- co-equals? The changes indicated' here will onlycome about MISCHKE: No. The computer isa tool.I think through the initiative of the engineering educational I made it clear that cookingup a merit function community, courage to experiment, hardheaded and figuring out how to generate alternativesare evaluation, and considerable effort whose onlysure design activities. The machine does the rest. reward is the excitement of doing and contributing. QUESTION: I wasn't disagreeing withyou, but I The dangers inherent in such changesare the same was trying to see whether you were saying that as in. any change.Perhaps the most important the purpose of introducing the computer into the necessary (but not sufficient) item is whole-hearted engineering curriculum at allwas for its use in faculty support of whatever plan is implemented. design courses. Nothing dulls a student's enthusiasmmore than MISCHKE: No, other people ought touse it, too. to have had one professorassure him something is At my school the amount ofuse by the disciplines important, only to have a subsequent professor within M. E. is exceedingly small at the moment. show disinterest (or ignorance). We hope to jar them a little. SUMMARY CRANDALL: I know in our particurlar depart- The computer can perform the design function. ment in which people take fouror five courses, we give them five hours of work and three hours An undergraduate student can be givena meaningful introduction to the computer's capability in worth of credit. Do you realize whata burden design if : this might put ona student as far as an academic load if every one of these classes requireda com- 1. He receives a programmingcourse in his first puter program? year. MISCHKE: One programper course per semester? 2. His mathematics sequence is enriched with It doesn't have to be very sophisticatedto teach ideas and approaches froma numerical ana- the ideas. In fact.that's the challenge. lysis viewpoint, and the computer is usedas (laughter) a teaching aid. CRANDALL: No, I'm thinking that inour basic 8. Every engineering course requiresone prob- computer course we require approximately five lem in which the computer is used. or six problems in the one-quarter system. Ifyou do 4. The faculty develops a design subroutine this in every quarter from then library in advance. on, they are in effect taking an additionalcourse load as I see it. 5. The faculty develops an efficient search sub- MISCHKE: No comment. routine. CARNAHAN: I think the new computingsystems 6. Consideration is given to formulating figures are going to make quite .a difference in what can of merit. be done in the way of assignments.That is, in 7. A design experience is permitted in which the the past it has takena minimum of two weeks to student writes executive programs and merit do a problem with any complexity at all, but with subroutines to solve design problems. new systems I think it will certainly not beun- Once convinced of the potential of the computer reasonable to ask a student to doa problem be- as a design tool, the engineering graduate will be tween Monday and Wednesday if hecan sit down motivated to learn on his own all hecan concerning at the console and get sixruns and debug his pro- methods and approaches useful in computer work. gram 58 THE SIMULATION SPECTRUM NORBERT HAUSER Polytechnic Institute of Brooklyn Brooklyn, N. Y.

This presentation is based on the Simulation MODEL VALIDATION Chapter of Reference (1), which has been distri- Since it is impossible to prove the validity of a buted toConference attendees.The following given model, the best that can be hoped for is that remarks confine themselves to a few aspects of itwill withstand a series of tests, performing model building and digital computer simulation. reasonably and producing reasonable results.In A process or system model is of practical value analytic models the danger arises from oversimpli- only if (a) it adequately represents theprocess, and fication introduced in order to make the model suit- (b) it is amenable to some form of mathematicalor ableto mathematicalanalysis. Insimulation, logical analysis leading to a better under tanding of however, the danger frequently is of the opposite the process. kind.The desire for realism may cause some im- Currently available methods to implement (b) port' nt features to be overwhelmed by a mass of frequently require violations of (a). When there realistic but trivial information.Assuming the is a conflict between realism desired andpower and primary purpose to be a gain of insight into a elegance of approach, the choice appears to be system's cause-and-effect relationships, it is essen- between seeking (1) the exact solution toan ap- tial that all material irrelevant to the particular proximate problem, or (2) an approximate solution investigation be left out.This introduces some to the exact problem. The first choice (1) is per- circularity :if we knew exactly which variables are fectly acceptable if the resulting distortion does relevant, we probably would not need to simulate. not significantly affect the problem's characteristics, Model construction thus becomes an iterative pro- or if the engineer is able to evaluate and counteract cedure consisting alternately of (a) selecting varia- the distortion's consequences.The second one (2) bles and specifying their effect on each other, and on the other hand implies the ability to duplicate (b) running the model and observing its behavior. an entire system (or to operate on it directly). If this behavior is unsatisfactory, either some rele- Even if this were feasible, the time and expense vant variables were excluded, or the interactions required to design and execute controlled experi- were inadequately specified. When the behavior ap- ments necessary to isolate relevant variables makes pears to be reasonable, the sensitivity of all variables this approach a last resort.In most cases the and parameters should be explored. Besides leading engineer looks for a point somewhere between these to the possible elimination of some variables, this extremes. He is quite satisfied with a formulation will indicate the precision required of parameter which has abstracted the system's essential 1 charac- estimates and other data used. teristics, permitting him to apply other than brute Not until we have reached this point are we force methods. ready to collect data from the system, and to deter- Simulation is one of the available methods.It mine how much time and effort should be expended requires no assumptions of linearity,differenti- in improving the accuracy and precision of this ability, or any other form of mathematically good data.The mistake of building the model around behavior.A set of mathematical and/or logical available data, or of collecting the data before statements describing the exact (in a deterministic constructing the model is frequently made.The or stochastic sense) relationships among all compo- previous discussion should make the improvidence nents is needed.The simulation then produces the of such an approach apparent. overall effect of these iteractions.It is important to realize that simulation need not be considered solely EXPERIMENTAL DESIGN an alternative to analytical methods. Many times Since simulation as such provides no algorithm leading to improvements in succesive runs, effective both approaches may be used concurrently.Indeed, one of the cardinal rules of efficient simulation search methods (machine or man-machine) have to be incorporated. design is not to simulate those parts ofa model This topic, including an extensive which can be treated analytically. bibliography, is discussed in Reference (1).As in all experimental techniques, thought must be given to designing a method which will yield maximum lEssential with respect to a particular application.Thus a model designed to answer one set of questions about a information for a given amount of computation.If system may be worthless when confronted with another set random events are simulated, estimates of the pre- about the same system. cision of computed output statistics are needed. 54 Alternately stated, how many runs are needed,or tor, lawyer, or teacher, simulation may achieve its how long should a single run be to achieve a given greatest success by making itself superfluous. interval of precision at a stated level of confidence? To answer these questions, the simulator must have SIMULATION LANGUAGE considerable knowledge of probability and statistics. A host of specialized digital-computer simulation languages, intended to facilitate the construction COST of programs ranging from discrete queueing models The cost of programming, debugging, and run- to analog representations of differential equations, ning a simulation are usually high.Furthermore, are currently in use, and are constantly being im- it invariably requires more timeplanning, testing, proved. The reader is referred to References (1) executingthan originally anticipated.The avail- and (2). ability of simulation languages may reduce programming time and cost somewhat. Simulation does offer some outstanding advant- References ages, however. It can be used effectively in training (1) Katz, D. L., Carnahan, B., Douty, R., Hauser, N., and demonstration.Sometimes, valuable insight McMahon E., Zimmerman, J., and Seider, W., Computers in Engineering Design Education, Volume 1,University of can be gained by simulating a system which could Michigan, 1966. also be treated analytically.This insight may lead (2). Bibliography on Simulation. I.B.M. Publication No. to a better analytic formulation.Like a good doe- 320-0924-0.

55 OPTIMIZATIONMETHODS A REVIEW AND SOME EXAMPIE APPLICATIONS BRICE CARNAHAN University of Michigan Ann Arbor, Mich.

INTRODUCTION those interested in details shouldrefer to the ap- During the 9-week 1965summer program of the propriate discipline-oriented report. Michigan Projecton Computers and Mathematical TechniquesinEngineeringDesignEducation, ENGINEERING DESIGN AND THE several days were devotedto formal presentations OPTIMIZATION PROBLEM on mathematical optimization methods. Principal Virtually every engineering designevolves in emphasis was given to thoseoptimum-seeking and an iterative or trial-and-error fashion. The design mathematical-programming methodswhich are best problem is often stated inan ambiguous and open- suited for implementationon digital computers and ended way.The designer's first task is to decide which in addition do notrequire a mathematical what the problem is, and what the background much beyond that requirements or of typical senior-level specifications for any proposed solutionare.The engineering students. designer then generatesa design concept, usually During the final four weeksof the summer pro- in the form ofa rough configuration, for the object, gram, the twenty-nine participatingengineering system or process being designed.These phases of design teachers formulated,solved on the computer the design activityare most difficult and require the (IBM 7090) and documentedone or more design greatest ingenuity and engineeringjudgement. The problems suitable for inclusionin tl eir engineerirr designer next attempts to "model"the object, sys- design courses. Most ofthem attempted touse one tem or process.This model mayassume a wide of the mathematical optimizationmethods to pro- variety of forms, but those usuallypreferred are duce a design thatwas "best" in some sense. mathematical in character.The engineer may use These problems have beenpublished in the five many weapons from his arsenal of mathematical discipline-oriented volumes ofthe final project re- and scientific tools, from themost fundamental of port' which have been distributedfor the first time nature's laws to the most empiricalof data correla- at this conference.In addition to these volumes,a tions, from the most rigorous analyticalmathemati- summary volume (Volume 1) willsoon be available; cal tools to the least formal cut-and-trymethods, it contains,among several others, a lengthy review to assist in the analysis of the proposeddesign. paper on mathematical optimization techniques. The The mathematical model normallycontains several paper includes a bibliography of 65 pertinentrefer- (say n) parameters whichwe will call the "design" ences which should prove useful to thoseinterested variables, x1,x,, ...xn.To simplify the notation, in introducing optimizationmethods into the designcourse environment. we will use the notion of a design "vector,"51= [x1, x29 Xn] We may view each design variable The present paper is,a considerably less detailed as one dimension in the "space" of the designvari- review of some of the betterknown mathematical ables and any particularset of values for the design optimization techniques and isdeliberately brief and variables (i. e., any particulardesign vector) as a incomplete. No doubt thereare exceptions to some "point" in thisspace.It may happen that certain of the generalizations whichappear.I hope that equality constraints these remarks will give the reader the flavor of the g1 (x) =0 subject and lead him to studythe source materials for details and rigorous mathematicaldevelopments. Some example applications ofthe methods will be gr (5f) =0 described briefly in the latterpart of the paper; and/or inequality constraints gr+1 (5f) 1Katz, D. L., Carnahan, B., Douty,R. T., Hauser, N., Mc- Mahon, E. L., Zimmermann, J. R.and Seider, W. D., Com- g, (Tc) puters in Engineering Design Education,Volume I- Sum- mary, Volume II-Chemical Engineering, VolumeIII-Civil are imposed on the space so that only pointsin a Engineering, Volume IV- Electrical Engineering, Volume portion of the totalspace may even be considered V - Industrial Engineering, VolumeVI - Mechanical Engin- as acceptable candidates for solution of the eering. design problem.Such a region is termed the"feasible" 56 space; points in this region are termed feasible solu- 3. Optimization methods for constrained objections or feasible design alternatives. tive functions. The equality constraints come from functional 4. Optimization methods which exploit system relationships among the n variables which must or information flow structure. be strictly satisfied; these often follow from elementary conservation principles.If there is to be a Optimization Methods for Unconstrained design problem, i. e., if there is to be a choice in Functions of One Variable the values which at least some of the design vari- Some optimization methods for unconstrained ables may assume, then r must be smaller thann. functions of one variable are listed in Table II. One The inequality constraints usually come from some criterion for locath,g a local extreme point of an un- specified design limitations (maximum permissible constrained, continuous and differential objective stress, minimum allowable temperature, etc.) There function f (x) is that its first derivative f' (x) van- is no upper limit on the value of m. ish.The traditional tool for solution of such prob- Briefly stated, the optimization problem is to lems is the differential calculus, but its applicability find that one design vector or point in the feasible in design problems is severely limited by the com- solution space which. corresponds to the "best" de- plex and nonlinear nature of most objective func- sign.In order to locate the best feasible solution tions. of the usually infinite number of feasible solutions using a mathematical optimization technique, it is TABLE II essential that a meaningful, computable, quantita- Optimization Methods for Unconstrained Functions tive, single-valued function of the design variables be One Variable f(x) formulated to serve as a measuring stick for com- I. The Calculus parison of feasible design alternatives.This func- II. Numerical Mathematics tion is usually called the "merit," "effectiveness," A. Root-Finding Procedures for f' (x) "criterion," "utility," or "objective" function. 1. Newton's Method One of the designer's principal responsibilities is 2. False Position Methods to construct a suitable objective function to be maxi- 3. Half-Interval (Bisection) Method mized or minimized. The objective function might 4. Methods for Polynomial Functions be quite simple and relatively easy to calculate (the a) Graeffe's Method total weight of a storage tank to holda specified b) Bernoulli's Method amount of some material), or quite complicated c) Iterative Factorization Methods and difficult to calculate (the rate of returnover a III. Searching Schemes 20-year period for a proposed oil refinery, taking in- A. Simultaneous Methods to account likely variations in markets, raw materi- 1. Exhaustive Search als, taxes, etc.).The objective function may be B. Sequential Methods very illusive due to the presence of conflicting or 1. Dichotomous Search dimensionally incompatible subobjectives;for ex- 2. Equal Interval Multi-Point Searches ample, one might be asked to design a plant of mini- 3. Golden Section Search mum cost which at the same time minimizes air 4. Fibonacci Search and water pollution and is aesthetically appealing. 5. Lattice Search In such cases, it usually will not be possible to find a quantitative objective function and hence to auto- Some tools from numerical mathematicsmay mate the selection of better design alternatives prove useful, particularly the numerical root-finding using one of the mathematical optimizationpro- procedures listed in Table II.Newton's method cedures.In what follows, I assume thata suitable requires the computation of the second derivative objective function has been formulated by the de- f" (x) as well as the first derivative f' (x). The signer. false postion and half-interval methods require only The mathematical optimization methodscan be that f' (x) be continuous and computable.Should classified in a variety of ways ;one such classifica- the objective function be a polynomial, thenone of tion is shown below in Table I. the listed root-finding procedures for polynomials might also prove useful. TABLE I In contrast with these numerical methods, the Classification of Optimization Methods searching schemes listed under parts IIIA and TIIB 1. Optimizationmethodsforunconstrained of Table II require only that the objective function functions of one variable. f (x) be computable; for most of these procedures, 2. Optimizationmethodsforunconstrained neither differentiability nor continuity of the objec- functions of n variables (n > 1). tive function is essential. 57 In each case, the optimization problem is to locate on the interval of interest asshown in Figure 2. the global extreme value of an objective function The function need be neither differentiable nor f (x) on some arbitrary starting interval continuous. such that final optimal x value found by the algorithm and the true optimal x value lie within f ( x) a specified "interval of uncertainty," i. e., as interval of specified length known to contain the extreme value.To simplify comparison of the methods, we assume that an arbitrary starting interval [A, B] has been mapped onto the unit interval [0, 1] by a suitable linear transformation.Hence, the initial interval of uncertainty for each method will have length 1.Other nomenclature used in common throughout this section is : a = length of finai interval of uncertainty N = total number of evaluations of f (x) required to reduce the length of the interval of un- Lip.X certainty from 1.0 to a 0 n = the number of iterations for the sequential Figure 2: A Function Unimodal on the Interval searching strategies (see below). [0,1]. The one-dimensional searching schemes may be These sequential methods all involve the computa- classified into two categories termed simultaneous tion of the function at one or more arguments and sequential.In the simultaneous schemes, all which, because of the unimodality assumption, per- points at which the function is to be evaluated are mit reduction of the length of the interval of un- selected a priori.One such approach termed "ex- certainty. A second set of functional values is then haustive search" is illustrated in Figure 1.Here established for the reduced interval and yet a smaller the function, which need not be differentiable or uncertainty interval found. The process is repeated even continuous in some cases, is evaluated at equally until the interval length is reduced to the specified spaced intervals, a /2, assumed to be small enough value a. All these schemes have the property that to catch all pertinent features of the function. The the number of function evaluations required to total number of function evaluations required is achieve a specified terminal interval of uncertainty is known a priori, although the particular arguments N 2 1 used cannot be. a Clearly, the evaluation of the function at just one Clearly, this is not a very effective approach for point on the interval yields insufficient information small a, particularly when the function is difficult to reduce the interval of uncertainty. At least two to evaluate. function evaluations at different arguments in the interval must be made. The location of these argu- f (x)j ments is arbitrary, but some choices are better than

I others. A few of these one-dimensional searching 1 schemes are illustrated in Figures 3, 4 and 5. 1 1 1 If on the first pass of the sequential interval- 1 I I I I I I slicing process we evaluate the function at two argu- I I 1 1

I I 1 1 I ments centered about the midpoint and separated by I I I I I 1 I II a small distance e, i. e., compute f (1/2-1-e/2) and

I 1 I 1 I I I I 1 1 I I I 0.6 0.7as0.9 I 0 0.1 0.2030.40.5 f (x) Figure 1. Exhaustive Search. When the sequential methods listed in Table II are used, the arguments forwhich the objective function will be evaluated cannot be known a priori; instead, the sequence of argument values depends uuon the already observedvalues of f (x) .All these methods require that f (x) exhibit unimodal character, i.e., exhibit just one local extreme point Figure 3: Dichotomous Search. 58 interval search is f (1/2 - e/2 as shown in Figure 3, then thesmaller A similar three-point equal sub-interval of length (1-e) /2 can be rejected on the shown in Figure 5 below. Inthis case, the function the is first evaluated at x=1/4, x=1/2and Y=3/4 (the basis of the functional values (the l's shown near inter- curve). In this case, the interval [ (1 e) /2,1] becomes l's near the curve) .That half of the original the new interval of uncertainty; two newfunction evaluations are made at arguments centeredabout its midpoint (the 2's shown near the curve)and the interval subsequently reduced againbased upon f (x) these new functional values.This process, which is termed the dichotomous search, reduces theinterval of uncertainty approximately byhalf for each iteration of the procedure.N and a are related to the number of iterations n as follows : X 0 1/4 1/2 3/4 1 1 1 1 ) e 5/8 7/8 a 2n `J- 211 Figure 5: Three-Point Equal IntervalSearch. N= 2n=21n =2.891n val centered about the argumentof largest function Other choices of argument locations for theevalu- value (x=3/4 in this case)is retained.In the illustrated in second pass of the sequentialscheme, the function ation of the f (x) on each iteration are In Figures 4 and 5.In both cases, the selected points is evaluated at x=5/8, x=3/4,and x=7/8. are equally spaced in theinterval.Figure 4 shows this case, and in fact for allthe odd-point equal the two-point equal interval method appliedto some interval methods, there will be onefunction value unimodal function. On the first pass, the function from the preceding iterationwhich will appear is evaluated at x=1/3 and x=2/3) the l's near the again and need not be recomputed[f (3/4) here]. curve).That two-thirds of the original interval The parameters a, nand N are related by : which must contain the extreme function valueis retained as the new interval of uncertainty for the a so(+)11 second pass ; for the function shown, this would be (a) the interval [1/3, 1]. On the next pass, thefunction N = 1 - 2.89 In is again evaluated at 1/3 and 2/3 of the span of the current interval, in this case at x = 5/9 and x =7/9 It can be shown that thethree-point equal interval search illustrated inFigure 5 is the most efficient (the 2's near the curve).Again, the 2/3 of the N interval known to contain the extreme function (i.e., requires thefewest function evaluations value is saved, in this case the interval [5/9, 1]. for a given a) of all possibleequal interval search procedures. Intuitively, one would guess thatrejection of one- f(x) half of the uncertainty interval periteration of a sequential interval-slicing procedurewould be the best one could achieve on the average,given no advance information about f(x) except for its unimodal character.And from the viewpoint of the "minmax" strategy, i. e., thestrategy which minimizes the likelihood of the worstpossible outcome, this is the case. However, toreduce the uncertainty 0 1/3 5/9 2/3 7/9 interval by half, using either thedichotomous or Figure 4: Two-Point Equal Interval Search. three-point equal interval search,requires two function evaluations. Are there otheryet more efficient The procedure is repeated until theinterval of interval-slicing methods? Yes! Two suchmethods uncertainty is reduced to the specified a .The described below are called the goldensection search parameters a, n and N are related by: and the Fibonacci search.

f 2 n The golden section search isillustrated in Figure 6.In this strategy, two selectedargUments are N = - 4.95 ln (a) positioned symmetrically about thecenter of the 59 interval of uncertainty in such a way thatv,the ratio of the lengths of successive intervals ofun- Fo = 1 certainty, F 1= 1 L. F= F "J 1-1+ F i >2.

where Lj-1 = Lj + Lj+1 The first few numbers of the Fibonacci sequence generated by this recursion formula are: remains constant and is approximately equal to Fi 0.618, the so-called "golden-section" number. Hence, 0 1 5 8 each iteration of this sequential technique reduces 1 1 6 13 the interval by the factor 0.618, which would 2 2 7 21 not appear to be aseffective as the interval 3 3 8 34 halving achieved by the methods described above. 4 5 9 55 However, ck 3er examination shows that, after the first pass, one of the two arguments chosen for the The arguments at which the objective function f (x) j'th iteration will coincide withone of the arguments is to be evaluated at each iteration of this sequential from the(j-1) 'thiteration(e.g.,x=0.618in strategy are determined by ratios of appropriate Figure 6).Thus, only one new function evaluation Fibonacci numbers.Given a starting interval of is required for all iterations after the initialone. length Fn, the Fibonacci search requires, at most, n function evaluations to reduce the interval of un- The parametersa, nand N are related by : ceriainty to length 1. a =(0.618)u N = n To simplify the explanation of the methodsome- 1 = 1- 2.08 1n (a ). what but without any real loss of generality, consider an objective function f (x) on the starting interval [0,13] shown in Figure 7.Suppose it is desired to reduce the length of the interval of un-

f (x) 2 3 4 5 6 7 8 9 10 II 12 13 Figure 7:Fibonacci Search. certainty to 1.Then one examines the Fibonacci sequence, finds the Fn which corresponds to (or is next greater than) the length of the initial interval, 13. Since F, = 13, no more than n =6 function evaluations will be required.The first two arguments are positioned by the ratios F4/F. and F5/Fe, namely, at 5/13 and 8/13 of the span of the inter-

0 0.382 0.618 1 val, or at x = 5 and x = 8 (the l's shownnear the 0.764 curve). Since f (5) > f (8), the new interval of Figure 6:Golden Section Search. uncertainty for the second pass of the iterative scheme is [0,8].The arguments in the new interval It has been shown that the very best strategy for are positioned by the ratios F3/F, and F4 /F5, i.e., optimizing a unimodal function is the Fibonacci at 3/8 and 5/8 of the span of the interval, or at x=3 search, which is based on the Fibonacci number and x = 5 (the 2's shown near the curve). In gen- sequence F1 where eral, the ratios involved on the j'th iteration are 60 given by Fn_i_i/F._i+i andFii_j/Fn_j+1. One of the GEOMETRY OF MULTIDIMENSIONAL arguments for the j'th pass will always coincide with SURFACES one of the arguments for the (j-1) 'th pass. Hence, Some of the geometrical properties of a two- just one new function evaluation need be made for dimensional function f (x1,x2) =. f ('1) are illustrated each interation after the Vit. On the last (n=5) 'th in Figures 8, 9 and 10. A pointin the two-dimen- pass, one of the arguments is located within a small sional space, i.e., in the x1-x2 plane of Figure 8, distanceEof the other to halve the interval of un- yields an observed value for the objective function certainty after the (n-2=4) 'th pass. f (54.--)which is displayed in the third dimension,

If the problem is reformulated in terms of the RESPONSE SURFACE, starting interval of length 1 used to illustrate the OBJECTIVE FUNCTION other searching strategies, the parametersa , n and N are related by: a = 1 /Fn N = n. Fori > 5,the ratio F,_,/F, = 0.618; thus, the method is closely related to the golden section search, and usually no more than one additional function z2 evaluation can be saved by using the Fibonnaci search rather than the golden section search. A closely related search procedure called the "lattice search" can be used to find the maximum of a discrete valued function which is unimodal.

A comparison of the number of function evalua- (a) (b) tions required to reduce an initial unit interval to Figure 8:Response Surface and Level Contours a specified value of a is shown in Table III.The for a Function of Two Variables. sequential strategiesare clearly superior to the exhaustive search. perpendicular to the space of the variables.The TABLE III locus of objective function values in the third dimen- Comparison ofNumber of Function Evaluations sion is sometimes cniled the "response surface." The intersections of planes parallel to the x1 -x2 plane Exhaus- 3-Point Dicho- Golden Fibo- a tive Even Int. tomous Section Theca (planes of constant function value) and the response 0.1 19 8 6 6 surface, when projected onto the problem space of 0.01 199 15 14 11 11 the x1 plane, produce curves of constant function 0.001 1999 21 20 16 16 value called "level contours."Assuming that f (x) 0.0001 19999 28 27 21 20 is everywhere continuous, the level contours will

*Depends onE. form connected curves ; should the function be everywhere differentiable as well, then the level contours One might reasonably ask why so much emphasis will be smooth curves (see Figure 8b). has been placed on the one-dimensional uncon- If f(Ye)is continuous and differentiable, it can strained strategies, when most engineering design be expanded about some point, say N, (see Figure problems will have many design variables and prob- 9a) in a Taylor's series.If only the linear terms ably constraints on the design variable values as TANGENT PLANE well. The principal reason, which will become AT 'I clearer later, is that many of the multidimensional TANGENT LINE IN THE x2.62 PLANE strategiesinvolvesequencesofuni-directional TANGENT LINE SLOPE 8f1, IN THE1,a, searches, and these one-dimensional methods can PLANE. be adapted very easily to handle the suboptimiza- SLOPE3lita (a) tion problem.In addition, although these one- dimensional searching strategies have been developed without reference to constraints on the space of the variable, the selection of a starting interval places an arbitrary constraint on both the largest X2 and smallest argument for which the objective function will be computed.Hence, the methods are directly applicable to constrained one-dimen- ) fit) sional problems if the starting interval is chozen Figure 9:Tangent Plane, Contour Tangent and properly. Gradient for a Function of Two Variables. 61 in the seriesare retained, then the equation of the T tangent plane of theresponse surface at N. will be 01 -10 A (SE generated as N), is definite.Here A is the matrix of second partial y(7) f(i) + (x1- al)fxi(i) + (x2- a2)fx2ral derivatives If the intersection of the contourplane f (N) and y(x), the tangent plane ata, is projected onto the fx x X1 X2 XX xi-x2 plane, a line tangent to the level contourat N 1 2 1n called the "contour tangent"is generated.Its equation is f A== x x1 X2X2 (x1- a, + (x2 - ao)f,(i)= 0. "" '41 "2 The projection onto thex, -x2 plane of that line in XX1 XnXn the tangent plane along whichthe function y(R) n increases m( st rapidlyper unit distance, is called the "direction of steepestascent" or simply the "gradient."The gradient at N isorthogonal to If A is positive definite, the extreme pointis a the contour tangent atN and its direction cosines minimum point; if negative definite, the extreme di and t:2 are given by: point is a maximum point. d 1= fxl(a) /s d2= fx (-a) /S The concept of unimodality carriesover from one dimension into n-dimensionalspace in a natural way; unimodality implies that the response surface has just one extreme point in the region undercon- where 2 2 s 1(fx(a)) + Cal) sideration.Three kinds of unimodalityare illus- 1 2 trated in Figure 10. A function is said to be uni- The n-dimensional generalizations modal if, given an extreme point "N, there issome of the tangent path from every other point 5? in thespace along plane and contourtangent are "hyperplanes,"the n-dimensional analogue of which the objective function values strictly rise a plane in three space. or fall (see Figure 10a). While impossible to picture Figure 10b shows the level graphically, their mathe- contours for some two-dimensional function. Given matical formulations followin a simpleway as : an extreme point a, it is not possible to find a strictly rising or falling path from point E to pointN; y(x)= f(a) + E (x- ai)fx (a) therefore the function is not unimodal. A somewhat 1=1 more stringent form of unimodality called "strong unimodality" requires that the straight line path and from every point x to the extreme point 11 be strictly rising or falling.This is illustrated in Figure 10c. 2 (x4-ai)fx 0 i =1 A yet more restrictive form of unimodality, "linear unimodality," is illustrated in Figure 10d.In this The direction cosinesof the gradient, which remains case, all straight line paths between any two points a line even in n-dimensional space, are given by in the space (e. g., 13, eq) must yieldobjective di =f(a)/s function values whichare unimodal. While a strictly unimodal function would where have just one "local" extreme point equivalent tothe one and only "global" extreme point, it is often possible to confine observation to a small region of interest 12.Cf.), g»2 = ,n in which the only concern is that the localbehavior i=1 i of the function be unimodal. Most of the optimization procedures to be described in the next section A point N isa "stationary point" when all the require that the function exhibit unimodal character, first partial derivativesvanish, i.e,when fx, = 0, i = 1,2, at least locally; some require more restrictivegeo- .. . , n. An "extreme point" is a station- metric properties.Virtually all are more efficient ary point at which the quadratic form when the function is strongly or linearly unimodal. 62 x2 x2 B. Sequential Methods A 1. Methods Requiring Evaluation of f (5E) only a. Imbedded Random Search b. Lattice Search c. Univariate Search d. Rotating Coordinate Method X I p. e. Direct Search Unimodality (b) Not Unimodal (a) 2. Methods Requiring Evaluation of x2 x2 fxi(R) a. Contour Tangent Method b. Steepest Descent Methods c. Optimal Steepest DescentMethods d. Partan Methods Since one of the criteria for finding a stationary point A' of a continuous and differential function is that all the first partial derivatives vanish there, (c) Strong Unimodality (d) Linear Unimodoiity it may be possible to use one of the numerical root-finding procedures to solve the generated sys- Figure 10:Unimodality for a Function of Two = 0,1 = 1,2, .. ,n Variables. tem of nonlinear equations fxi for the unknown ai,a2, ,an.The Newton-Raphaon procedure is a common iterative solution method for OPTIMIZATION METHODS FOR UNCON- systems of nonlinear simultaneous equations; it has STRAINED FUNCTIONS OF n VARIABLES excellent convergence properties(whenitcon- (n > 1) verges) and is probably the best method when applicable. Unfortunately, the method requires that Several optimization methods suitable for finding all second partial derivatives fxixj (Tf) be computa- extreme values for objective functions of more ble, preferably from analytical expressions. Numeri- than one variable are shown in Table IV. cal approximations of the required derivatives may The calculusis the classical tool for solving be adequate in some cases. extreme problems in n dimensions.Unfortunately, Fortunately, all of the simultaneous searching its usefulness in the engineering design area is methods and some of the sequential searching meth- rather limited because of the difficulty of evaluating ods listed in Table IV only require evaluation of high order derivatives and subsequently of solving f (R).Hence, these methods are particularly useful systems of usually quite nonlinear simultaneous where derivatives are difficult to evaluate, the usual equations. case in engineering design work.The multidimensional methods will be discussed in the order shown TABLE IV in Table IV. Optimization Methods for Unconstrained Functions The n-dimensional analogue of the one-dimension- of n Variables (n>1)f(R) al exhaustive search already discussed requires the evaluation of the objective function at all nodes of I. The Calculus a grid structure overlaid in the region of interest. For an n-dimensional hypercube (the n-dimensional A. Root-Finding Procedures for fxi(5t) equivalent of a rectangular area in two space or of 1. Newton-Raphson Method a cube in three space) wherea Iis the desired inter- II. Numerical Mathematics val of uncertainty for variable xi and ai-4x1--4-bi, the total number of function evaluations required is III. Searching Schemes A. Simultaneous Methods n r2(biai) 1. Exhaustive Search N -1 2. Random Search 08 Note that even for a three-dimensional problem Some values of p for selected values of a and sare with a rather modest starting hypercube(say shown in Table V. An illustration of this "random bial =1) and rather large interval of uncertainty search" approach for a two-dimensional problem (say a = 0.001), the total number of function is illustrated in Figure 11a. evaluations requiredis about 8 X 100; from a practical standpoint this approach is virtuallyuse- less. TABLE V For the general n-dimensional problem,assume p (No. of Random Points) that the feasible region is an n-dimensional hypercube with sides of length d, in variablexi, and that a s= 073i=0.9 s=0.95 s =0.99 the final desired interval of uncertainty in variable 0.1 1 22 29 44 xi iset:.Then the ratio of the volume of the final 0.05 32 45 59 90 "hypercube of uncertainty" to the starting hyper- 0.025 64 91 119 182 cube !Is given by 0,010 161 230 299 459 0.005 322 460 598 919 I ai 1=1 1 As described above, the number of function eval- If we now choose a random point X, i.e., a point in uations required for the random search increases the space for which the n coordinates have been exponentially with the number of dimensions, hence selected randomly, then a is also the probability that it is PO real improvementover the exhaustive strat- X is the hypercube of uncertainty containing the egy described earlier. extreme value of the function. The probability that a random X is not in this hypercube is 1- a; and the A great reduction in the number of random points probability that not one of p such randomlygener- required with only a :marginal decrease in theover- ated points will fall into the optimal hypercube is all probability of success can be effected by imbed- (1- a) P,Then the probability s that at least one ding the random search to give it the sequential of the p random X values will be in the optimal character illustrated in Figure 11b. Here the original hypercube is rectangle of uncertainty with sides of length 1 s = 1 - (1 - a)P. and area 1 may be reduced toa smaller rectangle

112 of uncertainty with sides of length 0.316 andarea 0.1, using just 44 function evaluations witha probability of success 0.99(see Table V).If this subregion centered about the best point (basedon da the function value there) in the firstpass is subsequently isolated and an additional 44 random points are generated in it, then a second rectangle of un- .xlX certainty with sides of length 0.1 andarea 0.01 can (0) be generated with a probability ofsuccess of 0.99. The overall probability for success will be degraded somewhat to (0.99)'0.98, but the total number AREA 0.1 0.316 AREA 0.01 of function evaluations required is only 88.To 0.316 achieve a comparable likelihood ofsuccess using just one iteration of the procedure would require 392 function evaluations. The imbedding process can be carried out indefi- nitely with some decrease in the overalls at each PASS 1 PASS 2 iteration. There are no restrictionson the character (b) of the objective function, i.e., differentiability, con- Figure11: (a) Random tinuity and unimodality are not required. However, Search. (b) Imbedded since the only criterion for choosinga given random Random Search. point as the center of a new block for further im- Assuming that the a are much smaller than the bedding is the value of the objective functionthere, corresponding the probability holds strictly only whenthe function value at every point in the hypercube of uncertainty 8), pss about the extreme point is larger than atevery a point outside this hypercube. 64 SEQUENTIAL METHODS REQUIRING EVALUATION OF f(x) ONLY :1° In the following section, the lattice, univariate, Po " )P1 PTV4 rotating coordinate and direct search methods, OI IO 1 which do not require computation of derivatives of the objective function, will be described briefly. such that RI., is the point of minimum functional value along the one in direction ill passing through Lattice Method: In this method a grid of selected point Rt.Thus coarseness is overlaid in the region of interest(see Figure 12). A node of the grid is selected as a alt + a1 31 I 0 1 ,n-1 starting point.The function is evaluated there and at the 311-1 "adjacent" node points where n is where I a IIis the distance between XI and XI. the number of dimensions.The point with the Although one need not compute the gradient greatest (or least) functional value is then selected inr(z141) = iri.i procedure re- as the next starting point and the to find atp a.tis determined by the relationship peated, except where the functional values already 15,Tfrif,=0 computed need not be recomputed. i.e., 1-5, is tangent to the contour at XI t,. After com- The process is continued until the greatest (or pleting one round, a second round is completed, etc., least) functional value occurs at the central point until allI a d are smaller than some corresponding (starting point). In this case, the grid size is halved tolerance values ri. in each dimension and the search begun with the The method can be defeated if the variables reduced grid mesh.The grid size reduction is re- x ,x interact strongly (Figure 13,) such aswith peated, when necessary, until the inter-point dis- a ridge-like structure for whichthe normalized gra- tance in each dimension is smaller than some speci- dient vector on the ridge has all direction cosines fied tolerance value. approximately equal in magnitude.If there is no interaction at all, i. e., if the principal axes of the surface are in the coordinate directions, then the MINIM2.83 method will find the extreme value in the least num- A B3 ber of rounds.

111131:1 8$ P'"'" AilW1111111111,. 4/101LOP.445.111M1 UI '41111MI WNW= 011MIIP START 1111111111111.1S'MILW11111Imsziantirsimm 116 111113110:211111411111111 011.4101111111MIIIMMII (a) Little interaction. (b) Ridge with strong Ell411AllEVANII interaction. 16. bron....111/11 Figure 13: Univariate Search. R.".. Figure 12: Lattice Search. Rotating coordinate method: In an attempt to overcome the ridge orientationproblem of the univariate search procedure, Rosenbrock 2 developed Univariate Search: In this search only one vari- a variant of the univariatemethod which involves able is changed at a time.Typically the variables rotation of the axes of the search. are changed in order, i.e., xx2, ... ,x", sothat the function is minimized (or maximized) in the coordinate directions.Starting with some arbitrary point X., successive intermediate points Tei,YE-2, 2 H.H.Rosenbrock, "An Automatic Method for Finding X. are found by carrying out linear minimizations the Least Value of a Function," The Computer Journal, in the directions Vol. 3, p. 176 (1960) . 65 Here the procedure begins with a standard one- The coordinate rotationtechnique used here round univariate search starting at point Yo (arbi- causes the vectors fink to align themselves along the trary), i.e., n successive linear minimizations are directionof asharp ridge, and hence the method carried out in the coordinate directions tends to be quite good for tracking along a ridge, particularly along a relatively straight one.The method can be defeated however, given a bad start- 3ci+1 ai 0, 1,..., ing point Z. suchthat

1)1g1+1=0

where 0 0

P1 1 + i'th position

Next, a new set ofsearching directions is generated, Figure 14: Rotating Coordinate Method. such that Direct search:The direct searchofHooke and Jeeves 3 is a "trial-and-error" technique which has "nv Xo pn been usedsuccessfullyby many investigators. There 11;171611 appears to be no one setofrules which constitute "direct search."All the approaches consistofa

and Fn+1 9 P2n-1 local "pattern search" followed bya global move called a "pattern move." satisfy the orthogonality conditions The exploratory search consistsofa restricted 0, otoolo pTjP- i,j n, n+1, 2n-1 univariate search, such that at mostone step (usually small)oflength 81 is taken in the j'th direction. and 1. Start with an arbitrary base point b1. -T - =0. Pi gi+1 2. Add (or subtract) di to xi at 131 and As before, evaluate the function there. Let

OW = rci+ai Pi 61 0 In general, each round of the method consists of 0 n one-dimensional searches in n orthogonal directions.In the first round the n directions used are the coordinate directions.In the second round the 0 first search is conducted in the direction given by a line passing through points M. andf.; the following and evaluate the function at bl +61 if n-1 searches are in directions mutually orthogonal and orthogonal to the first one as well. In succes- necessary at bl - 61 and at bl keeping sive rounds new sets of orthogonal vectors 15, , the "best" point. Let the new "tempo- are generated so that for round k 1 rary" point be called tn., i.e., -p-nk xnk Tell(k-1) k 1, 2, IXIlk XI1(k-1)II 9 3R. Hooke and T. A. Jeeves, "Direct Search Solution of The additional n-1 search direction vectorsare then Numerical and Statistical Problems," J. A. C. M., 8, 2, 212- generated using the Gram Schmitt method. 229 (1961). 66 { bi +d1 if f (b1+71) < f (b1)

{ tll { - if f a'1-71) < f(71)

{ if { 171 071) < min{071+61), 031-71)1

3. Next add (or subtract) 62 to x2 at 111 where '0- 62

62 0 0

MIL

retaining the best point. Call it t12, i.e.

62) T114-72 if f(111 + < 1 f(T11)

712=1711 751 if "111 E2)

each time finding a new temporary point tii where

{t1,3*-1+6'3if f(t1,j-146j) < f(t1,j-1)

71j={t1,3 *-1-"37* if f (t1 ) < f (T1 ,J-1)

{711 -1 if f < min{ (T1, j-1+75j )f (11,j-1+ 6j)/

0 0 j'th element 0 0

Let the second base point b2 be equal to ti,n. Note that the first

exploratory search around the starting point b1 requires at least

n + 1 but no more than 2n+1 function evaluations.

5. Next, make the first pattern move by moving in the direction 172-171

along the line from b1 to b2 a distance equal to twice the distance

from b1 to b2. Call this point temporary point t20, i.e.

67 -..1111.4MOr

1-20 = 2(62-'51)- 252

6. Now conduct an exploratorypattern search about the pointt20 as was

done around the original basepoint El (see 2, 3, and 4 above..) Call

MM. the terminal point of the searcht2n and let

713=72n

7. Make the second pattern move in the direction b3-72to locate a third

.11110.111 temporary point t30

T30=7;2 + 2(73,E.2)=

8. This process of conducting exploratory searches aroundthe points

tko and determining the k'thbase point as Ek.fr Tkn followed by

a pattern move to the point 1k+1,0according toT164,0= ii3k+1 bk continues so long as the function value at tk41 issmaller than at

7k+1

9. Should f(tk +l) ' f(7k+1),then the base point Elis dropped from

consideration, anda new search is started by designating

bl =;(41

and returning to step 2.

10. Should a local exploratorysearchnear the newly designated El

=I= fail to finda temporary point till--* different frm bt , i.e., should

=Et then the sizes of the searchincrements Sj must be reduced

and a new exploratory searchmade by returning tostep 2.

11. When the Si have been reduced sufficientlyto be "small enough,"

i.e., when

where e3 is some small number,then the search is completedand the

last designated bl becomes thedirect search approximationto the extreme point.

68 then the contour tangent line in twospace (a plane or hyperplane in higher dimensional space) sep- arates the space into two half-spaces (Figure 16). The functional extreme point will lie inone of these two half-spaces.One "blocking" approach, called the contour tangent method, begins with the arbitrary selection of a starting point Ro, computation of the first partial derivatives there, and subsequent determination of the contour tangent. Using the direction of the gradient as the criterion,one half of the total space, the halfon the "wrong" side of the contour tangent, can be rejected as not possiblycon- taining the extreme point. Anew point jc-i located in the "center" of the saved half-space is selected Figure 15:Direct Search Method. next and the process repeated.With each pass of The method appears to be very efficient formany the method part of the space is rejected. When the minimization problems, particularly those where bounded or "blocked" space known to contain the ridges aril encountered. The direction of successive extreme point is small enough, the search iscom- pattern moves tends to become aligned with the plete.At each point selected, R it isnecessary to ridge.Experimentally in such problems, the num- compute the equation of the contour tangent: ber of function evaluations required tends to increase directly with n (the number of dimensions) T_ (x-R. )g(Ri) rather than with some higher power of nas would = f(xi) be expected. A variant of the procedure described abovecom- There are two major difficulties encountered in applying the method, selecting the locations of the putes the temporary points as follows: R and maintaining a running record of the extent of the space known to contain the extreme point. tk+1, 0 = 2.17k+1 61 Wildesuggests four different points, the "mid- Thus all pattern moves are along lines with direc- point," "minimax point," "median," and "centroid" tion:, 151{,-E, which pass through .13,.The step sizes for From a practical standpoint only the mid- in the pattern moves are thus amplified more rapidly point can be calculated easily. Evenso, the number than in the standard method. of constraining hyperplanes may becomevery large, It should be noted that since the criterion for leading to an extremely complicated algorithmeven termination is failure of a univariate pattern search for the two-dimensional case. using sufficiently small 8,,the method can fail in those cases where a univariate search would fail, x2 i. e., for surfaces with sharp ridges having certain orientations with respect to the coordinate axes. SEQUENTIAL METHODS REQUIRING THE EVALUATION OF FIRST PARTIAL DERIVATIVES In the following section, methods requiring the computation of first partial derivatives or the gradient of the objective function f (x) will be discussed. Where possible, the derivatives should be computed from analytical expressions.Since this is usually riot feasible when dealing with complicated objective functions (e. g., the cost of a plant), onemust usually compute the derivatives numerically by perturbing each variable in turn and computing the value of the objective function there.To compute all n first x I partial derivatives numerically (n being the dimen- Figure 16: Method of Contour Tangents. sion of the space) will usually require n 1 function evaluations. Method of Contour Tangents: If the function to be 4D. J. Wilde, "Optimization by the Method of ContourTan- minimized (or maximized)is strongly unimodal, gents," AICHE Journal 9, 2. p. 186 (1963). 69 Method of Steepest Descent : The successive points The scaling problem can be illustrated by the fol- of the search RI}, are in the direction of the negative lowing simple function gradient at f (x,y) = x2 + 25y2 xi,-gi =-g(xi) which, with the transformation u = 5y, becomes

=MIND =MM. = u2. i.e., xi+1 = xi + ai pi, 1=0, 1,... f (x,u) x2+ In the x,y space the gradient direction is not nearly where pi = and ai is arbitrary the optimal direction(the direction toward the 11E111 minimum) while in the x,u space it is the optimal direction.It is wise to scale the variables to make but such that£(7ci+i)< f(xi) the contours as nearly circular, spherical, or hyper- spherical as possible.Since one can't very well (see Figure 17).Justification for moving in the know what the shape of the contours is ahead of negative gradient direction is that the greatest im- time, such a scaling is generally not possible.One provement in f (1) per unit length of travel, at least alternative is to choose a starting hypercube of for infinitesimal a is in the negative gradient direc- uncertainty and then use the length of the i'th edge tion. as a normalizing factor for the i'th variable. Scaling X2 difficulties are almost always encountered when the variables are dimensionally incompatible (pressure and temperature, for example).All the gradient dependent methods which follow also have scaling problems. Optimal Steepest Descent Method: This perhaps misnamed method is similar to the method of steepest decent, except that the ccI are determined automatically :

xi+1 =.111 xi + ai .111pi gi where pi =- [ii

XI - T - Figure 17: Method of Steepest Descent. pigi+1 = There are three principal difficulties with the Thus the gradient direction at Yet is followed until method : the gradient of f (R) is orthogonal to it (see Fig- 1. Criteria for choos-ing the cc 1 are not clear. ure 18). 2. Scaling can cause problems. 3. It may take a very large number of steps to X2 find the optimum value. A One suggestion made by Marquardt 5 which allows a more-or-less automatic step-size adjustment is to use the formula

aias a.1_1(ai +a2 cos40) where 0 is the angle between p, and 15,, and a, and a2 are constants (suggested values are a, = 0.5, a2 = 1.5). Note that this causes ai>a1_1 when the gradient direction is not changing much (when RI and RI-1 are far from the optimum).

xi 6Marquardt, Chem. Eng. Progress, Vol. 55, No. 6 (June, 1959). Figure 18: Optimal Steepest Descent Method. 70 A linear minimization in the direction Di must It can be shown relatively easily that if f (30 is be carried out starting at point Mi. Any of the one- quadratic, then the contour tangents at Yei and 5c-2 dimensional searching strategies may be employed are parallel.Conversely, it can be shown that, to find the point of minimum functional value if one has two parallel lines CTi and CT2 and in the negative gradient direction.The method locates the point of minimum function value on usually requires fewer functional evaluations than each (say it, and TO, then the functional minimum the method of steepest descent. This is particularly is on the line passing through RI and 5E2 (see Figure true for those problems which require that the 19). partial derivatives be computed numerically. In two dimensions the PARTAN method usually PARTAN Methods:Because the steepest des- takes on one of the following forms : cent methods often exhibit slow convergence to the extreme point, oscillating severely from one step to Method (a) (see Figure 19) : the next, particularly when the objective function 1. Arbitrarily choose a line CTi in the space of surface possesses deep, narrow valleys, so-called the independent variables. acceleration methods have been developed which usually speed up convergence greatly. The methods 2. Choose any other line CT2 parallel to CTi are motivated primarily by some geometric proper- 3. Find the "best" point on CTi using one of the ties of quadratic functions.Since most continuous one-dimensional search techniques. The best and differentiable functions assume a near-quadratic point (call it 5-ci)is the point for which the character in the immediate vicinity of an extreme line CTi is tangent to a contour of the objec- point, the methods can often be used for non-quad- tive function. ratic functions as well. 4. Repeat step 3 for line CT2 and call the best Consider the simple two-dimensional quadratic point x2. function : 5. Search in the direction of the line passing through ki and 5E2 using a one-dimensional f(x1, 2 2 search strategy.If the function is quadratic, x2) a11"1 a12 X1 X2+ a22X2 the best point on this line will be the functional where Xi = x1 xi ,X2 = x2 4 and extreme point. x* Method (b) (see Figure 20) : is the center (extreme) x2 1. This method begins with two passes of the optimal gradient method as shown in the fig.- point of f(x1,x2) X2 x2 A

4111111111. xl Figure 19: PARTAN Method.Method (a) Figure 20: PARTAN Method. Method (b). 71 ure. Let Fcc, be the initial point and 5f2 and The methods available to the engineering designer x3 be the terminal points for the first two for dealing with such problems vary in complexity optimal gradient searches. and power, principally with the nature of the objective function and constraints.Table VI lists a few of the methods available. x2 = xo + aoPo TABLE VI Optimization Methods for Constrained Objective go Functions 170 I. Linear Programming I t II. Nonlinear Programming A. Lagrange Multipliers 170T g2 =0 B. Iterative Linear Programming (Simpli- cial Methods x3=-2x2 a2 rn 2 1. Cutting Plane Method C. Adaptations of Methods for Unconstrained Functions P2 1. Gradient Projection Method II-1211 2. Multiple Gradient Summation Method T Linear Programming P2g3= A linear programming problem is a constrained optimization problem for which the objective function and all constraints are linear in the problem 2. A final linear minimization isnow performed variables.Very powerful algorithms have been in the direction reo3t3 along the line passing developed for solution of such problems.Linear through 34 and Ye, The minimum point re will programming has in the recent past been the most be the extreme point if f (R)is a quadratic used of all optimization methods, particularly in function.This follows because 132, the direc- such business-oriented applications as resource al- tion of the contour tangent at 573i is orthogonal location. to Iso and hence parallel to the contour tangent Manyindustriallyimportantproblemsare at Ro. amenable to solution by linear programming methods.Three of the more important typesare known OPTIMIZATION METHODS FOR CONSTRAINED by the labels "The Transportation Problem," "The OBJECTIVE FUNCTIONS Activities Analysis Problem," and "The Diet Prob- Figure 21 illustrates a typical optimization problem."A brief description of these three typical lem in two dimensions with one nonlinearcon- linear programming problems is given below. straint.Note that the unconstrained functional ex- a. The . Transportation .Problem:A company treme point lies outside the region of feasible solu- wishes to ship a number of units of some item from tions. its warehouses to its customers (for example,a rubber company shipping tires to its wholesale dis- tributors).Each customer has ordered a number of units of the item; each of the company'sware- houses can supply up to a certain amount of its current inventory of that item.What shipping schedule should the company use to minimize the total transportation cost, assuming that the total inventory available from the warehouses is equal to the total of the orders of all customers? b. The Activities Analysis Problem: Amanu- facturer has fixed amounts of severalresources available for his use (these might beraw materials, labor, equipment, money, etc.).The resources can be combined in several differentways to make one or more of several different possible products.It takes a certain number of units ofa particular re- ppo. X source to produce one unit of a particular product. Figure21: A Constrained OptimizationPrcblem. He knows how much product hecan make for every 72 unit of any particular product he manufactures. X2 How should he allocate available resources to maxi- 8 mize the total profit ? c. The Diet Problem :Given the nutrient con- 3x 1+4x 2=24 tents of several foods and a minimum daily dietary 6 requirement for each nutrient, minimize the total 5 food cost while maintaining at least a minimum 4 nutrition level. 17'7)N Each of these problems can be written in the form of a standard linear-programming model of the form 2 3 9 10 II xI 2x1 +5x2=10 n 4x1 t3x2 =12 +4x 2=11 maximize Z C i=1 Figure 22:Geometric Interpretation of a Simple subject to the constraints Linear Programming Problem.

x > 0 J= 1,2,In that the maximum feasible value of the objective ' function occurs at a vertex of the polygon (15/7,- 8/7). In fact, it will always happen (even in higher dimensional space) that the objective function for { a linear programming problem will assume its ex- a x. = }b., b. a 0, i= 1,2,..m treme value at a vertex of the space of feasible solu- jal J > 1 1 tions.All linear programming algorithms employ the strategy of examining the objective function only The objective is to maximize the linear function of at vertices of the region of feasible solutions. The the n variables xi, subject to m constraining rela- algorithm moves from vertex to vertex in such a tionships also linear in the xi ;the constraints can way that the objective function value at the suc- assume the form of equalities or inequalities cessor vertex is at least as great as at the predeces- In addition, each xi in the feasible solution must be sor vertex.In a finite number of steps the optimum non-negative. can be found. In two dimensions the linear programming prob- The algorithms required to solve a linear pro- lem has a very simple geometric interpretation. gramming problem are too complicated to describe Suppose the problem is to maximize the function briefly.Detailed descriptions can be found in many f (xx2) = 3x1 4x2 subjectto the linear con- texts, among the more readable being Gass,6 Had- straints ley,7 and Dantzig.8 2x1 5x2 G 10 4x1 3x2 -4- 12 Nonlinear Programming x1 0 In the nonlinear programming problem, the ob- x2 0 jective function or one or more of the constraints This is illustrated in Figure 22. assumes a nonlinear form; the constraint functions Notice that the inequalities require that the solu- are normally defined over continuous ranges of the tion be in or on the boundaries of the cross-hatched independent variables. When the objective function region.This region is termed the "feasible" region to be maximized is subject to equality constraints and any (x x2) in the region is said to be a feasible and both the objective function and the constraint solution.The linear inequalities enclose a feasible function can be differentiated, it may be possible region which has the form of a convex polygon where to use one of the techniques of variational mathe- "convex" implies that all points on the straight line matics such Lagrange multipliers.Unfortunately, joining any two points in the set of feasible points in most engineering design problems, constraints are are also in the set.This simply means that a linear of the inequality form and the objective function combination of two feasible solutions is also a feasible solution, provided the combination lies between 6Saul I. Gass, Linear Programming Methods and Applica- the two solution points chosen. tions, 2nd ed., McGraw-Hill Book Company, 1964. 7G. Hadley, Linear Programming, Addison-Wesley, Read- For the given problem, the objective function is ing, Mass. 1962. the family of straight lines with slope -3/4 ; its value 8George B. Dantzig, Linear Programming and Extensions, decreases as the lines approach the origin.Note Princeton University Press, Princeton, New Jersey, 1963. 73 is seldom easily differentiable. Consequently, La- problem becomesone of searching for the optimum grange multipliers have only limited utilityfor the in n-dimensional space.It frequently happens that more difficult engineering design problems. the structure of the systemcan be exploited to allow So-called simplicial methodshave met with a cer- the optimization problem to be tain success. reformulated as a These all transform the nonlinear stage-wise decision process, i.e., formulated in such programming problem into an iterative sequence a way that computation of the optimumsolution of linear programmingproblems. One of theseap- takes on a serial character involving proaches, known at each stage as the cutting plane method, is or step in the process the variation ofno more than described briefly below.Still other approaches to a veryfew ofthe n design variables at the nonlinear programming a time. One problem involve modi- powerful solution approach to suchproblems which fications of methods forunconstrained functions of may involve a very large number of design variables many variables, such as those alreadydiscussed. is called dynamic programming. Two which appear to have achieved a substantial The stages are usually numbered acceptance are the gradientprojection method,a in reverse order modification of the method of to the direction of informationor material flow (see steepest ascent, and Figure 23). the multiple gradientsummation method which isa modification of the direct searchof Hooke and Jeeves. PROFIT Cutting Plane Method:In this method 9 thegen- eral nonlinear programmingproblem maximize f FEED subject to xn+1 gi (3E) -40, i = 1,... ,m DECISION = [x1,x2, PCn] is rewrittenas a problem with a linear objective, Figure 23:Dynamic Programming. function and nonlinearconstraint by addingone new variable and one new constraint, gni+, = The input or state variableto the i'th stage is the Xn+i f (51) Then the problem is: maximize xn+1 output state variable for the (i+1)'th stage x1+1. subject to (xi+, might be a vector of state variables gi (R) ,i = 1, xi .) A ,m decision or design variable di (orvector of decisions gm+i-40 In the cutting plane di) is specified for the i'thstage.For each stage approach, each constraint there are two functional relationships. function, gi (Y), is replacedby a linear approxima- The first, tion obtained by expanding denoted hi describes the output statevariable for it about some starting stage i, xi in terms of the input state variable point, 27°, in a Taylorseries and dropping all xi+, terms and decision made di.The second, denotedpi, of order two and above.The nonlinear problem is then solved by first solving: describes the profit (negative ifa cost) associated maximize xn+, subjectto with the operation of the stage. Theobject of the dy- gi (50) gi (V) (3Er) _co namic programming algorithm isto determii: e how = 1,2,...m-F1 to operate the system (i.e., what decisions di,d2,., The solution, Tc*, of thislinear programmingprob- do should be made) to yielda maximum overall pro- lem (which mayor may not fail to satisfy all of the fit. original constraints)can be used as the base for new Taylor series expansions of theconstraints, Operation of Stage xi = hi (xi+l di) creating a new linearprogramming problem to be solved.When the solutions rk andrk+, for two Stage Objective successive iterations differ byless thansome ac- Function (e.g. Profit) Pi = gi(xi+10i) ceptable amount and Vk+1satisfies the originalcon- Optimal Cumulative Object Function: straints, a solution to thenonlinear problem has been found. fi(xj+i)= max. (pi(x d1) +fi_1(xj)) di METHODS WHICH EXPLOIT SYSTEM OR The procedure, which isa bit too complicated INFORMATION FLOWSTRUCTURE The methods described to detail here, is basedon Bellman's principle of earlier havean essential- optimality, which states: ly sequential point-to-pointcharacter; all the design variables are selectedsimultaneously fora given "An optimal policy has the property that, point, i. e., if thereare n design variables, then the whatever the initial state and initial decisions are, the remaining decisions must constitute an J. E. Kelley, Jr., "TheCutting Plane Method for Solving optimal policy with regard to the state resulting Convex Programs," J. SIAM,8, pp. 708 -712, 1960. from the first decision. As a consequence of this principle, it is possible TABLE VII to optimize a serial system using a tail-to-head Engineering Design Problems Using Optimization strategy, first examining a one-stage process, then a Methods two-stage process, etc. (Note) :The format adopted is: Provided the program can be appropriately formu- a. Optimization method. lated, an apparently n-dimensional problem can be b. Objective function. solved as a sequence of m-dimensional problems c. Design variables. where (m « n) ;in some cases the n-dimensional Chemical Engineering problem can be reduced to a series of n one-dimen- Volume II sional problems. The number of computations re- 1.Economic Optimum Conditions for a Staged quired can be reduced by several orders of magnitude Separation Process over that required by standardoptimum seeking a. Univariate Search methods when the dynamic programming approach b. Rate of return = is applicable.Even so, for problems with a large (Income - Product Cost) - Capital investment number of state and decision variables, at each Capital Investment stage the computations problem can become quite c. XD = Productpurity, MF = Multiple of formidable. minimum reflux, FR=Fractional recovery During the past year or two, and inspired pri- 4.Design of a Mixed-Suspension, Mixed-Pro- marily by the success of the dynamic programming duct Crystallizer concept, there have been several studies involving the a. Exhaustive Search decomposition of macrosystems, i.e., large and com- Income-Total Cost plex systems containing many components or sub- b. Rate of return Capital Investment systems. The object of the decomposition schemes is c./Lay,Average crystal size to allow effective suboptimization of individual subsystems, using virtually any of the various mathema.- Voiume IIICivil Engineering ticP1 optimization techniques, both simultaneous and 6.Optimal Design of Two-Way Reinforced Con- sequential, to reduce a problem of hopelessly large crete Slabs dimensionality to one of computable proportions. A a. Exhaustive Search recent paper by Wilde" is particularly well written. b. Either weight of slab or cost of slab DESIGN PROBLEMS c, t = Slab thickness During the last four weeks of the nine-week sum- 7.Decision Making in Building Planning mer program of the Project onComputers in Engi- a. Linear Programming,Simplex Method neering Design Education, each of the participating b. Cost of apartment complex faculty members generated, solved, and documented c. Number of apartmentsbuilt for each one or more design problems whichhe felt suitable, price range and size for use in his design course. Most of these problems 8.Optimization of a Two-Span Cover-Plated are included in the disciplinary volumesof the final Steel Beam report of the project. Each problem is stated as it Cutting Plane might be presented to a student in an engineering a. Nonlinear Programming, design course and detailed descriptions of the solu- Method tion methods, computer programs and computer re- b. Weight of the beam sults are included with each problem. c. bf = Flange width, tf =Flange thick- Most of the faculty participants attempted to ness, d = Web depth, t =Web thick- use one or more of the mathematicaloptimization ness,F= Left span length,t= Right methods of the previous pages to produce a design width, that was "best" in some sense.Table VII lists the span length, by = Cover plate volume number and general field of interest for t = Cover plate thickness, etc. those problems involving some optimization strategy. Volume I Summary The number of the problem used in the reports, its Volume II Chemical Engineering title, the type of optimization method, the nature Volume III Civil Engineering of the objective function, and the design variables Volume IV Electrical Engineering involved are listed under each.Those interested Volume V Industrial Engineering in details should refer to the appropriate volume of Volume VI Mechanical Engineering Final Report, Project on Use of Computers and Mathematical the project report.11 Optimization Techniques inEngineering Design, D. L. 10D. J. Wilde, "Strategies for Optimizing Macrosystems," Katz, B. Carnahan, R. Douty, E. McMahon, N. Hauser, Chem. Eng. Progress, 61, No. 3, pp. 86-93, March 1965. J. Zimmerman; and W. Seider, March 1966, University of " Computers in Engineering Design Education. Michigan (800 pages). 75 9.Design 4f Rectangular Tied Concrete Col- Volume VIndustrial Engineering umns 22, A Dynamic Programming Problem for the a. Exhaustive Search Assignment of Tolerances b. Cost of columns 11, Dynamic Programming c. Size and number of longitudinal bars, b. Cost per combination within overall size and .spreading of lateral ties tolerance specified 10.'Optimal' Design of Structural Steel Framing c. Component (stage) variances for Tier-Type Buildings 25.Optimal Assignment of Machines toan Oper- a. Dynamic Programming ator b. Cost of a tier-type building a. Exhaustive Search c. Joist spacing, joist span, number of beam b. Cost per piece manufactured panels, number of girder panels, col- c. Machine cycle time, operator idle time, umn specifications machine idle time 11.Minimum Weight Design of a Linear Elastic 26. A Dynamic Programming Approach toan Column Inventory Problem a. Univariate Search a. Dynamic Programming b. Volume of material b. Cost of maintaining inventory c. Column radius, length, and wall thick- c. Schedule for purchasing ness Volume VI Mechanical Engineering 12.Optimal Design of a Timber Warehouse Floor a. Dynamic Programming 29.Rocket Nozzle Design b. Cost of a warehouse floor a. Dichotomous Search c. Joist spacing, floor material, joist mate- b. Weight of nozzle wall rial, beam material c. Wall thickness, materials of construction 13.Optimization in Site Selection 30.Gear-Ratio Determination by Computer-As- a. Constrained Gradient Search sisted Search b. Cost of a static missile test stand in- a. Cartesian Walk cluding utility and road construction b. Precision of gear-ratio determination c. Distance of missile stand from water c. Number of teeth sources, drainage pond, barricades, and roads 31.Optimization of a Pneumatic Linear Actuator a. Direct Search 14.Design Force for Bracing in a Deep Vertical Cut b. Operating cost (compressed air usage) a. Direct Search usage) b. Bracing design force c. 11 variables including working area, c. Horizontal position of logarithmic spiral stroke length, spring rate, working pres- cut sure, opening area, etc. 32.Ejection Actuator 15.Optimal DesignofaWide-Flange Beam Using Standard Rolled Shapes a. Monte Carlo Methods a. Exhaustive Search b. Reliability of actuator system b. Weight of the beam c. Resistance force c. Beam dimensions 33.Determination of Optimum Pipeline Route a. Dynamic Programming Volume IV Electrical Engineering b. Cost of a pipeline route 17.Design of a Ferromagnetic Circuit c. Number of sites, pump, and pipe sizes a. Dichotomous Search 34.Cam Design of Minimum Size Based on b. Mass of the circuit Bearing Stress Limitations c. B = Width of leg 3, K = Width ratio- a. Direct Search leg 2/leg 3, etc. b. Contact stress 20.Class C Amplifier Design c. Prime circle radius a. Gradient Search 35.Optimum Reheat Pressure for a Steam Power b. A trade-off function including input and Cycle output signal powers, and plate efficiency a. Fibonacci Lattice Search c. Plate supply voltage, grid supply volt- b. Heat cycle efficiency age, minimum plate voltage c. Extraction pressure 76 THE SYSTEM APPROACH TO COMPUTERAIDED DESIGN: USE OF PROBLEMORIENTED LANGUAGES R. L. MOTARD University of Houston Houston, Tex. Modern engineering design continues to expand synthesis is largely a subjective process and analy- in complexity and depth under the impact of com- sis is largely objective. By this I mean that analy- puters.The use of computers in the control of sis, following the formal logic of language and chemical plant operations places an additional bur- mathematical symbolism, is an external medium of den on the engineering analysis of complete proc- expression useful for the recording of completed ess system.Adaptation is most certainly taking conceptions and for the social exchange of scientific place in engineering curricula to meet these re- information.Synthesis, on the other hand, depends quirements. A recent report in a survey of ac- on the infralogical processes of the mind and there credited engineering programs indicates an expo- are now some philosophers of science 2,' who main- nential growth in course changes.In some respects tain that such processes are not couched in the change and adaptation are addicting and the pre- external and formal logical language but depend vailing atmosphere in most academic institutions on unknown mental symbolism more closely related now favors continued evolution in the pursuit of to unknown modelling and heuristic maneuvers and excellence. relying on intuitive notions of analogy, symmetry, No doubt modern engineering design demands limits and visualization'I would hazard a guess higher sophistication in mathematical skills, more that many mental models in design synthesis spring intimate contact with computers and some atten- from physiological and visual analogs of the sys- tion to mathematical optimization; with further tem and its elements. emphasis on the systems approach, we complete The proper academic medium for the develop- the picture. Many engineering schools are strug- ment of synthetic skill in design must involve di- gling with the expansion of these topics in the under- rect experience with total systems since every de- graduate curricula.Hopefully these developments sign must always begin at this stage. Every de- will add to the young engineer's skill in design but, signer or problem-solver hungers for a perception as we have come to appreciate, something more is of his total problem. Systems science and engineer- needed, namely, a broader experience with opening offer him some guidance in structuring and ended design problem situations than is generally classifying the concepts which have been refined in available to the undergraduate.Perhaps it has past design work with total systems. However, such already been said enough, but you will forgive me terms as control, optimization, transfer function, if I repeat, that creative design is both synthesis feedback, time delay, dynamic response and com- and analysis.' The something more that I referred munication channel are almost meaningless to the to above is the missing component that mathe- average undergraduate student. Yet it is precisely matics, computers and systems do not confer upon such concepts or their analogs which he must con- the student. We need a component that adds syn- sider in modern design situations. We have trained thetical vision, creativity, intuition and inductive him rather well in analysis and he usually enters skill to the engineer's mental equipment. a senior design project very much task-oriented. Any mental process that can be developed and He is not prepared to innovate and create systems described as a step-by-step procedure is analysis. solutions to open-ended problem statements. If we It is the stuff of scientific verification and valida- borrow some more of Bruner's notions,' our aver- tion. Analysis formalizes for the benefit of scien- 1 Motard, R. L., "SYSTEMS ENGINEERING: Engineer- tific or engineering fraternity the structure of a ing Come of Age and Its Academic Image," Journal of Engi- solution and its relation to sound principles. Syn- neering Education, vol. 56, p. 198 (1966). thesis, too, must be grounded in scientific princi- Moles, A., La Creation Scientifique, Editions Rene Bider, ples that have been thoroughly assimilated, but it Geneva, 1957. relates differently to these disciplines. We must 3 Bronowski, J., "The Logic of the Mind," American Sci- entist, 54, 1 (1966). not confuse synthesis with analysis and vice versa, Bruner, Jerome S., The Process of Education, Random since they are distinctive modes of thought. Al- House (Vintage Books), New York, 1960. though it is an oversimplification, we can say that 5 Bruner, Jerome, S., ibid. 77 age student has beenbrought upon the reward and punishmentsystem and has process and callsupon the process-unit seldom experienced subroutines to calculations complete absorption,intensive curiosityand the carry out the chemicalor physical lure of discovery. transformations.All of theseare provided for the What is proposed, student. The keyto the great simplicity therefore, isa laboratoryex- of use of perience with totalsystems in the student's the language isthe limited amountof data to be field of supplied tocarry out a simulation. A interest whichare intended to heightenhis intuitive comprehen- grasp of systems and to sive thermodynamic-datasubroutine provides stimulate hiscreativity be- the necessary ,ail fore he is askedto producea new design. Many phase equilibriumand energy data. other approaches Only hydrocarbonsystems arecurrently are possible, but thisone lends it- grammed. pro- self well, I believe,to upper-division when the student design courses The details of the has alreadymasteredmany ana- subroutines andexecutivesys- lytical proceduresand understands tem are given inthe Chemical the elements of of the Report of Engineering Section his systemsreasonably well.The systems in the Universityof Michigan Proj- case are computer this ect on the Use simulations fromwhich the stu- of Computers inEngineering Design. dent can extractinformation quite In brief, the basicinput data fora simulation are ily, yet they simply andeas- shown in Table I. are of such complexitythat hecan never be quite assuredof a total The student understanding. TABLE I should experiencethe samesense of Basic Input to the wonder, mysteryand discoverythat he would Chemical ProcessEngineering find in System exploring theoperation of theactual hardware. His experiencewith the system 1. Job descriptionfor output two phases: the would consist of identification. first or familiarizationphase where- 2. Control constants. in the optimumoperation of Number of components is sought an existing structure per stream. ; and the second phasein whicha restruc- Component identificationand names. turing of thesystem is allowed.Such a problem New jobor continuation. should then parallel real-world experiencein design Sizes of principaldata matrices. but with theadded incentiveof "instant" Process unit numbers in allowing the experience in recycle loops. computer to performmost of the Tolerance. alysis which the an- student followsvicariously. 3. Process matrix,one row vector for each unit. 4. Stream PROBLEM-ORIENTEDLANGUAGE composition matrixfor eachprocess The simulation feed stream. of complexsystems in itself isnot 5. Equipment a task for theundergraduate student. parameters matrix,one row vector lems are simply The prob- for each unit. too vast anddetailed to be doneas classroom projects. 6. Process controlandconvergence forcing in- It is morepractical to provide formation. a computer-executivesystem which theuser can have easyaccess to and structure The structure ofthe process andstream matrix to his own needs. row vectors are shown The languageof communicationwith the computer in Table II andthe flowchart is problem-oriented for the executivesystem in Figure I. in that it isespecially apt in A typical cod- describing systems ing for a unit andits extensionare shown in Figure common to a givenengineering 2. discipline.The executive Further details ofthe system operationare dis- system used in theexam- cussed in the example ple discussedbelow isa major revision of problem and results. PACER executive the system whichwas originally de- TABLE II veloped by ProfessorPaul T. Shannonof Dartmouth Structure of College and is Process and StreamMatrix Row now implemented in theMAD lan- Vectors. guage on the Universityof Michigan computer- KPM (I, 1) onerating system.The MAD languagelends itself = Unit number, ISN (3,1)= Stream very well to this type of number, 3 development becauseof its KPM (I, 2)=Unit SR name SN (3, 2) great sophisticationand flexibility,its information- KPM (I, 3)= = Stream type processing ability External nameSN (3, 3)= Enthalpy and the simplicityof its input. KPM (I, 4)=Input stream SN (3, 4)= Temperature This chemical KPM (T, 5) process engineeringlanguage is =Numbers designed to fit thestructure ofan ordinary chemi- (positive) SN (3, 5)= Pressure cal plant in whicha set of streams of solid materials gas, liquid or SN (3, 6)= Total moles are heated, separated,compressed SN (3, 7) or transformed chemicallyin an = Moles matrix of interconnecting component 1 process units. Theexecutive systemper- KPM (I, -)= Output streamSN (3, 8)= Moles forms heat andmaterial balanceson the simulated numbers component 2 etc. (negative) etc. 78 Change Deck Print Report =MINMALIIMI Reports

Data Main Read Program Economic Saved Summary Results

Data Yes Check Complete? Print Data 4 Non Loop ....Let Up Propertiei Calculations

Process Yes No Enthalpy of Unit Converged?) Input Streams Calculations Loop Calculations I Convergence .41Thermo Forcing Properties Process Contro

FIGURE 1 Flowchart of Executive System

KPM = Process matrix SN = Stream matrix ENC = Equipment matrix KPM(8, 1) = 8, $EXTRA43, $DIST23, 1, 8, 4, -2, -5 SN(1, 1) = 1., 1., 0., 650., 150., 45.4., ENC(8, 1) = 8., 10., 5., 1., 10., ENC(4, 1)=4., KPM (4, 1)=4, $UNAME7$, $E1$, 2, 4 KPM(5, 1)=5, $UNAME4$, $V1$, 6, -8, -9 KPM(6, 1)=6, $UNAME1$, $D1V$, 9, -3, -10, KPM(7, 1)=7, $UNAME7$, $E2$, 5, 12, -7, -18 KPM(9, 1)=9, $UNAME4$, $V24, 7, -4, -11, ENC(5, 1)=5., ENC(6, 1)= ENC(7, 1)=7., ENC(9, 1)=9., FIGURE 2. Use of Problem Oriented Language 79 3

(1)

cyci.c etAS

10 VI (3)

CI (I)

C430460/140 LEVIca- 4qes se TOM( FIGURE 3 NAT(JRAL. 674501-WE PLANT ABSORPTION dECT/OA/ /7r4roZsamAldre40a4 es.) tim sr on.Givilma47 ANI4044 offs;en.)Amp ZS, 7.1444 MOsOM EXAMPLE PROBLEM erties for all streams in the process,a total of 49, The process chosen for classroom experimenta- and an economic summary whichcan be analyzed tion is a natural gasoline plant.The process flow- to determine the next stage in the coolingopera- sheet is shown in Figure 3. tion.(See Figs. 4 and 5.)No guidance is given to the students regardingany formal optimization Sufficient flexibility was allowed in the structur- procedures and this must be done externally by the ing of the process to permit the manipulation of team.It is anticipated that the optimization will nine basic variables. Four of the basic variables be done heuristically by most students.No class- involve stream splits and require the specification of room experience can be reported since the students the joint stream ratios.The object of the first are only beginning to use the system. Perhaps the phase of the problem, as previously stated, is to most disturbing aspect of the whole concept is the optimize the economic return from the existing tremendous amount of preparatory work by the plant.Each team of students, fouror five men to instructor and the great demandson the use of the a team, is supplied with a coding form and a change computer. The present project represents at least deck of variable cards. Only the variable cards for four man-months of instructor time and eachrun those variables being altered need to be submitted takes a minimum of fifteen minutes of IBM 7094 for any run, along witha binary deck representing time. For the entire first phase of the project it is the computer output of previous results froma past estimated that the four teams of students will need run. Using a previous output as a starting casere- ten hours of computer time. Very few universities duces the computer time requirements. could afford to supportmany such projects at the Each team is allowed about ten "plays"or runs, present levels of computing facility.Undoubtedly and any set of the basic variablesmay be varied any extension of educational effort in the direction from run to run. of more complex computer executivesystems will require more staff support for instruction,since no DISCUSSION OF RESULTS individual faculty membercan afford to carry the Output from the program provides streamprop- entire burden. VAR.7. VAR. 7.1 VAR. 6

. 5.2

VA . 5. VAR. 4

LNC(6,3) ll lal(21.11-5)=. YAR.

ali(21.6)= VAR. C. i(et. C4e2 )= VAR. 2

NC<1.4)- 9 VAR. 1 ,BLUE OIL CO.- NATURAL GASOLINE PLANT' VAR. 1 )CHANGE DECK - PRECAMBRIAN, TEXAS. TEAM CAPTAIN. MIKEJONES. TRIAL NO. 4 :III II III II II II II II 11 I I I I I III Ill I III I esessisssessolossmeloselememodusloommosommotommossomossm Ississysoolismenounassnansmonsonsmusasassir-iseaseuseumwsmussmunmempanosennunmennxim 1111111111111111111111111111111111111111111,111111111111111111111111111111111111 21222222221222222222222222222122222212222222222222222222221222222222222222i22222 133333333133313333133133313333333333313313331333133333333333333333331333333.33331 444144414444441414444444144144441411111111114114111111111141111111111111114441441m i$5515155155555155$555IssM155511151551111515515555111111SIMISiiiiiiiiiiiii 1111111111141iiiiisissolisSIIIIIIIIIIIIIIIIIIIIIMI4111111111111111111110111011 111111111111111111711111117111111111111111111117111111111111111111111111111111t1 sIsostoommommossoloommolammosoolsommommomissolossosts '

J11$11111$111111111111111111,1111111111111101111111111111111111111111011111.1111mmweet \,...11,1411soyfewswamennwinniessasnamnansimanissimmemesoommumussuenummismusumennnumniinew

FIGURE 4 Change Deck of Variable Card.

81 a.

STREAM NUMBER 39 40 41 42 43 EQUIP. CONX ION FR 14 TO 15 FR 14 TO 16 FR 15 TO 0 FR 24 TO 0 FR 25 TO 24 VAPOR FRACTION .0000 1.0000 .0000 ENTHALPY, BTU 1.0000 1.0000 4554.6212 36992.6064 2277.3106 77095.7373 85780.6436 TEMPERATURE, R 569.2714 569.2714 569.2714 646.3154 PRESSURE, PS I A 646.3154 80.0000 80.0000 80.0000 3000.0000 1000.0000 COMPOSITION, LB-MOLES/HOUR METHANE .0057 .5133 .0028 ETHANE 15.1846 15.1846 .0324 .7322 .0162 .4345 .4345 PROPANE .0965 .7295 .0483 .2414 I- BUTANE .2414 .2410 .7855 .1205 .0222 N- .BUTANE .0222 .3162 .7721 .1581 .0029 .0029 N- PENTANE .1876 .1609 .0938 N- HEXANE .0000 .0000 .1387 .0433 .0693 .0000 .0000 N- .H EPTANE .0893 .0110 .0446 N .0000 .0000 - OCTANE .0420 .0020 .0210 ,0000 TETRADECAN- LO .0000 .0125 .0000 .0063 .0000 .0000 TOTAL 1.1619 3.7498 .5809 15.8857 15.8857 FIGURE 5 STREAM PROPERTIES OUTPUT FORMAT

UTILITIES CONSUMPTION WATER STEAM ELECTRIVITY FUEL GAS GPH LBS/HR KWH SCFH UNIT COSTS .00003 .00040 .01500 .00035 EQUIP. NO. 1 177 0 .0 42 13 2067 0 .0 0 16 -0 0 .0 0 18 0 0 .0 1048 22 0 0 4 23 0 1419 0 .0 0 24 173 0 .0 28 45 70 0 .0 25 TOTAL USE 3905 0 4 1160 12 .00 .06 .41 TOTAL COST .58 PRODUCT VALUES, STREAM NO.41

MOLES/HR UN IT VALUE VALUE

METHANE .00285 .0000 .00 ETHANE .01619 .0967 .00 PROPANE .04825 .2088 .01 I- BUTANE .12052 .3585 N .04 - BUTANE .15808 .4960 .08 N- PENTANE .09382 .8244 N .08 - HEXANE .06933 .9354 .06 N- HEPTANE .04463 1.0494 N .05 - OCTANE .02099 1.1700 .02 TETRADECANE .00626 .0000 .00 TOTAL VALUE, DOLLARS .35

FIGURE 6 Economic Summary for Gasoline PlantSimulation

82 COMPUTERS IN ENGINEERING DESIGN AT UNIVERSITIES

SECTION D

CHAIRMAN: L. A. SCHMIT, JR., Prof. of Structures, Engineering Division, Case Institute of Technology, Cleveland, Ohio PRECEDING PAGEBLANK- NOT FILMED COMPUTER-AIDED DESIGN ALLEN B. ROSENSTEIN University of California Los Angeles, Calif. Foreword: A brief summary is appended of the Design has been defined as an iterative decision- courses with a computer-designorientation given making process.The computer-aided design course or planned by the EngineeringDepartment of is built upon the assumption that once the decision UCLA (Appendix I).Required undergraduate rules and procedures for the design of a product core curriculum courses are listedalong with have been completely formulated, all futuredesigns selected graduate courses. Course outlines or any using the same rules and information sources can additional information are available upon request. be produced faster and better by machine thanby hand. We can restate this assumption by saying INTRODUCTION that if an engineer can describe to another engineer The Commission on Engineering Education is tr) in an unambiguous fashion the procedure necessary be thanked for organizing a conference at this to design a given product, then he can alsotransfer to assess the impact of computer-aided design upon the same rules to a machine. engineering education. Design problems are almost never formulated Unfortunately, with only a few notable excep- with enough equations to cover all of the param- tions, engineering educators have been very late in eters.An educated search procedure is usually integrating the computer into the design process. In required.However, once the rules and logic for this day of high-speed information processing, inso- these procedures are formalized, they can bebetter far as we inengineering education emphasize performed by machine. The course, therefore, looks manual analysis and ignore problem formulation, for the generalized computer-aided design concepts syntheses, machine information processing, optimi- which are applicable in all fields of engineering. zation, and decision-making, we are turning out men We begin by attempting to give an organized who face early technological obsolence. overall view of the present state of computer-aided The course that I wish to describe in this paper design with some prediction of its future. The de- has been offered at UCLA as a graduate seminar sign process is reviewed and an analysis is made (Appendix II).It has been proposed as part of of the comparative present and future contributions the prtparation for a doctoral field in Engineering of the engineer and the computer to each element Design. The mass of material now becoming avail- of the design process (Appendix III). able leads us to believe that the course will even- The expectations for generalized computer-aided tually expand into two semesters with supporting design procedures become apparent when we ex- offerings in optimization and simulation.Itis amine the major factors of the design anatomy and expected that, as the material becomes more forma- catalogue the available design parameters, analyti- lized, a substantial portion will find its way into the calmodels, and optimizationtechniques. The undergraduate curriculum.As this happens, the limited classes of parameters to be varied in any arguments over design as a discipline will become design are identified as geometry, materials, and academic.I would predict that computer-aided energy (these sometimes appearing aspeople and design courses have a high probability of becoming information).The analysis problem hinges upon the key part of widely required undergraduate de- the selection of either a lumped parameter, distri- sign stems. buted parameter or probabilistic model.The opti- Our computer-aided design course is based upon mization procedures available are also not extensive. several assumptions.The first assumption is that Applications to illustrate computer-aided design design taken in its broadest sense is a formal dis- are organized under the generalheadings of : cipline. We take for granted the fact that the de- 1. Component Design. sign process possesses logic, structure, and meth- 2. System Design. odology.Furthermore, this process is amenable to generalization and therefore it can be described, From the standpoint of the overall design pro- taught, and put into practice. The morphology of cess, there is no differencebetween the design of the overall design process has been described by components and the design of systems. At present Asimow. I have called the basic elements of design, we think we see asubtle but significant distinction which are repeated over and over again in the design between the parameters of components and those process, the Anatomy ofDesign. of systems. 85 We generally find the design parameters ofcom- COMPUTER-AIDED SYSTEM DESIGN ponents and systems can be comparedas follows : System design is typified by largenumbers of Component Parameters System Parameters independent parameters. Theamount of data to be Materials Components processed becomes so massive thatgeometry and Geometry Topology materials can no longer be consideredin the equa- The distinction between thedesign procedures tions. The components of thesystem are then repre- for components and systemsseems to lie primarily sented by their through-and-acrossrelationship in the number of independentparameters that the (transfer or influence coefficient).The intercon- engineer or the computer is calledupon to handle. nection of the components is givenby topology. The The number of independentparameters that are system design consists of perturbationof topology perturbated in the design ofa component is rela- and components. tively small and an optimum solutioncan be sought. Although computer-aided systemdesign has made COMPONENT DESIGN BY COMPUTER significant advances in recentyears, and a great Component design is typified by the deal of systems analysiscan be done only by means direct per- of a computer, the surface of turbation of allowable materialsand geometries. this major application The component may intiallypossess a relatively has only been scratched. Thepresent state of the large number of parameters.However, statement computer-aided system design art resolvesinto a of the equation of constraint arisingfrom the laws man-machine partnership. The engineerwill select of nature, laws of society, a topology and initial components which willhope- customer specifications fully be capable of meeting and the house rules of the manufacturingconcern the system requirements. will generally reduce the numberof independent The computer will then perturbatethe components variables to a more manageable number. of the topology to performa quasi-system optimization. Today major components rangingfrom transfor- The resultant performance andcomponent values are then read back to theengineer, who can mers, motors, and filter networks to turbines, foun- either accept them dations, and the structure ofmulitistage rockets or suggest a new topology to are regularly designed by computers. To illustrate the computer.The dialogue betweenman and a general component-design procedure, the machine continues until the engineerhas exhausted course his repertoire and selected the investigates the computer design ofcommercial best of the quasi- three-phase induction motors. optimized alternatives. Induction-motor design requires the detailingof The computer-aided designcourse introduces a at least 26 parameters. The equationsof constraint system designed by studying thecurrent advances permit a reduction to five independentparameters. in electronic circuit synthesiswith computers. The One of these is a material parameterwhile the annual expenditure of engineeringman-hours on other four are geometric parameters.They are: the design of electronicsystems has spurred the extensive application of computers 1. Damper bar material Material. to this area.I 2. End ring cross section have been told thata substantial percentage of the 3. Punching diameter Minute Man Missile-Guidancesystem was produced 4. Core length Geometry. with the assistance of computer-aidedcircuit design. 5. Flux density- winding turns Computer routinesnow exist which only require the engineer to spell out the topologyby identifying The computer procedure beginswith the custo- system nodes, the connections between mer's specifications. A sizing nodes, and routine basedupon the type of components makingthe connections. the experience gained from pastdesigns determines The computer then converts this the maximum and minimum volumes input data into to be expected. the appropriate matrix equations.The equations Volume considerations and standardpunchings then are solved for any combination of inputs and determine the range of usable standard outputs laminations to give the d.c. performance, thea.c. performance and the maximum rotor length.With the design and the transient behavior. thus bounded, the computer then The componentsmay conducts a search be both active and passive, andmay also be linear for the optimum designas defined by the specified or nonlinear. criterion function.Full design and manufacturing In addition to the performance data are read out. to be expected from the computer componentvalues, the computer The gross similarities of componentdesign pro- will analyze effects of parameter cedures make the prospects for fully shift. The "worst generalized case" system performance is thenprovided with all component design routineappear excellent.The components taken at their worst procedures used to design extremes for an induction motor are temperature, aging, and manufacturingtolerances. applicable to other components, suchas a structural Since "worst case" for member. a large system is seldomrea- listic, Monte Carlo routinesare also available to 86 give the expectedperformance spreads of production lots. "This program providesa means of analyzing all load conditions ina trajectory and can select Electronic circuit designmay at first glance seem to be highly specialized. minimum component designweights. A complete However, the nonlinear, definition is provided automaticallyof the mini- lumped-parameter model ofthe electronic circuit has broad applicability mum weight vehicle system, thatcan be designed to other physical systems. with the input parameters.In addition the stif- COMPUTER -AIDEDPRELIMINARY DESIGN fened shell sections of thestructures are described A complicated proceisoften can be profitably by print-outs which definethe material used, skin categorized frommore than one viewpoint. Pro- gage, areas of stiffeners, and stiffenerspacing fessor Asimow has listedthe sequential steps inthe in both the longitudinaland circumferential Morphology of Designas determination of need, directions.The efficiency of thisprocedure is feasibility study, preliminarydesign, detailed de- illustrated by the factthat itis possible to sign, production,distribution, consumption, andre- synthesize a four-stage launchvehicle system by tirement.While repetitive detaileddesign is obvi- this process in ten minutesof computer time. ously within the provinceof the computer,pre- The same problem doneby hand would literally liminary design inmost industrial organizations take many man-months ofdetailed effort." usually requires their bestengineering talent.Yet THE FUTURE even here the computer has becomemore than a big The studies attendant adding machine. The upon the development of following quotationsare taken our computer-aided designcourse have given us from, a paper writtenby engineers ofa leading some rather definite ideas about aero-space concern.1 They the future of engi- are commenting upon neering. No singleinstrument, event,or idea within their corporation'suse of computers for thepre- the past 200 liminary design of multistage years has had the impactupon the rockets, but they also engineering profession, andconsequently engineer- document the applicabilityof computer routines to ing education,as has that of the computer. the design of otherengineering structures such as The physical originsof the equations it manipu- bridges or domes. Theyreport as follows: lates are ofno particular concern to thecomputer. "An example of thecomputer-aided designcon- The very limited number cept is the NAA LaunchVehicle Weight Syn- of analytical models available represent the realworld; the limited numberof thesis Program whichwas developed for the of criterion functions preliminary design weight and the limited numberof synthesis of multistage optimizing routinesguarantee the launch vehicles.The program is initiatedwith emergence of computer inputs relating very general automatic designprograms with ex- to performance, specifi- tensive interdisciplinaryapplications. cations, and design criteria.The programcom- putes the weight of With the advent oftime-sharing, the availability individual structuralcom- to even the smallest ponents, tank pressurizationsystems, propellant engineering office of large-scale systems, and insulationsystems for each stage computing facilities isno longer a question of cost ofthevehicle and accessibility.It is now possible inLos Angeles configuration. Other system to obtain your weights, suchas propulsion, flight control, guid- own console, a leased line, anda good ance, electronics and instrumentation block of timeon one of the largest commercialcom- packages puters for a total rental are handled as input numbers. Theprogram pro- that is well under$500 a vides a rapid month. The onlytemporary impedimentsto indus- means of optimizing structural try-wide computer-aided weight, includingassociated geometry weight, design are those ofsoft- penalties. ware and the lack of engineeringgraduates who have been educated to "The program consistsbasically of a sizing use this new tool. subroutine, a loads analysissubroutine, a tank Competitive pressuresnow guarantee that the pressure subroutine, various stress activities of the engineeringprofession must change analysis sub- drastically within the next routines, and a detailed weightanalysis subroutine decade.All professions which includes systemanalysis. are committed to the productionof decisions to In addition sub- optimize some values in routines are included whichinterrelate the vari- the face ofsome degree of uncertainty.There appears little ous analyses and accomplish theoptimization of question that the the stage mass functions speed, accuracy,memory and decision-makingcapa- and print out calculated city of the computer will numbers as wellas provide profile drawings of cause it to assume most the vehicle. of the routinizableanalysis and decisionmaking in engineering design. 1L. A. HarrisT. C. Mitchell,G. W. Morgan, "Computer- The role of the practicingengineer is goingto Aided Design for CivilEngineering Structures,"Technology become more professional.Less time willbe spent Status and Trend Symposium.Marshall Space Flight Center, on direct computation, with Huntsville, Alabama, April21-23, 1965. more effort devoted to modelling, criteriaestablishment, andthe design of 87 new design procedures.Optimization procedures Component-design computer programs will un- will be used with increasing frequency, particularly doubtedly be created to handle larger numbers of after some breakthroughs have been made in the parameters. The generalizedcomponent-design techniques for criterion function syntheses. programs will in turn spawn special problem-ori- Engineering schools will continue to provide a ented routines to facilitate communication with the good foundation in science and analysis, but in- machine. creasing time will be devoted to modelling, synthesis, The design information stored in machine memo- and optimization.Particular attention will be ries of material characteristics, component behavior, directed to the interface between a design and the subsystem performance, and useful topologies will society that the design purports to serve.Courses become extensive for every field of engineering. in "applied humanities" are beginning to appear. National design information centers will be linked to local computers.The system-design procedure General realization that the answers to any sub- written for machines will gradually becomemore stantial engineering problem must be a direct func- like that used by the engineer himself.Machines tion of a set of criteria and values will ultimately in their "off hours" will read the technical litera- direct a substantial concentration of engineering ture, analyze the papers, optimize the topologies research effort upon the problem of value system revealed, and store the resulting system-performance synthesis. data in memory (after perhaps writing a letter to The computer will have an indirect as well as the editors).Concurrently, programs will be de- a direct effect upon engineering curricula.Pro- veloped to allow machines to tear system specifica- gramming course requirements will not last any tioninto subsystems whose characteristics the longer than the old slide rule requirement.How- machine can then match against the subsystem ever, lower division mathematics will be substantial- characteristics stored in memory.At this stage ly reoriented to prepare students to think incom- the computer will be able to generate its own alter- puter terms and to obtain the most efficient use of native solutions to a system problem. the increase in availability of computer time and routines.Today we see a continuous increase in The major idea of the automatic factory with information which the engineer must handle, along computerized decision-making is probably somewhat with an increasing ability of the computer to process repugnant to many of us, but the elements are all available. information and .hake decisions.Thi5 combination It should be possible to deliver the insures that required courses in information-pro- customer's specification directly to the computer. cessing will become as commonplace in engineering The machine would then design, price allcom- curricula as are their present-day counterparts in ponents, and prepare a bid proposal. Upon receipt energy-processing. of an order the computer would complete the final design, produce drawings, check inventories, order Room for these new courses in computer-aided materials,programtheautomaticproduction design and information-processing will be found machinery, write work orders and factory releases, through more efficient arrangements of our present establish test procedures, program the automatic so-called engineering science courses. An intensive check-out equipment and update the design files study of upper-division engineeringcourses at with the results of the proof tests.All of this would UCLA found that, with few exceptions, therewas be done in addition to other routine decision-making, no new science introduced in the engineerirg science such as bookkeeping, payrolls, invoicing, etc. courses.All of the basic science had been intro- How much of the foregoing future possibilities duced in lower-division physics and chemistry. The will become reality is, ofcourse, a matter of con- engineering science courses were, in fact, courses jecture.It is, however, certain that its realization in analysis. A typical example would be circuit and direction will be the responsibility of the theory in which the ideas of inductance, capacitance, young engineers we are now educating. Weare beginning and resistance learned in physics are employed to to realize that engineering has progressed to the extend the student's ability to manipulate differen- point where the engineer tial equations. now has the ability to solve any engineering problem which hecan properly The industrial prospects for computer-aidedsys- define and to which society is willing to commit suf- tem design are particularly exciting.The general ficient resources. The designprocess has been well availability of large computers through leased lines defined in recent years and its major elementsre- and time-sharing will cause man-machine design vealed.It is only a matter of time before much of teams to become commonplace.Laboratory and our present-day engineering activities will be rou- testing time will be drastically reduced as models tinized enough to become subhuman activities which and routines become more sophisticated. we must relegate to machines. 88 APPENDIX I

UCLA Computer- and Design-Connected Courses tions of mathematical techniques for optimi- Quarter System zation with and without constraints. Linear quadratic, and nonlinear programming, dy- Engr. 8A Use of Digital Computer in Engineer- namic programming and stochastic processes. ing.Required Freshman Course - 2 units. II. System Simulation. Programming and utilization of digital computer with emphasis on the solu- Theory and methods applicable to the simulation of engineering problems. tion of a wide range of systems characterized by inherent stochastic elements. Engr.16B Introduction to Design. Simulation languages including SIMSCRIPT, Required Lower Division Course - 4 GPSS, and DYNAMO. units. Student term projects utilizing IBM 7094. Formal introduction to the design process. Problems requiring computer so- III. Design-Automation System. lution are usually included. Design-automation system for digitalde- Engr. 100D Information-Processing Systems. vices. Computer design of computers. Eequired Upper Division Course. Analysis, design, and utilization of in- IV. Computer-Aided Design. formation-processing systems; representation of information, processing of Study of computer-aided engineering design. information,digital systems - compo- Organization of the design process, its de- nents, building blocks, algorithms, pro- cision points and information sources for gramming systems. automatic machine-processing of premature inputs (specifications) to provide full design Other Required Undergraduate Experience data for any unique member of a designable Upper Division Design Course and Upper Division family. Preliminary design, component de- Design Laboratories have projects which usually re- sign, systems design, and the placement and quire some recourse to a computer. routine of problems.

Graduate Courses A major doctoral field in Engineering Design has I. Technique of System Optimization. been proposed with several of the above courses Theory, evaluation, and computer applica- expected to form part of the preparation.

89 APPENDIX II

COMPUTER-AIDED DESIGN Course Outline

I. The Anatomy of Design. B. Transformer. A. Review of major elements of designprocess. C. Waffle plate. B. Computer as an information-processing, decision-making device. III. System Design with Computers. A. Electronic circuits. C. Investigation of present and future roles of B. Structures. the engineer and the computer in the design process. IV. Placement and Routing. D. Component vs. system design. V. Preliminary Design.(Example- multistage rocket). II. Component Design by Computer. A. Induction motor. VI. Computer-aided Design Research Frontiers.

90 APPENDIX III

Anatomy of Design

Operation Performed by

Present Future I. Identification of need. Engineer Engineer II. Information collection, organization and stor- Engineer Computer age. III. Identification of system variables Engineer Engineer (inputs-outputs). IV. Establish criteria and constraints. Engineer Engineer 1. Constraints :allfactors which express + Computer Aid limitations. 2. Value system and criterion function. V. Synthesis of alternatives. Engineer Computer + Computer Aid + Engineer Aid

VI. Modeling of parameters of alternatives. Engineer Parameters Models Geometry Lumped parameter Material Distributed Energy parameter People Probabilistic

VII. Analysis-including test,evaluation and Engineer Computer prediction. + Computer Aid VIII. Decision steps. Engineer Computer Comparison ofanalysis of alternatives + Computer Aid + Engineer Aid against desired results using decision rules.

IX. Optimization. Engineer Computer + Computer Aid X. Communication, implementation and pres- Engineer Computer entation. + Computer Aid + Engineer Aid Paper work generation.

91 EBNER: What do you see as the proper balance and noticed that it required the students to derive between the courses in an engineering curriculum a particular integral. Just out of curiosity he and the synthesis courses? went around and asked all of his staff how they ROSENSTEIN: it depends on what you plan to do would derive that integral, and without exception with your output.Obviously, if the students are they said, "Well, I'd look it up in the tables." If going to become engineering professors you prob- so much of our analysis is going to go on machines, in other words, on mechanical tables, how ably don't have to teach them much synthesis. If, however, they are going to become practicing much time can we spend teaching the details of engineers, and if their main business is going to this?I think we should teach them something be synthesis, I think we have to strike a better that we basically don't teach, and this lack shocks balance than most of us, including my school, us when we analyze our engineering education. have struck so far. This is going to be particularly We give them a lot of workout on the analysis true if, indeed, we mean what we have been saying of problems which the professor has formulated here for the last few days, that analysis is going and which have nothing to do basically with real to become a sub-human routine and that the real life.Most of the time the boss says, "There's essence of engineering is going to be the assess- a problem over there," and if you're lucky you ment of need, the problem statement, the model- don't know where "there" is.But from there ing,' and probably the evaluation of the analysis on in you have to formulate the problem and the and the optimization which has come out of the analysis.You can usually get it analyzed if, you machine. can state it.In fact, this is the thing that gives As good an example as any I can think of is that me so much hope about what we're talking about, of Professor Kemeny at Dartmouth who was con- we have stated the computer-aided design prob- cerned at one time about the teaching of mathe- lem.I am completely confident now that 'the matics. He picked up a sophomore examination details will be taken care of.

92 THE IMPACT OF TIMESHARING COMPUTERS IN EDUCATION IN ENGINEERING DESIGN PAUL T. SHANNON Dartmouth College Hanover, N. H. It's a real pleasure to be here this morning. only brought my computer;I brought Mr. Mike Like Jim Reswick I also brought my own com- Juha, my programmer, and that really simplifies puter.I owe a great deal of thanks for this to things.Mike is a senior in Dartmouth College, the General Electric Company and the Illinois majoring in Engineering Science. Bell Telephone Company and I considerit an advancement in the state of the art that my re- D11-20 mote terminal teletype is operating out in the (IS Million Characters) hallway right now.I asked for it this Monday aim morning and it'shere, operative, on Thursday. In fact, it was operational yesterday.I now have my own computer here in Chicago.I can run in Chicago ;I can run at Dartmouth ; I can run Dual Access in New York;I can run in Phoenixall I need Controller is the magic set of numbers.This morning I'd DATANET-30* GE-235 15k-6.94gs 16k-5.94gs like to describe very briefly the Dartmouth-Gen- (with AAU) CIU eral Electric Time-Sharing System developed by 1711 Professor John Kemeny and Professor Thomas Kurtz and six undergraduate mathematics students (who did 80 per cent of the programming 0 work),to tell you a little bit about the system and how it operates, and then try to give you some 0 U appreciation of the tremendous impact this has Teletypes (33/35) had on our engineering education program at Dart- oss U mouth. DARTMOUTH TIME-SHARING SYSTEM 0 Time-sharing computer systems are certainly the Figure 1. The Dartmouth-General Electric 265 most significant development in computer technology Time-Sharing System.' in the last two years and will have a fantastic im- 2 G. E. Brochure CFB-440A. pact in the next five.The fundamental idea in * DATANET Reg. Trademark of the General Electric Co. time-sharing is the parallel servicing of multiple users, servicing them in such a fashion that they Our computer is a conversational computer so carry on a conversation with the computer so that the first thing we'll do after calling on the telephone each man believes he has his own computer. A good, is to say "hello."The system replies, "Who are concise description of the Dartmouth-General Elec- you ?", i.e., what's your user number ;what kind tric 265 time-sharing system has been given by of programming language did you want to use? Kemeny and Tribus. 1 The conversation is opened. The system is so fast "Time-Sharing Computing Systems" are just that I am not aware most of the time that there are merely words until you sit down at a teletype and use other users. the system. One of the remote terminal teletypes At Dartmouth we have some 30 teletypes located you see represented schematically in Figure 1 is around the campus. We have four in the engineer- located out in the hall.I hope as many of you as ing school which service 175 undergraduates, 60 possible, during our breaks and after the sessions, graduate students and 25 staff members. Our basic will get a chance to sit down and use the system philosophy is that Teletype time is "free," i.e., readi- because there is no other way to truly appreciate it. ly available for all to use. We seldom run into long I'm going to use it here in just a minute.I not queues to get on to the Teletypes. It is sometimes 1Kemeny, J. G. and Tribus, M., "The Dartmouth Time- a little crowded in the afternoon during the week Sharing System," Journal of Engineering Education, Vol. but computer use slacks off about 11:00 p. m. in any 56, No. 8, (1966) p. 826. event. 93 The system isactive 20 hours a day, seven days DEVELOPMENTS AT a week. The 30 Teletypeson the campus service DARTMOUTH about 1,000 One of the mostfantastic success stories users who have approximately5,000 of in the I know programs in core.The system is development of thecomputer systems memory bound. occured at Dartmouthunder the direction We have to "clean out"the 18-millioncharacter disc of Pro- once a month. If you haven't fessors John .1.Zemenyand Thomas Kurtz.In March, used aprogram within 1964, the General the month it getsthrown away. Electric 235 computerhardware was unloaded from thetruck and installedin the In a time-sharingenvironment if the basement of College doesn't talk to the Teletype Hall. We had theusual trouble user within 10 seconds(not 15 with air conditioning,etc.The equipment minutes or 3 days)he is unhappy. was in operation by April 1,1964. On May 3, 1964,at 3 :00 a. m., the time-sharingsystem operated with AN EXAMPLE BASIC for 30 seconds!By June 1 of thatyear both the Last February Ineeded an exampleproblem for 'time- sharing systemand the BASIClanguage com- a workshop session ofa course In digital simulation piler were operationalin phasezero.In early June for practicingengineers in Sarnia,Ontario. 8 We Professor Kemenyintroduced BASIC andthe sys- were using the Dartmouthcomputer in Hanover, tem to the faculty ofthe college. Thethree summer New Hampshire.A quick interrogationof my files k months saw the establishment ofthe computerexecu- produced a seven-stageextraction problem.Being tive system at theComputation Center andboth pressed for time, Iturned to Dan Frantz,a first- the time-sharingsystem and BASICentered phase year graduate student. Danwas a physics major at one of development.The systemwas then intro- Mkhigan. Idescribed to him in20 minutes how duced to the entirecollege in September1964 with an.. extractor works. Youknowsolidsgo in and some 20 Teletypes installed. mix with water; the solutiongoes one way and In Thayer SchoolI presented twoor three lec- the ganggoes the other. Dan leftme, went to his tures on the system desk and wrote and BASIC, andstaffed our his computer-unitcalculation in computation room containingfour Teletypes with one hour. An hour anda half later he hadpro- undergraduate andgraduate studentsto assist our grammed the unitcalculation, got itworking, im- students and staff ingetting to know and mediately fitted it use the within an executivesystem for new system. By February1, 1965, I removedthe this kind of problemand in three hours, students because they clock, he by the were no longer required ! The came back to me with theanswers.His system had becomean integral part of letter to me, the our educa- program and the answersare given tional and researchactivities at ThayerSchool. in Appendix I.This illustratesvery well the power At Dartmouth, of time-sharing in introductory programming,using a programming environment. the BASIC language,is taught to all ofour fresh- As an engineer doingdesign I'm alwaysgenerat- men in their mathematicscourse.About 90 per ing alternatives butI want a quantitativeevalua- cent of our students(an all-male population)take tion of them. Forexample, what happensif I take a year of mathematics.I mentioned yesterdaythat 80 poundsper hour as the amount ofwater into the this then ledus to delete acourse in numerical leaching system? methods and computingfrom our engineeringscience core curriculum and put in While thiscase is being run, I'd liketo give you a course in field theory, a little bit of the history electricity and magnetism.But let'ssee what is of the developmentof really the impact time-sharing at Dartmouthand referyou to the on the students. April, 1966, issueof the Journal ofEngineering Edu- STUDENT IMPACT cation which was published just in timeto alleviate In talking withMike and me from writinga whole lot of things down. some other students, the lead article by In they make statementssuch as, "Weuse The computer Professors DonaldKatz and for anything thattakes more than Brice Carnahan,they view the futuregreat impact 30 minutes to of time-sharing calculate by hand;it's easier." Sophomorestudents computers in engineeringeduca- in their first engineering tion. 4 The last articleby John Kemeny course, ES 21, Introduc- and Myron tion to Design, wroteand useda BASIC program Tribes 5 describesthis impact whichis here today. which related torque,angular position,velocity and acceleration for therotation ofa human arm in Nrohnson, A. I., "DigitalSimulation of Chemical raising itself anda weight in a vertical plane. Chemistry in Canada, Plants," They April 1966, p. 16. used these designcalculations in theirsophomore 4 Katz, D. L.and Carnahan, B., "ThePlace of Computers project work. We find in Engineering Education," that our studentsregularly Journal of EngineeringEduca- use the computer to solve tion, Vol. 56, No. 8,(1966) p. 293. course problems without 5 Kemeny and Tribus, having staff membersassign computerproblems. op. cit., p. 326. The students are using the computer notbecause they 94 were told to do so, but simply because they see it as group of practicing engineers in Sarnia, Ontario an easier way to do their work and they're always this fall and we ran on the Dartmouth computer as looking for an easier way to do things.They are part of the course work. 6I am teaching a course developing an aptitude so that getting a numerical in another industrial concern and will install my answer is trivial. The whole problem is : how to forcomputer terminal and operate.In the chemical mulate problems, haw to program them, how to set industry you have some companies right now who them up so that I know I will get answers. The stu- have extensive internal time-sharing capabilities. dents very rarely get into mathematical difficulties. Tremendous strides are being made in the automatic When they do, they head for the staff member for analysis of data from analytical expirements.If help and they get them solved. We have mounted one you look for repetition, chemical analysis is a fertile Teletype on wheels and we use it in the classroom. In field.So now we see automatic programming of several courses that meet for a couple of hours, ques- experimental equipment such as an electron probe tions raised in the first hour have been programmed microscope taking this information immediately into and completed in the fifteen-minute break between the core of the computer via telephone lines and classes and the answers discussed in the second processing it.I was at one research center where hour.Half of the students, in their courses in fluid a man made the statement that he can now generate mechanics and solid mechanics, write problems for in one week all the experimental data ordinarily data reduction and analysis of their laboratory found in a Ph. D. thesis ! work. We find a much closer tie between what is going on in the laboratory and the mathematical COSTS modeling because you can go back and forth so Time-shared means cost-shared and in calculating easily. the cost of a time-sharing system you have these three basic elements :the cost of the remote termi- STAFF IMPACT nal, the cost of the telephone link, and the cost of Now, what has happened to our staff?I'm not core time on your big computer.This opens up a so much impressed when Paul Shannon uses the tremendously increased number of possible users. computer as I am when some of our senior staff mem- However, while the rates go lower and lower, the bers who have never used a computer before do so. total price tag goes higher and higher. Two years For example, a senior professor in civil engineering, ago Dartmouth had an LGP 30 and an IBM 1620 whose field is water and waste treatment, has taught computer which cost us roughly $1,100 a month. Now himself BASIC and discovered that in 15 minutes we have 30 Teletypes on campus, each of which costs he can do simple linear regressions that used to take $130 a month for a total of $3,900 a month just him three to five days or three of our staff, having for our remote terminals. And we're contemplating no previous computer experience, who got together a next generation of computers in which the main- and offered a computing course for Dartmouth tenance cost is about $100,000 a year.Perhaps alumni in the construction field.It is this impact these few figures will highlight the very serious that impresses me.I did a quick tabulation of our questions about who is going to pay for this tre- staff members and found that a little over 50 per mendously increased functional capability that is cent of our staff members use the computer. Within being implemented in our colleges and engineering that 50 per cent there are five that use it very heavily schools. and within that five there is one who takes 90 per cent af the machine time ! CONCLUDING REMARKS Time-sharing man-machine interaction produces Well, I want to finish up very briefly by indicating a situation in which staff and students can imple- one type of design effort which is possible today ment their concepts, program their mathematical because of computers and one which I think is an model, debug the programs and obtain answers 10 indication of what is going to happen tomorrow. to 100 times faster than a person operating in an Last year I had the real pleasure of working with efficient batch-processing system. These figures are the five members of the Chemical Engineering de- not just based on looking at Dartmouth but also on partment at McMaster University and their eight some surveys that were done in industry.The seniorchemicalengineering students. Working industrial surveys indicated that it was 40 to 100 closelywithpeople from CanadianIndustries times faster by the clock between the statement of Limited, we set ourselves the task of mathematically a problem and the answers compared to a conven- modeling C. I. L.'s 300 ton/day sulfuric acid plant tional batch system. located in Hamilton, Ontario. We started in Decem- A computer is now as close as your telephone.I ber, got going by the middle of February andcom- have one here, thanks to the work of a lot of people. I helped to teach a course in digital simulation to a °Johnson, op. cit., p. 16. 95 TOL,S.,K, 1

pleted the study on April 6.7 The resulting PACER the wave of the future, but somebody's got to simulation represented the simultaneous solution of give some serious thought to ways of cutting the approximately 500 equations, many of them highly price down by orders of magnitude. non-linear, given about 200 constants and solving for about 1,000 unknowns. PAYNTER: Twoquestions, one serious, one humor- I will simply say that we did it andone of the ous. The serious oneare you people thinking things that gave me the most pleasurewas talking at all of really small, portable consoles that stu- to a student just before final presentation who said, dents could actually carry with them? "When I first started this project I thought it was SHANNON:No. We are looking very much at the impossible.I no longer do.It's just difficult." design of 1,000-bit per second remote terminals. In my graduate level design courses Iam now Now the question is: can we builda box that asking my students to realistically modelan entire sells for under $500 which will put out plus and desalinization plant within three to four weeks. This minus 10-volt analogs, plus and minus 10-volt is only possible in a time-sharing enivronment. electrical signal, so that I can plot ona standard Time-sharing is having a tremendous impact and I oscilloscope.We may have to buy amemory think that as we go into the next five years we will scope initially, so that we can drive a standard develop more users and terminals which have much analog plotter, but the feeling is that if the engi- more man-machine interaction, and thus evolve an neer can get an answer out to 10 per cent on an almost completely different mode of computer utili- x-y plot for a couple of variables, he won't want zation in engineering education and design. to see 90 per cent of the stuff he now prints out. And so it's the development of this nextgenera- BRAUN:We've had two terminals into the GE tion of program-controlled hardware that we're Datanet system for about five months.I'm not looking at. a chemical engineer and I don't really know what chemical engineers do, butone of our senior and PAYNTER:The humorous one was, has anybody most conservative chemical engineering profes- at Dartmouth suggested using terminals to test sors has brought it into his unit operations lab. and teach, thereby putting the faculty eitheron As I understand it, the students used to make sabbatical or emeritus? some measurements and sit down at a table in SHANNON:Well, not quite, because there's an the lab with their slide rules foran hour or two, awful lot involved here. I find myself using the make some calculations, and thengo back. Now, following words trying to get acrosssome of the every Wednesday and Friday afternoon the pro- ideas. The words are, "mathematical hardware." fessor runs down and putsa cart underneath our When Dean Tribus asks me what we do, Isay Teletype terminal andruns it into the lab, and the We build mathematical hardware. And computer students now make some measurements,run a programming is the development of mathematical couple of programs they have developed and ina hardware. And as we are able to define func- matter of minutes go back and modify their tionally the components ofour hardware and measurements.He is completely sold on this build up the unit calculations and have standard thing;I think probably if we got rid of these routines, we can then go to the building of auto- machines, he and a number of the other students matic machinery. Now automatic machinery is would pull the walls of the place down! modular, functionally designed, and highly speci- I would like to take some slight exception toa fic.So I think there's quite a good analog here comment made about the price of this.These two between the generation of problem-oriented lan- terminals were costing us $5,000a month because guages now under development and the sort of the machines were up almost twenty-four hours automatic 'machines we see sitting around in a day, six days a week.This is pretty nearly plants.Machines also have capital costs, design half of what it cost us torun our 7040, and that's ideas, maintenance costs, operating costs, all of only two terminals running almost twenty-four these things which are all true of computing work, hours a day. We now have a lockon the tele- and this is one way I try to get across, not toan phone dial.The lock distresses me, because the audience like this, but to those who haveno feel students are not using it nearly as effectivelyas at all for computers, what is actually goingon. the people at Dartmouth.We've managed with The whole PACER system might be said torepre- the locks to get the cost down to about $2,200a sent a format for mechanically implementing the month.That's the problem.I think that this is standard mathematical techniques known today. ',Shannon, P. T., et. al., "Computing Simulation ofa Sul- QUESTION: Youmentioned somebody who is gen- furic Acid Plant," C. E. P., Vol. 62, No. 6 (1966)p. 49. erating data in one week that originally took about 96 a year without the computer. I think I talked to generating ideas and translating them into com- the same man, and he also mentioned that, being puter languages.The man who generates the a Ph.D. candidate, his school refuses to accept ideas deserves the credit;the man who trans- this work for dissertation, or at least is very relates them into computer languages is merely luctant to do so.So my question is directed to performing the tasks that can be done in a pedes- you and all the other educators, is this going to be trian way. a problem? SHANNON: That's right.You don't want to get SHANNON: My answer to that is yes. Particularly, involved in coding.There's a great deal of dif- one has to come to the realization that a legitimate ference between coding, writing out the detailed product for an engineer today is information. In procedural instructions, and the process of trying fact, I think a tremendous percentage of our to generate out the algorithm, the functional form engineers are really technical information hand- of your system. How it's all going to be put lers, basically, in their job function. But if you together, how to analyze it and study the informa- say to a man who thinks only in terms of physical tion flow, all of these things are not coding prob- embodiment of his concepts, I've got a student lems. who has written this fantastic computer program, and he says, "Yes, but he's done no design," I say, QUESTION: I'd like you to extend that to one more what do you mean he has done no design?Well, question, please, because I think there's an im- he didn't conceive of anything and try it and portant factor left out in this time-sharing dis- modify it, generate alternative ways to do some- cussion.And I think the speaker should say thing and all the rest of this. We heard so much something about the considerations of the incor- of this yesterday but, in fact, this does involve poration of display scopes in the general time- exactly the design function, the design process. sharing system. Creating these programming languages, creating SHANNON: I visualize the next couple of genera.. these mathematical systems involves the same set tions of time-sharing equipment as going through of ideas and principles and techniques and prob- several orders of magnitude as far as the remote lems, but it is not recognized as such as yet. And stations. At level one, we had the DAC-I, "sketch- this is another reason I talk to my engineering pad" kind of console with the light pen, a million educator friends about mathematical hardware, dollars of hardware on the end of your remote to try to get them into this frame of mind. But station. For every one of these systems I visua- it is a problem and it is not recognized. Another lize ten small computers, actually complete remote thing that is not recognized is the fact that com- computers with very fast transmissions into puter programs should be treated exactly the larger processors. And for every one DAC and same way as any other scientific publication. The ten small computers, a hundred systems which first thing on it is the man's name, referenced, will have very limited facilities, say three-inch doc,-umented, and you don't pick up anybody else's scope, able to print at 100 lines a minute. And work and use it without giving credit. And this then we are going to see a thousand Teletype-like has got to go forward; it's moving much more terminals.So, we are going to get this triangle in this direction now. with decreasing costs and more users at each QUESTION: I'd like to draw a distinction between level.

97 APPENDIX

FEBRUARY 23, 1966

TO: PAUL T. SHANNON FROM: DANIEL R. FRANTZ RE: SEVEN STAGE LEACHERUSING 'PACER'

ENCLOSED ARE MATERIALS CONCERNING THE SEVENSTAGE LEACHER THAT HAS BEEN ADDEDTO THE 'PACERTIME-SHARING MANUAL': 1. STATEMENT OF PROBLEM 2. INFORMATION FLOW CHART 3. LISTING OF UNIT CALCULATION 4. SINGLE STAGE EXTRACTION A. PROGRAM AND DATA B. RESULTS 5. SEVEN STAGE LEACHER A. PROGRAM ANDDATA--INCLUDING AUXILIARYSUBROUTINE 'SETH2O' LISTING B. RESULTS

NOTE THAT SUBROUTINE 'SETH2O WAS WRITTENTOSPEED CONVERGENCE BY FIRSTFILLING THE EXTRACTIONCELLS WITH WATER AS THEY INDEED WOULDBE IN ACTUAL OPERATION. EFFECT THE SAME COULD HAVE BEEN PROVIDEDBY ADDING THE WATERTO THE DATA, BUT THISAPPROACH SEEMS SOMEWHATMORE FLEXIBLE. THE RESULTS OF THISSIMULATION ARE IN EXACTACCORD WITH THE ANALYTICAL RESULTS FORTHIS LINEAR CASE. IT IS NOTEWORTHY, HOWEVER, THAT WHILE ANALYTICAL SOLUTIONSTO SUCH A PROBLEM DO NOT EVEN EXIST FORNON-LINEAR CASES, ALLTHAT IS REQUIRED TO MODIFY THE 'PACER' SUBROUTINE FOR NON-LINEARITY IS THE REDEFINITION OF ONE EQUATION. (REFERENCE FOR ANALYTICAL SOLUTION: CHAPTER XI,"ELEMENTS OF CHEMICAL ENGINEERING", W.L. BADGER AND W.L.MC CABE.) ANOTHER POINT OF INTERESTIS THE AMOUNT OFTIME REQUIRED TO DEVELOP THIS SUBROUTINE. THE SELECTION AND DISCUSSION OF THE PROBLEM WITHA NON-CHEMICALENGINEERING STUDENT TOOK HALF HOUR; ONE CODING INTO AN 'ALGOL'v 'PACER' TYPE CALCULATION OCCUPIED UNIT ONE AND A HALFHOURS; TYPINGAND DEBUGGINGAT THE TELETYPEREQUIRED ANOTHER HOUR TOTAL OF THREE AND AND HALF--A A HALF HOURS FORA STUDENT WHO HAD NEVER HEARD OF .A LEACHERBEFORE.

98 W",,,=

LEACHING

AN EXTRACTIONBATTERYCONSISTS OF 7 DORR THICKENERS. THERE IS FED INTO THE NO. 7 THICKENER (STRONG SOLUTION END) 20 TONS PER HOUR OF MATERIAL TO BE LEACHED, ANALYZING: 80 PERCENT BY WEIGHT INSOLUBLE GANGUE 9 PERCENT HIGHLY SOLUBLE SALT C A 28 PERCENT SOLUTION IS SATURATED AT THE TEMPERATURE OF LEACHING) 11 PERCENT WATER

THERE IS PUT INTO THE NO. 4 THICKENER PER HOUR 15TONS WEAK MOTHER LIQUOR ANALYZING: 3 PERCENT SALT 97 PERCENT WATER

THERE IS PUT INTO THE NO.1 THICKENER PER HOUR 40 TONS OF WATER. EQUILIBRIUM BETWEENUNDERFLOW AND OVERFLOWMAY BE ASSUMED TO BE 2 TONS OF WATER (EXCLUSIVE OFDISSOLVEDSALT) PER TON OF INSOLUBLE GANGUE. LIQUID AND GANGUE MOVE THROUGH THE BATTERY IN CONTINUOUS COUNTERCURRENT FLOW.

CALCULATE

A) QUANTITY OF STRONG SOLUTION PRODUCED PERHOUR. B) CONCENTRATION OF STRONG SOLUTION. C) WEIGHT PER HOUR OF SOLUBLE SALT DISCHARGED FROM NO.1 ALONG WITH INSOLUBLE GANGUE.

99 40 H20 14.55 H20 0.45 SALT 0 11111I 0wi I 4 5 7 is -644-(5/ 0 16 SOLID® 1.8 SALT 2.2 H20

FIGURE 2 SEVEN STAGE LEACHER

100 EXTR 14:51 JUNE 17,1966

PROCEDURE EXTRAC; BEGIN REAL FR,RAT,WAT,SLT,SOL,A,B; INTEGER I,J; COMMENT THIS PROCEDURE SIMULATES AN EXTRACTION CELL. IT MIXES ALL THE INPUTS AND DISTRIBUTES THE RESULTANT TO THE OUTPUT STREAMS AS FOLLOWS: FIRST OUTPUT: OVERFLOW STREAM, RECEIVES THE EXTRA WATER AND SALT, IF ANY. SECOND OUTPUT: THE CONCENTRATED OUTPUT. TAKES A PORTION OF THE SOLID (AS DEFINED BY THE EQUIPMENT PARAMETERS BELOW), SOME WATER AND SALT. THIRD OUTPUT: (IF PRESENT) RECEIVES THE REMAINING SOLID, AND A PROPORTIONAL AMOUNT OF WATER AND SALT

IF THERE IS NOT ENOUGH WATER TO MEET THE REQUIREMENTS FOR OUTPUTS 2 AND 3, THEY WILL BE PRO-RATED ACCORDING TO THE THE AMOUNTS OF SOLID EACH IS TO RECEIVE.

STREAM FORMAT EQUIPMENT FORMAT 1. STREAM NUMBER 1. EQUIPMENT NUMBER 2. AMT. SOLID 2. FRACTION SOLID TO OUTPUT 2 3. AMT. WATER 3. RATIO OF WATER TO SOLID 4. AMT. SALT 5. FRACTION SALT SOLUTION;

SLT: :WAT : :SOL : :0; COMMENT TRANSFER TO PROGRAM VARIABLES AND FIND TOTAL SOLID, WATER AND SALT INCOMING; RAT:=ENINE,31; FR : :IF NOUT :3 THEN EN[ NE,2] ELSE 1; FOR I:=1 STEP 1UNTIL NIN DO BEGIN SOL:=SOL+STRMI(I,2); WATI=WAT+STRMIII,31; SLT:=SLT+STRMIII,41; END;

STRMO[2,2]:=A:=FR*SOL; STRM0(3,21:=B:=(1 -FR)*SOL; IF RAT*SOL4=WAT THEN BEGIN COMMENT DESIRED RATIO CAN BE SUPPLIED, DO SO; STRM0[2,3] :=A:=A*RATI STRM012,41:=A*SLT/WAT; STRM013,31:=B:=B*RAT; STRM013,41:=B*SLT/WAT; END ELSE BEGIN COMMENT NOT ENOUGH WATER, PRO -RATE IT AND THE SALT; STRMOU2,331=FR*WAT; STRM0(2,41:=FR*SLT; STRM0[3,3):=(1 -FR)*WAT; STRMOI3,4]:=(1 -FR)*SLT; END;

COMMENT FIRST OUTPUT IS OVERFLOW; STRM0(1,21:=0; STRM011,31:=WAT-STRM0[2,3]-STRM013,31; STRM0[1,4]:=SLT-STRMOI2,41-STRM013,41; COMMENT NOW CALCULATE SALT CONCENTRATION FOR ALL STREAMS; FOR I: :1 STEP 1UNTIL NOUT DOSTRMOII,51:= IF STRMOII,41=0 THEN 0 ELSE STRMOEI,41/(STRMO(I,4] +STRMOII,31); END OF EXTRAC; 101 1,11111111.110

SETH2O 14:55 JUNE 17,1966

PROCEDURE SETH2O; BEGIN INTEGER I,J,N; REAL A;

COMMENT THIS SUBROUTINE ISUSED TO INITIALIZE THE WATER CONTENT OF STREAMS IN A SERIES OF EXTRACTIONCELLS.

IT TAKES THE WATER THAT IS CONTAINEDIN THE INPUT STREAM AND FEEDS IT TO OTHER STREAMS, AS DESCRIBED BELOW.

THE STREAMS TO BE " WATERED" ARE GIVENIN THE SPECIAL DATA BLOCK "SWAT" IN THE FORMAT: DATA EWAT:=N,S1 ,S2,...,SN; COMMENT WHERE N = NUMBER OF STREAM NUMBERSFOLLOWING SJ = THE NUMBERS OF THE STREAMS TO BE WATERED THE STREAM FORMAT ASSUMED IS: 1. STREAM NUMBER 2. AMT. SOLID 3. AMT. WATER

4. AMT. SALT 11 5. FRACTION SALT SOLUTION;

COMMENT A IS AMOUNT OF WATER TOFEED; A:=STRMICI,3);

COMMENT RESTORE DATA BLOCKIN CASE THIS EQUIPMENT IS USED MORE THAN ONCE; RESTORE(EWAT); READATA(EWAT,N); FOR I:=1 STEP 1 UNTIL N DO BEGIN READATA(EWAT,J); SNIJ Olt:A; END; END OF SETH2O; LIST

LEACHR 17:40 JUNE 9,1966

! 100 PROCEDURE EXTRAC;BEGIN REAL FR,RAT,WAT,SLT,SOL,A,B;INTEGER I,J; 110 SLT:=WAT:=SOL:=0;RAT:=ENINE,31;FR:=IF NOUT=3 THEN ENINEal ELSE 14 120 FOR It :1 STEP 1 UNTIL NIN DO BEGIN SOL:=SOL+STRMICI,21;WAT:=WAT+ 130 STRMIII,3];SLT:=SLT+STRMI(1,41;ENDISTRM0(2,21:=A:=FR*SOL;STRM0(3,21:= 140 B:=0-FR)*SOL;IF RAT*SOL4=WAT THEN BEGIN-STRM0(2,31:=A:=A*RAT; 150 STRM0(2,41:=A*SLT/WAT;STRM0(3,3):=8:=B*RAT;STRM0(3,41:=B*SLT/WAT; 160 END ELSE BEGIN STRM0(2,3):=FR*WAT;STRM0(2,41:=FR* 170 SLT;STRM0(3,3):=0-FR)*WAT;STRM0(3941:=0-FR)*SLT;END;STRM0(1,2):=0; 180 STRM0(1,31:=WAT-STRM0(2,31-STRM0(3,311STRM0(1,41:=SLT-STRM0(2,41-

190 STRM0(3,411FOR I:=1 STEP 1 UNTIL NOUT DO STRMOII,51:=IF STRMOEI,41=0 200 THEN 0 ELSE STRMOII,41/(STRMOII,3)+STRMOII,41);END; 300 PROCEDURE SEIH20;BEGIN INTEGER I,J,N;REAL A;A:=STRMI(1,3); 310 RESTORE(EWAT);READATA(EWAT,N);FOR I: :1 STEP 1 UNTIL N DO BEGIN 320 READATACEWAT,J)1SNIJ,31:=A END; END; 400 PROCEDURE EQCAL;IF NECALLINE1=3 THEN EXTRAC ELSESETH20; N

OLD OLD PROBLEM NAME--D1LECH***

READY.

LIST

D1LE0 17:42 JUNE 9,1966

410 PROCEDURE PRUN;IF NECALLINE1=3 THENPRINTC","EXTRAC","") ELSE 420 PRINTC","SETH20",""); 1000 COMMENT SEVEN STAGE LEACHER; 1010DATA D1:=21,50,1; 1020DATA D2:=8,6,3,17,5; 1030DATA D3:=8, 10311,3,1,16,-5,-4,0, 10322,3,5,15,-6,-16,0, 10333,3,6,14,-7,-15,0, 10344,3,7,13,17,-8,-14, 10355,3,8,12, -9, -13,0, 10366,3,9,11,-100-12,0, 10377,3,10,2,-3,-11,0, 10388,4,1,-1,0,0,0; 1040DATA D4:=8, 10411,1,2, 2,1,2, 3,1,2, 4,1,2, 5,1,2, 6,1,2, 7,1,2, 8,0,0; 1050DATA D5 : :3, 10511,0,40,0,0, 10522,16,2.2,1.8,.45, 105317,0,14.55,.45,.03; 1060DATA D6:=S-4,$-4,$-4,$-4,$-4; 1070DATA D7: :8, 8,1, 7,2, 6,2, 5,2, 4,2, 3,2, 2,2, 1,3; 1080DATA EWAT:=12,5,6,7,8,9,10,16,15,14,13,12,11; 1090END;

108 EDIT WEAVE PACER-***,LEACHR***, D1LECH*** READY.

RUN

D1LECH 17:47

DARTMOUTH ALGOL.

DATA FOR RUN NUMBER 21

PROCESS MATRIX

1 EXTRAC 1 16 -5 -4 0 2 EXTRAC 5 15 -6 -16 0 1. 3 EXTRAC 6 14 -7 -15 0 4 EXTRAC 7 13 17 -8 -14 5 EXTRAC 8 12 -9 -13 0 6 EXTRAC 9 11 -10 -12 0 7 EXTRAC 10 2 -3 -11 0 8 SETH2O 1 -1 0 0 0 CALCULATION ORDER EQUIP MODE 8 1 7 2 6 2 5 2 4 2 3 2 2 2 1 3

EQUIPMENT PARAMETERS

1 1 2 2 1 2 3 1 2 4 1 2 5 1 2 6 1 2 7 1 2 8 0 0

104 STREAM VARIABLES

1 0 40 0 0 2 16 2.2 1.8 .45 3 0 0 0 0

4 v . 0 0 0 0 5 0 0 0 0 6 0 0 0 0 7 0 0 0 0 8 0 0 0 0 9 0 0 0 0 10 0 0 0 0 11 0 0 0 0 12 0 0 0 0 13 01 0 0 0 14 0 0 0 0 15 0 0 0 0 16 0 0 0 0 17 0 14.55 .45 .03

,i CONVERGENCE CRITERIA

.0001 .0001 .0001 .0001 .0001 4.

CONVERGENCE IN 39 LOOPS

RESULTS FOR RUN NUMBER 21

STREAM VARIABLES

1 0 40 0 0 2 16 2.2 1.8 .45 3 0 24.75 2.0962 7.80819$-2 4 16 32 .153149 4.76312$-3 5 0 40 .191436 4.76312$-3 6 0 40 .430732 1.06536$-2 7 0 40 .729875 1.79199$-2 8 0 54.55 1.50538 2.68553$-2

9 0 54.55 2.06038 3.63958$-2 , 10 0 54.55 3.00668 5.22385$-2 11 16 32 2.71024 7.80819$-2 12 16 32 1.76377 5.22385$-2 13 16 32 1.20866 3.63958$-2 14 16 32 .883084 2.68553$-2 15 16 32 .5839 1.79199$-2 16 16 32 .344586 1.06536$-2 17 0 14.55 .45 .03 CONVERGENCE CRITERIA .0001 .0001 .0001 .0001 .0001

TIME = 37 SECS.

105 A.m. t..01. *NO VW- 40.6

fometowara"nows.e...... ,113.07,4.

THE NEED FOR COMPUTERS IN TEACHING AND RESEARCH H. M. PAYNTER Massachusetts Institute of Technology Cambridge, Mass.

INTRODUCTION ized Department of Electromechanical Systems En- As high-speed computers seem here to stay, albeit gineering. PRO has just returned to XY Tech from becoming ever larger, faster, more hybridized and the Conference on the Impact of Computers on Edu- interactive, it now seems useful and necessary to cation in Engineering Design. attempt to assess their place in society.If they are, The scene opens : indeed, as revolutionary an innovation as gun- powder, movable type, the steam engine, etc., then PRO: Well, thanks for sending me to the confer- in particular we should try to forecast their impact ence.At least this one was concerned with new upon our engineering schools. uses for computers and not just new models, pro- Now I happen to belong to that strange cult which grams and languages. conceives the design mystique as the natural highest CON : You don't have to go to Chicago to learn calling of the engineer, and, in truth, of all pro- that, I hope.Did anybody say anything signi- fessionals and artists.So I cannot view the teach- ficantly different or differently? ing of design as somehow distinct from teaching, PRO: At least a theme was well set at the start. research, and scholarship in general. And for me, Dean Everitt charged the conference with the at least, computers in various guises are as simple professional obligation to crystallize the funda- necessities as pencils for the poet and scalpels for the mental concepts of computer-aided design.Lick- surgeon. lider and Miller said we must learn how to teach To avoid some of the stigma of presumptuous pre- students now to fully exploit the computers of tense, and also simply to have fun, I have chosen to decades hence, which will be unlike anything now cast my pearls in the form of a fictitious dialog. existing.Licklider went on to say these machines Perhaps thereby I can express with impunity some undoubtly will deal with more than numbers and strong feelings on the evolving role of computers- alphanumerics, and will rather be true informa- on-campus. tion networks. Chuck Miller described the newer Prologue and Cast of Characters new-man of 1976: a master designer more akin The following imagined conversation takes place to the architectural master planner-builder than between two protagonists : the design engineer of today. ARDENT PROPONENT: a young man of CON : Wait a moment!Surely these ideas didn't the new breed, born to a world in which tele- all go unchallenged.I have heard Licklider wax vision, space-flight and large computers seem poetic over the on-line intellectual community as natural as air and water; and a more attractive wording than Martin Green- berger's information utility.But aren't there a SKEPTICAL CONSERVATIVE: an older few steps along the road which are not even now man, mentor of the foregoing, alert to the ex- equipment-limited?You know that I think the ploding technology and the main thrust of the book was the greatest teaching machine ever in- systems approach, but whose natural milieu is vented! Maybe Lick thinks people have forgotten philosophy and scholarship. how to read as they have forgotten how to spell? Henceforth they are called simply PRO and CON, PRO :That's not the point!Lick, and others, too, respectively.Together with some of their school made much of interactive graphic displays used colleagues, they are responsible for undergraduate for self-communication and for inter-communica- and graduate curriculum development and other tion.I don't see how any ordinary sort of book teaching and research activities in the areas of sys- can fulfill that role.And you, yourself, know tem dynamics and control, and the applications of what we are doing with computer-aided teaching computers to engineering analysis and design. Both of dynamic system behavior.Incidentally, much have spent countless hours over the last decades in talk went on regarding on-line dynamic modeling interdisciplinary workshops and on faculty commit- and system simulation.This was put in contrast tees devoted to such subjects. PRO is Director of the to the conventional, more static, view of engi- Computer Center; CON is head of the newly organ- neering. 106 CON : But a passive attitude and a static view have CON: And also, I would suppose, Hank Paynter. never been natural either to children, or to Ameri- But, to go on, surely among the most remarkable cans, for that matter! Change and dynamism are developments of the last few decades has been not exactly new. What may possibly be new is the incorporation of computers, large and small, the ability to describe and to control change; to analog, digital, hybrid, directly into prototype appreciate and to model dynamic situations amidst systems, frequently as special-purpose devices. complexity and confusion.Certainly for us who For the life of me, in this connection, I find it were teethed on analog computers and even on hard to distinguish between computers, proper, thinking by analogy, on-line dynamic modeling and such elements as instruments, signal condi- and real-time simulation are not new experiences. tioners and controllers.Certainly all must play We have attempted to see behavior before building a role in design. for some time. PRO: As I said, Jim Reswick in particular made And another point before castigating statics : this point at the conference.But I would like remember that dynamics in 3-space becomes sta- to get across another recurrent point:nearly tics in 4-space; the Hamilton-Lagrange dynamics everyone viewed present and future computers as of n-degree-of-freedom systems becomes statics exemplary means to enhance the creative talents when time is adjoined as a state-variable. and decision-making skills of the engineer, both in school and in industry.Through computer PRO: Of course, I generally agree with you. You simulation the design engineer (or engineer-in- must think that most of such people are mere training) could rapidly acquire these skills,ex- Johnny-come-latelies who only associate Lagrange plore a wider range of alternatives, and substan- with multipliers and Hamilton with quadternions. tially optimize a tentative or final design. But another important point was raised :that there is an analogy between the man-computer CON: I am always bothered by this word "creative" and the heuristic-algorithm situation.In particu- which keeps cropping up at all meetings on design. lar, Licklider set design in the context of problem- If we are being honest about designas it has solving and in turn emphasized that problem- actually been practiced, why attach such a pre- solving ultimately demands successful formula- mium on creativity? Or, similarly, on novelty in tion :the asking of the right questions. research?Shouldn't the goal of honest research effort be better understanding, at least for the CON: Whoa there!Don't credit the problem-solv- researcher, if not for the world? ing or heuristic approach to Lick or even to Mar- Are not then old things and old ideas valuable? vin Minsky, for that matter.I seem to recall Just because the simple lever was invented (or that it was the mathematician George Polya, pre- better, just thoroughly understood) by Archi- sent elder statesman of analysis, who made the medes more than 2,000 years ago makes it no word "heuristic" fashionable.If the computer- less valuable or so much easier to improveupon bugs find his ideas useful, so much the better for (perhaps we could make it 200% efficient!). My them! guess is that when Licklider's on-line community But his reminds us of our recurrent fear :the gets going there will justly be a number of voices possible adverse effect of computers on analytical from the grave! skills.Aren't we always likely to fall into the Aren't all you people pussy-footing around the trap :"Why think, if we can compute?" legitimate meaning of creative engineering and design ? Wouldn't it be more accurate tosay that PRO: But you know yourself that computers are a prime task of the engineer is to produce (or at no more at fault on this count than logarithms least help to produce) acceptable, efficient, and or nomograms. There is no apparent reason why competitive designs, whether aided by computer the mere existence of one tool should necessarily or not.Novelty enters the picture only because stultify another. Moreover, any artful use of com- the task changes and new needs and possibilities puters should enhance the analytical gifts of the present themselves. But no single engineer should user. take unto himself too much credit for that! CON :Well, I agree with you in part.But I'll It seems to me that the distinction whichsome use your point to make another. You talk as if people make between ordinary engineering and everyone at the conference thought always of creative engineering is that which I would make computers as large, general-purpose, digitalcom- between bad engineering and engineering. puters... And another point, if you don't mindmy shout- ing you down. You let slip with that nastyword PRO:...except for Jim Reswick. "optimization."Shouldn't the rule of parsimony 107 Ockham's razoralso apply occasionally? Why PRO: Nothing I heardor saw would cause me to shouldn't we search more often for minimal! declare war yet!Both industrial and academic models :models which lead us to the correct people foresee benefits far outweighing liabilities. decisions most efficiently, with the least (and not I'd like to summarizesome thoughts of mine the most!) degree of elaboration.Surely, if I which were certainly sharpened by this meeting. granted you a sum of money, you wouldn't in- (1) In the absence of clear understandingof stinctively want to spend it as quickly as possible? collective, organic design procedures, theexperi- PRO: I'm not so sure about that!If I were too ments in design on campuses yield at leastas sig- miserly you would cut my budget nextyear ! nificant informationas industrial counterparts. Which reminds me :it was the question of bud- Typical school design projectsare smaller, direct- geting for computer needs which sentme off to ly involve far fewer people, andare much more the conference.It strikes me that nobody at- closely monitored (contrary to belief insome tempted to come to grips with the question of quarters). computer costs in academic budgets for the future. (2) Computers have been usedon campuses for The speakers and audience seemed still basking in formal scientific purposes far longer thanin in- a honeymoon glow. Nobody seems to have men- dustry.By natural evolution, then, computer- tioned figures like the oneswe have come up with : aided, rationally analyzed studentdesigns have $100 to $400 per year of computer cost forevery become more commonplace than is generallyrea- Tom, Dick and Harry at XY Tech. lized. (3) Schools have, generally speaking, CON: Before we get back to that item, I would like done to summarize our thoughts before much better than industry in maintaininga you left for the balance between analog and digital hardware and meeting and then ask you ifyou have changed utilization. your opinion regarding any of these. But, with growing appreciation of the singular benefits of hybrid computation,a We had generally agreed that the primaryre- new rebirth of significant school-based hardware sponsibilities of the school lie in theareas of teach- activity is a likely event. ing and research. Relative to engineeringdesign, (4) Schools are woefully remiss in laboratory this implies that wemay consider the teaching instruments and control relative to counterpart of design and the design of teachingmethods, industry. likewise research in design and design of But campuses will soonsee the start re- of revolutionary developments in on-linecom- search procedures and experiments. puter control of research experiments. As to computerswe felt that, directly related to (5) Clear patterns have been established for computer-aided design, are computer-aidedteach- successful on-line, computer-aided teaching and ing, and computer-aided research, includingex- the design and use of upgraded, open-ended teach- periments in fully automated experimentation: ing machines.Only the overwhelming costs of machines to produce fundamental scientific know- utilization, followed by more modest requirements ledge without direct human intervention! of faculty indoctrination, presently limit theim- What are the objectives in all this? Simplythe pact of such developments.Remove these bar- goals of every university:the education of stu- riers and some of Licklider's most colorfulpre- dents, researchers, and faculty.Particularly we dictions may yet become understatements. felt it important to stress thata school must help (6) Finally, via computers, several aspects of each student, as soon as he showsup at the door, the real engineering worldmay be brought into to cultivate two attitudes (withor without the the classroom, exposing students to situations assistance of a computer!): which would otherwise be impossible to experi- 1. A strong, self-confident, personalstyle ence. In particular, reasonable, flexible, articulate and deliberative purposefulresponse:in models of large and sophisticated systems (for all, a sense of unity and integrity; and transportation, power, etc.)can be made readily 2. A lasting awareness and speculativecuri- available. osity:in sum, a life-long zest for learning CON :I could comment endlesslyon this but we and accomplishment. both have classes coming up, so let's get back to I think both of us felt that, shouldcomputers work ! ever interfere with these objectives, thenyou and I, at least, must set about to developsomething FENVES: One of your pointswas that there is a better and more wholesome whichcan ultimately kind of osmosis from researchuse of computers supplant them. What doyou say now to these to teaching use and computer-aided design teach- lofty ideals? ing. Are you sure that there is sucha thing, or is 108 this just wishful thinking? My experience has ground-background where you either have it avail- been that there is something the psychologists able or have it planned so that, in effect, you can call a "logic block." We get bright young Ph.D.'s run a batch from the console and this allows you who have used the computer for their research to use the time-sharing console as a monitor in work and who continue using the computer exclu- control for a low priority batch running.So, it sively for the analytical research work, but the seems to me that almost the converse is true. The thought of using the computer in classrooms !tut ideal way to learn programming, if you can pos- doesn't seem to occur to them.Would you like sibly afford it, is in real-time time-shared mode. to comment? MILLER: The trend is very much in the direction PAYNTER: We really have nothing to do with that of answering that question.It's not a question process once you put the tools in their hands. of how large you have to be, or how much use Once every freshman at a universitywhich isa you have to have, before you start time-sharing liberalarts schoolhas accesstocomputing in the sense of multiple access.The newer sys- methods, I think you can leave it safely up to tems coming out now for computer-aided instruc- him to exploit these rather fully.There's an- tion right down to the high school level consist other aspect of the research, and that is the really of a very small computer with a number of type- close interplay between the research-experimen- writers hung on it.One doesn't have to think in tation, instrumentation-type use of computers terms of a giant, or even a medium-sized com- and the direct use in teaching, and here again I puter, to have multiple access. think a lot depends upon how strong analog indoctrination is on a university campus.If re- SHANNON: One comment along thisline.At searchers, both students and faculty, are familiar Dartmouth, I'm a big user and I've gotten shoved with instrumentation techniques in general, then off because these thousand little users have come they're pretty capable at building bridges. in. One of my students said, "I'll fix you up," and I think by next week we will be able to come in SMITH: In relationshiptobatch-processing vs. the background mode and run automatically time-sharing, about how much use should be chained programs in background.I'll be able to being made of a computer before it becomeseco- put in a program which occupies ten core loads nomical to use the time-sharing technique? of 6K each, and swap back and forth, and just PAYNTER: Your option may be partly removed by come on at night and run.But I do have that the manufacturers themselves, because within functional capability within the system and this two or three years the equipment at your campus is what's going to happen in the next generation is bound to permit some modicum of time-sharing of time-sharing systems; you will be able to anyway and the interface here is getting less and develop the programs in foreground, stick them less distinct.Most time-sharing systems now in the background, give them a low priority and have something equivalenttoso-calledfore- say, "Send me the answers in the morning."

109 A LANGUAGE PROCESSOR FOR INTRODUCTORY ENGINEERING COMPUTATION RICHARD W. CONWAY Cornell University Ithaca, N. Y. The use of computers in engineering design is mation, and requires sophisticated systems for clas- already impressive and the prospects along this line, sifying, linking and searching these files.In other with the advent of remote-access and graphical com- applications, the computer is made to implement a munication ,devices, are very exciting. The rapidity special algebra, manipulating symbols according to with which the importance of this technology has operators that have nothing to do with familiar been recognized by engineering education so that arithmetic processes. This is particularly important it has been given a place in already hard-pressed to the engineer in applications involving simulation curricula is equally impressive.However, I rather of dynamic systems or the automation of the design doubt that this time is being as well used as it might process.However, current instruction is concen- be.I think that some of the time and effort now trated almost entirely on the calculator view of this being spent on programming instruction could be many-faced machine.And even this is not being better used to give a broader introduction to com- don very well.It is losing sight of the key points puter science, and that this could be done without in a letho2a of technical detail and taking too much reducing the students' programming facility simply time in th.process. by employing a more appropriate language and pro- Probay the fundamental concepts in this calcu- cessing system than is typically used. lation are the assignment of value to variables and In principle, these questions are really quite inde- the conditional control of sequence.These are, of pendent.If present instruction in programming course, present in FORTRAN and can be properly can be improved, that is presumably worthwhile.It underscored by a knowledgeable instructor, but does not immediately follow that any time that is there is a tendency for them to be obscured by the gained should be invested in a broader exposure to mechanics of the process.For example, the dis- computer science.In practice, it is much easier tinction between fixed- and floating-point numbers to shift emphasis within a course than to change contributes nothing at all to the understanding of the time allocated to a general area. the basic process, nor does it add to the students' The question of what material in computer science problem-solving ability. It is strictly a question of should be provided to all engineering students is, of machine convenience and efficiency, yet time is spent course, the more important and more difficult prob- defining and explaining these forms and conveying lem, and I treat it briefly in the following only be- the timeless truth that integer-valued variables must cause I am less confident of my answers.I believe have identifiers beginning with I, J, K, etc. that the computer field is in real danger of becoming There is no question that students can learn and as much involved in the peculiarities of current use FORTRAN, or more typically a subset of technology as engineering in general was some fif- FORTRAN, in the time allocated in introductory teen years ago.Especially in this rapidly changing courses.However, there is also no question that field there is real necessity to identify the basic comparable power andunderstandingcanbe concepts concerning information and its manipula- achieved by the same students in a fraction of the tion, as opposed to the details of a particular machine time by the use of a language specifically designed and a particular language.The obsolescence of for this purpose.This has been proven in practice these details is incredibly rapid and complete. For with a language called CORC at Cornell since 1962, example, estimate how much of the material of a and more recently with BASIC at Dartmouth. There typical computer course of the mid-fifties involving are two entirely different computing worlds.One the IBM 650 is still relevant today. is the world of the experienced programmer who is It is, of course, far easier to point out what is not concerned with compactness of expression, efficiency fundamental than to state what is, but as a starting of execution, and to whom the many rules and con- point I think that it is necessary to come to regard ventions of a production language become second the computer as a more general device than a calcu- nature through continuous use.The other world lator.A student does not really understand the is that of the neophyte and intermittent programmer character of the computer until he appreciates that, who is struggling to learn or to remember the essen- in some applications, the machine is really a device tial form of the language and who will sacrifice for storing and retrieving large volumes of infor- much convenience in favor of simplicity. With such 110 different circumstances and objectives, it would in- time limits), a dump of variables and final values deed be surprising if the optimal design of a com- is automatically provided and a tabulation of the puting language turned out to be FORTRAN in both frequency with which each label was encountered cases. during execution is given.This latter has turned Similarly, the translators and monitors that are out to be invaluable diagnostic information. typically used for introductory instruction were not This processor was constructed with the assump- developed for that purpose and do not do all that tion that this would take more machine time, but they could for the student.The CORC processor that this was a price worth paying for more ef- can illustrate some of the services that canbe pro- fective assistance in instruction.It turned out that vided, but it does not exhaust the possibilities by any there was no price to pay.By recognizing that means. CORC begins by scanning the sourcefor the programs for which this was intended would syntactic errors, probably more exhaustively than only require a small fraction of the 32,000-word any other such processor. Forexample, by examin- storage of the CDC 1604, we made the compiler ing context and usage, CORC will often conclude permanently reside in core, (and suitably protected that an identifier is actaally a misspelling of another it from errant student programs). Since the setup identifier and does not represent a distinct vari- time for each program in FORTRAN under the able.The important innovation is that CORC not CDC 1604 Monitor is 25 seconds, this meant that only discovers errors and reports them to the pro- our average time of about twoseconds for com- grammerit attempts to correct them.For ex- pilation and execution of these programs compared ample, it performs spelling correction by looking rather favorably with the 26 to 27 seconds under for transposition of adjacent characters, omission FORTRAN. (Card-to-tapeandtape-to-printer of terminal characters, concatenation errors, key- operations are handled by a satellite 160A.)In ad- punch shift mistakes and confusion between zero dition, the number of machine approaches to achieve and the letter "o."It supplies operators, paren- successful execution of a problem was just about theses, labels, reserved words, and even statements halved in comparison to our previous experience in some special cases.In fact, no matter how badly with FORTRAN and Burroughs ALGOL. Finally, the sourcedescription may be written, CORC the system proved sufficiently adept at correcting restores it to a form that is syntactically correct, keypunch and other spelling mistakes so that we generates the object code, and begins execution. The could abandon key-verifying of programs.All of correction is attempted not so much in the hope of this is bonus and I would like to believe that we being able to reconstruct what was intended, as sim- would still be using the CORC system even if it ran ply to produce anexecutable program.There is more slowly than FORTRAN. much diagnostic information that cannot be ob- We are aware that some of the necessity for this tained until execution, so that refusing execution type of processor is eliminated if a student can when errors are found in translation increases the write, test and execute his program in conversa- number of machine approaches that are required tional mode at a terminal of a time-shared computing and lengthens the testing process. To our surprise, system.However, we are convinced that there is the corrections often turn out to be valid and at still a modest future for a well-conceived, batch- least one pass at the machine is eliminated.I would processing system for student programs. It is going estimate that at least one-third of the final programs to be some time before the luxury of time-shared submitted to me have at least one instance where computing is available to all of the institutions that CORC has successfully corrected some minor error must provide this introductory instruction, and, of spelling or punctuation and the student did not even for institutions that have large-scale time- find it necessary to resubmit the corrected program. shared installations, it is not at all clear that they During execution, the monitor communicates with can afford to be entirely terminal oriented forhigh- the programmer in source language terms, referring volume introductory work.I think that a good to identifiers and statement numbers. This is done, multi-programming system such as the one at Case, without recourse to interpretive operation, by re- in conjunction with a CORC type of language and taining the identifiers in the symbol table and by processor, would be a very effective and economical inserting statement numbers in dummy instructions means of handling high-volume instruction.At in the object code.This exacts a minor price in least the fully conversational systems should be re- space and time that aproduction-processing system quired to justify their existence against this alter- would be unwilling to spend, but it is entirely ap- native, rather than against the current typical propriate here.It means that the programmer never batch-processed FORTRAN. requires, or is confronted by, information in a for- The same situation holds with greater strength eign form.Finally, after execution is terminated for computer use other than numerical calculation. (either naturally or by exceeding error count or Languages for non-numerical and list-processing 111 are even less suitable for introductory instruction, 2. R. W. Conway, J. J. Delfausse, W. L. Maxwell andW. E. and have effectively limited this important typeof Walker, "CLP-The Cornell List Processor," Communica- computing to advanced students.In 1963 we ex- tions of the ACM, Vol. 8, No. 4, April 1965. panded CORC in a manner patterned after 3. W. C. Lynch, "Description of a High-Capacity, FastTurn- SIM- around University Computing Center" (Case Institute), SCRIPT to a language called the Cornell ListPro- Communications of the ACM, Vol. 9, No. 2, February cessor.This provides a slightlymore advanced 1966. student with access to a language that isconvenient for such problemsas the simulation of dynamic sys- PASKUSZ: I think you'rea little pessimistic as far tems, game playing and heuristic programming. as the time is concerned when we go to a remote Programs which constituteda master's thesis before console, time-shared operation.I can't really CLP are now given as homework problems inun- see any difference in time-per-student involve- dergraduate courses. ment on a console versus the time-per-student We are currently developingan instructional ver- involvement on, say, a keypunch machine.I sion of PL/I. We have extracteda subset of state- think they would be comparable.There may be ments consisting of assignment, READ,WRITE, a factor of two, or something like that. We have GO TO, IF ALLOCATE and STOPand have adapted about 2,000 engineering studentsnow and a the PERFORM statement from COBOLto serve in couple of requiredcourses which they all take; place of PL/I's DO, CALL and PROCEDUREstate- and we've also got a couple of electivecourses ments.This PERFORM modification makesour which many students take. With the halfa dozen version not quite aproper subset but we believe keypunch machines we have there isn'tany great that the resulting simplification is wellworth the amount of queueing, so I don't think that there departure. The languagegoes beyond PL/I in the will, be a great problem of howmany of these sense that is has fully dynamic storage allocation remote consoles you're going to need ifyou use and direct matrix operations.For example, in them for basic instruction. X = (A+B) *C or IF Y*X = A A...the operands CONWAY: I agree that this will take the can be scalars, vectors or matrices, limited only by same order f magnitude of time. We find thatit's not the usual algebraic rules forconformability.This much more expensive to add Cornell version of PL/I is being implemented a girl to the key- with punch, and that we actuallycan cut down on the a translator and monitor that provideseven more number of punches we would otherwise require ambitious assistance than CORC. This will be at very little extra cost, and I think this isa available for "small" IBM 360s (Model30 or higher, reasonable thing to do for with disc and 64K of main storage) late a student.After all this year. we're not trying to teach him to punch cards;the During 1967 an extended version includinglist-processing and non-numerical capability way our operating system works, it's being done will be re- for the studenton the first pass and he makes cor- leased.Shortly after Cornell installs a time-shared rections. So if there is something aboutgetting 360/67 in September, 1967, the languagewill be one's hands on a keypunch, the students do adapted for remote-terminal operation. experience it for small bits of time.I wish I could be- In summary, I am not trying to sella particular lieve that we could provide enough terminalsat language but rather to make the followingpoints : essentially $1,000 rental apiece to do this forthe 1. It is at least as important for engineering students. undergraduates to understand the fundamental ROSENTHAL: I'd like to make character of the computationalprocess and be a very quick defense able to view the machineas more than a calcu- of FORTRAN as the introductorylanguage. lator, as it is for them to achieve proficiency Namely, you don't haveany textbook available in programming ina particular language for for the other language. What Imean is an avail- a particular machine. able, undergraduate, introductory textbookto in- 2. This broader objective troduce the system and the language.That I can in part be covered think is the primary prerequisite for wideme of during time that could be salvaged fromexist- the language. ing courses by the I agree that you need something use of a language and an other than FORTRAN, but not untilyou produce operating system that has beenspecifically a textbook. designed for the needs of undergraduatein- struction. CONWAY: I'm not really tryingto sell anyone CORC for the simplereason that most of you References: couldn't use it even if I did convinceyou. CORC 1. R. W. Conway and W. L. Maxwell,"CORC-The Cornell was done on a shoestring by some people in the Computing Language," Communications of theACM, Vol. Industrial Engineering Department, andit was 6, No. 6, June 1963. done over the objections of the ComputingCenter 112 .11,01.- =

ismvw141itit'art.34

at Cornell and the Mathematics Department.It CONWAY: I am in complete agreement. My point took a little less than a year before it reached the mainly is that the design problem for this large point where all introductory instruction on the and important segment has been neglected in the campus was done on CORC. Students graduate past;that we have been given a second -hand into FORTRAN and ALGOL as soon as they product, we have had an adaptation of something have some significant computing to do, something that was never intended for this purpose.Cer- that will run into a dozen minutes or so.I would tainly the complete system, that we're designing like to make you unhappy with certain of the now includes a central processor, various remote things that FORTRAN is and does, but I don't devices and a lot of language processors. I would really have anything to offer you in the mean- be willing to conclude simply that itgoing to time, because you couldn't use CORC even if we take more than one language processor to serve did have an those textbooks. a highly disparate group of users adequately and fairly. FENVES: To answer Mr. Rosenthal's comment, with due apologies to Dr. Organick who is sitting ORGANICK: I was very excited about the very in the audience, I 'maintain that there is no ac- beginning of your paper in which you spoke about ceptable FORTRAN textbook available. the three points of view of the computer and I realized you simply didn't have the time to de- MOTARD: Listening to these various comments, I velop the explanation of how you were going to think what we're trying to do here collectively is provide, in an introductory course, these other like designing a, system, and we on the team have two views. I would like to call the audience's at- different attitudes because we haven't really come tention to a very interesting report, Monograph to grips with the question of what is the unit 7, of the Discrete Systems Project. In this report module that we want to use as the unit of com- they attempt to provide their view of how and munication between the man and machine, and what they would do to introduce a student to we have also found out that this is going to vary computer science.They make a rather strong from one user to the next. For instance, the most point, as I'm sure you would, if time permitted, sophisticated users want unlined, time-shared con- that there are at least two equally important intel- soles because they've mastered the artificial lan- lectual processes in the programming of the solu- guages and all the sophisticated techniques and tion of a problem on a computer.The one we they think now that the limitation is accessibility have been guilty of overemphasizing to the ex- and feedback; the neophyte hasn't even gotten to clusion of the other is, "How do you organize the that stage, so we have to take him by the hand steps of the problem that you want solved." The and thus need a different kind of system.It's a second very important one is, "How do you con- red question of whether we are going to be able ceive of the data and its organization, first at the to design the ideal system to satisfy all needs, abstract level and then at the level of machine and I think we can, and we will need to have a representation?" And I would agree with Fenves range of responses according to the units of com- that all of the FORTRAN books written for munication input, the time-scale response that beginners so far neglect the second aspect.I we want, and the degree of analysis that is to be think this is where we have been weaker; in our done, engineering introductory courses so far.

118

1144444,04.1~440wortoviwipmeemolsnmeeprimp44.440.p.f mr.44,14,4s,..04#0 PRECEDING PAGE BLANK- NOT FILMED

NM11110.0111111111I

MANUFACTURERS' FORUM

SECTION E

Round-table presentation by computer manufacturers on products and concepts pertinent to theuse of computers in engineering design.

CHAIRMAN: S. G. CAMPBELL, Assistant Vice President, Xerox Corporation, Rochester, New York

- PRECEDING PAGEBLANK- NOTFILMED AUTOMATED DESIGNWITH HONEYWELLSYSTEMS R. RODERICK BROWN Honeywell, Incorporated Waltham, Mass.

design has 1961 the statistical analysisof component failures The use of computers in engineering design for proliferated rapidly in the last decade.The dom- was placed onthe computer, as was the fabrication of all wiring.In 1962 the utilization inant characteristic of this growthhas been the applications, such as increasing number of different applicationsto which of computers for production inventory control and themaintenance of service computers have been put.Electrical apparatus, proportions, and in communications and power systems, structures, con- records, reached significant 1963 and 1964 these areas werelinked to the engi- trol systems, and other topics discussedat this con- had also been interference, have all evolved in practical,computerized neering applications, which connected by then. During thistime, the use of the form only recently. solving local prob- With but few exceptions, each ofthese applica- computer by individual engineers "single-minded- lems amenable to formulationin FORTRAN also tions shares a common attribute : increased, primarily in areas whichhad not been ness."That is, each program has beendesigned to prob- previously exposed to computeraids.In pardllel solve but one aspect of the overall engineering of the company lem.For example, the electrical-apparatusdesign with these activities, other areas performance of the were continuingthe development of employee,finan- program might calculate the As would device, but would provide no directly usableInfor- cial, and marketing-programsystems. primary be expected, the latter did notinfluence the engineer mation on its fabrication. In each case, the for a share objective has been to solve the outstandingproblem: directly except through the competition of the limited computer timeavailable. to reduce the burdensome calculation orimprove from that of other the accuracy (or speed, or cost) inaccomplishing a From this experience, and large r'imber of repetitious operations. users of computers, we candraw several significant with conclusions. Most apparent is theevolutionary ta- As each engineering discipline fortifies itself computer to design. this essential tool, a trend is emergingtowards a ture of the application of the Within an area of an engineeringdiscipline, the broader view of the application of the computer. personal, This view, which may be termed the "systems" ap- first contact with the computer is a very individual affair, concerned onlywith a very lim- proach, recognizes the role which the computer can The initial display in the conduct of an industry's business.The ited but highly irksome problem. concepts of the "total systems approach" and "inte- course is very awkward,partly because the problem vigor- must be translated into alanguage which already grated management systems" are now being situation, and ously pursued in the areas of financial and produc- exists but is not truly suited to the outstanding partly because the engineer must expresshis prob- tion control, where solutions to the In either problems have been achieved, and the integration lem to another person (the programmer). of single-minded programs into a cohesive structure case, the engineerwill suffer some initial frustra- These same concepts are tion because he will not have expressedhis problem is being implemented. the require- now entering the engineeringdomain the future with sufficient preciseness to satisfy ments of a computer.With time and patience, will see their extension across the organizational In marketing,engineering,and however, he will be rewarded with an answer. barriersbetween single- production. this way, the programs which I have called One of the more prominent examples of the real- minded are created.If thciy are successful, they ity and practicality of this approach exists in the will be used over and over again, and willbecome a area of the engineeringdesign of digital computer working tool in the design organization's reper- systems. An example taken from Honeywell's ex- toire. perience as a manufacturer will serve to illustrate When there are several such programsin exist- this trend in action.Prior to 1958, the application ence in an area,the thought comes to mind that of computers to aid in their design was sporadic they should be integrated into a "totalsystem." and rudimentary.At that time, two applications This is economically feasible onlywhen approached were developed, one for logicdesign analysis, and in the proper manner. The term "linked" has been one for the design of wiring suitablefor fabrication used above to describe the evolution of some phases by a numerically controlled machine tool.In 1960 of design at Honeywell. By this is meant a joining these two applications were linked together and a together, as two islands might be joined by a bridge. materials-explosion application was initiated.In The analogy is apt, for the traffic (data flow) across 117 these links is usually aperiodic,independent of the he must communicate his results existence of the link, and usually to others and in restricted to a turn receive their critique of hisdesign.At this small portion of the population.This is particularly stage, the computer isa tool, off in a corner out of true when different engineeringareas (such as elec- sight, out of mind. With the linking tricaldesign and production and integra- engineering)are tion of the programs used by thecompany, the com- linked, for each needs onlya distillation of the data puter now becomesan integral part of his working required by the other. To providemore would be environment.It is used daily, aiding in the uneconomical'(higher accom- transmissionorstorage plishment of design and automaticallyproviding the costs) as well as chaotic (notethat the link pro- communication links to other vides information, rather areas.If he was a than data, which is quite good designer before, henow has less such work to different).These links are economically justifiable do; if he was mediocre before,he must do more, but because the information capturedin one area is the quality of his results is increased. made available to the other This latter without requiringa results not because thecomputer can necessarily man-machine interface with all ofits attendant produce good designs, but because problems. it can evaluate their quality and report to the chiefdesigner more Within a given area of engineeringdesign, the than was heretofore possible. Thus links should be much stronger, the engineering and may evenap- designer finds himself relyingon the computer in proach full integration, whereindata flows are fre- his work as muchas he does on his car to get him quent, automatic, and prolific.Such integration is to work. He could get along necessary when large and very complex designs without it, but would are rather not.The nature and quality of hiswork is produced by biggroups of people working together. known better by others, and Here data once captured is of recognition and reward utility to all con- based on more knowledgeableevaluation is possible. cerned and, indeed, the computerperforms a true service by providinga uniform representation of it One further point must bemade.It is the con- to all who must work with it.When these single- tention of this writer thatthe evolutionary nature minded programsare integrated, we also see the of the designprocess using computers and the major emphasison the development of a problem- changing environment of theengineer representa oriented language forcommunicating with them. pattern which we shallsee repeated many times in The objective of sucha development is clearly to the future.Each major new idea ina discipline, remove the frustrations which plaguedthe early each technological breakthrough,will start theo,n- development of theprograms, and which linger on ; tire processanew. The transitional phaseswe have and to improve the efficiencyof describing problem described will bemore easily and quickly accom- statemnnts by increasing precisenessand concise- plished, but they will be there. ness and eliminating the unessentials. What does this allmean to the colleges and uni- Another major conclusionwhich can be drawn versities which teach from our experience is the our engineers and to Honey- direction in which the well as a manufacturer ofcomputing systems? engineer's working environmentis changing.Ini- Each in our tially, the designer using own way must be prepared to cope computers does soon an with a wide diversity of problemsituations. Versa- occasional basis.Only a part of his time isspent tility, adaptability, and getting a solution to that special high capabilities must be problem, while the prime attributes ofour products, both men and others less difficultare done by hand; and, of course, machines.

118 AUTOMATED TECHNICAL DOCUMENTATION SYSTEMS AND ENGINEERING DESIGN PAUL H. ROSENTHAL Univac Division Sperry Rand Corporation El Segundo, Calif.

The current pace of technological development is a high-speed printer at the remote side.Extensions causing an information explosion that is severely to the current system will, of course, incorporate taxing current documentation methodologies.The remote typewriters for input and output.The automated man-machine, engineering-design sys- documentation systems runs on the UNIVAC 1108 tems being discussed at this conference will not Computer,' which is a large-acale multi-processing reach the performance levels hoped for unless the system designed for computer utility operation. documentation bottleneck is broken.This paper will therefore outline some of the automated-docu- ME= ISMIDEL EMS mentation system work being performed at the Univac Division of Sperry Rand Corporation as a Text 8 I portion of the total effort in the engineering design Commands II I I area. *Corrections The hardware and software systems currently be- INPUT ing marketed and installed by the Univac Division DATA will allow the full facilities of a computing center UNIVAC 1108 PROCESSZ to be utilized from the desk of the engineer or exec- Editing MASTER FILES utive. A limited automated-documentation system *Correcting is currently in pilot test for those hardware systems Ccespoition which will process the mass of documentation produced by our development groups. Later on, extensions of this system will give it the capabilities needed by complete man-machine design systems. Text Figure 1 outlines the capabilities of this initial doc- Comend PROOF IN/ LISTING umentation system.Current information-systems Index methodologies are used by the system to automate To TYPEIETTINO software and hardware equipment manuals. These MACHINES manuals are subject to constant revision and are distributed widely both inside and outside of the company. Figure 2. Figure 3 defines the two types of input data to the UNIVAC AUTOMATED documentation system.Commands are separated DOCUMENTATION SYSTEM from text by a starting and ending command-delim- iter code that is specially assigned to each type of UNADS input unit.In the examples that follow, left and *Maintains Documents on Magnetic Tape Files right brackets will be used to represent these de- limiters. *Inputs Text and Corrections *Outputs Proof Listings and Final Documents INPUT DATA TYPES TEXT *Performs Line and Page Composition *Information to be printed on Final Document Figure 1. COMMANDS *Control Composition *Control Correcting Figure 2 is a schematic of the flow of information *Information needed for UNADS Program in the computer system. Current input is via paper Figure 3. tape or cards, usually from a remote site. The primary output is aproof listing containing the text of 1 UNIVAC is a registered trademark of the Sperry Rand the document and an index.It is now produced on Corporation. 119 Proof Listing ems) vertically.Typical action commands include Text those for pagination, graphics, tabs, quadding, and Index Commands flushing. THE TEXT MATERIAL 241 0 PAR Editing commands control the correction and ISLISTED, COMPOSED 242 merging of master files with data from the input AND HYPHENATED AS 243 stream. The 'ADD' command illustrated will insert IT WILL APPEAR ON THE 244 after word 4, line 3, section 27 of the current master FINAL DOCUMENT. 245 file the word "with." Typical editing commands in- COMMANDS ARE EX- 246 0 PAR clude the 'COPY' command, and delete andrevise. TRACTED AND SHOWN 247 A special command, 'REPLACE,' will searchthe 24 master file for a phrase of text and replace it with IN RIGHT COLUMN. 8 another phrase.This allows selected text to be Figure 4. tailored for a product subsystem by changing each occurrence of specialized specifications or names. Figure 4 illustrates a proof listing.Since most Information commands supply parametersto the remote- terminals do not have upper/lower-case UNADS program and do not haveany visual effect printers on-line, we slash the characters to represent on the composed output document. The illustrated an upper-case character.Except for justification, `In Library Tape' (INLIB) command containsthe the proof listing simulates the final document as information needed by theprogram for checking the closely as possible. The index shown is used for cor- label of the magnetic tape being usedas input.In- recting and editing the text and commands shown. formation commands specify parameterssuch as Correcting can be done at the section level (illustra- change letter, end of data, and input/outputma- tion at 24), at the line level (illustrated at 1 through chines being used. 8), or at the word level. The commands shownare preceded by the word number which they follow. COMMAND STRUCTURE They are removed from the text and rsted to the right to improve the proof-listings simulation ofa USER DEFINED COMMAND final document. *DEFINING The data stream through the UNADS systems [PAR =LINFED; TAB] [CHAP= consists of data and commands. Figure 5 lists the SECTN; PAGE ; RESTOR; LIN- four types of commands and an example of each. FED, 6; HA; QUAD, CENTER] COMMAND STRUCTURE *USAGE When Starting Paragraph BASIC COMMANDS [PAR] The... When Starting Chapter [CHAP] Intro- *FORMAT [INDENT, LEFT 3] duction... *ACTION [LINFED 3] Figure 6. *EDITING [ADD, AFTER 27-3-4, "WITH "] *INFORMATION [INLIB,2457, "PERT MAN- UAL REV. 1C"] The entry of the format andaction commands which control compositioncan be very redundant. Long streams of commandsare required to define Figure 5. the start of chapters,sections, and paragraphs. Each type of document will requireextensive set-up Format commands modify the composition of text for page formats.The redundancy problemis from the point of entry until theyare modified. solved by allowing theuser to define his own com- The 'INDENT' command shown will move the left mands. Figure 6 illustratesthe defining andusage Margin 3 spaces (in printing nomenclature 3em of user-defined commands. spaces) to the right. At the start ofa docu- mypical format commands in- ment all of the standarddefinitions to be used in clude one each for page size,a margin set, line spac- the document are entered,thereby substantiallyre- ing, a column set, anda tab set. ducing the typing requirements. 4") Action commands modify compositionatheir Figures 7 and 8 illustratethe contents of point of entry but haveno lasting effect. some The of the composition andediting commands.The iT 1,IFED' command illustrated willterminate the complete set of commands include current line being composed and the capability for move the paper directing most of the machineswhich are used for or film 3 spaces (in printing nomenclature lead3 the production of printedmatter. 120 SAMPLE COMPOSITION COMMANDS A command library entrycan be entered into the PAGSIZ, WIDE 6, HIGH 11 input stream simply by giving itsname within de- limiters. IDFORM, TOP .5, FACING Brass width tables give the size of the characters in each of the various type fontsm ail- GRAFIC, HIGH 3.5, WIDE 3.5, LEFT, TOP able on the typesetting machines.An input and QUAD, CENTER output processor exists for each kind of standard FLUSH, ".". input-outputequipmentavailabletoindividual users.They translate specialized external formats Figure 7. to the standard format used by thecomputer in- ternally and stored on master files. SAMPLE EDITING COMMANDS The proof listings and control tapes forprinting COPY, 2457, FROM START THRU 121-48 final documents are producedon high-speed printers REVISE, 21-7-5 WITH "REGISTER" and paper-tape punch located at theremote site. REPLACE, "DOW JONES" WITH "A.M.C." Figure 10 is a photograph of theUNIVAC 1108 ADD, AFTER 120-38-4, "RESULTING," computer used by the UNADS system.It is a very DELETE, 120-42-7 high-speed, large-scale system currentlyin use with remote terminals.The UNADS system is being Figure 8. coded and tested Li Los Angeles througha remote link with our department's UNIVAC1108 computer A diagram of the hardware configuration used by in Minnesota. the UNADS program is shown in Figure 9.The UNIVAC 1108 processor will Tun the program in time-sharing and multi-processing modes.Input data normally is received from a remote paper-tape reader. The input and output master files are handled by tape units at the computer site.

Input Text and Commends

Magnetic Tape rives

Command Libraries

Brass Width Tables Input Master

Input Pr.-Processors Files UNIVAC Output Post-Processors 1108

PROCESSOR

Figure 10.

Output Master Files The UNADS system is being developed in three Proof Listing Control Tapes phases.Phase I is designed for the needs ofour For Typesetters own documentation problem.It will be used to High Speed Printer produce the documentation forour hardware and software systems.Phase II incorporates a wide range of pre- and post-processors for usage bya Figure 9. wide variety of computerusers. Phase III incorporates remote correcting and editing by theconsole The mass storage drums on site contain pieces of of master files contained in d,ta banks. This phase the UNADS program called by individual jobs. meets all the requirements for documentationsup- Command libraries are groups of user-defined com- port of automated engineering-design systems.Fig - -Mands defining a specific document type and typing u. 11, 12 and 13 list the features of these phases format. and demonstrate the eventual scope of the project. 121 FEATURES UNADS I FEATURES UNADS III * SINGLE COLUMN PAGE COMPOSITION * REMOTE EDITING * TABULAR MATERIAL COMPOSITION * AUTOMATIC TABLE OF CONTENTS * MULTI-COLUMN PAGECOMPOSITION * MAGNETIC TAPE MASTER FILES * AUTOMATIC INDEXINGTHROUGH * LINE JUSTIFICATION WITHOUT HYPHE- THESAURUS NATION * MASTER FILES ON DATABANK or * INPUTS PAPER TAPE, CARDS MAGNETIC TAPE * OUTPUTS TYPEWRITERS Figure 18. Figure 11. The Univac Division of SperryRand Corporation FEATURES UNADS II will probably use UNADSPhase I for approximately one year, gaining experience * GRAMMATICAL HYPHENATIONALGO- and modifying RITHM the system for optimum productiveperformance, * LINECASTER POST-PROCESSOR before UNADS Phase II is releasedto the field. We * PHOTO-TYPESETTING POST-PROCES- are just now starting our training ofpersonnel for SORS usage of Phase I and hope to be in productionin the * AUTOMATIC FOOTNOTING near future.We hope that the UNADSproject and others of its kind will appreciably * AUTOMATIC PARAGRAPHNUMBERING reduce the cost- and time-scales of documentationcurrently Figure 12. caused by the information explosion. COMPUTER CONTROL COMPANY PRODUCTS AND THE ENGINEERING DESIGN ENVIRONMENT R. A. COWAN Computer Control Company, Inc. Framingham, Mass. The use of computers as an aid in engineering de- used, which is extremely flexible and allows ready sign is a step toward the development of an inte- expansion of the I/O capabilities.The basic soft- grated complex of men and machines. The purpose ware package includes the DAP assembly language of this complex is to use the creative and imagina- program, FORTRAN IV (ASA FORTRAN Stand- tive powers of the man and the analytical andcom- ard), and executive routines.Also included are a putational powers of the computer to efficiently selection of utility programs for aidi ig thepro- carry out the iterative process of creative design. grammer in debugging, a complete I/O library and The key to the use of computers in the designproc- a mathematical library, ess is communication between the designer and the computer. DDP-124 To fulfill the needs of the engineering design ap- The DDP-124 is a fast, general-purpose digital plication, the computer should havea comprehensive computer featuring p:PAC monolithic, integrated- instruction repertoire and flexible high-speed I/O circuit construction. The standard mainframe in- (Input/Output).It must be easily expanded to cludes 8K of memory with a 1.75 microsecondmem- provide for increases in the requirements of the ory cycle, hardware index register, fast multiply/ engineering designer.It is especially important divide hardware, electric input/output typewriter, that it have graphical I/O capabilities suchas CRT 300 cps paper-tape reader and 110 cps paper-tape displays or plotters.It should also possess hard- punch.It is fully program-compatible with the ware interrupts for use by the designer in communi- DDP-224.Modular design and broad I/O options cating with the computer and for providing efficient enable the DDP-124 to be tailored toa variety of I/O communication with the peripheral devices. applications on or off line.The standard main- The software associated with the engineering-de- frame is competitively priced at $65,000. sign computer system should be designed to sim- The flexible 124 command repertoire works with plify the process of stating the problem and should sign-magnitude arithmetic and contains 48 instruc- minimize the tasks of set-up and modification of tions.It includes, for example, logical commands, design parameters.It is desirable that the com- an executive command and a step-multiple preci- puter should be valuable to the neophyteprogram- sion command for very simple and fast multiple- mer as well as the expert. For this purpose, higher precision routines.The jump-storage and jump- level languages such as FORTRANare necessary return commands provideeasy subroutine linkage to simplify the programming process. A super- and nesting of subroutines.Also provided are a visory program should be provided, structured in variety of conditional program-transfer commands. such a way that either modification,or addition, to The combinations of input/output capabilities the design programs is possible ina simple fashion. that are provided are specifically adapted tothose In this way, the designercan build upon previous applications which require fast input/outputdata programming and designing experience and expand transfer and easy adaptation of the computerto his capabilities. allow efficient operation inmany different system Computer Control Company (3C) presentlypro- configurations.For this, the DDP-124 providesa duces three digital computers whichare applicable variety of input/output channels, four different to the engineering-design field.They are the DDP- input/output modes of operation,easy modular ex- 124 and 224, both 24-bit computers, and theDDP pansion of its I/O capabilities andhigh data-trans- 116, a 16-bit computer.These computers are all fer rates. Data may be transferred directlyeither modularly designed, high -speed computerswith into or out of the main arithmeticregister or any parallel organization. They havea variety of I/O memory location with a single command. Transfer capability and are easily expandable toincrease the of character inputsor outputs into or out of the capacity of the computer. Arange of peripheral arithmetic registercan be controlled by a mask equipment is offered with them, plusmany internal included with the instructions.Any combination options.As standard equipment, they includein- of up to 6 bits of the characterdata can be read direct addressing and indexing capabilities andpri- or generated to allow flexible formatting. ority interrupt schemes. Input/ The I/O bus system is output rates of up to 285,00024-bit words persec- 123 and are possible using the standard input/output DDP-224. Some typical execution times are ADD commands. 3.8 psec., MULTIPLY 6.5 iksec., and FLOATING- Any input/output channel can be operated in POINT MULTIPY 10.3 Asec. either Ready Mode or Interrupt Mode.In the Input/output rates of up to 325,000 words per Ready Mode, the computer program tests periodic- second can be handled in the DDP-224 by using the ally for channel-ready signals.If the test is suc- fill-memory block instruction. By using the Direct cessful, the program will transfer to the proper Memory Access option, transfer rates of 285,000 subroutine ; if not, the program carries on with the words per second are possible in the normal mode current routine.Simultaneous testing of up to 12 of operation, or 525,000 words per second in the I/O channel-ready signals is possible.In the In- hog mode of transfer.Data transfer rates of up terrupt Mode, the computer program is automati- to 525,000 words per second are possible using the cally interrupted and a program jump takes place to fully buffered channel option.The FBC permits the memory location directly related to the interrupt fully buffered data transfers to occur simultane- concerned. On completion of the interrupt subrou- ously with program execution whenever a fully buf- tine, the main program resumes at the interrupted fered channel addresses one memory module while point.Single bits of input or output information the control unit of the computer addresses another may be transmitted via the sense lines or output- module. control signals respectively. Special options make it possible to combine sev- The Direct Memory Access is an optional input/ eral DDP-224's into integrated, large-scale com- output mode for direct access into or out ofsuc- puter systems with functionally common and/or cessive memory locations, independent of thecom- private memory, control, arithmetic, system in- puter program after being initially setup. The put/outputfacilitiesand peripheral equipment. DMA channels have priority over the control unit These options are the fully buffered channel, the of the computer for memoryaccess, and will steal access - distribution unit and the time-multiplex unit. cycles from the processing of the computer at rates The access-distributionunitpermits processors determined by the devices being serviced.Data and/or fully buffered channels to share a memory rates of up to 570,000 wordsper second are pos- complex.It also contains the logic and priority sible. system to the common memory and the distributing system for data flow to and from the processors The standard 8,192-word corememory is readily and common memory.The time-multiplex unit expandable to 32,768 direct-addressable words, 64 permits multiple processors to communicate with sense lines and 64 output-control pulse lines.Also available are character input/output channels. a common set of peripheral devices. ThinP.116 DDP-224 The DDP-116 is a 16-bit general-purpose com- The DDP-224 is a high-speed, 24-bit, general- puter with a 1.7 microsecond memory cycle time purpose computer constructed with S-PAC digital- and a standard 4,096-word core memory.It has logic modules. The DDP-224 includes all those fea- a fully parallel machine organization, uses two's tures which are available on the DDP-124, plus complement arithmetic and includes both indexing those features described below. A standard main- and multilevel indirect addressing.Standard fea- frame includes 4K ofmemory with a 1.9 microsec- tures include a flexible instruction repertoire of 61 ond cycle time, three hardware index registers, commands, a powerful I/O bus structure, a stand- floating-point arithmetics and input/output type- ard interrupt line and a keyboard and paper-tape writer, and is priced at $96,000. I/O unit.The DDP-116 is priced at $28,600. The DDP-224 instruction repertoire includesmore The command structure of the DDP-116 is un- than 70 commands. In addition to those commands usually powerful for a machine of its size.In- included in the 124 repertoire, it includes absolute.. cluded are a three-way branch instruction, an in- value add and subtract commands, normalize and crement memory and skip on zero, plus a full set scale instructions and the repeat command which of arithmetic logical and rotational shifts.The allows table-look-up, block load andmany ether basic. input /output commands test the status of powerfuloperations. For data transfer from T/0 devices before transfer.If the test is not memory at very high rates, the dump-memory block successful, the next instruction is executed.If the and the fill-memory block commandsare available. test is successful, the data is transferred and the They allow asynchronous inputor output of data next instruction is skipped.Also available are from a block of successivememory locations con- optional hardware multiply /divide commands. Some trolled by ready signals from the external system. typical execution times are: ADD 3.4 microseconds, Also available is a comprehensive set of floating- MULTIPLY 9.5 microseconds, DIVIDE 17.9 micro- point instructions as standard equipmenton the seconds, interchange memory and ADD 5.1 micro- 124 seconds and increment replace memory and skip standards and conventions. Thesame high degree 5.1 microseconds.Most non-memory referencein- of quality that is in the structions are carried out Computer Control hardware in one memory cycle. is provided to theuser in the software. The standard DDP-116 All soft- core memory is option- ware programs for the 124 and 224are fully com- ally expandable to32,768 words.The memory patible. The DDP-116assembler and compiler have is divided intosectors of 512 words eachand a sec- been specially designed tor flag is used to control to eliminate theconcern which sector is addressed. of the programmer formemory sector boundaries' When the sector flag isa one, the address portion The sector boundaries of the instruction are accounted for by the refers to thesame sector as that compiler or the assembler,and the proper inter- addressed by theprogram counter. Whena sector sector linkages establishedby the desectorizing flag is azero, the address portion of theinstruc- loader. tion refers to sectorzero.Thus, 1,024 words can A FORTRAN IVcompiler is provided whichcom- be addressed bya single-word instruction.Access plies with ASA to additional sectionsis obtained by either standards, including Booleancapa- indirect bilities and complexand double-precision arithme- addressing or indexing. tic. A real-time The basic I/O system FORTRAN IV compiler withre- of the 116 consistsof a cursive call subroutinesis provided for the DDP- general input/output bus. This bus is used to trans- 124 and DDP-224computers with three indexregis- fer full words inand out of the computer.In ters.FORTRAN IV object addition, programs can also it contains lines whichprovide timing be run on the DDP-116in real-time applications, and command signalsto peripheral devices.Each provided that recursive peripheral device tied to calls are not made to stand- the I/O bus has itsown ard subroutines. TheDAP symbolic-assemblypro- buffer and control logic.This feature permitsa gram which is provided conforms high degree of flexibility with SHARE in using multipledevices standards and acceptsall basic machineinstruc- concurrently and inhandling multipledevices tions, plus extendedmachine and pseudo operations. through priority interrupt.The four basic modes Object output in which data may be either relocatableor absolute. can be transferred back and forth Subroutine linkage isprovided which will allow between peripheraldevices and the 116are single- FORTRAN IV compatibility. word transfer, single-word On the 24-bitcom- transfer with priority puters, the DAPprogram includes MACROcapa- interrupt, direct-multiplexchannel and direct-data bilities. channel. The utilityprograms supplied include The direct-multiplex a debug channel permits datatrans- program which works throughthe typewriter, fer between peripheraldevices and thememory dumps, loaders, updating concurrent with computation. programs, a comprehen- In this mode, the hive I/O libraryand a mathematics library.An starting location to whichthe block of informa- executive-controlprogram is available for the 24- tion is to be transferredand the final locationat bit computers which which the block interprets commands through transfer is to beterminated are the typewriter andcan load selected programs from set up underprogram control.The data transfer magnetic tape. is then initiated by In those configurationswith three the program.Once this has or more magnetic tapes,load-and-go capability is been done, transfersoccur independently of the available. A Real-Time program control until the Monitor willsoon be avail- block has been filled. able which will allowthe user to share thecompu- Data transferscan occur at a maximum wordrate tational capabilities of of over 145 kilocycles the computers betweensev- with this mode. eral programs inmemory. The Real-Time Moni- The direct-data channel(DDC) isa high-speed tor will schedule andcontrol the execution ofall channel capable ofachieving input/outputtransfer programs and also control all I/Ooperations and rates in excess of500,000 wordsper second.It interpret interrupts. can operate in a time- sharing mode with the basic PERIPHERAL EQUIPMENT computer.Program control ofthe DDC isaccom- plished entirely through'the device with whichit Computer Control Companyoffers a selection of is interfaced. peripheral equipment whichwill fulfill the needs of most engineering designsituations.The stand- SOFTWARE ard peripheral devices A comprehensive offered are : software package which willsup- 300 cps paper-tape reader ply the needs of eitherthe experiencedprogrammer or the occasional user is supplied 110 cps paper-tape punch with the SCcom- 200 card/min card reader puters at no extra cost.The programming languages and utility routines havebeen designed to conform with the ' C. W. 'Walker, "ProgrammingThe Compacts," Datama- most widely usedprogrammer tion, April, 1966,pp. 31-34. 125 100 card/min card punch veloped our digital-computer product line and the 300 line/min printer (120 column) capability to design and build large-scale complex 300 increments/sec digital plotter computer-based systems. This complete digital cap- Dual and triple density magnetic-tape units ability provides 3C with an effective means of de- In addition to the standard equipment just developing new products.In addition, this capability scribed, 3C is able to offer a wide range of special facilitates the expansion or modification of 3C equip- equipment through the facilities of the Systems ment with compatible logic and memory modules Engineering Department. In this department, com- directly by the customer. puter systems and related equipment are custom- Within the near future, 3C will expand its com- designed to meet special customer requirements. puter product line with more integrated-circuit These special designs have ranged from simple computers which will offer significantly more com- modifications to standard peripheral devices, and putations per dollar than the present day discrete- also to complex multiprocesSor configurations such component computers. In addition, more and lower- as the Apollo Mission Simulator.Systems Engi- cost peripheral devices will become available as neering provides those devices which are not yet standard equipment, including bulk storage devices, offered as standard equipment on 3C products be- such as discs or drums, and display units.The cause of diverse customer requirements.Typical memory speed of these new computers will be in the of this work has been the design of several different 1-microsecond range.In the software area, more communication interfaces for linkage with either user-oriented application packages will be provided, high- or low-speed lines in large-scale communi- in addition to the continuing 3C effort to improve cation network and time-sharing computer centers. the quality and capabilities of the general-purpose Computer Control Company has also designed software programs. several different cathode-ray tube displays with The industry trends are definitely pointing to- speed ranges from 10 to 100 microseconds per point. ward extremelyhigh-speedcentralprocessors. These displays have included such features as char- Memory speeds within the range of 100 to 200 acter generation, line and circle generation, size nanoseconds should be obtainable within a few control, light pens, buffered memory for refreshing years.The size of the central processor will be of the display and keyboards.Each display is de- insignificant compared to the equipment around it. signed specifically to meet the individual customer's Minus the power supply, memory, and I/O equip- requirement. ment, the central processor could be as small as a The previous description provides a brief glimpse cigarette package. of the present 3C product line.These products, combined with the special Systems Engineering Communication between man and the computer capabilities, allow 3C to meet the requirements of through the peripheral devices has not seen com- the engineering design field. parable advances in the state-of-the-art compared to the advances in central processor designs ; how- FUTURE PRODUCTS AND TRENDS ever, we at 3C feel that this situation will change. Computer Control Company entered the digital More emphasis will be placed upon increasing the product field primarily as digital module suppliers. efficiency and lowering the cost of the peripheral Through continuing research and development have devices wheil the cost of the central processing unit come many improvements in 3C products such as approaches a small percentage of the total system the highly reliable and versatile S-PACs. The new cost.Two areas which show high promise for monolithic, integrated-circuit p-PACs are a direct tomorrow's communication with the computer are result of a continuing research program being car- oral communication and pattern recognition.An- ried out in our Micro-Techniques Laboratory. With other area that will show marked advances is the that line of versatile logic modules, 3C'has devel- area of higher-level languages which work directly oped the memory product lines and many special with the English language and have the feature of hardwired systems.From this base, we have de- self-modification to improve their capabilities.

126 1.11.0 T

THE COMPUTER AS ADESIGN TOOL J. WORTHINGTON Data Processing Division IBM White Plains, N. Y.

The 1960s will probably be remembered asthe is comparable to the large machines which were decade in which the computer came of age as an available a decade ago. engineering-design tool for the broad spectrum of The Model 1130 and its associated support com- problems. The papers presented at this conference bine to form an excellent tool for an engineer. The are proof that this is true.In projecting trends, price is such that it can be justified by a smallengi- it is desirable to consider those that have brought neering organization, or by larger organizations the computer to the forefront as a design tool. with systems located at the points of usage.Our experience indicates that this will be a popular ma- One essential is the availability of computers at considers reasonable. One chine and will c 'joy broad usage in engineering a price that the buyer design. means of calibrating the worth of acomputer offered The largest system in the product line is the today is to make a comparison with what was avail- System/360 Model 91,` which by any measure is a Today's machines can execute able ten years ago. large machine.It is particularly well suited to at least ten times as many instructions per unit of solving problems that would require an excessively time as did the 1956 machines, and they sell for a long time to solve on lesser machines.Because comparable price.Machines with arithmetic cap- of the size and power of the system it is attractive ability equal to the 1956 systems are available for to organizations with large problems and/or a large about a tenth of the price of the earlier systems. number of people who are solving problems. The buyer not only can get more capability per The Model 1130 computing system and the Model dollar, but also has many more options in size, 91 represent opposing strategies for satisfying come speed, and accessories today.In 1956, the num- putational requirements.The small machine can ber of options offered was a little reminiscent of be effectively used by an engineer sitting at its con- Mr. Henry Ford's famous comment that the buyer soln and interacting with it as the problem is solved. could obtain a Model Tin any color that he wanted The largest machine, because of its relative speed as long as he selected black.The Model T era of with respect to the human reaction time, is more computers is past.The number of sizes of com- suited to having the engineer work separately from puting systems and the options available for fea- the system, either through batch processing of work, tures and accessories is somewhat overwhelming. or through suitable multiprogramming. In April, 1964, IBM'' 2 announced its new product Some users desire a system which offers Lime line which is based on solid-logic technology. machine performance, and the close man-machine The product line at announcement consisted of coupling of the small machine without undue loss six different models of processors with 15 options of system capacity.The Model 67 Time-Sharing of memory which offered 19 different configurations. System' has been designed to service interactive, Since that time additional models and memory op- or conversational, terminals which give an engi- tions have been announced, bringing the number neer the full functional capability of a large ma- of models to nine and the configuration total to 34. chine at a response rate commensurate with his re- The smallest member of the line is a low-cost, action time.Furthermore, it will handle batch compact computing system, listed as Model 1130' processing as background work. The machine's at- It is designed for general-purpose computing and tention is focused on the active terminals so that is particularly oriented towards operation by the each user receives reasonable service.The time individual requiring the problem solution. The in- not required for servicing the active terminals is put/output equipment attachable to the system in- available for batch processing.Another essential cludes a typewriter, paper-tape reader/punch, mag- to easy utilization of computers by the design engi- netic disk storage, card reader/punch, line printer neer is the availability of 'terminals or transducers and a plotter. The capability offered in the system which facilitate the man-machine communication. The more popular interactive terminals are type- G. M. Amdahl, G. A. Blaauw, and F. P. Brooks, Jr., writers and cathode-ray display units.The major "Architecture of the IBM System/360," IBM Journal of Research and Development, April, 1964. IBM System/360 Model 91 Functional Characteristics, IBM System/360 System Summary, A 22-6810-6. A 22-6907-0. ' 1130 System Summary, A 26-5917. System/360, Model 67 Time-Sharing System. 127 7,1FRoweiwViVIPONIMMR"Vat

le0014000/0601e.000400000_

trend in these devices is towards improving their developed by Professor Miller.'The engineer who function and packaging. uses COGO needs little or no knowledge of com- The utility of a terminal is principally determined puting and can solve problems on a computer with on a systems level where the combination of pro.. quite nominal instruction.The emergence of ap- gramming support, the processor. and the terminal plication-oriented languages represents one of the determine the capability available to theuser.As major trends occurring in the use of computers. an example, a cathode-ray display device attached to a processor is quite worthless except whenpro- The multiplicity of computers and options in grammed. A FORTRAN compiler with graphic which they are available gives rise toa new de- extensions can transform this combination of hard- sign and selection problem. The selection ofa com- ware into a rather capable tool for the design engi- puter involves matching user requirements against neer. many alternate configurations.In a sense, the new Perhaps one of the most important ingredients freedom afforded by many options hasa price, but in hastening the general acceptance of computers it is offset by being able to optimize selection based is what might be termed themacro- or application- on the user requirements. oriented level of programming. Toprogram a com- From the foregoing, it is evident that computers puter either at the symbolic-languageor at the algo- have arrived in the engineering world.It is diffi- rithmic-language level requires that theuser must cult to imagine an engineer who, in 1970, will not already possess thenecessary knowledge about the have a direct involvement with computers forex- machine and the programming language. By intending his ability. The emphasis in education must serting an application languageon top of the hard- be to expose engineering students to the potentials ware and algorithmic compiler, it is possible to of the computer and the techniques for itsusage greatly reduce the level of knowledge requirecto so that they may in turn build upon the knowledge utilize the computer. A good example of thDi is that has been set before them. found in the coordinate geometryprocessor, COGO,6 7"Computing Report," The Scientist, and Engineer, Vol. COGO Application Program 1130, H20-0148. I, No. 3, December, 1965.

128 memAirriMallat .asuaLLameiti...whok

ANALOG AND HYBRID arviPUTERS IN EDUCATION IN ENGINEERING DESIGN G. G. MOSER AND P. E. HUBER Electronic Associates, Inc. Princeton, N. J.

INTRODUCTION solved economically by purely digital or analog tech- Other papers in this session have placed emphasis niques. They combine the high speed of the analog on the value of digital computers in engineering computer and the program flexibility end precision education.Although Electronic Associates, Inc., of digital logic and the digital computer. does manufacture digital computers, this paper will The simplest application of a hybrid computer is briefly highlight the equally important role of ana- the simulation of a system which is naturally hy- log and hybrid computers in education in engineer- brid.Examples of such systems include : attitude ing design. control of a space vehicle, certain functions of the A report by the committee sponsoring this con- nervous system, and direct digital control of a ference says, "Familiarization with techniques for chemical process. using both Analog and Digital computing equip- Optimization problems requiring automated, high- ment as a simulation tool can largely be integrated speed, iterative solutions are particularly suited to into the existing course structure of the curricu- hybrid computers. Dynamic models are simulated lum."The ASEE in its report on Computers in on the analog computer, and evaluated and modified Engineering Education makes the following recom- on the digital computer. Problems in this category mendation : "A recommended lower-division intro- often aemand hundreds of thousands of integra- ductory computer-use course for digital, analog, and tions of differential-equation sets.Only a hybrid hybrid machines is needed which should be equally computer can do this within a time which will satisfactory for students entering all engineering make the design economically feasible. disciplines." The solution of multi-dimensional, partial-dif- These direct quotes from the reports of the Com- ferential equations is also only practical when ana- mission on Engineering Education and ASEE pre- log (parallel) integration supplements the digital sent a clear challenge to both educators and com- computer.Consider a particular example, that of puter manufacturers.They indicate the need for a heat exchanger involving three states (liquid, training in analog and hybrid computation for all vapor and heated vapor) with variable boundaries. engineering disciplines. A power company evaluated programming and de- This paper will briefly address itself to three ques- bugging time at six months if the system were simu- tions : lated on a large digital computer. To arrive at an 1. Why are these recommendations important? acceptable design, over 100 solution runs would be 2. How are engineering schools implementing necessary.Prior calculations showed that each these recommendations and what other action digital run would require up to two hours. On a can be taken? hybrid system it took less than a month to program 3. What analog and hybrid hardware and support and debug the problem, and only two days to arrive are available to the educator who is implement- at a satisfactory design. ing the recommendations? There are already over 150 hybrid installations and the rate at which hybrid systems are being HYBRID COMPUTERS purchased is at an all-time high.These hybrid Analog and digital computers made it possible to facilities have grown from a need to simulate and solve problems that would have been impossible or study systems which could not be simulated eco- economically impractical before their advent.In- nomically on either digital or analog computers, deed, they have stimulated new thinking and have although most companies with hybrid capability made it feasible to define new problems in order to also have separate digital and analog facilities. push forward research frontiers.Many excep- In these companies the design engineerscan choose tional achievements in the life sciences, the process the most appropriate "tool" to solve their particular industry, and the automotive and aerospace fields problems. are the direct result of using the powerful tools of In education, hybrid computers and computation digital and analog computation. concepts are in their infancy.The interest on the Hybrid computers have grown out of a need to part of educators, like yourselves, has been intense, analyze and solve new problems which cannot be beck ase of the recognition of the excellent potential 129 1014., AU.*

of the hybrid computer as a tool for engiaeering and sees an immediate response.He has an ex- design. A number of schools already have taken, cellent feeling for the system because there is a the initiative and acquired hybrid-computingcapa- one-to-one relationship between the physical sys- bility.These include schools like the University tem, the mathematics, and the computer model. of Minnesota, Worcester Polytechnic Institute,the Instructors are constantly relating stories about University of Texas, Harvard, the University of the enthusiasm and interest of students whoare Michigan, Stanford, UCLA, and the University of taught with the aid of analog computers.The Wisconsin, just to namea few. poorer student is able to absorb basic concepts which These schools and others have recognizedthat are difficult for him without the aid of the com- problems demanding hybrid computationare not puter, while the bright student can get answers in the future, butare being defined and must be to much more complex questions. solved today.Industry must have engineers who A significant point about the use of the analog are capable of using these systems, and today's stu- computer as a lecture aid is the fact that the students and practicing engineersmust be taught hy- dent does not have to know how toprogram the brid concepts in undergraduateand continuing edu- computer.A short lecture supplies all that is cation programs.Recent workshops sponsored by needed to relate the computer solution to mathe- universities and industry have beenfilled to capac- matics and the real world.The implications for ity, a testimonial to the need forcontinuing educa- teaching differential equationsas early as the fresh- tion in hybrid simulation. The designengineer who man year are particularly appealing. By develop- does not appreciate this type of simulationis se- ing an early appreciation of mathematics, thestu- verely limited in his conception ofproblems.He dent is ready for exposure to designcourses early runs the risk of solving problems in increasingly in his academic training. inefficient ways, or not at all. The activity of the Engineering Concepts Cur- To date, courses in hybrid computationhave been riculum Project of the Commissionon Engineering confined primarily to graduate work.A small num- Education has actually introduced analog computers ber of courses have been openedto upper-division at the high school level.The computers are used undergraduate students, but there isa need for more to give the student an appreciation of dynamic than a slight increase in the trendand the num- systems and the concepts of engineering design. ber of theseprograms.The ASEE report calls In the laboratory, the course in analog computa- for a lower-divisioncourse to acquaint students of tion often gives the engineering student his first all engineering disciplines withhybrid computation; experience with the design process.After acquir- this recommendation underlinesthe importance of ing the fundamentals of computer programming, early action by educators andindustry. the student is usually given the opportunity to choose one or more design projects involvingsyn- ANALOG COMPUTERS thesis, programming and optimization throughex- perimentation on the computer. The substantial number of analogcomputers al- If it is important ready in use in classrooms testify totheir value as that we teach engineering design, then I suggest teaching tools.In the past few years, EAI alone that the analog computer has been, and will continue has supplied more than 600 machinesto colleges to be, one of the most powerful teaching tools. and universities. The analog computer with its peripheral equipment is also an excellent tool for research. The computers serveas mobile classroom demon- The stratora which permit teachers to "hands-on" use of the computer with theman in demonstrate dy- the loop provides incomparable insightinto the namically solutions of complex,non-linear differen- operation of a physical system. It is tial equations.Analog computer simulation an economical makes form of experimentation offeringthe flexibility it possible for the student to appreciatethe values of described physical systems of continuous adjustment ofparameters with im- mathematically.It mediate display ofnew solutions.Programs can can, and does, penetrate the barrier which theaver -. be changed directlyon line by repatching all or age engineering student finds when he tries tore- parts of the program on the computer. late mathematics to the real world. The analog computer is often used The analog computer to examine answers the student's ques- problems that are poorly defined, that is,to develop tions about how the physicalsystem behaves un- insight into physical systems that the designer der different conditions. does Instead of equations that not really understand and cannot describemathe- are impossible to solve, he seesa real solution. matically.Through intuition and "gut feel,"the More important, his questionsabout changes in designer begins to guess at the mathematicsunder- parameter and initial condition valuesare answered lying the physical system. Hemay construct sev- immediately.The instructor,or student, simply eral subsystems on the analog computer turns a potentiometer and at- or throws a function switch, tempt different types and amounts ofcross-coupling 180

rlf tt'-+*W7WmrplrInOrrrr.WW,W1gMjMPPIRPFM *Nam V60.11....C' . 114,11.

Airl&s20111.10111misiiiiiiiaftwarmwomagrurkUMOKANIIIMINW11.1

until his solutions begin to resemble the behavior, ANALOG/HYBRID HARDWARE or predicted behavior, of the physical system. AND SUPPORT When his results agree, hecan go back and determine the mathematical description. The next step Modern analog and hybrid hardware for educa- might bea larger, more complex analog analysis, tional applications employs solid-state design and or the use of a hybrid or digital computer to refine does not require special environmental control. The his analysis. systems have been designed for low maintenance and "student-proof" operation. They employ modu- The strength of the analog computeras a teach- lar construction to permit maintenance of modules ing tool for the student of engineering design should without disrupting system operation. be apparent.It is economically feasible to have four or five, or more, consoles in the laboratory, EAI offers a complete line of analog and hybrid particularly with EAI's liberal educational contri- computing systems to serve both education and bution program: Smallgroups of students, or indi- industry. The TR-20, and its predecessor, the EAI vidual students, can work on a design project with PACE TR-10, are the most popular analog compu- maximum man-machine intimacy through thepower ters for educational use. They are used extensively of CRT displays, XY and strip-chart recorders, and in teaching basic analog computation to undergradu- digital voltmeters.The students' curiosity, drive, ates in engineering and science. They are also used creativity, and desire to investigate furtherare in- heavily in lecture demonstrations and in both under- spired by the immediacy of hands-on parameter graduate and graduate laboratory research and de- adjustment and instant readout of complete sets sign projects. A special educational contribution of solutions. policy makes the TR-20 particularly desirable for The analog computer not only offers the student schools. the opportunity for involvement and participation The EAI TR-48/DES-30 Analog/Hybrid Com- in engineering design, but also teaches him disci- puting System is the standard for both industry plines of analysis which are very valuable incon- and education in desk-top class equipment.More structing and conceiving realistic models.In ad- than 500 of these systems are in operation in class- dition, he acquires basic concepts which will be rooms and laboratories, serving principally as tools necessary in teaching him the fundamentals of hy- to aid in engineering design and research. brid computation. The list below shows the departments already The EAI 680 Analog/Hybrid Computing System using analog computers in design education in engi- is the latest addition to EAI's product line.It has neering: been designated for tie-in to a general-purpose digital computer, and offers powerful hybrid computing Aeronautics and Astronautics capability at a price within the university budget. Agriculture Chemical The EAI 8900 Hybrid Computing Systemcom- Civil bines the EAI 8800 Analog/Hybrid Computer with the EAI 8400 Digital Computer.This system is Computer Science designed to satisfy the needs of large simulation Electrical/Electronic laboratories in government and industrial installa- Mechanical tions. R is the type of equipment, along with the Nuclear TR-48/DES-30 and EAI 680, that today's student will be using to solve engineering design problems This list clearly indicates the need fora basic, after he graduates. sophomore, introductory course in analog simulation, as recommended by the ASEE subcommittee In addition to hardware, catalogs ofcourses are on computers in engineering education.This ap- available to educators, as well as to industry. These proach would certainly be more efficient thanpro- educational programs start with basic programming grams which require basic analog-programming and extend to sophisticated computed application courses within all the different departments. The concepts in most disciplines. EAI offersa library basic course opens doors for educators in all de- of laboratory problems for student work in most partments, not only in the student's major, but also engineering-design areas.This reference library in mathematics, physics, and chemistry.If all of has a number of pre-programmed demonstration the students receive basic training, the instructor problems as well.Programmed student workshop can employ the analog computer whenever it can material, slides, films, and other teaching aidshave best fulfill a teaching requirement in the classroom also been developed by educators forclassroom or laboratory. use. 131 SUMMARY These computers serve equally v.- all as mobile class- Many people are using analog and hybrid com- room demonstrators, student laboratory instruments puters in engineering design. The use of these tools and as effective tools for design and research.In is growing rapidly as new simulation areas are line with the ASEE subcommittee recommenda- discovered. The design engineer who cannot appre- tions, the universal value of analog and hybris'com- ciate, and use analog and hybrid computers risks puters in engineering design strongly suggests the solving problems inefficiently in time and cost, if, need for including analog and hybrid concepts ina indeed, he can solve them at all. lower-division undergraduate course. Engineering educattys in all disciplines use ana- Economical, modern solid-state hardware, to- log and hybrid computers as aids in teaching engi- gether with available educational software, make neering design, as well as in teaching students the analog and hybrid computers particularlyappro- concepts necessary to use these tools in industry. priate for educational use. RECENT DEVELOPMENTSIN ON-LINE,REAL-TIME TIME-SHARING ANDMULTIPROGRAMMING COMPUTERCAPABILITIES A. M. ROSENBERG Scientific Data Systems Santa Monica, Calif.

Computer time-sharing and on-line use,basic tools not actually using itall the time.An obvious for education in engineering design, are areasin course was to share andinteract with the computer, which we at Scientific Data Systems,and myself still giving the DES-1 user prioritybecause once in particular, are focusing specialattention. he had started problem execution,the time cycle On-line use doesn't necessarilyinvolve time- became very demanding and critical.The problem sharing; in the early days of computerdevelop- was that there wasonly one channel available for ment, whenever a person went up tothe computer the typewriter, and a second user wouldhave to use and used it, he was on line, and hecould do what- the same typewriter.If the second user had had the machine. But as comput- another I/O (input/output) channel, hewould have ever he wished with because any ers became larger,faster, and more expensive, this been able to time-share the machine, procedure became increasinglyimpractical; and on-line use of a machine involving a human opera- practical and tor involves delays.If all of the core memory is time-sharing evolved naturally from avail- economic considerations. not being used, the other user can take up One definition of time-sharing is thefollowing: able CPU time.With the DES-1, the user had be- a number of usersshare a computer in an interac- conversational use of his simulation program tive environment. At SDS we takeinto considera- cause he had on-lineediting capability.In addition the entire spectrum of time-sharing,which tion, he had on-line intervention by meansof the includes critical real-time operation-(such as pro- DES-1 console. cess control), interactiveconversational use, and Most of our machines at SDS are beingused in background production (a job-stack operation). on-line scientific applications, and, morespecifically, There's nothing wrong with having acomplete in real-time data acquisition, telemetry work, nu- mix of all three, particularly if they arerelated ; clear work, biomedicine, etc. it's only a question of scheduling and responsetime. One of our creative customers, the Universityof Our new product line, the Sigma series, is spe- California at Berkeley, purchased an SDS 930 com- cifically designed to process simultaneouslybusi- puter and modified it for multiprogramming.This ness data, solve scientificproblems, perform real- modification allowed dynamic relocations, memory time control, and interact with multiple-usersta- protection, and a number of features that would, tions. for example, allow the use of common public rou- First, I'd like to review what SDS has done in tines.This machine is available from SDS as the the past with regard to time-sharing or on-line use, 940 computer. particularly in the area of design.SDS has one All aspects of time-sharing are featured in the product, the Differential Equation Solver (DES-1), SDS 940 including multiprogramming; real-time which is based on a general-purpose digital com- processing; on-line, remote data processing; and puter but which has a special consolefor on-line simultaneous access to the central computer by manipulation of problem parameters and timing several users.The SDS 940 time-sharing system cycles, giving the user a combination of capabili- provides up to 128 users with simultaneous access ties. to the problem-solving capabilities of the high-speed The language is based on FORTRAN, and either digital computer. A response time of two or three the console typewriter or the card reader creates seconds facilitates rapid communication between tha the operating program. The analog inputs are con- user and the 940. trolled by the operator at the DES-1 console. This So far we've discussed hardware but in the console permits on-line monitoring and control of time-sharing business, it's the software that counts. problems during execution, along with strip-chart Berkeley developed outstanding software and it's recorders, scopes and plotters. possible to see it, as well as to use it, in opera- The flaw in this approach (one-man use of a tion today.The Berkeley system has been in general-purpose machine) was that the user was continuous use since April, 1965, and demonstra- keeping the entire machine tied up, although he was tions of its ability to service multiple, remotely 133 located users have been made by linking the Language 1 (PL/1) or PL/2, which they calla computer with teletype consoles at the Univer- FORK. By means of the FORK facility,programs sity of Illinois, Urbana, Illinois, and at Stanford can call other programs, jump back and forth from Research Institute, Palo Alto, California.It is a one program to another, and take any kind ofpro- general-purpose, time-sharingk stem in the sense gram (for example, one that is a design tool) and that there is fullmemory protection, and the user put it under the control ofa teaching mechanism can work from the assembly language level onup. such as a programmed instruction package.To gain As a result, they've builtmany sophisticated tools, experience, the programming toolcan also provide and have done this easily because ofa set of com- the mechanism for guidanceactually leadingthe mon routines, involving text-handling, etc., which student by the hand. provide building blocks for other applications. At SDS our manufacturing personnelare running The Berkeley time-sharing system hasa macro- circuit-design problems on-line usinga typewriter assembler, a very sophisticated debugging package, as a communications tool.They are also using a a sophisticated editing package, LISP, SNOBOL, logic design program that is basedon a version of FORTRAN, and ALGOL, in addition to Auto-Secre- FORTRAN to which have been added logicalopera- tary, which is a documentation editor. ti Theywere tors such as : AND, OR, INVERT, BUFFER,mid able to create these andmany other programs very so on.These programs allow labels to be much quickly with the on-line, time-shared facility. longer than 26 characters, and theuser can actu- A major feature of the Berkeley system,from the ally insert a logic equationas a label. point of view of teaching and ease of use, is full- The next important step for the future is duplex communication. We all wantto have our pack- cake and eat it too. The aging, which involves putting the modulestogether user would like to talk to the co,npactly so thereare no timing problems caused machine, but perhaps he is nota good typist, or by separation of the modules. perhaps he is lazy. Yet he wantsto see his output, and he'd like tosee it in English rather than in Documentation information is storedon magnetic tape and used in testing the final product hard-to-read shorthand. The full-duplexcommuni- for custo- cation allows every character tobe processed and mer service needs, and any other documentation to turned around, and that processing support a product we develop. In thismanner SDS can actually is taking advantage of the change the input character in dramaticways. advanced techniques which evolve from the development ofour own With the Berkeley system,any kind of executive- products. system or subsystem commandcan be typed in; when the system recognizes whatthe user is really A new item on the SDS marketinglist is our trying to say, it will finish it for third-generation computer, Sigma 7, which hasbeen himit's almost designed for the full spectrum of alive.If the user doesn't require thishelp, he can time-sharing. tell the system he's For real-time applications, Sigma 7hardware ca- an expert rather than a novice, pabilities feature high-speed computation and it will react by merelyaccepting the shorthand to keep notations without completing the pace with real-time processing demands; flexible full text. input/output to permit handling Every on-line system and a variety of types every on-line applica- and sizes of data easily; almostinstantaneous re- tion is a potential learningmechanism because when the user is interacting, he sponse time with an improved priority-interrupt is learning, even when system that always reacts withinmicroseconds ; he makes mistakes. There isa teaching mechanism and the ability to change the called HELP that has been built entire state of the into the Berkeley machine from normal processing tointerrupt proc- system as software.It takes natural English text essing, while saving all in a question-answer form and necessary status informa- looks for key words. tion and interim results. Asingle instruction re- It also recognizes the questionin tree context of what quiring only six microseconds provides the user is asking about. this context- For example, if he is ask- saving capability.In addition, high reliability is ing about string manipulationfor SNOBOL, apro- achieved throughuse of SDS-designed, monolithic, gramming language especially suited to operations integrated circuits and specialproduction and auto- on strings of characters, HELP knows(in context matic checkout equipment. of SNOBOL rules) whatthe user is asking and will give back the appropriate Field-proven peripheral equipment foruse with answers.Every sub- Sigma 7 includes keyboard printers,card readers, system, every service system,and every language card punches, line printers, system can also be connected magnetic-tape units, to HELP so that the rapid- access datafiles,paper-tape readers and user can always have HELP either directly,or while punches, graph plotters, display he's working with equipment, and some language or process. data communications equipment.SDS graphic The Berkeley system forthe SDS 940 capitalizes scones will be particularly useful in on-line design on the tasking mechanism similar toProgramming applications. 134 ..414,-1410.tignakkia-*Strar...c.aokira. ikakssii

We feel that the system should be centralized be- and comprehensive hardware/software package and cause of a relationship between what the user may includes FORTRAN IV and PL/1 compiler in both be doing, for example, and a real-time process, a standard and high-efficiency versions, because SDS conversational process, or perhaps a secondary recognizes that hardware and software have become background process. There is no single job that is so completely interdependent that comprehensive necessarily restricted to only one mode of operation, software is now essential to full-system utilization. and the jobs in various modes may be related. The difference between conversational and batch For example, suppose the user has a real-time processing makes little difference because we use control process and wants to improve it or wants essentially the same compiler. We provide on-line to change some parameters, developing the change editing similar to the Dartmouth system with one first and debugging it before actually controlling exception : with Sigma we prov ode for closer user something.The mechanism allowed in the hard- interaction. We do not wait until the user finishes ware and software can break up such a process into, typing in all his statements to have the compiler first, the very critical, responsible, sensitive part look at them and say, "You have made the following dealing directly with the input/output; and, sec- mistakes." There is a connection between the editor ondly, the part which is the normal computation or and the syntax analyzer of the user's compiler so processing program, which could be full of errors that, as the user enters each text line, he is told and ought to be caught and trapped if it is making immediately what is grammatically wrong.This a mistake. provides a step-by-step analysis and reporting of At SDS we do this by the programmed operator errors. The program can then be compiled and ex- mechanism.In the Sigma series, just as in the ecuted and the input cycle repeated if the user finds SDS 940 computer, there are programmed opera- logical program errors.This interactive relation- tors that are public routines operating trap loca- ship will be available with FORTRAN and PL/1, tions.These operate in the master mode, and any and the mechanism is general enough so that any machine operation can be legally performed. Some language that we decide to add to the software sys- of these trap locations can be dedicated or allocated tem (such as LISP II or COBOL) can be handled by to a real-time process by saying, "Please load me the same mechanism by merely plugging the syntax- in."The user can specify what part is sensitive analyzing portion of the language processor into the (for example what part should operate directly with on-line editing package.The on-line editor is es- I/O and be plugged into interrupt lines) and what sentially what makes the languages conversational. part is a normal kind of process to be treated as a With Sigma 7 we are also planning to allow the slave or user program. The latter, of course, can be DES-1 concept to be expanded, but in a time-shar- kept in core memory and not swapped out, while ing environment, because it is not efficient to tie up normal jobs may be swapped out because of multi- an entire machine unnecessarily.Because of time programming activity. constraints and scheduling, it may be possible to The Sigma 7 has selective memory protection.It have many on-line users getting one- or two-second has dynamic relocatability through a map which responses at the same time a DES-1 user la operat- also allows fragmentation of programs. The pieces ing, but batch jobs can be run in the background. of a user's program can be moved around anywhere This is why we say that the whole spectrumcan be in physical core memory. Input/output bufferscan accommodated ; the facilities are there and theuser be left in core while the rest of theprogram can be can operate a Sigma 7 system to accommodate his removed at will.Through the software, tasking needs. facilities are provided, which means thatan entire A frequently asked question is, "How many users program need not reside in core unnecessarily. can you put on your time-sharing system ?"The Sigma 7 can also control up to eight input-output answer is, "How many users doing what, with what, processors in various combinations of two types. and requiring what kind of response time?" Each multiplexor I/O processor can accommodate Tf it's a real-time process that has certain de- up to 32 I/O channels operating simultaneously. mands on the CPU and core storage, some capacity Each selector I/O processor can accommodateup to will necessarily be lost. We are generalizingour 32 very high-speed devices, only one of which can scheduling mechanism so that the number of pri- be operating at any given instant. Combining these ority queues (that is, the rules for operating in each two types of processors permits the user to con- queue : how much of a slice of time should be given, figure the exact combination of simultaneity and does it run to completion, does it takeover the speed required.The I/O processors operate semi- whole machine, etc.) are the parameters thatthe independently of the central processor, leaving it user will have to determine it terms of hisown in- free to provide faster response. stallation and applications. SDS Sigma 7 software offers a totally integrated At this time users haven't clearlydefined their 135 -

own needs. Some say that a response time of ten a busy signal) can be establishedso that no new seconds is sufficient.The Berkeley designers, for users would be accepted by the system. example, won't tolerate anything over one second The teaching of engineeringdesign will be greatly for a conversationalprocess; whereas, in the System enhanced by recent developments Development Corporation system, in .on-line, real- the range, which time, time-sharing, andmultiprogramming,com- was constantly changing, was anywherefrom one outer capabilitiesdimensions keyed second to approximately 15 seconds. to the needs It is my view of dynamic change. Theseneeds have been reflected that consistencycan be even more important than in the design of the SDS any particular set of times, and Sigma series computers. every user will have With the moduliSigma concept, system configura- to determine for himself whetheror not he wants tions can be expanded to fit gradual degradation. To the needs of virtually prevent gradual degrada- every operating environment fora variety of creation of response time,a cutoff point (indicated by tive applications.

136 GENERAL ELECTRIC REACTIVE DISPLAY SYSTEM MARVIN T. S. LING Graphic Techniques Project General Electric Computer Equipment Department Phoenix, Ariz.

GENERAL ELECTRIC REACTIVE DISPLAY cates with the computer by means of textual as well SYSTEM as pictorial languages. INTRODUCTION OBJECTIVES Perhaps one of the most important developments The GERD system has two major objectives: connected with the computing field during the last 1. To expand and maximize the total user value few years is time-sharing. By this we mean that a derived from present and future computer systems. number of users share the same computer simulta- To achieve this, GERD should provide many remote neously and the response from the computer is fast display terminals time-shared with a central com- enough to satisfy the needs of all users.Essentially, puter system so that the user accessibility can be it implies that every user of the computer has di- increased.The communication languages used at rect access to the system. The development of time- each display terminal should be both textual and sharing is very important, since it offers at least graphic.Graphic communication is important in the following advantages : that human beings understand graphics quickly and 1. It eliminates the inefficiency c4used by the the presentation is concise.Furthermore, each present red tape programming of a hierarchy of terminal and the system should be designed in such programmers. a way that users of various backgrounds may use 2. Because of fast turn-around time, it improves the system on-line with a minimum of learning re- human productivity. quired. 3. It increases the consumption of computer 2. To ensure that the basic GERD software pack- power by an increase in the number of problems ages be general, flexible, efficient and open-ended, submitted to machine solution which were never GERD should provide a horizontal software pack- previously submitted because of the inconvenience age which allows users to generate their own speci- of using the computer. fic application programs within broad compatibility 4. It provides the opportunity to work with a boundaries. The total number of applications users class of problems in which the interaction between may make at the display terminals is practically un- computer and user may facilitate or lead to a more limited. GERD initially categorizes those applica- creative solution. tions in two areas : the engineering and manufac- Focused on human direct access and the on-line turing applications, namely, design and numerical use of computers, the following matrix of techniques control; and the information storage and retrieval for man-machine interaction is provided : applications for management. A. Keyboard SYSTEM DESCRIPTION Mediapunched cards, alphanumeric texts in To achieve a system such as that described above, hard copy proper organization of information within the com- Machineryk7punch, printer, Teletype puter is of critical importance. A general data equipment structure should be developed for efficient retrieval, B. Visual/Keyboard updating, and storage.In particular, it should be Media--A + alphanumeric texts in display able to make associations easily between stored in- MachineryA + display tubes formation and new ideas.With this kind of gen- C. Control Wand/Visual/Keyboard eral data structure, it is possible to operate every MediaA + B + pictures application problem on the common data base by MachineryA + B + control wand common primitives and subroutines. For example, D. Voice/Control Wand/Visual/Keyboard the information as to how many bolts are usedon MediaA + B + C + microphones part A, or how many cars are red, will be stored in MachineryA + B + C + microphones and the same general data structure as the information speakers for what line and curves should be machined next by The General Electric Reactive Display (GERD) a numerically controlled machine tool. system is an experimental system focused on the use GERD should be designed for users of various of technique C above. By using a hand-manipulated backgrounds.Mechanical and electrical engineers control wand and display scopes, the user communi- may use it for their various design problems. The 137 le..±.-,r4+

manufacturing engineermay use it for 2-D or 8-D Finally, GERD should beeasy to use. No special numerical control of parts production.Or he may training should be required.Users should not be make use of an X- Y plotter located near the con- required to have any knowledge of sole for more precise drawings computers, al- after the desired though this may help them to makemore creative part has been sketched at theconsole.Draftsmen use of the system. The design of the system should may use the system to generate new drawings and take human-factors engineering intoconsideration. store them, as well as to obtainhard copy. Or they The number of pushbuttons should be may make use of the image-processing unit limited, so to bring that users will not have to spend timesearching for an existing blueprint into the computer andanalyze particular buttons.Consequently, a large number or modify the drawing at the display console.Engi- of functions should be displayed, witha particular neering drawingsmay eventually be stored in the function being selected by the controlwand. The mass storage media, with hard copiesobtainable re- languages should havea syntax and vocabulary motely through the terminal%When GERD is to be complex enough to permita flexible range of alter- an on-line managementor command-and-control natives. To assist communication system, the users should have between the user complete controlover and the system, GERD shouldcontinuously display the system performance. Athis option, theuser the next level of languages ininquiry form. That may select various data formats for display.He is, the system should continuouslyassist the users may analyze the displayed data and up-dateit at in forming the expressions they the console. wish to communicate. Furthermore, theuser should not be required Since the total amount ofinformation that .can to keep track of the syntax hehas formed atany be displayed atone time on a display console is lim- time ; and he should not beworried as to syntax ited, GERD should allowan initial display of the restrictions and other possibleconstraints. outlines of entities anda further display of the de- GENERAL DATA STRUCTURE tail of a particular entityselected by the control The importance of having wand. a very general data The display scope should betreated as a structure in any powerful systemcan never be window that can look eitherat any particularpor- overemphasized. tion of information, In order that GERD be general or at more information with less and flexible, the display terminalsshould be able detail.GERD allows the displayof perspective to handle all possible kinds ofentities. An entity views of three-dimensionalobjects.Rotations of is defined as anything in this the displayed object real world which has on three axes can be doneeas- certain properties. These propertiescan be sepa- ily by the users. rated into three categories.First, an entity has The graphic language to bemade availableon some properties which are unique to itself.A tri- GERD should include basicfunctions such as draw- angle may have been drawnby the designer to ing lines, circulararcs and mathematically de- represent a support ofa bridge with infinitive ri- gidity. scribed freehandcurves; it should also providecon- This information,as well as the date, pro- straint capabilities (makingone line parallel or ject number, andname of the designer, are data of perpendicular to other lines, tangentto a curve, or the entity notcommon to others.Second, an entity of a certain length, etc.)and such functionsas may contain other things in variousways. If lines erasing, and moving, etc.It should also be capable and points are regardedas entities, then a triangle of generating mathematicallydescribed three-dim- contains three line- and threepoint-entities. Finally, ensional surfaces, givensome boundary conditions. an entity may very well be containedin some other The users can easily makemodification of surfaces things. From the previousexample, the line isan through a control wandor through pushbuttons. entity; and it is containedin other entitiesnamely, The users of GERDmay have the choice of eliminat- points and triangles.Consequently, an entity al- ing hidden linesor having them remain displayed. ways associates with other entities,if one wishes to Since the hidden-line eliminationprogram is rather define them as atomic levels.It is thus very impor- complicated, the system shouldnot impose extra tant that the general datastructure should be de- computer time on theusers if they can accept the signed in sucha way that entering properties of hidden lines. things and makingassociations between thingscan be easily accomplished.The general data structure In addition to graphic languagesand languages used by GERD is shown for computer-aided design and in Figure 1. for information stor- The square blocks represententities, while cir- age, rerimal, and control, GERDshould provide cles represent "conjunctions" logical alfir arithmetical operators which contain the de- so that users may scription of the relationshipbetween two related name and execute procedures, andthen store and entities (theusage description). The mechanism retrieve them. This facilitycan be constructed out used to link entities and of all existing languages. conjunctions isa "ring" structure, although other typesof structure could 188 Entity

40..4. A A + 1 ....04 A + 1 A 4' 1 1 40... A + 2 44..404 A + 3 11 + 1 .1

Conjunction Cadvaction

1 +1 C + 1

1 + 2 C + 2

1 + 3 C + 3.

Entity

.0.... D C+3 ....01,.., D + 1 1+3 1 40... D+2 D+3 ...411. D + 3 D+3 '1

FIGURE 2 - RING STRUCTURE EXAMPLE Since the number of required languages is already large, and will grow Pq more and more application FIGURE 1. EXAMPLE OF A GENERAL DATA STRUCTURE programs are built in, tio is importantthat the con- also be used. A ring is a special form oflist tie. sole should be designed in such a way that the user Each element in a ring consists of forward,back- will not be required to search for the required ward, and head pointers.Forward and backward function keys. Basically there are two ways to com- pointers close themselves and form a ring. municate the desired languages to the computer : by Basically, any entity will have two types of rings, function keys or by function pads.Function keys those that relate entities it contains and thosethat are pushbuttons, eachassigned a particular func- relate entities in which it is contained. We callthese tion by the system designer. Function pads replace two types "call-out" and "where-used"rings, re- function keys by displaying symbols representing spectively.The head pointer of a call-out ring is functions in the oscilloscope, and the selection of then pointed to the head element of its where-used each function is made by the control wand. ring, and the head pointer of a where-used ring is GERD will have both function keys and function pointed to the head element of the call-out ring(for pads.The total number of function keys will be our data structure, wewould never search for the rather limited. The use of languages will be in the head element of the same ring). The ring structure conversational mode, and they will be in hierarchal is shown in Figure 2. The unique properties of an order. entity are stored in the entity block, or its exten- The following are brief overall descriptions of sion.The example shown in Figure 2 indicates languages designed for GERD. Each of these lan- that an entity A consists of 2 entities B, 1 entity C, guages represents a group of finerlanguages. 4 entities D. An entity B consists of 3 entities E, 1. CREATEThis allows the user to create new 2 entities F, and so forth. entities.These entities may take the form of two- LANGUAGES dimensional or three-dimensional pictures or sur- On an experimental basis, part of the languages faces.Textual information is considered two-di- outlined here have already been implemented for mensional.The user may indicate either outlines the GE-435. It is not our present intention to cover or details of entities, and may give them symbolic all possible languages which all future users would names.It also allows him to create new display like to see available at the console. Languages for formats. a particular application, such astool selection, spin- 2. STOREThis allows the user to store the in- dle speed, coolant on/off, etc., in numerical control, formation (entities or data) into mass memory. are not included. The operator may be requested to indicate if the 139 11,

entity should be stored in the where-used or call-out a. To rotate the picture around a specified axis. rings of other entities. b. To scale either the specified individual picture 3. DISPLAYIt allows the user to display three or an entire screen. types of information : c. To turn the pages vertically or horizontally, a. Previously stored entities.If entities are not primarily for textual information in table form. in core memory, the system will bring them into it d. To select and display a particular view of a from the mass memory. The user may specify if three-dimensional object. outlines or details of entities are to be displayed. 10. CONSTRAINTSThis allows the user to ap- b. All constraints applied to the current pictures. ply geometrical constraints to the drawings, inas- This is to enable the operator to erase some of the much as the inaccuracy of the display scope, plus constraints and further assist maze-solving for a human errors, prohibits creation of accurate draw- constraint-satisfaction program. ings.The constraints include : c. Information satisfying given conditions. This a. Vertical f. Fix is to allow display of all information, satisfying the b. Horizontal g. Tangent given conditions expressed in Boolean algebra form. c. Perpendicular h. Length 4. DRAWThis allows the user to draw points, d. Parallel i, Angle lines, circular arcs, freehand curves, and surfaces, e. Merge on the display scope.Points can be defined by con- 11. VALUEThis allows the user to specify the trol wand alone or by keying in X, Y with reference nature and ranges of ordinate and abscissa in Car- to a given point.If the control wand is used, any tesian coordinate systems and further allows them point on the display can be selected from those to request the display of a characteristic curve of points physically located on existing things, such some system.It also displays data of particular as the intersection of two lines. A circular arc can points on the curve and the value of things, which be defined either by three points on the arc, or by include the dimension of a line and an angle. the center, starting, and ending points of the arc. 12. MATH OPERATORSThis providesthe A filletthat is, a circular arc tangent to two exist- user with a set of mathematical operators.These ing linesis also definable.Freehand curves are operators include *,-F, /, 2, NEGATE, IN- mathematically defined.Surfaces may be defined VERSE, ABSOLUTE. either by sketching boundaries or keying-in bound- 13. LOGICAL OPERATORSThis provides the ary conditions. user with the following set of logical operations : 5. REPRODUCEThis allows the user to call LOAD, STORE, >, <, = AND, OR. upon previously sketched entities either in the form 14. MATH FUNCTIONSThis providesthe of copy or instance. If it is copy, the internal data user with the following set of functions : MAX, structure is accessible by the control wand; and MIN, MEAN, SQUARE, SQRT, SINO, LOGE, EXP, the data structure of the copied entity is merged ARCTANO. with that of the current picture.If it is instance, 15. SYSTEMThis allows the user to start and the entire entity is treated as one item; andaccess stop the display terminals, dump the data struc- to its internal data structure is prohibited.RE- ture, obtain a hard copy from the precision plotter PRODUCE can be in its original form or in mir- and convert the information read into the data rored form with respect to its origin, the X-axis, structure by the image-processing unit. or the Y-axis. 16. MISCELLANEOUSThis allows theuser to 6. MOVEThis allows the user to move ap- request removal of hidden lines, hatch the specified pointed things either in the association or the dis- area, specify the scale of things, designate attachers, association mode.In the association mode, when and undo the last function. a "thing" is moved, all others depending on that REFERENCE ring are moved correspondingly.On the other John W. Weil, "The Impact of Time-Sharingon hand, in the disassociation mode, only the thing Data Processing Management," Data Processing, pointed at by the control wand is moved, and is Vol. 9, 1965. made independent of all other things. ACKNOWLEDGMENT I would like to expressmy deep appreciation for 7. MODIFYThis allows the user to modify the the encouragement and support given to this project things displayed. The modificationmay be reseg- by the General Electric Computer Equipment De- mentation of lines or reshaping of freehandcurves or surfaces. partment; especially, to Dr. Leonard Maier, Dr. John Weil, John Couleur, Larry Hittel, Line Young, 8. ERASEThis allows the user toerase the and Frank Yao. Many ofmy colleagues have con- entire screen, an individual thing, oran entity. tributed ideas in developing the GERD system. The 9. WINDOWThis allows the user to view the concept of reversing the head pointer in ring struc- objects in the following various ways: ture was originated by Mrs. Sandra Palais. 140 CONTROL DATA'S ACTIVITIES Basic Hardware and Software Peripheral Equipment Application Program W. BEILFUSS Meiscon Corporation Subsidiary of Control Data Corporation Chicago, Illinois

With the expanded product lines of today's com- tion is shifted from program toprogram based upon puter manufacturers, it becomes difficult to de- changing priority. scribe briefly all of the products that will affect 2. Multi-programmingtwo ormore programs engineering design.These product lines include active in central memory.Using various internal a range of computers which vary widely in both and external criteria, control is transferred rapidly size and function; a growing list of peripheral de- from one program to another. vices for different forms of input and output as well 3. Multi-accessmany devices havingaccess to as for data storage ; and a broadening software and the system.Terminals include customary peri- application program library.To cover adequately pherals ; unit process control devices; or people in all of Control Data's activities and their implica- a remote, on-line, real-time basis. tion would take more than the time allotted.I Each of these activities require high-speedin- will try to touch only briefly upon those aspects of ternal processing which can significantlyoutper- our activities that seem to be most widely known, form the I/O times, interrupt capacityto break and to dwell somewhat longer on those less widely program processing when an external signal calls publicized, especially as they are concerned with for service, and an address transfer schemeto re- graphic processing and engineering program de- locate programs being processed when theymust velopment. be transferred in memory. Control Datacurrently BASIC HARDWARE AND SOFTWARE approaches time-sharing with two offerings:the 3300 and 3500 computers, and the 6400 and6600 ACTIVITIES computers. The goals to be achieved are reduced cost of com- In each of these systems, the wholerange of time- putation per job, reduced turnaround time, enlarged shared problems can be intermixed and problem-solving capabilities, and greater capacity to.. handled through a combination of hardware andsoftware. systematize the man-machine loop. To achieve these The 3300-3500 computers goals we are developing faster and more efficient use interrupts, registers and software to process theprograms in small seg- CPU's enlarged parallel-processing units (such as ments called "pages."The 6000 series, with their the 6600 which has ten separate processing units, extended memories and faster transfer rates, all of which share an eleventh), and a greater use utilize a "memory swap" arrangement to exchangecore of time-sharing.Problems which were beyond the in large chunks.Regardless of the technique, capacity or economical handling of previous com- the engineer's reqeest for computation from hisremote puters are currently being attacked on 6600's. console will be honored automatically, with Worldwide weather forecasting, in which the earth due priority.His requests will be handled alongwith all is being thermodynamically modeled in increasingly other programs that are in line for processing, greater degree, illustrates this enlargement of prob- just as if the engineer were on-site.The cost, due to lem capacity.Probably the computer advance efficient sharing of the machine with which currently generates greater interest is time- others, will be based solely upon computation time,and that sharing, for this concept attacks the problems of time is much reduced. computational cost, turnaround time and machine In support of its hardware, accessibility. Control Datacur- rently provides complete operatingand extended Control Data's approach to time-sharing includes compiling software, including thatnecessary for the following capabilities : time-sharing operation. In additionto these stand- 1. Multi-processingautomatic operation of com- ard items, software suchas APT, process control puter systems with two or more programs of differ- software for the 1700 computer, andcertain special ent priority being submitted in batch mode. Opera- software, such as management "RequestLanguage," 141 ,,... -

_ r; k.4.1.4141414esi

is being developed.The latter is an English lan- a photo-multiplier tube in the console cabinet. guage command-set to "call-out" management data Twenty-four function buttons are used to communi- from a company central business file and display cate with the Functional Control Program (FCP) it on either typewriteror CRT console. which controls time-shared operation and provides PERIPHERAL EQUIPMENT basic graphic manipulation capabilities. AFOR- The peripheral equipment that provides the great- TRAN-oriented interface is provided by FCPto test assistance to engineering problem solverscan facilitate communication with the user's application be classified in two general categories:mass stor- program. age, and input/output. In the former, Control Data FCP works with a two-level data base. The prime provides large disks ofup to 200 million bits of level, called the Digigraphic List, is created bythe storage, 4 million character drums with 2 million computer system as a direct result of light-pen and characters-per-second transfer rates and the fami- function-button manipulation. This,concurrent liar disk packs.Continued advances in thisarea with manual light-pen and function-buttonopera- are planned. However, it is in the area of graphic tions, is used to generatea secondary data base I/O devices where new ground is being brokenfor which creates a graphicresponse on the cathode- design engineers.Document readers, high-speed ray tube and verifies the input commands.The CRT graphic hard-copy devices, filmscanners and geometric precision of the primeDigigraphic List conversational mode light-pen andscope devices are is high.It is equivalent to constructing graphic currently being installed,or, in the case of the film images on an X-and-Y coordinate gridstructure scanner, are being made ready for production. with approximately eight millionintersections on a Control Data's man-machine graphic systemsare side.If the coordinate grid structure isdesignated being developed atitsDigigraphic Laboratory as having spacing of 0.0001 of an inch between where the aim is to developan information-pro- each coordinate line, the master grid isequivalent cessing system which wouldmerge the best attri- to a working area of approximately800 square butes of film-based graphic systems with inches with an addressability down to0.0001 of an those of inch. digital-computer based systems. Thecross-over me- The secondary data base, used forconsole dia between the image world and thedigital world display and film recordingpurposes only, consists evolved as the cathode-ray tube.These display of a coordinate grid structure with4,096 intersec- images are suitable for film recordingand man- tions on each X and Y axis. image communication, and yetare easily driven The dual data-base concept facilitatesstandardiz- by digital devices. ing the prime information-manipulationtechniques The term "Digigraphic "---a registered trademark by separating the output representation(CRT dis- of Control Datarefers toa broad range of hard- play, X-Y plotter, photo-recording)from the main ware and software systems which usually involves: graphic data base, the Digigraphic List. 1 On-line, real-timeuse of a computer system by Digital-graphic techniquesare generating appli- a man using a cathode-ray tube with lightpen and cations in three majorareas : computer-aided de- function-control buttons, frequently operatingin a sign, computer-based research activities,and spe- time-shared mode with other consoleson the same cial types of command and controlsystems. computer while workingon the same, or a differ- While extensive use is currently beingmade of ent, data base with thesame, or different, applica- computers in engineering design, mostconventional tion program. computer applications are in post-designanalysis 2. Digital manipulation ofa data base contain- and evaluation. Few existingapplications involve the actual creation of ing geometric descriptions, alphanumericsand logi- a design on-line in real-time. cal linkage information pertainingto the geometric Indee 1, the extensiveuse of computers in the post- configuration of the image,as well as category indesign role of evaluation has createdmany "closed- formation about the characteristics ofthe physical shop" computer complexes wheredesign function entity the image represents. users are, in many cases, forced into situations in which 3. Extensive use of graphics andsymbols as the parameter-evaluation processing requires primary turnaround times ranging frommany days to, at man-system,conn-communicationlan- best, overnight. guage. The use of Digigraphics The Digigraphic System 270 consists in computer-aided de- of a console sign is aimed at providingengineers and scientists utilizing a large, 22-inch fiatfacecathode-ray tube, with graphic communication a controller and a combined auxiliary memory/buf- and computational back-up to perform both theactual design and the fer combination.The console is equipped witha design analysis and evaluation light pen consisting of in an economical a thin, 48-inch fiber optics manner while on-line with a time-sharedcomputer cable which pipes light from thescreen face into system. 142 ..1.640.160.11.114,5116&1116

FCP eliminates the need for a user to learn the the management of a three-dimensional cubs struc- complex technology of display programming. Users ture will, in the future, make feasible significant easily establish communication between the console expansion of the application base.The construc- display and their application program through sim- tion and manipulation of three-dimensional repre- ple FORTRAN statements to FCP. Without such sentations of physical entities must be considered a combined hardware and software approach, a from two aspects:(a)the management of the 4. user could become buried under a graphic-program- fundamental data base which includes the coordi- ming requirement which might exceed the complex- nate point (X, Y, Z) information about the physi- ity of the very application program it is designed cal entity; and (b) the creation of suitable images to be used with. for viewing various projections of the physical en- In computer-aided design application, teams of tity within its 3D, cube-like coordinate structure. engineers and scientists, working in separate geo- (a). Data-base management. graphical areas, work on a common geometric rep- The data-base model is created by engineers resentation of the end item being designed.This and scientists, working with a three-dimensional "model" constitutes the prime data base (Digi- coordinategridstructure, who specifygeo- graphic List) on which all parameter testing is ac- metric parameters and pertinent notes required complished by the users' application programs. by the end user of the model to establish the The data base also contains pertinent notes, cate- proper linkages between related image items. gory information and geometric-linkage data so In addition, the creators specify the proper cate- that various subsystems can be called up and dis- gories for determining the functional or end-use played on the console without unnecessary clutter. classification of all items in the model. Since the model is maintained in a digital form, (b). 3-D display. specific views or sections can be transmitted long It is important to recognize that in working distances over conventional telephone lines for re- with 3-D techniques, the secondary data-base construction and review at remote locations. manipulation necessary to present a 3-D perspec- The Digigraphic List also serves as the main tive on the screen of a 2-D cathode-ray tube is source for documentation on the item created : a separate problem from the management of the A. It makes feasible the use of soft-copy devices 3-D data base itself.Currently there appears to (CRT's) to view the model in any desired level of exist much more know-how on how to manage detail from remote locations thus reducing the re- and process 3-D coordinate point information and quirements for maintaining hard-copy references at its logical linkages than there does about how to end-user locations. generate accurate views of this same information B. Thelist-structure techniqueslinking both on a CRT console. Such demonstrations as rotat- graphic images of an end item with a narrative on ing a 3-D object on a CRT may or may not its physical characteristics and identification num- provide insight into the structure of the real en- ber make parts-accounting an automatic by-product tity, depending upon the simplification required of the graphic creation process. in order to generate the display image when, for C. With only one small model, represented by the example, the view involves complex surfaces. Digigraphic List, to which all end users will have 2. Scanning.Precision cathode-ray tubes Wriii immediate access, regardless of their location or exceptionally high resolution now make it feasible the nature of the documentation they desire, engi- to, in effect, give the computer an eye. In this tech- neering change-order procedures become signifi- nique, a film clip or aperture card is placed in front cantly simpler. of a suitable optical system and cathode-ray tube. D. By saving all changes in part descriptions or The computer places the CRT beam in a certain X- their images in chronological order, through the and-Y location on the face of the cathode-ray tube. Digigraphic List techniques, the user may recon- The light passes through an optical system, through struct the history of the design for reliability analy- the film image, and into a light detection unit be- sis and for field modification. hind the film.If the beam encounters a dark area on the film, this change in light transmission is SIGNIFICANT NEW DEVELOPMENTS perceived by the light sensor. The information for IN THE DIGITAL-GRAPHIC FIELD each discrete location which the computer program There are two major new concepts currently ma- probes is passed back to the computer program so turing which will have a significant impact on Digi- that it can make decisions on the next location it graphic applications : wishes to examine.Selective searching, based on 1. Three-dimensional data-base manipulation and some pre-programmed plan, soon produces enough viewing. Expansion of the initial Digigraphic List information for the scanner program to begin creat- concept from the flat-grid coordinate structure to ing a Digigraphic List containing pertinent coordi- 143 nate-point information about the image on the film. munity.Several of these program systems have Thus, in theory, it would appear feasible to create been developed on an R-and-D basis to evaluate Digigraphic Lists representing existing graphics the usefulness of computers to as yet untried engi- recorded on aperture cards.In truth, however, the neering activities.As a means of assuring that scanning procedure amounts to actually scaling the these program systems can be truly evaluated, Meis- film image in order to obtain pertinent dimensional con also performs contract consulting engineering information.Hence, the information is not valid to prove their worth. The largest program of this enough, dimension-wise, to be used in constructing type to date is called Constructs. a high-precision geometric model of the part de- Upon completion, Constructs will be an inte- scribed by the drawing. Since the drawing has been grated-design and fabrication-detailing system for digitized, however, it may be displayed on a cathode- steel structures.The design portion of the system ray tube and conventional light-pen operations begins after a framing plan has been developed by may be executed to produce changes in the image. the engineer.The output of the design becomes This modified image can subsequently be recorded input to the detailing phase whose function it is to on film from the face of the same, or an equivalent, generate drawings equivalent to those produced precision cathode-ray tube.As a drawing-change manually for shop fabrication.Use of the Digi- technique, considering the amount of capital equip- graphic capabilities has not, as yet, been integrated ment involved, this type of application is considered into the system planning, although it will be essen- by many to be economically marginal. More prom- tial for the extension of the system to include the ising applications in scanning film images exist in important phase of structural layout.Under cur- the analysis of photographs where precise coordi- rent development are the following three parts of nate-point information is not critical or where the design system. advantage can be taken of the ability of the system Part 1 provides structural analysis for two- and to detect tonal changes, as, for example, in the some three-dimensional indeterminate frames with analysis of bubble-chamber particle traces, com- both horizontal and vertical loads.Until implemen- puter-based analysis of chest X-rays and in the tation of Digigraphics, the system requires that the analysis of slides containing organic tissue sections. engineer locate the positions, but not the size, of the Real-time scanning of dynamic phenomena also structural members through card input. Both pinned promises some fascinating new applications for digi- and fixed connections will be included within the tal computers. Here an actual occurrence is analyzed scope of the analysis.Resulting from the stability by a computer and a precision cathode-ray tube, analysis are joint moments and shears, which are in real-time, as, it happens.The computer pro- rearranged so that they will be suitable for com- gram examines the changing phenomenon and cre- bination with the results of Part 2, the gravity ates a Digigraphic List so that a human operator floor-load distribution program.Adequate print- can subsequently view the sequence of occurrences out procedures allow for inspections of the data to on the face of a console CRT at a later time. Since assure the user of the proper program response each light probe by the scanner requires only about when necessary, and to allow the user to conveni- twenty-five millionths of a second, a considerable ently modify the system decisions. number of moving phenomena can be observed in Part 2 of the system distributes both live and real-time and later regenerated on a console CRT dead gravity loads from floor to beam and from at a considerably slower rate for analysis purposes. beam to column.To accomplish this, the program In some cases, the computer program may also elect develops a digital model (Le., data list)of the to activate a photorecording device to make a film structure so that loads may be transmitted between image, at some point in time, of a continuous oc- all affected members of the structure.All of the currence which it has been continually monitoring. loads distributed to the members from the floor will In one such application, a computer, through a pre- be combined with loads from the stability analysis cision CRT, watches a bubble chamber and makes of Part 1 to form a compilation of all loading con- pictures of particle traces which are of significant ditions for the design of each member. interest, but ignores those traces its computations A column-schedule program will be implemented indicate are of no significance. to calculate preliminary loads from bents and floors for the selection of column sections between each APPLICATION PROGRAMMING CONSTRUCTS pair of floors.The printed results allow the engi- In addition to these activities of Control Data, neer to set splice points and pick desirable column there is one other that should be of interest to this sections for the whole structure.It is at this point conference.It is the operation of an engineering that optimizing techniques may be applied to select software development subsidiary, called Meiscon the least-cost column arrangement.The column Corporation, whose purpose it is to prepare useful schedule is produced at an early date so that the application programs for the engineering com- substructure design may proceed. 144 Part 3 of the system analyzes each member of the as a draftsman draws upon his experience, toselect structure as a free body with both input and previ- and design connections.In phase three, the drafts- ously calculated loads applied to the member. From man decides how to portray schematically thevari- this loaded free-body condition, shears and moments ous pieces and dimensions of the structural mem- are calculated and an economical section ,is selected. bers in a manner commonly understandable by If the engineer chooses, he may ask for a range the fabricating shop. Here he selects the schematic of members rather than for one. pictures, dimension lines, notes, etc., that are re- Part 3 can also handle beams, columns and brac- quired to produce the complete detail.Constructs, ing.It is comprehensive enough to meet all of the again through stored program logic, provides these requirements of the new AISC code.It has been same functions. The last phase involves the physi- implemented to provide the engineer with either cal movements that are necessary to portray the final selection as to member size, or a range of feasi- draftsman's mental picture by movements of his ble sizes for the engineer's selection. pencil on the paper.Constructs now generates the plotter commands which, by guiding a drafting de- AUTOMATED DETAILING SYSTEM-- vice, produce these pictures in exactly the same CONSTRUCTS manner. The details of the final drawings are com- The second portion of Constructs is primarily of parable in all respects to those produced manually. use to the steel fabricator.Inpu* may be the out- The quality of the automatically generated draw- put of the design phase or it may be independently ings exceeds that of the manual, both in clarity generated.The general concept of the Constructs and regularity. system is to provide by computer, program and The major program functions are as follows: plotter the capability of preparing structural steel INPUTThe basic data for Constructs is that fabricationdrawings. These drawings require information portrayed on the engineer's design more than the simple placement of lines on paper. drawings.Data regarding each member's relative The engineering concepts to fit the structural ma- location within the framing plan (i.e., 6' from Bent terial and attached secondary connection material C; 2'8" from Floor 7; etc.) are input along with a are an important part of detailing and are included declaration of the absolute location of the structure's in the Constructs system.The inclusion of these bend and floor plans ; the user need only input what detailing operations within the program allows for he sees on the design drawings, and does not have more efficient use of a computer and plotter by to make tedious and error-prone mental summa- avoiding wasteful man-machine communication and tions to compute coordinates.The man preparing thus provides maximum return on the use of the the input need not be concerned about connection equipment. design or dimension determination since the system Specifically, the Constructs system is capable of handles these functions. handling the main members of AISC, Type-2 steel The location of each member, its shape, size and structures.The basic input describes the general orientation are also input.In addition, a declara- features of the structure, the specific detailing ay tion of the members which support the piece in connection design criteria, and the shape, location question is also supplied.Subsidiary information and support conditions of each member. After the regarding the shop and field methods of fastening initial input, the system is designed to operate with is also input.This data is written on preprinted little, or no, user intervention or additional input. input forms.It is then keypunched and a mag- The user is only required to review error messages netic tape is prepared for input to the system. at intermediate points during the operation. VALIDATIONAn extensive series of valida- The Constructs detailing system can be thought tion tests are supplied to the input data to keep of as being divided into four basic phases.Each detectable errors from passing through the system. of these phases simulates the thought processes Member sizes are checked for validity against the and physical actions of an experienced draftsman. AISC table.Abnormally long, short or skewed In the first phase, the draftsman studies a set of members are brought to the viewer's attentionas engineering drawings, assimilating and categorizing possible error conditions. The support declaration what he sees and tentatively planning what he will made by the user is verified for consistency, andan do.Constructs looks at a set of plans by reading "internal model" of the structure is created to de- punched cards and storing into memory pertinent termine the design sequence of the members. The data regarding the structure. The next phase, that declaration of shop- and field-fastening modesfor of determining how the members can be intercon- the various classes of membersare also cross-vali- nected, is essentially a handbook and mechanical- dated to avoid inconsistencies in fabricationpro- fitting operation for the draftsman.Constructs cedures.If any error conditionsare uncovered, uses stored program logic and AISC tables, just printed messages indicating thesource and cause 145 a, ail,

of the error are given. When necessary, the user SCOPE OF CONSTRUCTS adds correction cards to the original input and re- Constructs was developed to provide a broad abil- runs this phase until the data is correctly formed ity to handle the major members of beam-and-col- and the system acknowledges this condition. umn type structures.All of the specifications, de- CONNECTION DESIGNAfter the input has sign procedures, and rolled sections as found in the been assimilated, it is sorted by the system into the American Institute of Steel Construction, 1963ver- proper detailing sequence for submission to the sion, are incorporated in the Constructs logic. The design phase.In this operation, each member is system allows for any combination of field bolting interrogated as to shape, location and supporting with shop welding, riveting or bolting. members.The individual joints are designed by the design-and-detailing procedures recommended There are over 2,400 combinations of input al- by the AISC code and by additional parameters of lowed which can generate tens of thousands of detailing criteria input by the user.All the basic picture combinations. In addition to preparing the facts regarding each member, its end and interior drawings, Constructs will produce a computer-listed connections, cuts, copes, fasteners, etc.,are deter- bill of material for each drawn sheet anda listing mined and are stored in preparation to plotting. similar to a shop list or mill order for all members GRAPHIC PREPARATIONThis phase inter- input to the system. prets the stored data for each member and itsasso- The drawing procedure, including both pictorial ciated connection iv.terial, and determines the representation and dimensioning method, which is manner in which the member should be represented used in Constructs attempts to satisfy the majority graphically. Each piece of data is "looked" at and of all structural steel detailing systems. a judgment is made as to which subroutine will have to be used to provide thenecessary graphic In addition to Constructs, Meiscon has prepared views.The cumulative effect of all the "data-to- a highway design system, extended from some of picture" determinations provided the total detail the techniques of the DTM system developed at drawing. The output of these subroutines consists MIT. Of interest is the terrain definition technique of simple declarative commands suchas "draw clip," which allows for relatively close definition of the "draw weld symbol," "draw hole pattern," and indi- central ground, no matter what alignment is chosen. cates where these actions are to take place.The This system produces, as output, drawn pictures of effect of this phase is thesame as speaking to a terraincrows-sections,roadway alignment and draftsman at some remote locationover the tele- earthwork-mass diagrams.Study is currently un- phone and telling him the standard draftingcom- derway to generate an even more precise model ponents which will make up the picture. from aerial photos so that final designcan be ac- DRAFTING TRANSLATOROnce the basic complished from the same terrain data. drafting components have been selected, theyare Besides these design systems, a number of other then transformed into actualpen movements as developments are being pursued utilizing computers required by the language of the plotter.Since to generate drawings for process piping and for each plotter has an individualized commandstruc- manipulated three-dimensional objects. ture, a general plotting language capable of manipu- As indicated previously, the restriction of time lating graphic entities, suchas "rotate clip," along precludes a discussion in depth of anyone of the with a post-processor for the particular plotter, has many activities being pursued at Control Data.I been developed. The general plotting languagecon- think this short synopsis will indicate toyou that sists of all the basic operationsany plotter would the foundation of problem-solving capability of tobe called upon to perform. day's engineer is already broad and is growing.

146 4/..diksi.1.134,ftadma

CRT DISPLAYS INENGINEERING DESIGN M. A. FORD Digital Equipment Corporation College Park, Md. barrier has the INTRODUCTION The third solution to the language very interesting sidebenefit of providing a means of The digital computer is a very powerful engineer- instantaneous feed-back. A designer maysketch ing-design tool. As the cost of high-speed scientific his structural steel assembly on the faceof the CRT computers comes down towards the $20,000 figure, and place test loads where he wishes.If the stresses and the speed and software performance of such on the individual beams arenot to his liking (a computers increases to 1 or 2 iisec add-time, it be- fact he can ascertain right then and thereby read- comes almost credible to thinkin terms of one coming the values displayed next to thebeams), he puter for each engineer.Similarly, as time-sharing can change the designwith the light pen and key- becomes more in vogue, and very large, very high- board until he sees that his stressrequirements speed computers become simultaneously accessible have been met.This on-line design feed-back or to as many as 100 or 200 engineers, one can certainly design evaluation is possible with a CRTdisplay envision the possibility of one time-shared terminal terminal connected to a computer. for each engineer. In fact, only by means of a CRT andlight pen (or The usefulness of any computer or time-shared an equivalent interactiveI/O device) can the man- terminal to an individual engineer is limited by the machine language barrier be overcome and the com- language barrier which exists between the engineer puter truly be used by the engineer as adesign tool. and the machine. This language barrier might be Without such I/O equipment, the computer is no overcome by any of the following means : more than an expensiveslide rule. A. Make each engineer as proficient in program- Where does the Digital Equipment Corporation ming for problem-solving as he is in calculus. fit into this philosophy?Strangely enough, DEC B. Construct problem-oriented programming lan- has been manufacturing and supplying CRT sys- guages which permit basic designproblems to be tems to the engineering and scientificcommunity expressed by the engineer in a familiar language. for over five years. In fact, DEC was the first com- FORTRAN is a general example of this approach. puter manufacturer to sell CRT displays tothe non- C. Utilize a language common to both computer military market and is now one of the leading sup- and engineer which is a drawing or sketch, i. e., use pliers of display systems to scientists and engineers. CRT displays with light pen and keyboard as a In addition to CRT systems, DEC alsomanufactures computers. problem input device. a line of small scientific The first solution is neither possible nor desirable. THE PDP-8 In a scientific world where there is an information As I mentioned earlier, there are two possible explosion taking place, it seems to defeat the pur- means by which anengineer can have hands-on pose if all engineers have to becomeprogramming access to a computer : atime-sharing terminal, or experts in order to realize the advantages of com- a small scientific computer.DEC manufactures such puters. The ?.._% cond solution to the language barrier a computer which iscalled the Programmed Data is a practical one and one which is, in fact, being Processor-8 (PDF -8). used today. However, its practice does r.ot take full The PDP-8 makes available to engineering, scien- advantage of the computer as a design tool.Use tific, and educational users a compact but complete of problem-oriented languages implies the use of general-purpose digital computer with a high-speed, the computer as an expensive slide rule. A particu- random-access, magnetic-core memory.It is in- lar problem is given by the engineer to the com- tended for use in data processing or as a control puter and the computer calculates the solution or elementinon-linedata handling,experiment solutions. The engineer must then place this solu- monitoring,or a process-controlsystem. The tion in the context of his design and decide what standard PDP-8 is a complete hardware and soft- to do next.In practice, this may result in a new ware system, ready toperform general-purpose set of parameters and another pass through the computations and/or control operations on the day computer. The feed-back time most likely is on the it is delivered.Optional data-processing peripheral order of days. Whereas this is a vast improvement equipment or special control devices are easily con- over actually using a normalslide rule, it is hardly nected to the PDP-8 to form a special-purpose instantaneous feed-back. system.Instruction format is simple and logical. 147

iarfcCO-,APPMFARARMIRiit One instruction format exists foreach of four THE PDP-7 groups, so that format of individual instructions The PDP-8 is a 12-bit machine andthus has some need not be memorized by theprogrammer. A limitations as to programmingease and computa- complete software system accompanieseach hard- tion precision.Although double- and triple-preci- ware system and contains FORTRAN, MACRO-8 sion arithmetic routinesare available when needed, assembler, a library of mathematicalsubroutines, the 18-bit PDP-7 ismore suitable if a large portiov and utility and maintenance programs. of the engineering computationrequires such pre- The PDP-8 performs binaryoperations on 12- Gr cision. 24-bit 2's complement numbers.The 1.5-psec cycle DEC's Programmed Data Processor-7(PDP-7) time of the machine providesa computation rate is a general-purpose, solid-state,digital computer of 333,333 additionsper second.Addition is per- designed for high-speed data handlingin the scienti- formed in 3.0 psecs (withone number in the accumu- fic laboratory, the computingcenter, of the real- lator) and subtraction is performedin 6.0 psecs time process control system.PDP-7 is a single (with the subtrahend in theaccumulator).Multi- address, fixed 18-bit word length,binary computer plication is performed in approximately315 psecs using l's complement arithmeticand 2's comple- by a subroutine that operateson two 12-bit num- ment ,rotation to facilitate multiprecisionoperations. bers to producea 24-bit product, leaving the 12 most Cycle time of the 4,096-word,random-access, mag- significant bits in the accumulator.Division of two netic-core memory is 1.75itsecs, providing a compu- 12-bit numbers is performed inapproximately 444 tation rate of 285,000 additionsper second. Nees by a subroutine which producesa 12-bit The bale PDP-7 includes theprocessor (with quotient in the accumulator anda 12-bit remainder operator console) ; 4,096-wordcore memory; input/ in core memory.Similar multiplication anddivi- output control with device selector(up to 64 I/O sion operationsare performed by means of the connections) ; information collector(seven 18-bit optional extended-arithmeticelement in approxi- channels) ; information distributor (six18bit chan- mately 21 and 37psecs, respectively. nels) ; program interrupt; datainterrupt ;f I/O skip facility; I/O status check; and real-timr., Flexible, high-capacity,input-output capabilities clock. A of the computer allow it high-speed paper-tape reader (300 cps),high-speed to operate a variety of paper-tape punch (63.3 cps), and peripheral equipment. Inaddition to standard Tele- KSR-33 teletype and perforated-tape printer (10 cps)are standard input/output equip- equipment, the system is ment with the basic PDP-7. capable of operating inconjunction wth a number of optional devices suchas high-speed perforated- Interface to the PDP-7 allows fast parallelinfor- tape readers and punches, mation transfer between thecomputer and a variety card equipment,a line of peripheral equipment. printer, analog-to-digitalconverters, cathode-ray In addition to the tele- tube displays, magnetic printer, keyboard, and high-speedperforated-tape drum systems, andmag- reader and punch supplied with the netid-tape equipment.Equipment of special design basic computer, is the PDP-7 optional peripheralequipment includes easily adapted for connectioninto the PDP-8sys- tem. The computer isnot modified by the addition magnetic-tape equipment, card equipment and line of peripheral devices. printers, serial magnetic-drum storage,cathode-ray tube displays, a data communicationsystem, and PDP-8 is completelyself-contained, requiringno analog-to-digital converters.Special purpose I/O special power sourcesor environmental conditions. equipment is easily connected usingan interface A single source of 115-volt,60-cycle, single-phase of 'standard DEC modules. power is required to operate the machine. Internal CRT DISPLAYS- THE 338 power supplies produce all of the operatingvoltages required. FLIP CHIP modules The 338 Programmed BufferedDisplay is a pre- utilizing hybrid sili- cision, incremental display con circuits and built-in provisions formarginal system. There are three checking inaur,:, reliable operation elements in the 338: The PDP-8,the display proces- in ambient tem- sor, and the display tube and its associated peratures between 32 and 130degrees Fahrenheit. electronics. The combined capabilitiesof this buffered The low price ($18,000and up) and highper- display with processor otter thedesign engineer an formance capability of thePDP-8 certainly places unusual degree of versatilityand economic ad- it in a position ofconsideration for any engineering vantage over conventional buffereddisplays. design requirement wherelarge-computer power is Interactive console not required.In fact, it becomes economicalto As I mentioned before, the have on hand such language barrier a computer with FORTRAN is really overcome when the designercan receive capability for thepurpose of experimentation with instantaneous feedback froma change in design design-problem solution techniques. and can readily express this change indesign by 148 using the light pen and keyboards connected to of 250 itsec per point.Magnetic-deflection and a display processor to alter his design picture. The focusing techniques result in uniform dynamic reso- 338 may be used extremely effectively as an inter- lution over the entire usable area of the tube face active display input/output device, either connected and maximum spot size of approximately 0.15 of an to a larger computer, or standing by itself.The inch.Construction is solid state throughout with existence of the PDP-8 processor between the dis- excellent stability. play and. a *ger computer offers several other The 338 is a stored program system that was advantages : designed with the programmer in mind.Some of 1. Hardware interface costs are reduced since the the display capabilities are: job of controlling the interface between the 338 and 1. Data-acquisition cycle stealing - The display re- the larger machine can be accomplished by PDP-8 ceives 12-bit data and control words from the PDP-8 programming, rather than by more expensive hard- memory via the PDP-8 data break channel.The ware control logic. data break channel is a high-speed (1.5 osec/word), 2. Routine tasks such as light-pen following, direct-access channel that passes words to the dis- character generation, rotations, and elementary-data play transparently to the program in execution. structuring can be, handled by the PDP-8 program- 2. Automatic scissoring - The display can be pro- ming, eliminating frequent interrupts to the main grammed to represent a 9-%-inch square window computer. viewing a 75-inch square drawing.This window 3. In addition to these advantages, the 338 offers can easily be moved around the drawing by simple even more usefulness in the case of remote display translation of coordinates.Only the area of the terminals to a large time-shared system. Since most drawing corresponding to the display window will teleplivne line connections between remote terminals be seen. and large computing centers have a very low band- 3. Multilevel subrouting - The control state per- width, it would take almost a minute to transfer mits the display to jump from accessing one loca- a complicated display file from the central machine tion in the PDP-8 memory to any other. When it is to the display if the information were formatted for desired to jump to a display subroutine, the return the display prior to transmission to the remote dis- address is automatically stored in a push-down list. play.If the PDP-8 processor can be used at the 4. Man-machine interaction - The use of push remote terminal to appropriately format the display buttons allows the operator real-time control of the data as it comes off the line, the quantity of data display and computer programs.The high-speed and hence the transmission time can be reduced light pen permits the operator to establish a refer- by several orders of magnitude. ence within his program to a specific point displayed These are some of the reasons why DEC feels on the screen. The full duplex TTY Model ASR-33 that the 338 is extremely useful to the user who is provides a keyboard and printer, and also a 10-cps concerned with computer-aided design. Let me now paper-tape punch and reader. describe the 338 in a bit more detail. 5. Character generation - In addition to line gen- 338 Details eration, the display hardware can display characters The 338 Buffered Display System permits rapid specified by 6-bit codes. Each character is displayed conversion of digitalcomputer data into graphic and in an average of 22 Nees. tabular form.Its combined capabilities offer the 6. Communication with display registers - The user an unusual degree of versatility and accuracy. contents of the X-Y position-registers and the display address-counter can be read into the accumula- A self-contained unit with built-in control and tor of the PDP-8 via IOT instructions.Likewise, power supplies, the 338 requires only logic-level the PDP-8 can load certain registers in the display inputs for operation and may be easily connected which are used for starting the display or setting to any digital_ system as a buffered display with up initial control conditions. processor, or it may stand alone as a powerful computer-driven display system.Location of any de- Display modes sired point may be specified by any of the 1024-X The display logic is divided into the control state and 1024-Y coordinate addresses contained in a 9-%- and the data state.In the former, the 12-bit words inch square on the tube face.Discrete random are decoded as instructions to the display logic. points will be plotted at a rate of 35 pm per point, These instructions can set the parameters and mode in point- or graph-plot mode.If the sequences of which affect how the data is displayed; or they can points lie near one another, the plotting rate will cause a jump, jump to subroutine, or skip on display be about 6 psecs per point. In the other modes, the flags.Data-state words direct the deflection ampli- plotting rate is 1.2 psec per intensified point. Non- fiers as to the number and direction of mode. Ex- intensified points, however, are plotted at a rate amples of data modes are:increment mode, point 149 mode, vector mode, and graph-plot mode.Each In both applications it affords the engineeran oppor- data mode is also equipped withan escape mechanism tunity to present his problem to thecomputer in a to return the display to the control state. most efficient way, and to have his resultsindicated to him in a most meaningfulway. This truly repre- In summary, the 888 Buffered Display Systemis sents elimination of the man-machinelanguage a powerful scientific computing system by itself,or barrier and opens theway to effective use of the an economic terminal for a larger computer system. digital computer as a design tool.

150

,16.--v.+411. - PROSPECTS AND REQUIREMENTS

SECTION F

CHAIRMAN: N. M. NEWMARK, Heal, Dept. of Civil Engineering, University of Illinois A

PRECEDING PAGE BLANK- NOT FILMED COMPUTERS IN ENGINEERING DESIGN vs.DESIGN OF COMPUTERS L. A. ZADBE University of California Berkeley, Calif. I regret very much that the necessity of attending systems, nuclear reactors, etc.Finally, there are another meeting has prevented me from coming those who feel that we should go much farther than here earlier and listening to the many informative giving a little flavor of design in various engi- papers which have been presented on the subject of neering courses and should make a concerted effort the impact of computers on engineering design.I at teaching design per se, realistically and syste- ti believe I can take it for, granted, however, that the matically.I take it for granted that this point of Conference has established beyond any doubt that view is well represented at this Conference. As for the advent of the computer age is having a great myself, I fall somewhere betwevi the first and impact on engineering design, and, consequently, second of the above groups, since I am far from on engineering education. convinced (perhaps I would be more convinced The acceptance of this fact does not imply, how- had I listened to the papers presented here during ever, that we have a clear view of the role of com- the past two days) that Zesign is a teachable sub- putiers in engineering education, nor does it settle ject. the question of what place, if any, courses in Having defined my prejudices, I should like to engineering designwhether computer-oriented or make a few remarks concerning the role of com- nothave in engineering curricula.I may be puters in engineering design and then turn to a making myself a persona non grata at this Con- question that is closer to my heart, namely, the ference by raising the latter question. Be that as training of engineering students in computers and it may, I must confess to some doubts that design their use. can be taught effectively in the classroom, even Extrapolating present trends, it appears certain when it is heavily spiced with computers.My that in a few years' time engineering design will doubts are based on the fact that realistic engineer- become a highly computerized process with the ing design requires a great deal of know-how, and design of a standard system reduced to locating that, in an era of rapid technological advances, an appropriate program in a file or a library, feed- know-how can be acquired only through practice ing the system specifications to the computer and and experience, and not through teaching in the letting it produce a detailed design or, perhaps, classroom. instructions to computer-controlled tools or assembly As I see it, there are roughly three schools of lines.In the case of a design of a non-standard thought in regard to the place of design courses in system, the computers are likely to be used in an engineering curricula.First, there are those educa- interactive mode, with the designer employing the tors who argue that, because of the rapid deprecia- computer in a sequential manner (probably on a tion of specialized engineering training brought time-shared basis) involving back and forth flow about by changes in technology, engineering cur- of information and commands between the computer ricula should aim primarily at providing the student and the designer.This, of course, is an oversimpli- with basic tools of long-lasting value, such as mathe- fid view of the future, but it will suffice for pur- matics, physics, chemistry, mechanics, thermody- poses of our discussion. namics, system theory, etc., and only secondarily Unquestionably the computer-based methods of aim at giving him competence in specialized subject design will be much more powerful and effective areas in his major field of study.Implicit in this than the techniques employed today.Does this view is the assumption that the training of an mean, however, that design will become a teachable engineer does not stop upon graduation, and that subject? Not necessarily, in my opinion, since it the know-how and experience needed for design can isand will continue to bevery difficult to ab- be acquired much more efficiently in an industrial stract from the varieties of specialized design rather than an academic environment.Second, procedures a body of concepts and techniques with there are those who argue that we should go beyond wide applicability.Without such a body of con- providing our students with a broad training in cepts and techniques, engineering designwhether the fundamentals and give them some exposure computer-oriented or not--could not qualify as a to design in various specialized subject areas, e. g., subject that could or should be taught in the univer- in courses on electronic circuits and devices, thermal sity. 158 -,,.

What effect, then, will the computerization of audience, Newman Hall, its ExecutiveDirector, who design processes haveon engineering education ? was instrumental in organizing the Committee and Clearly, all engineering students, regardlessof their providing it with support by theCommission) to field of specialization, will have to be proficientin aid and guide electrical engineeringdepartments in the use of computersas information-processing the strengthening of theirprogram and course systems. What is not clear, however, is theextent offerings in computer and informationsciences. to which engineering students should be taught not It is interesting to speculateon the effect which just how to use computersas aids to design, but also about their internal structure, about the the establishment ofa computer science department principles at a university is likely to haveon the electrical of numerical and non-numericalprogramming, department at that university. about automata theory and formal In my opinion, the languages, about two can coexist if there isa framework for coopera- heuristic programming and artificialintelligence, tion between them in the form of and, more generally, aboutan array of subjects joint appoint- ments and a reasonable division ofresponsibilities which comprise what are currently referredto as for teaching and research in computer sciences. computer sciences and related fields.However, at those universities in This question is not yet a pressing one in fields which the main responsibility fortraining in infor- other than electrical engineeringa fieldwhich mation processing will be in thehands of computer bears a special relation to computersciences by science departments, electrical virtue of its long history of involvement with engineeringdivest- com- ed of its activities in informationprocessing, except munication and information processing.Indeed, all possibly on the hardware levelislikely to be a large-scale information-processingsystems are, in much less important and viablefield of study than the main, electronic in nature.This is a conse- it is at present and has been quence of the fact that, in an information-proces- in the past. Putting aside jurisdictionalquestions, I should eing system, a datum is representedas a state of like to draw attention to one or more of its elements.Thus, to process inan area in computer formation at a high speed it is sciences, namely, heuristicprogramming and artifi- necessary to have cial intelligence, which is a system whose states can change rapidly, which in certain to have anap- preciable impacton engineering design in theyears turn requires that the information-storingelements to come. Some of of the system have low inertia.These considera- you may be puzzled somewhat tions explain why electronic components by my mentioning inone breath artificial intelli- and circuits gence and engineering design, since, having extremely short transition timesfrom one on the surface state to another play such at least, there does notappear to be any connection a basic role in high-speed between them. information processing. In fact, I believe that thereis a very close relationship between thetwo and that By virtue of their special competencein the heuristic programming design of electronic systems, electrical and artificial intelligence engineers willin the not distant futureplayan important are naturally concerned not only with theuse of role in engineering design. computerstas are all other branches of engineering Let me explain why I feel but, more importantly, with their this way.In recent design and years, engineering and, in particular,electrical operation.However, althoughno one questions engineering, has tended the central role of electrical engineering to become increasingly in com- abstract and mathematicallyoriented.This trend puter hardware, there isa lack of unanimity on the is especially pronouncedin such areas of electrical extent to which it shouldconcern itself with com- engineering as system theory, puter software. information theory, There are someand I fall into control theory and networktheory, with the result this categorywhoargue that, because of their that most workers in these long-standing involvement in information areas, whether in uni- proces- versities or industrialresearch laboratories, tend sing,electricalengineering departments should to concern themselves onlywith those aspectsof greatly expand their teaching andresearch activities system design and analysis in computer sciences and give which are susceptible their students a of mathematization, that is,of yielding provable thorough preparation forcareers in this and related assertions in the form of fields.There are others who maintain lemmas and theorems. that com- Unfortunately, the real worldwe live in is much puter software should be theresponsibility of too complex and fuzzy to admit _computer science departments, of which of being fitted to several are a precise mathematical framework.In practice, now in existence and more are being setup.It may this means that thereare many basic problems not be inappropriate at this pointto mention that in system design and analysiswhich are beyondthe a committee called the COSINE Committeehas reach of classical mathematical recently been set up under the techniques, and that sponsorship of the we should strive to develop otherapproachesnot Commission on Engineering Education(I see in the necessarily appealing toa classical mathematician 154 for their solutions. This thoughtwas expressed very computer- oriented or or not, should be included in succinctly by Professor J. Tukey of Princeton Uni- engineering curricula; I am convinced, like most of versity and Bell Telephone Laboratories, when he you, that all engineering students should receive said, in paraphrased form, "Whatwe need now are substantive training in the use of computers; and not so much precise solutions to approximate prob- that electrical engineering students, in particular, lems as approximate solutions to preciseproblems." should receive specialized training in computer Clearly, in saying this he implies that toomany of sciences and related fields, so that they could not us have been and are busy solving artificial prob- only make effective use of computers in their work, lems which bear little relation to reality and that but also serve as designers of computers andcom- it is time for us to get down to the less intellectually puter-like information-processing andcontrol sys- satisfying but more challenging problem of devising tems. approximate approaches to the solution of realistic problems. It is at this point that heuristic programming ROSENSTEIN: While Zadeh questions whether and artificial intelligence make contact with engi- there is anything to teach in engineering design, neering design, since, inessence, they represent an his comments have crystallizedsome misgivings attempt at using computers to "solve" problems that I have had forsome time. I now wish to which do not admit of precise mathematical treat- question whether there isany such thing as comment or do not have computationally feasible and puter science. By your definition, whatwe con- mathematically rigorous solutions.For example, sider computer science is 80per cent engineering, many' of the problems in pattern recognition, (and I hope you will concede forthe purpose theorem proving, game playing(checkers and of discussion that engineering isnot science) and chess),information retrieval, system identification, the other 20per cent is mathematics.I don't machine translation of languages, etc., fall into this pretend to be a very sophisticatedmathematician category. The common characteristic of these prob- but I do remember reading BertrandRussell who lems is that they involve inan essential way the said, "There areno truths in mathematics." procPss of abstraction, that is, the process of deduc- ing a property which defines the membersof a ZADEH: In the first place, letme say that com- class of objects.It is our lack of understanding puter science is a somewhat unfortunateterm. of the subconscious ability of the human mind to When you use it, it givesyou the impression that this set of subjectareas has to do primarily with perform abstractions whichare not well-defined the computer. I think that this is mathematically that is at the root ofour inability not really the to make substantial progress toward the solution case and a better term might be information of the problems cited above. sciences, or information processing.Computers as we know them today are just one example of A heuristic program is,as its name implies, a information-processing machines, and, in the fu- less than rigorous or invariably effective algorithm ture, transmission and processing of information for solving a particular problem.Generally, it is will be playing a tremendously importantrole in a program that is arrived at by heuristic reasoning our technology and in everyday life, including and is not guaranteed to always yielda solution, perhaps the kitchen, too. Let's put asidethe term when one exists, or givea solution that is optimal computer sciences for the moment and in some specified sense. say that there will be a complex of subjects tendingto do I believe that the really challenging problems in with processing, generation, andtransmission of engineering design in the futureare likely to involve information, and, if you take that field, the design of large-scale systems such I think as air-traffic you will find that a relatively small part of itis control, information retrieval for libraries, banking, susceptible of mathematization. Nowthere are space communications, etc.Heuristic programming a few subject areas of mathematical nature, like is certain to play an important role in the design automata theory, formal languages, and of such systems, because they so forth. are much too complex But there is something about these fieldsin which for analysis, not to speak of synthesis, by precise there is discreteness as contrastedwith continuity algorithmic techniques.It is for this reason that I which makes them less susceptibleto mathe- believe that although heuristic programming and matization and it is for thisreason that it seems artificial intelligence do not haveas yet much to to me that engineers arevery well suited to work contribute to engineering design, they will,even- in those areas.If you look ata typical course, tually, Abe accorded an important place in the design let's say a course in programmingsystems, it is of large-scale systems of various types. really very muchan engineering course. In fact, In conclusion, unlike most ofyou, I have some it is so much ofan engineering course that it doubts that courses in engineering design, whether sticks out like asore thumb because in it there 155 7V1P.M17YRIOMPWIFWPTIPINITIMPOPPIPPIrer...17.11,77s.

isn't a single equation,or theorem, or proof, variables involved.I think somebody who de- nothing of that kind.It is simply a. matter of signs a coffee pot is likely tobe much more describing ingenious ways of putting together conscious of the fact that he isdealing with heat, any info-I/nation-processing system. electricity, materials, stresses,strain, etc., than I think that it is clear that thatset of subject somebody who designsan air-traffic control sys- areas will be playing a very important role, but tem, or an automatic ticket-reservationsystem. there is some questionas to the extent to which Another point is that in thecase of large-scale engineering students should be exposedto that systems, there is a crucial problem field; of how to pass should they be given justa smattering of from the overall objectiveof the system to ob- training, or should they be givenquite a bit? jectives for the components,so that if you have And I don't really havean answer to this question a large system which hasa certain function to in my mind. perform (let'sassume that function is well-de- 1 ROSENSTEIN: I do agree withyou about informa- fined), then the question arises,"How does this tion.I believe it is one of themore important well- defined objective influence local objectives subactivities of the designprocess and we should for the components of thissystem; how do you teach it as such. set specifications for thecomponents in sucha way as to realize specifications for thesystem as SAUNDERS: My question hasto do with large a whole?"This is a problem that versus small systems. is present I'd like to ask Zadeh if in the design of subsystemsalso, but there it is he sees any really basic differencein the exercise much less acute. one goes through in designing avery small system I could go on listingmore differences, but I such as, say, an oscillatoron the head of a pin, think I'll stop at these two. versus the design of a very large system suchas a global communication system. Isthere really any KARNAUGH: I'd like to secondZadeh's plea for basic difference inyour mind between the intel- more understanding of large systems.One of lectual exercise that the engineerperforms? the features about largesystems which perhaps has not been emphasizedvery much is the practi- ZADEH: Yes, I think that thereare basic dif- cality of the computer ferences. As a system becomes heuristics or algorithms larger and larger, that are applied. Myrecent experiences in this the chances are that it will includea larger and field indicate that most ofus need a great deal larger number of differenttypes of physical sys- more experience with numerical analysis tems.It may include mechanical pro. or electrical grams. The ones we havevery often don't work systems, or there might be economicfactors, and when they're needed.Not only are there bugs so forth.In other words,as you increase the in the programs, but size of a system, the person using them does you also increase the variety of not really understand theirlimitations, and when its components. So that makesit more and more you try using half a dozen essential to be able to deal or more of them in with large-scalesys- tandem, in some kind ofiteration, you have noth- tems without going into the physicsof the coming but wasted timeand trouble.One of the ponents, and, to deal. with 'theecomponents es- things I feel the engineering sentially on a black-box community as a level.There is an whole really oughtto work on is satisfactory inevitable tendency on the part ofpeople who are libraries of suchprograms with rather clear concerned with large-scale systemdesign to be statements about the conditions abstract, without questioning what under which they is inside the may work, and tests tosee whether they meet black box or what is the physicalnature of the these conditions.

156

di.enawk PROLEGOMENA TO ASYSTEMS ANALYSIS OF EDUCATION R. E. MACHOL University of Illinois at Chicago Circle Chicago, Ill.

I understand that the subjectof this conference comparable in dollar cost, it isvery much smaller is education, and I should like to talkabout that for in terms of the number of peopleinvolved. a little bit. As you will see, computers will become I should like to examine the educational system involved at the appropriate time. de novo, as a systems engineer.Normally we At the beginning of the presentcentury, men should first examine the functions to be performed, were attempting to design flying machines, andsome but these are a little complicated andwe will touch people copied the techniques that birdshad used, on them below.Education is a communications with flapping wings whichwere appropriately problem, and as suchwe should look at the signi- warped to change their aerodynamiccharacteristics. ficant communications parameters. Some of you may haveseen movies of these orni- The first of these parameters is bandwidth.Of thopters, which seem quaintly amusing.It was course there is a significant difference,as every shortly recognized that technology hae,made avail- experienced teacher knows, betweenthe rate at able devices different from those whichhad been which we teach and the rate at whichour students made available to birds.I assert that education is learn. Our speech contains informationat the rate in a state analogous to the ornithopter.We have of a few bits per second, whichmust be reduced been excessively traditional in acceptingthe man- by redundancy due to our repetition ofsignificant to-man system of education which hasbeen handed points. Actually the performance ofa good teacher down from past ages. includes a good dealmore than this, in terms of Two or three thousandyears ago, the only feasible his inflections, emphasis, movements,and the like, method of educationwas a group of students listen- all of which add, at least insome cases, to the ing to a master who addressed themon the subject value of the educational process. On theother hand, of his personal knowledge, and who mightoccasion- the student probably does not learn inexcess of ally make a drawing in the sandto illustrate a one bit per secondanyone who thinks that the point.Subsequently a vertical surface, such average student learns more than three thousand as the bits of new, significant information blackboard, became available to replace thesand, in the classroom but we retain thesame basic environmental struc- lecture of one micro-century durationis unduly ture of a classroom in which thereare a body of optimistic.In any case, the bandwidthrequire- students being lectured to bya teacher. Our modern ments are modest; with television, magnetictapes, educational system still consists basicallyof this films, or even a comparatively slowcomputer out- classroom environment, supplemented put mechanism, there ismore than enough band- by something width available. called homework which is comparativelydistinct from classroom study. To the bestof my knowledge Next we should look at S/N, the signal-to-noise the only new technique which has ratio.Noise in this communicationsprocess takes had a significant many forms : impact on education is the so-calledobjective test, it may be an instructor mumbling which may in somecases be machine scored; and into the blackboard, his back to theclass ; a car- there is serious question inmy mind whether this penter working next door;or a colleague whispering is not, on certain occasions at least,a step back- to the student about another matter.Mostly, how- ward rather than forward. ever, it is simple inattention.How many ofus have ever looked at a class whenwe were discussing The educational system isan enormously vast a technical point and seen one hundredper cent and complex one.The largest industrial complex alertness, with bright eyes andeager facesor even in our nation, the American Telephoneand Tele- as high a level of attention as we receive when graph Company (which incidentlyinvented Sys- we are telling them what questionsare going to tems Engineering at the Bell TelephoneLabora- appear on a test ?The fact is that it isdifficult tories) has a total capitalization whichis less than to concentrate for fifty minuteson new material, the annual expenditureon education.The only much less to do this formany hours a day, day after system comparable in size is the totaldefense sys- day.This is one of the reasons forthe high level tem of the United States, which ofcourse uses of redundancy in teaching, althoughperhaps thousands of systems engineers; and while a it is more important one is that the understandingof 157 concepts (as distinguished from the memorizing close. We need to locate error patterns and make of facts) generally requires that the point be made appropriate repairs. over and over again.This need for repetition The next question concerns feedback,a signifi- leads to one of the fundamental inefficiencies of the cant factor in the control of any complex system. present educational process, becausesome students Feedback at the present time is given largely require the point to be repeated less often and others through tests and through correcting of homework more, with the result that we bore some and con- (when homework is in fact corrected). Even when fuse others.This is not merely a question of the this is done, the feedback is much delayed, andwe bright students versus the dullones, although of know from the work of B. F. Skinner the extreme course this is a significant factor.It is also because importance in learning of rapid feedbackor, as he the students have had differentexposures and back- terms it, "reinforcement." grounds ; in particular, for example,some of them Finally we should talk about turnaround time. have had other courses whichare peripheral in This is an important measure of effectiveness inany nature, but none the less pertinent toa particular computer operation, and in this connection I should point. like to make a comment concerning remote-access, What is the transfer function ofa student? The time-shared computers which everybody isnow input consists of facts, together with relationships talking about.It is well known that there isa among them and appropriate explanations and em- sharply convex relationship, perhaps quadratic, be- phases which hopefully lead to "understanding" tween the performance of the computer and its cost; and, in the words of William Linville, "portablecon- that is, if one pays twiceas much for a computer, cepts."The output is somewhat more difficult to one gets perhaps four times as much performance define or to measure.It consists of some process out of it, or, putting it differently,one gets twice as which takes place between theears of the student, much performance fromone large computer as from and which we identify by such wordsas "retention," two small ones of equal total cost.I recently heard "comprehension," and "understanding."There is the assertion that remote-access time-sharingis not a famous professor who had a committee meeting, going to work outas well as everyone thinks, be- and therefore left on the podium in his classroom cause the advantages of the large remote computer a tape recorder with a little note to the effect that will be lost through the disadvantages of thenum- it should be turned on at the beginning of the hour erous input-output channels. This logic is based on and that a fifty-minute lecturewas present on it. the fact that at the present time the centralpro- Coming eagerly to the same classroom two days cessor is a comparatively small part of a total com- later to find ort. how his experiment had worked, he puter installation.I think this misses the point. found on the seats in front of hima whole series The major benefit of remote-access time-sharingis of tape recorders on each of whichwas a little note not economic so much as it is thevery quick turn- saying that it should be turnedon at the hour and around time. For the large bulk of computerusage, that there was ample tape to recorda fifty-minute which consists of small problems, turnaround time lecture. There is also the famous definition ofa lec- is far more significant than otherfactors, such as ture as the process by which the notes of thepro- whether the computation takes two secondsor 500 fessor become the notes of the student without milliseconds.What we are trying to minimize is passing through the minds of either. And finally a function which depends primarily on the overall I should like to tell Professor Newmark'sstory, time from the occurrence ofa problem to receipt about distinguishing the graduates from the under- of the answer, withsome modifications involving graduates in a classroom which contains both. On convenience and effort.If one has to walk to a the first day you walk in and say "good morning"; distant computer center,come back because he finds the undergraduate students reply "good morning," he forgot to get a charge number,go back to the but the graduate students, who have hada little computer center and leave his cards,come back the more experience in the school system, immediately next day to find they wouldn'trun because one of write it down in their notebooks.At any rate, the control cardswas incorrectly punched, and so while we may have made little attempt to define forth, one is not going touse the computer very the output of the teaching process,we have made much. many attempts to measure it.These measurements As far as educationaluses of the computer are are called "examinations," and unfortunately their concerned, we must similarly optimizethe overall primary purpose in most cases has been tomeasure man-machine interaction. Moststudent use of the retention rather than comprehension.I shall re- computer falls into the class of small,one-shot turn to this point below, but I wish to stress here programs written in FORTRANor a similar lan- that what we really need are what the computer guage, and there is a tremendous opportunityfor man calls "diagnostics." The analogy is really very much greater utilization of thecomputer if only. 158

is rapid turnaround time could be achieved.Turn- individual there is ample feedback and control of the around time as it applies to the educational system process ; and turnaround time is minimal. that we are examining has significance when itcon- cerns the length of time between a question popping Now let us talk about some of the possible objec- into a student's mind, or a mistake being made, and tions to this system.The first objection that will the opportunity to obtain an answer or a correction. be raised, of course, is that of motivation.This It is true that in so-called recitation sections stu- argument presupposes that the student is primarily dents are supposed to have the opportunity to ask motivated by a personal contact with a teacher in questions, but in many recitation sections such ques- a classroom. But let us remember that in our great tions are not welcomed ; even where they are, the schools the bulk of the contact which undergraduates majority of students do not ask them.In large have in classrooms is with graduate students who lectures, of course, there is no possibility of ques- are only a few years older than they are.It is true tioning, and in the homework (which theoretically that we have professorial people doing most of the represents more than half of the student's learning teaching in the smaller colleges, but we tend to time) the turnaround time is at least a day or two look down on these smaller colleges and feel that at a minimum. the student is not getting as good an education there. Based on the above criticisms of the present edu- If a really great professor gives a really cational system, I wish to recommend what I con- great lecture, it seems a shame that only twenty sider a better solution or system design. This would students (or at most a few hundred) should hear be based on an integrated set of lectures, books, it. Why can this lecture not be recorded andpre- and computer programs. The lectures would be of sented to thousands or millions of students ?There the "canned" variety, on magnetic tape or possibly is undoubtedly a certain magnetism in being in the movie film.The lectures can be corrected or same room with a flesh - and -blood speaker, but I changed, as experience dictates, or updated, by consider that the advantage thus represented is far splicing in appropriate bits of tape. Of course there outweighed by the advantages of a great scholar is no necessity for a lot of students to sit simultane- and teacher giving a thoroughly prepared lecture ously in a single classroom with many television with first-rate teaching aids, as against what hap- receivers staring at them, as is frequently done to- pens today in the great bulk of our classrooms. day for reasons of tradition. We visualize a stu- Furthermore, we will tend to avoid the boredom dent in a carrel watching a lecture, with the of the student who is ready to proceed faster and capability of stopping it and repeating any part the confusion of the student who has to proceed which he wishes to go over.The books might be more slowly, and boredom and confusion also have conventional text or programmed text, but in either significant effects on motivation. case would be thoroughly integrated with the lec- What about questions ?I repeat, "What happens tures and computer programs and designed to be now ?"The student cannot ask questions in large used with them.The computer programs would lectures, he cannot ask questions when he is doing be in the form of dialogues, between an individual his homework (and there are two hours of home- student and the computer, of a type that has been work for every hour in class), and even in the so- frequently worked out in the past.See, for example, called recitation sections very few students do ask the discussion on PLATO (IRE Transactions on questions.I think under this new system we can Education, Vol. E 5, pp. 156-167, 1962). See also a have greater accessibility to questions by the stu- fascinating dialogue between a computer and a dents.Teachers who are freed from the tedium of physician studying diagnosis in Computer-Aided giving the same course over and over will be avail- Instruction, John A. Swets and Wallace Feurzeig, able as consultants for the students who have Science,, 150, 527-576 (1965). reached an impasse.It is somewhat of a principle This type of dialogue is not new. Turing wrote of system engineering (see H. H. Goode and R. E. an incredible example of one such dialogue many Machol, System Engineering, p. 314) that automatic years ago ("Can a Machine Think," A. M. Turing, systems should have a provision for manual inter- Mind, 1950). vention in order to obviate the necessity for com- It should be clear that this system would satisfy plicated and/or expensive provisions for unforeseen all of the requirements we have listed above. There events or those of low probability. When the stu- is plenty of bandwidth ; the signal-to-noise ratio dent gets into trouble in his computer dialogue, he is unlimited ; we always have the appropriate error- presses the "Help!" button, and gets aid from the correcting codes in the ability of the student to go teacher-consultant.And, incidentally, this kind of back with his lecture, his book, or his dialogue with consultantship, which would be the principal teach- the computer; there is immediate reinforcement; ing function for most of the faculty, may keep them the amount of redundancy is optimum ; for each in closer contact with the students than the present 159 method; that is, it will improve the feedback to the many studentsstudents were taking because it was the teacher. easiest way to satisfy a "science requirement." One What about testing? There are three basic func- of the students in this category was a 55-year-old tions of the examination :motivation, assessment, woman. On a certain test, my friend gave them and feedback.Motivation is not an insignificant appropriate data asked them to compute the function; we know that many students will study age of the universe.This woman came up with harder when they know they are going to be tested. the answer of 37 years. My friend asked her if Assessment implies that we haw to decide which she was not in fact older than 37 years herself, and students will get the financial aids, the honors, the she admitted with a blush that she was. He then admissions to our better schools, the better jobs asked her if this did not appear to her to be incon- after they graduate.Feedback is, however, the sistent with her answer for the age of the universe, significant function of testing.This kind of test- but she could see no such inconsistency.I realize ing can be very much better done in the pro- that it is desirable to have laboratories to correct grammed text or the computer dialogue.Presum- this type of situation.I realize that we occasionally ably the student would be allowed to "cheat" on all get students into an electrical engineering labora- such testing, but he would be cheating no one but tory who have never before realized that it takes himself.Occasional testing for motivational and two wires to carry most types of electrical signals assessment purposes could be performed in conven- or power.Nonetheless, I assert that a laboratory tional ways; possibly these could be standardized is a simulation of the real world, and as such it is on a large scale, in the manner of the present most efficiently performed by simulating. And the College Entrance Examination Boards, or the Gra- tool for simulation is the computer, rather than duate Record Exams. unrealistic hardware. Does a circuit man in today's And advisement?I do not think advisement is engineering worldbuildcircuits? Rarelyhe done very well now (for example, we are continually changes parameter cards.The same thing is true finding students who have failed to take their phy- of aircraft: eventually, the designer must build an sics sequence and so are unable to begin their engi- airplane and prove that it can fly, and even before neering courses), and I do not think it will be very that he must put the airplane (or parts of it) in good under the proposed system.However, I do the wind tunnel, but the great bulk of his choices not think it would be any different under the pro- are made by analytical, computational, and simula- posed system, and if anything it should be better. tional techniques.Most of our laboratories are Good advisement is based primarily on the contact fraudulent or spurious in any case.Did anybody between an interested student and an interested ever see anything in the real world that looked like instructor; because so much of the work will be a mass, a spring, and a linear dashpot?No, of done on an individual basis, it will probably be course not.Real vibration problems are made of necessary to schedule more frequent interviews be- complex interconnections of many masses, many tween the student and his adrisor. compliances, and many frictional resistances, most Discipline? We should face the fact that dis- of which are nonlinear. The linear approximation cipline is now one of the significant factors in the is a useful approach to the problem, but it is mis- classroom situation.The teacher must be there to leading to tell the student that this approximation ensure that the students are paying attention and is an exact replica of anything in the real world. are keeping out of trouble.I recognize that this Why, then, not admit the approximation, and work will continue to be a significant function.I believe with the approximation on an analog device, namey we should hire professional disciplinarians rather an electronic computer? At many places (e. g., at than fragile Phi Beta Kappas to perform this work. Brooklyn Polytechnic Institute under John Truxal) They would be more efficient, and certainly far less they have had considerable success with this kind expensive. of thing. Finally, since this is a conference on engineering In this connection I believe thatmany laboratory education, we must refer to the question of lab- experiments do more harm than good. We givea oratories, which are a significant portion of that student a capacitor box whichgoes inductive at a education, and so we must ask, "What is a labora- few thousand cycles, and thenwe make sure that tory ?"It is a modeling, to reinforce and relate none of his inputs are above a few hundred cycles. learning to the real world.There is a significant Is this realistic? danger that the student in a lecture course ina On some occasions when Iwas an undergraduate scientific discipline will not recognize that what I wrote up an experiment which I had not done.I he is learning must be related in his thought pro- wonder if there is anybody in thisroom who could cesses to what is real.A friend of mine once not or would not make thesame statement.I define taught an introductory astronomy course, which a good laboratory experiment as one that cannot 160 4*-

be written up without having done it.How many munication LT in one direction only. What about of our experiments are of this type? The simulated the professor learning from the student? laboratory will certainly be less expensive than our present laboratories, and I think in most cases it MACHOL: It turns out once again, in our great will be more effective. universities, that we're down to about one course Of course, in the case of the laboratory as in all per professor per unit time.The professor then others, there is no reason to go to extremes. We is in contact with his students for about three can still supplement the laboratory with occasional hours a week and the rest of the time he spends real experiments, just as we can supplement with doing all the other things which professors do, occasional real lectures, with reference to library which are very much worth doing.There seems material, and the like.I do not advocate mono- to be a feeling that the contact with the student mania. is a comparatively small item. Now I feel that I suspect a number of other people have advocated this professor can perhaps be that consultant who more or less this type of thing in the past.If it is on the other end of the panic button and he will were subsidized by the government, there is the get a much more direct contact with his students. possibility of savings of really large sums of money, He will, of course, occasionally make one of these in the billions of dollars.If it were done by a book canned lecture sets, but he will still be there as an advisor and as the question-answerer, and I think publisher, there would appear to be the possibility that his contact with the student will be greater of incredibly large profits.It seems to me remark- than it is today.The professor who gets up in able that it has not already been done. front of a large class of 400 and lectures to them has no feedback to speak of, and this is what the PAYNTER: I want to say I find the controversy big name, the greatest professors do. delightful.But I feel, and perhaps all of us feel, threatened by part of what you're saying. NEWMARK: I'd like to ask one question.I'm Because we are physical and we operate by chemi- more concerned about another type of interaction cal process, I wonder if we couldn't apply a little which I don't see as being very convenient on the physical chemistry also, as well as systems en- computer, and that is the interaction between the gineering, and consider the fact that in a way students themselves.I think what makes some the lecturer and the teacher play the role of the of our great institutions great is the fact that catalyst, or at least they like to think they do. they have an incredibly large number of very And the catalyst, oddly enough, is defined as a excellent students and the students would get component which operates in a reaction in such a along fine if they had much poorer professors way that there's no change of mass or energy than they've got. How do you achieve this with of the component, and it's my clear understanding the kind of set-up you're talking about? that the role of tLe catalyst in the chemical process industry remains the No. 1 mystery. Would MACHOL: I don't see any significant difference. you like to comment on that? The student now sits next to his colleague in the classroom, but hopefully doesn't talk to him dur- MACHOL: It turns out that I took my doctorate ing the lecture. In the proposed set-up he will sit in physical chemistry, and I would like very much next to his colleague in the carrel;he can stop to comment on that. I would say that this sounds the tape recorder anytime he likes and turn to like something you learned in high school be- his colleague and say, "Hey, you, what do you cause I remember that they told me this in high think of the Sox?" or whatever he does say to school and it's just as false as can be. We now, his colleagues.I see no loss of interaction be- in most cases, understand how a catalyst works. tween students. It is not a mystery as far as the chemistry of the thing is concerned. However, I will agree there KARNAUGH: Has anyone proposed instantaneous is some mysterious way in which a professor acts feedback from the students, whether of the silent- as a catalyst and I thought I had basically com- button type or any other, which would, when mented on this when I discussed the subject of correlated with the tape of the lecture, allow a motivation.If you really have a good guy up certain amount of insight into what was done there, he does this, but I would guess that in 99 badly before the lecture might be canned? per cent of the cases there's very little catalysis taking place. MACHOL: I would first comment that one of the advantages of the canned lecture is thatyou can REILLY: In examining the educational process correct it and make it better and better.But I as a communications systems, you've made a would also like to ask somebody from Urbana to very basic assumption and that is that the com- talk about PLATO. 161 FENITES: PLATO isstill very much an experi- time we are going tohave to be on-line with mental device.At this stage of thegame, we a can only run one particular large number of studentsand constantly update course; of course, and correct theprogram so that, while leaving it doesn't matterwhere any individualstudent is within that the student quite freeto go in any particular course. PLATO hasgone through direction he wishes, several cycles of evolution. we can produce some kind The first onewas of convergent patternso that the objective of the nothing muchmore than a programmed learning lesson is accomplished. book, a linearsequence; then they started jumping out of sequence, backingup and so on. Now they are working on several different PAYNTER: Answeringthe question raisedspecifi- patterns of cally, some of Sheridan'speople at MITran teaching andone of them is inquiry teaching. number of experiments Once this gets implemented, for the better part ofa and implemented for term on instantaneousresponse and they played the college level, I thinkwe are going to have the both button things,with an indicator infront kind of interactiveenvironment that Macholwas talking about. Namely, of the, teacher, andthey tried lights, theytried there is no lesion plan. meters, they tried everything;one of the most It's a pure simulationsituation where the student carves out his own path, successful thingswas a sort of joy stick in front program in any direc- of the student.He could manipulatehis own tion that he wishes,can ask questions, backup, or go into side issues, and subjective understandingand, when he felt he eventually, hopefully, was slipping out of the programconverges and pushes him toward gear and not following, he the final result. would deliberately pushthe thing over, and, if he Rut there is stilla tremendous felt he was following amount of work to be along with the teacher, he'd done, and fora very long push hardover to the other side.

162 ti

EMBEDDING A DIGITAL COMPUTER IN A NEW CAMPUS AND SCHOOLOF ENGINEERING ROBERT M. SAUNDERS University of California, Irvine Irvine, Calif. New universities usually have not started in on the campus as far as scientific computing was recent years from "scratch"; typically they have concerned. We also realized that we had an op- representedexpansionsofresearchinstitutes, portunity not accorded many campuses to really specialized instructionalactivities, or other en- integrate the computer into the total activities of deavors.In the University of California twonew the campus. We noted that in the university of the campuses began instruction in September, 1965, future the computer is to play an increasingly im- having no constraints on their planning placed by portant role.Therefore, early in 1964, we made the any of the existing physical entities just mentioned. decision to install a digital computer which would The Irvine campus is one of thesenew ventures. serve the following functions: (1) scientific com- Located on one of the largest undeveloped parcels putation, (2) computer-aided instruction, (3) reg- of land (some 88,000 acres) in Southern California istration and student record keeping, (4) library on the one hand and the Los Angeles metropolitan search and retrieval, and (5) accounting. area on the other, a unique opportunity has been Of these (1) and (5) are well established in most presented us to build a 27,500-studentcampus in campuses today. As far as record keeping for the the midst of, and simultaneous with, the develop- Registrar is concerned, (3) is in partial existence ment of a city of 100,000. on most large and many small campuses today. An urban university of these dimensions calls However, the whole registration procedure, the for a liberal arts program to provide the background optimization of student programs, and the develop- for the successful transformation of students into ment of methods for the utilization of the physical contributors to society.Thus all students at the plant are still in formative stages. Similarly library University If California, Irvine, will takea mini- research and retrieval methods (4) need a great mum of two years' work in the College of Arts and deal of engineering to get them off the ground. Sciences before going on to more specialized work Computer-aided instruction (2) is a new area, in that College or in one of the professional schools, not well understood, where a tremendous payoff of which Engineering is one. can occur, and we intend to try to bring this about. The Irvine campus is a general campus in another At this time we really do not know very much about 1 sense as well.Along with other campuses of the computer-human interaction. Research and evalua- University we should in clue course produce for tion on computer-aided instruction will in time, we every 100 B. S. or B. A. degrees, 150 M. S. or M. A. are convinced, result in a radically new instructional and 25 Ph. D. degrees (100:159 :25). At the present system. time the University of California grants degrees To begin this endeavor there is now installed at on a 100 :186 :12 ratio.Nationally comparable Irvine an IBM 1410/1440/1448 system with card ratios are 100 :131 :5.Thus we shall be deeply readers and card punches for each of the two pro- involved in the complete spectrum of collegiate edu- cessors. We have five tapes and 11 disc packs. The cationfrom freshmen to Ph.D. students.Since 100,000 character 1410 is the control processor and our campus must ultimately achieve this complete is directly coupled to the 16,000 character 1440/ spectrum, we have started with freshmen to doctoral 1448 message processor and switching center.At students. the present time we have eighteen 1050 terminals The School of Engineering is one of the profes- connected to the system. Eleven of these terminals sional schools building upon the two-year liberal are in direct contact with students for at least 12 arts structure.In our junior-senior and graduate hours per day and several others are located around programs we expect to have 500 students in 1970, the campus in the registrar's office, in class andcon- and an increase by 100 students per year untilwe ference rooms, and in one of the psychological labor- reach around 2,500 in 1990. What the upper limit atories.Time-sharing operates under a monitor will be is unknown at this time; however, current appropriately called "Dean" and runsmany routines planning calls for a ceiling of 27,500 for thecam- simultaneously.For student scientific computing pus as a whole. we currently run an Irvine version of JOSS (we Very early in our planning we noted that digital call it JOSSI).We utilize a course-writerprogram computation was to become an essential fact of life for developing computer-aided instructioncourses. 163

,11,0*,43. We also use FORTRAN for faculty andadvanced- be charged one pointper minute.If they elected student research activities.The registrar's activi- to use the consoleor the computer between 3:00 ties, experimental accounting procedures,and some and 4:00 in the morning, theymight get 60 minutes library work are also programmedon the computer. for the same point.This method would distribute We have a research contract withIBM so that computer use over the 24 hours ofthe day. they share in both the cost and theyield in the development of these precedures. Underthe terms In general we feel that computer-aidedinstruc- of this joint endeavor, tion, program debugging, and inquiryon the part we provide the real estate, of the library, the registrar, the terminals, and some of themanagement person- or the business office nel and programmers; IBM provides some of the ought to be time-shared froma console, but that equipment and alsosome programmers.Thus we scientific-production computing oughtto be batch are engaged in a truly cooperative venture. processed, and runas either background or at a non-prime hour. To involve Irvine studentsat the outset, and to provide a basison which to build computer-aided Several speakers have alluded tothe cost issue, courses later on, we started a freshmancourse but none have reallycome to grips with the actual in digital computing emphasizingalgorithm con- costs of a computing facility.Professor Shannon struction.All freshmen in engineering, thesocial commented on behalf of the Dartmouthinstallation sciences, and the physical sciencesare required to that they have approximately 50students per con- take this course. The classesare composed in ran- sole.This has beenvery close to our experience, dom fashion of students from thevarious disciplines, where we find that thereare no queues for 40 stu- and we do not try to keep specialsections for social dents per console but there might beif we went to scientists or for engineers.In my class last fall 75.Considering that a terminal costs $1X100per quarter I had social scientists,physical scientists, year, and the central processor from $100,000to engineers, biological scientists,and even a French $500,000 a year, fora 10,000-student campus with student.Because the emphasis in thecourse was half of the campus active incomputing, $500,000 to on algorithm construction and the breadthof stu- $600,000 per year should be availablefor a comput- dent interest, wewere very severely taxed to think ing facility.For a 10,000-studentcampus this is of problemswe could give students who had hadno approximately equal to the librarycost per year. college mathematics. Cooper offers encouragement to theacademic com- Relatively little real instructionis given on pro- munity when he shows that theAmerican Electric gramming as such. To de-emphasizeprogramming, Power computing facility indicatesa trend toward we elected to use a student-orientedlanguage. The saturation in terms of costs while at thesame time one adopted was that developed forthe Johniac obtaining an increase in computingoutput.If this trend as at all predictive, there is open-shop system at theRand Corporation, JOSS; perhaps more this was adapted to theIrvine computeras JOSSI. hope for the campus computer facility. The feature of this language is that it has immediate Presently university computing facilitieshave accessibility. You simply type your name and iden- three principal sources of funds: tification number and (1) university you are in business.JOSSI funds, (2) federal grants, and (3) privateindus- is easy to learnbecause it interactsvery readily try.Grants from government agencies and private with the students.Its errormessages are clear and industry have been of inestimable help give clues in plain language in getting as to what is wrong started and in bringing the level of computeractiv- and what should be done.Furthermore, the instruc- ity in the universities to what it istoday. What the tion set is very small, beinghandled on a single role of these two outside forces willbe in the future 83/2x11-inch sheet ofpaper. is a matter of public andcorporate policy.It is For large-scale scientificcomputing we utilize not inconceivable that bothsources will be curtailed the Western Data ProcessingCenter at UCLA via to some extent and it therefore behoovesuniversities telephone links and courier. to look for internal support.I believe it to bea Our current policy is not tocharge anyone for safe prediction that the percentagesupport contri- computing time ifwe can possibly help it, although buted by the universities themselveswill increase we do expect those with research funds topay their in the future.If this conclusion is accepted, univer- share of the processing time.We are toying with sity operating budgetsare in the last resort the the idea of using prioritypoints either in place of, place from which the funds will derive.This means or together with, grants of computingtime. Under a higher student-faculty ratio and less support that system, applicants,no matter whether they money per student than is enjoyed today, since most were poor or rich, would be givena certain number universities, be they privateor public, are reaching of priority points. If theyelected to use the console saturation in the support theycan muster per stu- between 10 :00 and 11:00 in themorning, they might dent. 164 1-g

With respect to computer-aided instruction, the to expand and extend this study to the nonlinear School of Engineering will have a strong program. case. At outset, computer-aided instruction techniques In conclusion, I would say that we must integrate will be applied to laboratory equipment orientation. with the campuses of our universities across the Should a student come into a laboratory and not country in the same way libraries have become be conversant with the operation of a given piece synonymous with teaching and research. Similarly, of equipment, he would be asked to address himself we must plan to finance more of our computer facili- to a computer console.It would be his job to learn ties out of operating incomes for the campuses, even from the console what he is expected to know about though this means that we will have less money a piece of laboratory equipment before using it.In for equipment and faculty.However, the payoff this fashion we free the instructor for a more potential is extremely great since a well-embedded Socratic interaction with his students in the labora- computer system which is designed to be an integral tory and thereby place the onus for rote instruction part of a campus does more than provide a facility on the use of laboratory equipmentit the computer. it provides an exciting atmosphere and makes for We also expect to make extensi use of analog a climate in which we should all be able to produce computation for undergraduate instruction, and to better students who, in turn, will improve the engage in hybrid-computer research so that an quality of engineering practice. analog computer can be matched with a time-shared digital computer. ORGANICK: I'd like to get a little information on Ware also preparing engineering design courses the economics of the 7702 transmission.., Where in which we will utilize extensively optimization we looked at the problem -oftransmitting'. be- and simulation theory.These courses will be for tween Houston and College Station, Texas, which fourth-year students and will include not only the is about 100 miles away, we kept returning to the theory of design but also an actual study carrying fact that the bus is cheapest way. the design through to a prototype product. SAUNDERS: I don't think there is any way to We also expect to do research on computer-aided refute that.It's the turnaround time that we design and, in fact, at the present time are so en- would like to have. You can get 24-hour turn- gaged.Current work involves the development of around time with our bus system.We would optimization algorithms for systems described by like to have four hours turnaround time, which continuous and discrete variable equations of con- is about all you WA buy with that 7700-type straint and restraint for linear systems. We expect system.

165 I A PLAN FORIMPROVINGINTERDISCIPLINARY COMMUNICATIONFOR ENGINEERINGDESIGN BERTRAM HERZOG The University of Michigan Ann Arbor, Mich. INTRODUCTION There commenceda further de-emphasis in the Today's session closesa conference emphasizing the impact of computers teaching of technology andspecialized skills,ac- on education in engineering companied by a continuing risein the teaching of design.Many speakers havegiven case histories basic knowledge in scienceand mathematics.Al- and examples which makeit crystal clear thatcom- puters are an aid in though some of theaforementionedcourses again design and particularlyin fell victim, this timea large number of courses hav- teaching design. For thepurpose of my remarks, I assume that computers ing design emphasis alsogot the ax. They deserved play a useful role. My it.Design courses es inmany instances had stagnated concern, however, is that asa community of educa- and did not match the tors we have not taken students' broadening in- the impact of computers terests which had beenaroused by the ekploding seriously enough.Nor havewe really captured -the magnitude of this technologies of the last twentyyears. Yet design still-growing impact.In- is an important activityof industrial life. Greater dustrial managers sufferfrom this myopia. My knowledge of the liberal theme is that of improved arts is desirable,' A sound communication between education in the basicprinciples is mandatary. departments of colleges,between specialists in vari- Surely all will ous disciplines, and between industry agree that the changes have been and university. for the better; butnow we must bring design back This improvedinterdisciplinary communicationmust into the picture. The be brought about if question is: How to do this we are to meet the challenges without expanding thealready crowded undergra- implicit in thenot-so-distant future activitiesof duate curriculum? engineering education. This problem The problem is to get was faced before, when the greater a program underway which emphasis on science engineeringfirst began. Core will meet the needsof the undergraduatestudent, the engineer employed courses or curricula were invented bythoughtful in industry who isworried teachers who attemptedto tie together similar sub- about obsolescence, thefirst-year graduate student, and the faculty research ject matter having differentapplications within the and advanced graduate traditional fields of engineering.Just one such ex- student. These four classesof people form partially ample is embodied in distinct groups which taken the course in Linear Science together are the ingre- and Engineering whichis part of the Master-Course dients for success ofsuch a program of designedu- concept at John Hopkins cation.It is my contention University recently des- that deliberate and cribed by ProfessorGrayson.1Its content empha- speedy actions must betaken to meet the needs sizes the concept of of present and futuredesigners. Unearity, but examples of There are two factors applications reflect theparticular interest of the that contribute to the student and teacher.Professor Grayson speaks of urgency of the situation.About two decadesago the success of the approach engineering education witnessed and also warns ofsome the beginning of of the dangers of removingpedagogic repetition. a change.During the mid-fortiesa considerable In short, new emphasis outcry arose regarding need not require throwing the engineer'spoor state out useful and testedmaterial. Before proposing of preparation to meetthe problems of theworld and his role in society. a method for design education, letus consider the Consequently,more non- other pressures creatingthe need for anew look technical courseswere pumped into the curriculum at design education. and certain technicalcourses were discarded. Some deadwood was trimmed;surveying campswere THE NEED FORRENEWED INTEREST dropped; drawingcourses were reduced in number; The computer revolution,to use the popular and other separatecourses disappeared which, I phrase, has been accompaniedby associated changes am sure, you can name easily.Expediency did not in engineering andscience.Although high-speed permit reorganizationamong the topics of impor- electronic computerswere invented to solve engin- tance, although severalinstitutions triedsome novel eering or scientificproblems in ballistics, they and exciting approachesto the problem. Then approximatelyten years ago, 'Grayson, L. C., TheMuter-Course Concept at The Johns a new on- Hopkins University, Journal slaught of change facedthe engineering curriculum. of the Americas Society for Engineering Education, 56, 5,January 1966. 166 bz.

eventually found their greatest application in the modity, is very impressive. Variousestimates of the business comunity. However, engineers have found compression ratio of research orproblem-solving an ever-increasing number of computerapplica- times range from ten to fortyaccording to some tions, which have usually been typified by complica- workers. The possibilities in the designer'slife will ted calculations for the solution of partial differential be mentioned later. equations or other problems of advanced mathemati- For our present discussion, it shouldbe noted that cal analysis.Even more advanced problems are conversational interaction with a computeris exactly solved by researchers and new insights are gained what has been needed to solve the designproblem with the aid of computers. and the problem of how design is done. Recent developments in computer technology and So we arrive at the state where designhas been their applications foreshadow a significantly greater largely neglected and computer technologyhas ad- increase in computer use by engineers.I refer to vanced. What about design? What is it?To spend the developments originating at MIT's Lincoln Lab- much time on this would be redundant,but a few oratory and General Motors Research Laboratory. words of clarification are necessary. Sketchpad' and DAC-I have demonstrated with A SIMPLE VIEW OF DESIGN great clarity the ability to communicate graphic or All writers mancge to find their own wayof geometric sketch information to a computer. That dividing up the subject of design in their questfor graphic communication is an engineer's favorite a rational description.I believe Professor Asimov's form of communication need not be debated. model, given in his monograph "Introduction to De- Furthermore, computers can be programmed to sign," 4 will suit our purpose. A basic knowledge understand these geometries. Figures and objects of physics, mathematics, chemistry, or material can be manipulated, parts canbe machined, and science is required.This, together with an idea or other features such as stress distributions may be a specification for a product,initiates a process of computed. synthesis and analysis.This process, in turn, in- Such calculations are done most successfully when- volves assumptions, analysis based upon assump- ever the designer has theopportunity to interact tions, and calculations of results. These calculations directly with the computer. This method of working produce the first evidence of a trial solution to the on-line with a machine is not entirely new ;in the design problem. The trial solution is then tested ; early days of computing it was quite usual for the if it is rejected, new assumptions are made and user to sit at the computerconsole and to alter, at the design function is repeated until a successful his direction, any part of the computation. To do solution is obtained, whereupon the design is com- so graphically, however, is new.As machine speeds plete. increased and man's reaction time stayed constant, This model of the design process appears to be it became uneconomical to permit this extremely applicable as much to the design of a part as to the satisfactory communication. design of a whole assembly. One might also argue But computer technology again came to the res- that this model is an adequate description of the cue via time-sharing, asdemonstrated by Project trial-and-error method of design! MAC at MIT and the BASIC System at Dartmouth. Over the years a mystique has developed around It is now possible for many users to communicate the methods of design.Arguments regarding the simultaneously in an on-line fashion with the mach- relative merits of art versus science in design have ine. The machine shares its power with all of them raged for years.Please allow me to avoid this by capitalizing on any one man's long reaction time argument for it could consume us, but let me ac- (during which he requires no computation) to do knowledge that both art and science are part of the the computing chores of the other users. An eco- design process today and will, in all probability, nomical state of affairs is thereby achieved. Those always remain so.This does not mean that we of us who have had the privilege (or is it a right?) should not seek out the rational methods of design. of using such a system have found even greater eco- It also does not mean that one can ignore the nomies in the use of our time. The potential savings rationally unsolvable problems which have been "ar- in time of technical personnel, our most scarce com- tistically" resolved for years' Several significant events have occured to dispel 2 Sutherland, I. E., "Sketchpad: A Man-Machine Graphical Communication System," AFIPS Conference Proceedings, some of the magic of design methods. TheJacquard Vol. 28, pp. 828-28 (1965). loom demonstrated our ability to program product Johnson, T. E., "Sketchpad III: A Computer Program for Drawing In Three Dimensions," AFIP Conference Proceed- 4Asimov, M., Introduction to Design, Prentice-Hall, Inc., ings, Vol. 28, pp. 347-58 (1968). Englewood Cliffs, N. J., 1962. "Jacks, N. L., "A Laboratory for the Study of Graphical 'Automobiles are built daily at ever faster rates. To find Man-Machine Communications," AFIPS Conference Pro- a description of rational methods of analytically based de- ceedings, Vol. 26, pp. 843-50 (1964). sign for automobiles is an unrealistic objective, 167 "..srrywrr-sTr77,-.$7

variety.Numerical-control machining isnow an The modern designer should havea good under- accepted form of manufacturing fora certain class standing of the technology of materials, production of goods and its effectsare being felt, upstream in methods, and cost to be effective. But he should also the industrial process,at the designer'sdesk. know something about problem definition' (perhaps Numerically controlled drafting methodshave been system analysis) and optimizing methods.Obvi- used to speed up some of the design tasks. ously, universities will not be able to provide him The machines have joltedus out of some cherished with all these tools for his trade.The technology beliefs about the capabilities ofmechanical aids. of manufacturing materials changesso rapidly that Now we are about to receiveon intellectual jolt. it has long been hopeless to givea complete treat- Computer-based methods will permitus to attack ment in the undergraduate classroom.Educating design to seek out its rational characteristics. engineering students in rational design, that is, optimizing methods with proper cognizance ofmanu- My apparent obsession with the rationalaspects of design, or with design alone facturing techniques, is aproper activity for the at the expense of educational institutions. engineering as a whole, is entirely withpurpose. Advances in materials technology andexperimental Appropriate cooperation between industry and prototype development resulting innew products university could provide "real-world" data forthe such as transistors playan important part in engin- classroom. Visualize the industrial engineeringstu- eering and its contribution toour economy. How- dent investigating a problem of assemblyline ever, our society is not without people devoted to balancing; via computer-based communication, he this objective and adequate attentionis already would be permittedaccess to actual data to test his given to such developments inour universities. Con- model. The student would enjoya much more signi- sequently, special emphasis is needed forthe study ficant understanding of the problem.Industry of design methods. would actually receive the benefits ofa study not beset by assumptions which void its applicability. Our world has many extremely cleverdesigners who over the years have built These, then, are thepressures causing the need up an astounding to initiate a deliberate attack seeking the rational ability to optimize. However,many designers deal methods in design. New solutionsare hard to find, with extremely difficult problems bymeans of com- but by harnessing all paratively poor rational methods,and thus we have our available resources in a the phenomenon of "little change from concerted plan we may havea greater hope for suc- last year's cess.It is first necessary to study design methods model or design."Again, my hat is off to thesuc- cess of these methods. Today I have and to integrate them intoan educational process for very little to students and graduate engineers. offer which mightcause good designers to fear their replacement by a computer. The following four ingredientsmust be considered : The demands of our economy for technically 1. Undergraduate education. trained people is tremendous and designorganiza- 2. Continuing education. tions feel this shortage. Furthermore,many of our designers must spend deplorably little 3. First-year graduate education. time being 4. Facultyresearchand advanced clever and a large amount of timeperforming cleri- graduate cal or booking tasks.° studies. I will try to separate their respectiveroles and Many university graduates avoid designactivities you will see that I will fail. But do not be alarmed: because they have been bored by dulland unchalleng- from this failure we shall drawan extra conclusion. ing design courses in schoolor by equally distasteful activities in industry. In the explodingtechnology, UNDERGRADUATE EDUCATION universities are forced to utilize The primary goal ofany university is the educa- trite problems to tion of its undergraduate students. illustrate design situations.Frequently theyare Agreement on limited by the unavailability ofreal life data. The this point appears to be unanimous.The baccalau- real world, however, has itsproblems too ; designers reate engineers form the majorityof those who in operational situations inexisting' departments practice engineering, particularly inthe engineering- operate daily under such greatpressures that they design functions.It is well known thatyoung are denied the opportunity to examine thedesign engineers are the useful nuisances whofrequently process from a new point of view,one which might bring about change and innovationin practice. result in increased effectiveness. Their training differs sensibly fromthat of their colleagues, twentyyears older. 6 General Motors, in the film presenting theirDAC-I Sys- Many engineering studentsnow learn computer tem, say that as muchas 90 per cent of a designer's time programming as part of theireducation.While is devoted to these clerical tasks. this is essential today to enableany computer .168 zation, it has been difficult to implement such courses quoted as stating, "In every company I know of, and I do not feel that computer programming as there are mature drop-outs with college degrees such is a proper part of each engineer's education. because they haven't kept up with a changing world, Using computers, on the other hand, is important. they've cut their career potentials short."This Because of the benefits of computer use, an easier statement certainly can be applied to engineers. Not way of communicating problems to the computer only have they cut their career potential short, but and, frequently, an easier way of integrating com- probably they have hurt their employer's position puter use into the educational process is needed. as well. Several examples of "easy-to-use" computer sys- Continuing education for engineers in industry tems already exist to point the way, but much more is essential.Teaching rational and computer-aided work remains to be done in this direction? design methods to this group should be a most re- The lack of a complete set of special computer warding challenge.These professionals have had programs should not prevent teaching ofproblem a rea ltaste of the problems of design. To overcome specification, decision theory, optimizing techniques, some of their specially tuned prejudices may require and the philosophy of rational design.Contribu- special efforts; but, when alerted to other points of tions from research should amplify the repertoire view, they will probably be the most enthusiastic pro. of useful methods and applications to assist under- ponents. graduate education. This educational process will have another special reward.The practicing engineer who engages in CONTINUING EDUCATION Several years will pass before our prisent or additional education is the best source for communi- planned efforts in undergraduate education will cating some of the bothersome details of the actual bear fruit.The current needs of industry and our design world to the academic staff.Without these economy will not wait that long for advancedand inputs, the university's role in design education This is one of the bene- new design methods.Since engineering, with its would become most sterile. which need fits of interdisciplinary communicationthat it is own peculiar needs, has design problems a two-way process between the industrial engineer solutions.The wrong solutions have cost the tax- payer many millions of dollars. Theconsumer-goods and his university counterpart. business seeks ever better refinements in quality, Some serious problems arise in connection with accompanied by improved cost savings to offset com- continuing education. My colleague Professor Mar- petitive pressures. These goals have a direct impact tin Warshaw" in a recent talk here in Chicago on the design engineer.Consequently, the practic- addressed himself to diese problems.I borrow ing engineer is not only concerned with his loss of freely from his remarks. He suggested the notion knowledge of technology, but also with his lack of of a sabbatical leave.The engineer would be paid knowledge of modern methods. by his employer while he spends an extended period Self-education via home reading or night courses of time on the campus. The period of time should has been only moderately successful.The design be at least several months or, preferably, on6 year. engineer's daily pressures are tiring and are fre- Formal study in quantitative methods for design, quently carried home.Evening courses are thus balanced with some project or research work to conducted at the student's worst hours. Top man- sharpen the effect of the formal studies, would be agement attitudes have often been neglectful, and appropriate.This course of study could lead to a Master's degree, although this should not be the consequently the atmosphere for self-education is main purpose. An employer's main benefit would not always provided.8 be the increased value of his employee.After all, Re-education has been emphasized for the obso- we spend money to renovate machines, why not lete executive.Recently one business leader was men ? A stay of one year might be suitable for specialists TCOGO, STRESS and DYANA are typical examples: and middle engineering managers.It is difficult Roos, D. and Miller, C. L., "COGO-90: Engineering Users Manual," Department of Civil Engineering, R64-12, MIT to see how one could gain the attention of senior Press, April 1964. engineering executives for this length of time. How- Fenves, S. J. et al, "Stress: A Users Manual," MIT Press, 1964. 0"Obsolete Executives," The Wall Street Journal, Jan. 24, Theodoroff, T. J., "DYANA-Dynamics Analyzer Program- 1966. mer Part I, Descriptionand Application," Proceedings of the 10"Education for Physical Distribution Management,"a Eastern Joint Computer Conference, pp. 148-51, 1958. talk delivered by Professor Martin R. Warshaw, Graduate 8While many companies have generous reimbursement School of Business Administration, The University of Michi- schemes for after-hour education, concurrence of a super- gan.Seminar on physical distribution managementspon- visor is required.Often supervisors only approve those sored by the Chicago Association of Commerce, Chicago, courses they believe tobe useful on the job. Illinois, February 24, 1966. 169

nn rtt S, DIY w

ever, we must get their attention, for without sup- content of engineering drawings and their role in port from the top there is little hope of implementing the industrial process would form a formidable and an effective specialized program.To obtain this interesting study. support, a special continuing educational effort must The role of information and its processing is only be sustained.This effort, ranging from one-week one of the many fascinating problems to be solved. orientation courses for engineering vice-presidents Others include stress analysis methods, resource al- to one- or two-month courses for other executives location, scheduling, and optimization of multivari- and managers, must be a consciouo part of thepro- ate problems.The need for research obviously gram. These courses need not be relegated to the exists. summer since many universities are leaning toward year-round operation. IMPLEMENTATION THE FIRST-YEAR GRADUATE STUDENT These four ingredients, undergraduate education, Graduate study to obtain additional or specialized continuing education, first-year graduate studies, training is becoming an increasingly important part and faculty and advanced graduate student research, of the academic output.This third ingredient is must be blended together into one functional unit likely to offer many fruitful results in the design- which will devote its main thrust to design educa- oriented education program.Designers will need tion. The demand for modern design education and some additional course work in their areas of spe- research will increase. To meet the demand and to cialization, and advanced courses in optimization. exploit the available resources most efficientlyre- Analysis of information systems associated with quires, in my opinion, the formation ofa special design are most suitably presented at this level. activity.For the present purpose I will call this Some project experience or research should bean activity the Design Center. integral part of this training. Research conceivably The Design Center is organized for thepurpose could be sponsored by industry.The solution of of identifying the methods of implementingdesign industrial problems might be achieved by teams education at all levels.Furthermore, the Center composed of these students working in conjunction forms a rallying point for coordinatingresearch with participants of the continuing educationpro- efforts and for obtaining support for this research. gram and under the guidance of member? of the Participating faculty and students wouldstill be faculty. What better arrangement than to havea identified with regular department structures.De- given industry's employee solve its problem ina grees earned would satisfy the basic requirements different and, hopefully, constructive environment? of the department, althougha significant part of the Probably the earliest useful results for industry student's work would be influenced by the Design would be effected from this aspect of the educational Center's staff. This formal affiliation withthe usual situation, which is also the easiest to implement. departments would appear to be less importantfor FACULTY RESEARCH AND ADVANCED the graduate student.Indeed, engineers returning GRADUATE STUDIES to participate in the continuing educationphase, Research by faculty and advanced graduate stu- logistically speaking, would bea mixture from many dents is the fourth ingredient.Results of great disciplines.Many would be practicing ina field significance and lasting impact can be obtained only other than the one in which they receivedtheir by the type of effort traditionally associated with training. The needs of these designersare best met research in graduate schools. by an interdisciplinary approach. Not onlywould Assuming that the other aspects of this educa- the Center's design emphasis be attractive,but its tional process are successful and thatour industries interdisciplinary organization would beappropriate. are staffed with engineers having the new design Having drawn together the facultyfrom the look, then a whole host of new problems arise. The various departments, onecan enhance it even fur- more rational design becomes, the more readily will ther with recent graduates from institutionswhere its results be integrated with specification writing, design has been emphasized.Qualified industrial manufacturing engineering, purchasing, and all designers might also be attractedto participate in other attendant functions which make the finalemer- the Center's activities. gence of the product possible. Obtaining qualified faculty membersis generally Information transmission bcomesa significant a problem and I do not expect it to beany easier problem. Who shall get what information when? for the Design Center.Special efforts must beex- Many existing information systems survive only by pended. Whether or notone organizes such a de- bootlegging methods and one may seriously wonder sign center, it appears that"major efforts are still what will happen when computers speed the infor- required if engineering facultyis expected to use mation around. Investigations into the information computers creatively and naturallyin their teach- 170 ing."11 The Center would lend additional emphasis list.These suggestions alone have spent the total to this need and be able to meet it. budgets for many departments for a few years What will the faculty teach? Design is frequently hence.Prices of equipment must and will come associated with the role of engineering graphics. A down. Technological developments have always freshman-level course emphasizing the information looked upon us favorably in this regard. content and the value of the engineering drawing However, should a pricing miracle occur over- should be part of each engineer's training, provided night so that we could purchase all this equipment, broader aspects of the engineering profession are it would not do us much good. We would not know also included.In any event, such a course would how to use it. May I be permitted a rash generalizer, be part of a design education.During the sopho- tion?Equipment now being produced challenges more year, concentration on experimental and crea- our ability to use it effectively.The research and tive aspects of design is appropriate. Upon entering development required for efficient computer use for the last half of his education, a student will have gross design applications must be done soon. Con- completed his training in basic courses in mathe- sequently, today's higher prices for equipment must matics and engineering sciences and will be thus be paid in order to proceed with research.Here prepared to pursue advanced studies in a specialty again, the Design Center's combined staff an com- and to integrate these with material emphasizing plement each other and use the equipment efficiently,. rational methods of design. Again, proper utiliza- The supporting staff, out of necessity, would tion of existing course material, combined with include electronic technicians and equipment oper- suitable emphasis on the rational aspects, must be ators, programmers and clerical staff.All possible juxtaposed with the importance of the consequent freedom from distracting, although often fascinat- enriched and increased information flow. ing, detail work can thus be provided to permit rapid Curricula are undergoing change for many rea- progress on the main thrust toward improved design sons and the need for emphasis on design provides methods. yet another. The program can be achieved within Increased costs are unavoidable as we meet the the framework of a typical undergraduate program, challenges of education for modern society. Many provided this effort can be combined cooperatively have already suggested that computer services with other curricular revisions. should be reckoned in the same way as library costs. The graduate and continuing education needs out- The additional burden of the Design Center should lined before may be more easily implemented, since also be figured in this way. graduate curricula are usually less rigidly structured than the undergraduate program. This flexi- Looking a little further ahead, it would appear bility could facilitate installation and experimenta- that a successful center would attract workers in tion with the Design Center activity.After all, other disciplines.Business school professors are graduate programs provide benefits to the faculty, also concerned with decision-making processes and as well as to the student! optimization of business systems.Social scientists, The impact of computers has contributed to this architects, and other professionals could become renewed emphasis on design.The Design Center participants as the problems of environmental de- must be appropriately equipped with computers or sign begin to play an increasingly important part in computer service. A large time-sharing system pro- our lives and their professions. vides the most suitable service environment for the Industry's stake and contributions have been men- establishment of the Design Center. Assuming that tioned before. The need and benefits for interdis- a computer center can supply themajor equipment ciplinary communication at all levels is urgent.I and associated software, then the Design Center believe the proposed scheme offers a mechanism need only concern itself with the acquisition of ter- for, at least, a partial solution for engineering minal equipment. design. The manufacturers gave their reports to this CONCLUSION conference a couple of days ago. What we require My remarks have been less than specific with is a variety of terminals including touch-phones, regard to courses and topic outlines.The four in- keyboard terminals, cathode-ray tube terminals for gredients I have chosen to illustrate the problem of graphics work, and audio-response devices.Curve design education have not been described in the digitizers, text digitizers, and automatic drafting most precise manner,I indicated earlier that suc- machines are other useful devices.Film scanners cessful separation should not be expected today. and even more elaborate devices can be added to the Possibly such distinctions are artificial in any case and only served the discussion. The interrelation- 118atz, D. L. and Carnahan, B., "The Place of Computers in Engineering Education," Journal of Engineering Edu- ship of several activities is probably the essential cation, 56, 8, April 1966. element for success. 171 The lack of precision may also be thought to indi- ZADEH: At the risk of soundinga perhaps some- cate our lack of understanding of the intricacies of what dissonant note, I mustsay I was a little the design processjust anotherreason for starting surprised when I heard Herzog equate design work on the problem. with computer sciences. Now itseems to me that There are enough reasons for emphssis and delib- design, which is the primary function ofan en- erate action. As educators weowe it to ourselves gineer, is really the totality of the educationthat to take a good, new look at design.The need for we give our students and so that, in a certain industry-university cooperation isnecessary to suc- sense, it is much broader than computer science. cess and can be obtained. Let us work together to Computer science is justa small segment of sub- meet the needs of this problem before the impact of ject area which plays, perhaps,an impoitant role computers beats us about the ears. in the training of students for design.This is one point.Another point is that, in the talks that have been given here, it has beenshown SCHMIT: I'd like to ask a question related toac- that computers are more effectiveat solving cer- complishing interdisciplinary design activitieson tain design problems than classicaltechniques campus.One of the problems which I have which do not use computers. Butonce this stage sensed at Case is that it is not entirely successful. is over and our students andwe ourselves have It seems to me that interdisciplinary designac- learned to use computers in that fashion,then the tivities on a campus are handicapped by the lack next stage will be really theone that we are not of pressure of a total system design which is well prepared for and theone we don't really what is forced into interdisciplinary action in quite understand how toprepare our students for. industrial situations.I would like to ask if you What bothers me is that itseems that we might see any way of artificially creating the pressure get overenthusiastic about what, in reality,con- that seems to be necessary to bring aboutgen- stitutes a fee,le application ofcomputers, and the uinely interdisciplinary design inan academic stronger applications are theones that are going situation? to give us trouble in theyears ahead. HERZOG: The advantage I experienced inan in- HERZOG: I don't think I equated designto com- dustrial environment was that I had fifteenpeo- puter science.In contrast, if 1may now throw ple working for me and itwas up to my skill to in a little bit more controversy,I will argue get them working with me; that is, the organiza- strongly, and have for quitesome time, that pro- tion provided the for in thiscase, and it was up gramming courses perse, certainly as program- to me to provide the with.I think we charac- ming courses are today, haven'tany place in terize universities by the fact thatwe have a engineering education. That is notthe purpose of department of so many people working with each an engineering education, although people will other but there is very little staff,pressure, or argue that there are benefits to be obtained in structure which encouragesus to work for each problem definition, problem solution,and so forth, other. Now I'm not suggesting thatwe invert the by learning standard typesof programming organizational structures of universities to this courses. I'm much more concerned withthe use point, but there are certain advantages inan of computers in engineeringand I do agree that industrial structure that might be advantageously at the present time the onlyway I can get com- implemented in education.I'm not suggesting puters used in engineering is that I can organize research this to teach a program- way. We're not ming course, but I don't considerthat a legitimate getting the output from ourmany research proj- objective in the future.There are people who ects that are being talked about here,or the have questioned, in the lastcouple of days, what industrial implementation, intocourses with any the differences are in engineeringdesign and com- kind of real meaning.We're getting a lot of puter science.I don't thinkwe have resolved duplication, and also isolated instances.In our that, nor shall I try to doso at this time. But I own school we have this, and I'm sure you do. think that what is important in Since you've made a larger attempt already, I engineering today is to implement theuse of computers whereneces- bow to your experience in this regard. Iknow in sary in a most painless way, ratherthan in a a similar sense we have had this problem in inmost painful way. It is in that dustry and one has to work sense that 1 see very hard to pull a lack of involvement, not of individualengineer- these things together.The tendency to get ing faculties at the present time,but of engineer- individual spikes of design done and then thread ing schools as a whole, in thequestion of attacking them together into a system obviously leaves the problem in design and something to be desired in how to implement a system analysis computero here.I think we allagree that com- approach. puters are providingus with a vehicle for looking 172 - -

AerineVOig.

at design problems.All those illusive, intuitive I think we might consider doing is givingan things which we have alwaystossed under the awful lot of thought to exactlywhat our educa- desk when we were frustratedby not being able tional objectivesare and seeing if we can then to discuss them with each othercan now be at reach outside the classroom for problemsto be least be brought to the fore-front.I suggest to brought in and serveas a vehicle. In this regard, you that implementation is a problem of educa- one of the things I have been very muchcon- tion, as much as it isa problem for industry. ,vinced of in the last couple ofyears is that you The problems are thesame. The computer is never tell faculty what to do.You only offer going to be 'the crutch, ifyou will, that will allow them opportunities to become involved.I think us to do it. that the first thing any engineer needsis a client; PAYNTER: I'd like to pickup from one of the and, if we can select from industry,or from gov- comments Zadeh made earlier having to do with ernment, certain clients, and offerour staff mem- the fact that, as systems becomelarger, they bers the opportunity to becomeinvolved in the necessarily tend to become comprehensive and solution of their problemsas a vehicle for the edu- in- cational objectives that they terdisciplinary, and go back and try tosuggest want to attain, I for the record some specific think that this is oneway to start to get this inter- answers to Schmit's disciplinary work as question. It seems tome that, at the end of the an automatic consequence. undergraduate curriculum, weare dealing with HERZOG: It is difficult toargue with the success a student who is stiii. kind of, ipso facto, broadly of your efforts in this regard andI'm glad that interdisciplinary. He still is receivinga lot of you mentioned them.There are, I think,some basic instruction over avery wide range of fields. other problems whichare perhaps native to larger Often he hasn't even madeup his mind that he is institutions of education thatare not as evident going to be a specialist,or what specialization he in the case of the small institution. will follow.1 think we have. a very real oppor- SCHMIT: Zadeh mentioned that he tunity to present to sucha student unstructured felt there was problems in which he is forced to an emphasis with regard to computer scienceand use, even in its impact on or the abilityto apply it to design. thinking about them, a greatmany different dis-. 1 find this incorrectly exaggerated. ciplines. These problemsare not two masses and I feel that two springs, for instance, but problems much of the material in thecomputer science where he area provides maybe only a small input,but at must distill something out of these.They don't least a beginning, toward have to be big, becausewe can rig the stage and attacking the real design problem.I would like to suggestan idea present very simple problems thatare necessarily (which may be controversial), comprehensive. namely, that what we teach in engineering science representsa HERZOG: No argument here. However,there are very small but important underlying ingredient some difficulties about large problems whichwe for what is needed for actualengineering analysis have had to defer dealing with forso many years when we get out and start lookingat a real sys- because we haven't really been ableto tackle tem.In other words,we don't teach modeling them. or interpretation, those very vital ingredientsof the analysis of real problems, butwe do teach the SHANNON: We produce in engineeringeducation somewhat idealized aspects of engineeringscience, a product which goes out into the environment, and these may beara parallel role with respect and by the environment Imean our whole com- to what computer sciencemay have to offer to plex of the United States, andone of the things the whole design problem.

178 CONCLUDING REMARKS N. M. NEWMARK University of Illinois, Urbana, Ill. This conference was set up by the Commission engineering faculty. We have to do something to on Engineering Education for the purpose of giving shortcut this.In the past this wasn't such a great whatever assistance it could to faculty at engineering deficiency because things didn't change rapidly schools who are responsible forcourses and curricula enough in engineering to make the obsolescence of in engineering design.Our long-range objectives the faculty such a serious mattpr. Now it is be- are to set directions for engineering education as coming so serious that we have to give great atten- affected by its interaction with computers.We tion to ways of buildingup the ability of the faculty want your suggestions, comments and ideas ofways to use new techniques, and one of theways perhaps in which we can help the profession and engineering is to develop such simplified software thateven the education reach its long-range objectives, whether faculty can communicate with the computer! this may be in the form of conferencesor in terms There is also a very serious problem to which of other things that we might do.Please give me only passing attention has been givenso far: that any suggestions that you might have, or transmit is the matter of financing. Financing ofcomputers them to Newman Hall, the Executive Director of the is a very difficult problem. A few institutions, be- Commission on Engineering Education. cause of their small size and good support, or even I am glad to hear the suggestion thatwe have in spite of the fact that theyare relatively large in another conference. The Commissionon Engineer- size but have unusually good support, have beenable ing Education is, of course, concerned with where to solve the problem reasonably well. But thereare do we go from here and how do we get there, and if many schools that cannot afford the kind of equip- enough of you feel that a conference like this is ment and the ways of using this equipment that desirable in a year or two, we shall certainly do could really make great inroads into the problem something about it in the closing discussion. of doing a better job of teaching engineeringde- I would like to summarize my impressions from sign. We have to findsome solution to this problem. this conference.There are several factors that I Finally, there is the most important andmost would like to mention concerning, first, the types exicting matter of the revision of the curriculumto of use of computers, and second, the topics thatwe take advantage of the fact thatnow there are com- have to be concerned with in the development ofour puters available. Theways in which we have done future program. The types ofuse that have been things traditionallyare nut the ways in which they mentioned which are pertinent are:as a teaching should be done to take advantage of the factthat machine; for classroom demonstrations whichre- there is a computer. The kinds of things thatone quire input-output equipment, particularly display learns, the algorithms andso forth, are different equipment, and immediate interaction witha com- for an approach that does notuse a computer computer so that when a problem arises in class youcan pared to one that does, and a complete reorganiza- have it shown on the display ; in the solution of prob- tion of the engineering curriculum isone of the lems, which is a sort of traditional use (although things that we must face in the future. we've been doing this for a long while, thereare I would like to close by acknowledgingthe sup- certainly better ways of solving problems); in per- port of the National Science Foundation in helping forming logical operations, which perhapswe have us present this program, and I would like especially not done as much in the past as we should, and to acknowledge the work that has been done by the which we are certainly looking at for the future; steering committee in bringing together thisgroup for such things as filing and information retrieval; and arranging for the program and the detailsof it. and as a tool (although Bert Herzog doesn't like That committee consists of Steve Fenves,chairman, that word) in an integrated systems approach. Lucien Schmit, Sully Campbell, Frank Branin,Brice To accomplish these objectives we have to be Carnahan, and Sam Shapiro.They all have done concerned with hardware, especially input-output an excellent job and I would like to expressour equipment; software, and this is perhapsour big- thanks to them. gest deficiency; and software as it interacts withour I would like also to expressour appreciation to staff. We have problems of communicationnow be- the Extension Division of the University ofIllinois cause of the fact that it usually takes something and especially to the staff of the ChicagoCircle like a generation to change the background ofan campus for their fine cooperation and hospitality.

174 ATTENDANCE: CONFERENCE ON THE IMPACT OF COMPUTERS ON EDUCATION IN ENGINEERING DESIGN

SECTION G

'r- penrwrmum=oriPnvgqrawswimi:77

iII IIII ItZ I it MDu 2.41

PRECEDING PAGE BLANK- NOT FILMED

ANDERSON, ROBERT B. COWAN, ROBERT A. GLASER, ALAN A:. Carnegie Inst. of Technology Computer Control Co. Univ. of Pittsburgh Pittsburgh, Pa. Framingham, Mass. Pittsburgh, Pa. BALDWIN, GEORGE L. CRANDALL, KEITH C. GOLDSMITH, C. W. Bell Telephone Labs. Univ. of Washington So. Methodist Univ. Murray Hill, N. J. Seattle, Wash. Dallas, Tex: BAMPTON, MERVYN C. C. DAVIS, RALPH H. GOTTFRIED, BYRON S. The Boeing Company Westinghouse Electric Corp. Carnegie Inst. of Technology Seattle, Wash. Pittsburgh, Pa. Pittsburgh, Pa. BARKER, CLARK R. DAWKINS, GEORGE S. GRIFFITH, DEAN E. Univ. of Illinois Univ. of Houston Univ. of Texas Urbana, Ill. `Houston, Tex. Austin, Tex. BARTHA, TAMAS I. DI MEO FRANK N. HALL, NEWMAN A. Pratt Institute Drexel Inst. of Technology Comm. on Engr. Education Brooklyn, N. Y. Philadelphia, Pa. Washington, D. C. BAUMANN, DWIGHT M. DIBOLL, W. B., JR. HALL, WILLIAM J. Mass. Inst. of Technology Washington Univ. Univ. of Illinois Cambridge, Mass. St. Louis, Mo. Urbana, Ill. BEGG, JOHN DIETMEYER, DONALD HANCOCK, JOHN Univ. of South Carolina Univ. of Wisconsin Purdue Univ. Columbia, S. C. Madison, Wis. Lafayette, Ind. BEILFUSS, CHARLES W. DIXON, JOHN D. HART, DONALD E. Meiscon Corporation Univ. of North Dakota GM Research Lab. Chicago, Ill. Grand Forks, N. D. Warren, Mich. BENNER, RUSSELL E. Dix, ROLLIN C. HAUSER, NORBERT Lehigh Univ. Ill. Inst. of Technology Polytech. Inst. of Brooklyn Bethlehem, Pa. Chicago, Ill. Brooklyn, N. Y. BLATHERWICK, ALLAN DOUTY, RICHARD T. HAWTHORN, QUINTIN J. Univ. of Minnesota Univ. of Missouri Tri-State Coll. Minneapolis, Minn. Columbia, Mo. Angola, Ind. BRANIN, FRANKLIN H. DYBCZAK, Z. W. PAUL HAYS, ALAN. S. IBM Corp. Tuskegee Inst. Xerox Corp. Poughkeepsie, N. Y. Tuskegee, Ala. Rochester, N. Y. BRAUN, LUDWIG EVERITT, WILLIAM L. HERMAN, HARRY Polytech. Inst. of Brooklyn Univ. of Illinois Newark .Coll. of Engr. Brooklyn, N. Y. Urbana, Ill. Newark, N. L. BROWN, R. RODERICK FELLING, WILLIAM E. HERRING, BRUCE Honeywell EDP The Ford Foundation Auburn Univ. Waltham, Mass. New York, N. Y. Aubuin; Ala. CARNAHAN, BRUCE FENVES, S. J. HERZOG, BERTRAM. Univ. of Michigan Univ. of Illinois Univ. of Michigan Ann Arbor, Mich. Urbana, Ill. AnnArbor,Mich.. CHASEN, S. H. FORD, JOHN HUBER, PAUL E, Lockheed Georgia Co. Univ. of Illinois Electronic Aspic.; Inc. Marietta, Ga. Urbana, Ill. W. Long Branch; N.. Y.. CONWAY, RICHARD GALL, DONALD A. JUHA, MICHAEL Cornell Univ. Carnegie Inst. of Technology Dartmouth College Ithaca, N. Y. Pittsburgh, Pa, Hanover, N. H.' COTTINGHAM, WILLIAM B. GARDNER, NOEL JOHN KARNAUGH, MAURICE Purdue Univ. Univ. of Western Ontario ,... Bell Telephone W. Lafayette, Ind. London, Ont. Murray Hill; N,. J,..

,"" `-"i"'4,;" , rc"Agwinp-";4T

KATZ, DONALD L. MILLER, CHARLES L. ROCKHILL, RICHARD A. Univ. of Michigan Mass. Inst. of Technology Marquette Univ. Ann Arbor, Mich. Cambridge, Mass. Milwaukee, Wis. KIELING, WILLIAM C. MISCHKE, CHARLES R. ROSENBERG, ARTHUR Univ. of Washington Iowa State Univ. Scientific Data Systems Seattle,. Wash. Ames, Iowa Santa Monica, Calif. KINNEN, E. MORF, THEODORE ROSENSTEIN, A. B. Univ. of Rochester Ill. Div. of Highways UCLA Rochester, N. Y. Springfield, Ill. Los Angeles, Calif. KREPS, S. I. MOTARD, R. L. ROSENTHAL, PAUL H. Newark Coll. of Engr. Univ. of Houston Univac Systems Programming Newark, N. J. Houston, Tex. El Segundo, Calif. LAWLOR, J. J. MOWLE, FREDERICK J. Rossow, E. C. Stevens Inst. of Technology Purdue Univ. Northwestern Univ. Hoboken, N. J. W. Lafayette, Ind. Evanston, Ill. LAWRENCE, WILLIAM N. NEWMARK, NATHANM. SARAJOTL, AMPHORN Univ. of Michigan Univ. of Illinois Univ. of Colorado Ann Arbor, Mich. Urbana, Ill. Boulder, Colo. LEHMANN, JOHN R. NORTH, WALTER SAUNDERS, ROBERT M. National Science Found. Univ. of Windsor Univ. of California Washington, D. C. Windsor, Ont. Irvine, Calif. LEY, B. JAMES 0/PFNER, DAVID H. SCHICK, WILLIAM New York Univ. Univ. of Illinois Fairleigh Dickinson Univ. New York, N. Y. Urbana, Ill. Teaneck, N. J. LICKLIDER, J. C. R. ORGANICK, ELLIOTT SCHILLING, CHARLES H. IBM Corp. Univ. of Houston U. S. Military Academy Yorktown Hghts., N. Y. Houston, Tex. West Point, N. Y. LING, MAR'VIN T. OwEN, C. L. SCHMIT, LUCIEN G. E. Computer Equip. Dept. Ill. Inst. of Technology Case Inst. of Technology Phoenix, Ariz. Chicago, Ill. Cleveland, Ohio LONERGAN, WILLIAM R. PASKUSZ, G. F. SEWER, WARREN D. RCA Corp. Univ. of Houston Univ. of Michigan Cherry Hill, N. J. Houston, Tex. Ann Arbor, Mich. LOWE, THOMAS C. PAYNTER, HENRY M. SEIREG, ALI Univ. of Pennsylvania Mass. Inst. of Technology Univ. of Wisconsin Philadelphia, Pa. Cambridge, Mass. Madison, Wis. LUEBBERT, WILLIAM F. PEARSON, JOHN E. SHANNON, P. T. U. S. Military Academy Univ. of Illinois Dartmouth College West Point, N. Y. Champaign, Ill. Hanover, N. H. MACHOL, ROBERT E. PRYWES, NOAH S. SHAPIRO, SAMUEL E. Dept. of Systems Eng. Univ. of Pennsylvania Univ. of Illinois Chicago, Ill. Philadelphia, Pa. Chicago, Ill. MArrocK, ALAN H. REILLY, Maoism J. SHERIDAN, M. L. Univ. of Washington Carnegie Inst. of Technology Bucknell Univ. Seattle, Wash. Pittsburgh, Pa. Lewisburg, Pa. MEIN, JOHN W. RISWICK, JAMES SIDDALL, JAMES N. Univ. of Illinois Case Inst. of Technology McMaster Univ. Urbana, Ill. Cleveland, Ohio Hamilton, Ont. MER, CHARLES L, at. RIAZ, MAHOUD SMITH, PAUL Fairleigh Dickinson Univ. Univ. of Minnesota Prairie View College Teaneck, N. J. Minneapolis, Minn. Prairie View, Tex. 17$

bti-g 71-r.: =1; Ki

rt,

Soon, GERHARD W. MITER, ELMER J. WORLEY, W. J. General Motors Institute Marquette Univ. Univ. of Illinois Flint, Mich. Milwaukee, Wis. Urbana, Ill. STAGO, GLENN W. WELCH, WALTER R. WORTHINGTON, JOHN H. Am. Elect. Power Service Corp. Prentice-Hall Inc. IBM Corp. New York, N. *Y. Englewood Cliffs, N. J. White Plains, N. Y. VIDALE, RICHARD F. WOLFORD, JAMES C. ZADEH, L. A. Boston Univ. Univ. of Nebraska Univ. of California Boston, Mass. Lincoln, Nebr. Berkeley, Calif. WANG, C. K. WOODSON, THOMAS T. ZIMMERMAN, JOHN R. Univ. of Wisconsin UCLA Pennsylvania State Univ. Madison, Wis. Los Angeles, Calif. Univ. Park, Pa.

179

, J 1