On the Evolution of CAE Research

Claremont CollegesScholarship @ Claremont

All HMC Faculty Publications and Research HMC Faculty Scholarship

1-1-1994

On the Evolution of CAE ResearchClive L. DymHarvey Mudd College

Raymond E. LevittStanford University

This Article is brought to you for free and open access by the HMC Faculty Scholarship at Scholarship @ Claremont. It has been accepted for inclusionin All HMC Faculty Publications and Research by an authorized administrator of Scholarship @ Claremont. For more information, please [email protected].

Recommended CitationC. L. Dym and R. E. Levitt, “On the Evolution of CAE Research,” Artificial Intelligence for Engineering Design, Analysis andManufacturing, 8 (4), 275-282, Fall 1994. DOI: 10.1017/S0890060400000950

http://scholarship.claremont.edu

http://scholarship.claremont.edu/hmc_fac_pub

http://scholarship.claremont.edu/hmc_faculty

mailto:[email protected]

Articial Intelligence for Engineering, Design, Analysis andManufacturinghttp://journals.cambridge.org/AIE

Additional services for Articial Intelligence for Engineering, Design, Analysisand Manufacturing:

Email alerts: Click hereSubscriptions: Click hereCommercial reprints: Click hereTerms of use : Click here

On the evolution of CAE research

Clive L. Dym and Raymond E. Levitt

Articial Intelligence for Engineering, Design, Analysis and Manufacturing / Volume 8 / Issue 04 / September 1994, pp 275 - 282DOI: 10.1017/S0890060400000950, Published online: 27 February 2009

Link to this article: http://journals.cambridge.org/abstract_S0890060400000950

How to cite this article:Clive L. Dym and Raymond E. Levitt (1994). On the evolution of CAE research. Articial Intelligence for Engineering,Design, Analysis and Manufacturing, 8, pp 275-282 doi:10.1017/S0890060400000950

Request Permissions : Click here

Downloaded from http://journals.cambridge.org/AIE, IP address: 134.173.130.137 on 31 Jan 2014

Artificial Intelligence for Engineering Design, Analysis and Manufacturing (1994), 8, 275-282. Printed in the USA.Copyright © 1994 Cambridge University Press 0890-0604/94 $5.00 + .00

On the evolution of CAE research1

CLIVE L. DYM1 AND RAYMOND E. LEVITT2

'Department of Engineering, Harvey Mudd College, Claremont, CA 91711-5990.department of Civil Engineering, Stanford University, Stanford, CA 94305.

(RECEIVED May 26, 1994; ACCEPTED June, 18, 1994)

Abstract

Less than a decade ago it seemed that a new paradigm of engineering—called computer-aided engineering (CAE) —was emerging. This emergence was driven in part by the success of computer support for the tasks of engineeringanalysis and in part by a new understanding of how computational ideas largely rooted in artificial intelligence(AI) could perhaps improve the practice of engineering, especially in the area of design synthesis. However, whilethis "revolution" has failed to take root or flourish as a separate discipline, it has spawned research that is verydifferent from traditional engineering research. To the extent that such CAE research is different in style and par-adigm, it must also be evaluated according to different metrics. Some of the metrics that can be used are suggested,and some of the evaluation issues that remain as open questions are pointed out.

Keywords: CAE Research; Computational Paradigms; Designers' Assistants

1. INTRODUCTION

It seemed in the early 1980s as if a new wave of compu-tational power was about to sweep across — and radicallychange—the face of engineering research, education, andpractice. The wave derived in part from the considerablesuccess achieved in computational support for engineer-ing analysis in domains as diverse as the finite-elementanalysis of structures and the nearly automatic layout oflarge-scale integrated circuits. The new computationalwave also derived from and paralleled the success that ar-tificial intelligence (AI) was expected to have in solving"real" problems that were not susceptible to numerical oralgorithmic solution. The "hype" attached to the rapidgrowth of this branch of computer science was also feltin its application fields, and there was widespread opti-mism, at least among its adherents, that some AI para-digms-the most notable being knowledge-based (expert)systems—would provide extraordinary leverage for solv-ing engineering problems. To be sure, there was (and stillis) a countervailing view among traditional, engineeringscience-oriented researchers and practitioners that theseemerging paradigms lacked the rigor needed to make a se-rious contribution, and history may prove that in some

Reprint requests to: Prof. Clive Dym, Department of Engineering,Harvey Mudd College, Claremont, CA 91711-5990. Telephone: 909-621-8853, Fax: 909-621-8967, E-mail: [email protected].

1 This paper is an adaptation and extension of Dym and Levitt(1994).

sense these skeptics are closer to the mark. However, atthat time the pioneers of these new paradigms felt confi-dent that their broader vision would emerge triumphant.

Along with the optimism and the hype, there developedthe expectation that the emphasis of many engineeringgraduate programs would shift from an analytical, engi-neering science style to one that emphasized both numer-ical and nonnumerical computation to formalize a broadrange of engineering tasks. The programs in computer-aided engineering (CAE) would stress software engineer-ing, computer languages, programming environments,and a general understanding of software and hardware is-sues as the counterparts of traditional engineering knowl-edge in much the same way that many engineering scienceprograms stressed applied mathematics as the essentialimplementation ingredient or "language" for doing anyparticular branch of engineering science. And it shouldbe recognized that the thrust of such programs would beaimed both at engineering design perse and at synthesiz-ing engineering design, analysis, and manufacturing. Thatis, the thrust assumed that computational tools to supportengineering analysis were already accepted as an integralpart of engineering practice. In fact, the success of thesecomputational tools provided both a measure of credibil-ity for the further application of computers in engineer-ing and, as noted below, a need for still more powerfulcomputational tools to support the tasks of design syn-thesis, documentation, assembly and manufacturing, andplanning and control.

275

276 Clive L. Dym and Raymond E. Levitt

It was also anticipated that the growth of CAE wouldbe fostered by a continuation of the decline in the cost ofcomputing power. This cost trend would have the resultof making workstations, high-end personal computers, andtheir associated computing environments readily andwidely available. As a result of the increased availabilityof cheap computing power, graduate students and theiradvisers became advocates (and, in some cases, zealots) forthe increased use of computers to support or automate en-gineering tasks. In addition, this cost trend influenced thedevelopment of techniques for distributing and coordinat-ing complex engineering tasks. In particular, powerfullocal and wide-area networks linked engineers one to an-other and made it possible to decompose complex tasksorganizationally and geographically. For example, largeAEC firms such as Fluor, Bechtel, and CRSS routinelypartition large design projects and use computer technol-ogy (e.g., different layers in shared three-dimensionalCAD models) to allow teams of specialists to design dif-ferent parts of the projects from geographically remoteoffices around the world. While such computer-aidedproject decomposition made it easier to employ distrib-uted teams of specialists on complex projects, it also cre-ated significant new problems (and opportunities) in theareas of version control, data and information represen-tation, database accessibility, and propagation of designstandards and techniques.

The interest in the new Al-based paradigms and the de-clining cost of computing naturally fostered research intoapplications of these new styles of engineering comput-ing. However, what developed less rapidly was a concom-itant development of assessment and evaluation standardsfor these new areas of research and practice. Traditionalengineering science research quite easily adopted the ba-sic scientific approach wherein reproducibility was the keyevaluation standard — that is, the ability of investigatorsto reproduce in their own laboratories the work of oth-ers done elsewhere. While developed primarily for exper-imental work, the reproducibility standard was alsoapplicable to analytical work and, after suitable bench-marks were designated by the appropriate communities,to numerical computational work.

The situation for the Al-based paradigms did not al-low a simple extrapolation of the reproducibility standardbecause many of the problems being encapsulated in Al-based programs were viewed as ill-structured — a view-point that is itself open to debate (Reich, 1994)—and theirsolutions as idiosyncratic because the expertise of par-ticular domain experts was being used to develop newknowledge bases (a term, by the way, that did not evenexist before the emergence of AI paradigms because nu-merical- and graphics-based programs were not thought ofas repositories of "domain knowledge"). The applicationof the reproducibility standard was further complicatedby widespread use of different programming languagesand environments, and even more so by the different rep-

resentation formalisms (e.g., rules and objects) as they be-came available.

Given the problems attendant to applying the standardof reproducibility, are there metrics that could be used toevaluate CAE research in general and Al-based researchin particular? In the discussion that follows, four criteriafor evaluating CAE research as an identifiable intellectualapproach to engineering will be discussed and examined.They are as follows:

1. The routine use of KBESs and other Al-based pro-grams in industry to do identifiable and meaning-ful tasks.

2. The ability of such programs, whether or not inpractical use, to point toward useful ways of iden-tifying and formalizing engineering knowledge.

3. The ability of such research to provide further un-derstanding and clarification of the structure of en-gineering knowledge.

4. The ability of such research to extend and enlargethe engineering profession's understanding of engi-neering computation.

It is worth noting that these criteria are not easily quanti-fiable—that is, they cannot be subsumed within one ormore "hard numbers." Thus, at least at this stage of de-velopment of the field of CAE, the community will haveto accept qualitative assessments of the value of the ideasand techniques that are being developed and applied. Ina sense, these criteria can be said to amount to a standardfor assessing the reproducibility of ideas, rather than ofspecific answers or programs.

2. A HISTORICAL PERSPECTIVE

As noted, one of the forces pushing toward the increasinguse of computers in engineering was the success achievedwith analysis tools, most notably those numerical algo-rithms used to analyze complex designed artifacts such asaircraft structures, skyscrapers, integrated circuits, andsophisticated machinery. These algorithmic programswere based on procedural paradigms that employed nu-merical representations of various kinds of formulas. Arelated development, albeit one that came much later andrather more slowly, was the ability to draw and visualizedevices of all kinds. These two developments made it pos-sible for engineers and designers to consider, analyze, andevaluate a much broader range of design alternatives thanwere hitherto possible. That is, the cheapness of computer-based analysis allowed engineers to consider alternativeconfigurations —especially in structural engineering—toa degree that simply wasn't possible when the extensivecalculations needed to be done by hand, with slide rules,or with the aid of old-fashioned, mechanical calculatingmachines. The free-form, thin shells used in the TransWorld Airlines (TWA) terminal at John F. Kennedy Air-

On the evolution of CAE research 277

port in New York City became possible as alternatives tobarrel vaults and hyperbolic paraboloids because of theavailability of reliable computer-based analysis techniques.Other interesting structures — for example, the Frenchheadquarters of the International Innovation Institute(Texedeau & Souchet, 1991) —also became practical asstructural analysts were freed from the need to rely onclosed-form solutions.

In fact, the rapid spread of numerical algorithms in do-mains such as structural engineering has not only led tothe consideration of more design alternatives — whichmight be considered a good result—but has also led to theincreased use of analysis "black boxes" in rather mind-less ways that waste both time and other resources. Some-times, in fact, the mindless use of such tools is dangerousas well as wasteful, as in the recent instance of the fail-ure of an oil platform under construction in the North Seathat was traceable in part to the inappropriate applicationof a finite-element code to analyze a complicated structure.Engineering educators see a microcosm of this problemon a daily basis when students deliver computer-derivedsolutions to problems: solutions whose physical dimensionsand units are wrong, whose variables take on inappropri-ate values, and which show in other ways that studentssimply accept without question the results churned out bywhatever software they are exercising.

In practice, of course, this meant that engineering anal-ysis was being reduced to a technician-level job, whereinthe practitioner was expected to know more about theways to enter inputs than about the meaning and valid-ity of the output. The ease with which computer-basedanalysis could be done thus produced both a burdensomemultiplicity of needless, seemingly endless numerical anal-yses and a new bottleneck in the design process in whichthe engineering talent required to synthesize and evaluatemeaningful solutions became increasingly less commonthan the talent available to generate analyses. The rapidspread of analysis tools was, of course, also fostered bythe proliferation of cheap hardware and software.

Meanwhile, in academe and to a lesser extent in indus-try, researchers were experimenting with the new AI-based concepts—especially those falling under the rubricof knowledge-based or expert systems (KBESs)—to modelvarious engineering tasks. Moving beyond the proceduralparadigms and their numerical representations, research-ers found that knowledge-based systems could be used todeploy engineering knowledge through declarative para-digms based on symbolic representation schemes (Dym &Levitt, 1991 fir). From the start, KBES technology wasattractive because of its capacity to solve ill-structuredproblems requiring the formalization and application ofheuristic knowledge, including that which is more scien-tifically based, even if not representable in formulas andalgorithms. It was recognized that the rule-based para-digms which were the mainstays of early KBESs couldrepresent deep knowledge compiled from first principles,

as well as surface knowledge represented in associativeheuristics. While such technology has begun to find agreater level of acceptance within the broader engineer-ing community, it has required a relatively hard sell bythose who saw its virtues early on.

As mentioned earlier, the contention that the problemsand solutions encapsulated within such KBESs are ill-structured is itself open to debate (Reich, 1994). Indeed,since the knowledge being thus encapsulated is often deepknowledge based on well-understood physical principleswhose application is also well known, one could arguethat the problems are ill-structured only to the extent thatthey do not readily fit into the traditional differentialequation or matrix algebra formalisms for the expressionand application of such knowledge. In this view, it maybe that the debate over the degree of "ill-structuredness"is not especially telling because the focus of discussionmight more properly be on the understanding of the newformalisms and their limits, much as the understandingof the roles of mathematical, numerical, and graphicalmodels are now well known. However, one related aspectof this is the depth of understanding that the developersof engineering KBES-based problem solvers have of thedomain knowledge being encapsulated in these programs.There are several aspects of this issue that are worth ex-ploring, including the achievements of AI researchersworking on their own, the efficacy of collaborations be-tween AI researchers and engineers and designers, and thedomain backgrounds of newly-minted Ph.D.s in the in-fant discipline called CAE.

The first aspect is easy to deal with because there do notseem to be any success stories of engineering or designsystems developed by AI researchers working entirely ontheir own. Indeed, once the developers of "smart" pro-grams moved beyond solving classical puzzles (e.g., mis-sionaries-and-cannibals, the 8-puzzle, etc.), they foundthat they needed to work with domain experts to achievesuccess in any field, whether the domain was medicine,geology, computer configuration, or engineering. This isa lesson that industry seems to have learned and ab-sorbed, perhaps more so than the academic community.The most notable success stories, measured in terms ofdaily and consistent use of KBES technologies by com-panies that are interested in improving their efficiency andbottom-line performance, all involve active and deep col-laboration between researchers versed in KBES ideas andtechniques and respected domain experts. Three recent ex-amples of this drawn from the authors' experience—thereare certainly others—are as follows:

1. The PRIDE system for the configuration design ofpaper-handling subsystems of copiers was developedthrough a collaboration between a respected AI re-searcher, an engineer interested in exploring the rep-resentation of design knowledge, and a team ofengineers and designers who were very experienced in


the design domain (Mittal et al., 1986). The PRIDEsystem is now in active use by designers at the Xe-rox Corporation.

2. The HAZTIMATOR system for configuring soil va-por extraction (SVE) treatment of contaminated soilwas developed in a collaboration between research-ers at Stanford University and hazardous waste re-mediation engineers at CM2 Hill (Oralkan et al.,1994). This system has been validated and will serveas a prototype for a full-blown commercial system.

3. The DEEP system for the configuration design ofelectric service for residential plats was developedby researchers at Harvey Mudd College collaborat-ing with designers and planners at Southern Cali-fornia Edison (Demel et al., 1992). That system iscurrently being integrated into the day-to-day designenvironments used by SCE designers because SCEmanagers anticipate a substantial improvement inproductivity as a result of this integration.

Thus, in terms of the first criterion of success mentionedabove, it is clear that there are examples of success sto-ries, and more are being developed everyday. It is worthasking, moreover, why these development projects pro-duced successful results, whereas research projects under-taken by many academic researchers have not "takenoff," that is, they have generally not led to some obviousreal-world application or integration. In fact, it is inter-esting to note that, while these and other collaborativeprojects were being undertaken, a few advanced researchprograms were turning out Ph.D.s with an almost messi-anic zeal for the technology. This next generation of re-searchers should have formed a corps of shock troopswho should have successfully propagated this new visionof engineering computing. But there was no explosion ofgraduate programs in the CAE arena. Although strongand viable groups were established at a few schools—themost notable one being that led by Professor Steven J.Fenves at Carnegie Mellon University (CMU)—the ideasimply didn't take off. Indeed, while researchers in thisarea can now be found in many American engineeringschools, it is rare indeed that there is more than one fac-ulty member in a department whose principal focus isCAE. Indeed, even the Ph.D. recipients of the few pro-grams that are active are finding it very difficult to findgood academic positions, since they are often viewed aslacking sufficient depth in an engineering domain area(Fenves, 1993).

The lack of programmatic success was not due to ashortage of ideas or vision because, as researchers exper-imented with merging different computational paradigmsand their representations [e.g., Levitt (1990)], an intellec-tual framework was also beginning to emerge, built on thenotion that the Al-based work would serve as the "glue"for truly integrated computational environments for en-gineering [e.g., Dym & Levitt (19916)]. There were early

examples of such integration to lead the way [e.g., Dar-wiche et al. (1989) and Dym et al. (1988)]. Nor was thefailure to propagate and spawn similar academic pro-grams due to the expense of computing. After all, the costof computer power was declining as, over time, KBES de-velopers were able to migrate away from expensive Lispmachines and even more expensive KBES developmentsoftware. In fact, commercial vendors began to produceKBES software that did not require a knowledge of Lispor a deep understanding of underlying representation par-adigms. These expert system shells were relatively easy touse, transparent, and portable to mid-level personal com-puters (such as Mac Us and 286-based PCs). Thus, itappeared that all the ingredients were falling into placefor the heralded computing revolution: cheap machines,software that was both more powerful and more "userfriendly," highly visible research and graduate programs,a wave of graduate students who would spread the gos-pel to other institutions as new faculty members, andgrowing industry interest and support. Thus, CAE shouldhave been a growth discipline in academia, and yet it hasnot turned out that way so far.

Further, in industry and in government, a parallel sit-uation was unfolding in that the anticipated boom inKBES technology simply did not happen as expected(Feigenbaum, 1993). This in spite of incredible declinesin the cost of workstations and software. In fact, even af-ter expert system shells became relatively easy to use,transparent, and portable to high-end personal computers,there was no mass proliferation of KBESs that performedor assisted with engineering tasks which even remotelyparalleled the proliferation of numerical analysis andgraphics packages.

It is worth wondering, therefore, whether the failure ofthe academic discipline to stand on its own reflects —orcould even be caused by—a lack of clear-cut, recogniz-able standards for evaluating success. Thus, it is worth ex-amining what did not happen to see how the lack ofrigorous standards for evaluating research may have in-fluenced the course of development of CAE as a separateacademic field.

3. WHY DIDNT CAE(S) TAKE OFF?

Why did the anticipated boom never happen? One ma-jor reason is buried in a distinction drawn above, namely,that the real focus of these new approaches to CAE wasdesign and design synthesis, what might be called CAE(S).That is, to the extent that experienced engineers feel thatsynthesis is the core part of what they do—the very "soul"of engineering—they are not willing to surrender it to acomputer, at least not without a serious fight.

3.1. The lack of senior management "champions"

In industry, top management sees the value of reorganiz-ing the work process to automate all routine work and,


quite often, semi-custom work as well. Driven by the needto "re-engineer" their work processes in order to remaincompetitive, managers in large engineering organizationsare more than willing to explore alternative ways to re-duce the costs of doing their engineering work. Inasmuchas the kind of work is very labor intensive, the obviousmanagerial solution is to search for ways to reduce thenumber of engineers on staff. Thus, increasingly, man-agers are launching proof-of-concepts studies that focuson using KBESs that perform design tasks now done byexperienced designers [e.g., Demel et al. (1992) and Mis-helevich et al. (1991)]. While the engineers and design-ers provide the domain knowledge encapsulated in theseKBES-based design systems, they also recognize thatthe success of such systems may make their own workredundant. Indeed, as a result, it may well be the casethat engineers would viscerally oppose the thrust of thistechnology—as indeed all the books on labor relations saythey should! In fact, the introduction of KBES applica-tions has been quite successful only when senior manage-ment has been strongly committed to the idea (Riitahuhta,1988).

In fact, notwithstanding the success stories presentedin Section 1, there has been much more success in devel-oping these approaches to design synthesis in Europe,with a concomitant increase in design efficiency and aconsequent reduction in engineering staff. The story ofthe Intelligent Boiler Design System (IBDS) is instructivein this regard. Developed by Tampella Power Industriesof Tempere, Finland (Dym & Levitt, 1991a; Riitahuhta,1998), IBDS is a knowledge-based design system used forthe semi-custom design of power generation boilers. De-signs that would take months are now done in days, andTampella's engineering staff now focuses on product re-search and development rather than on semi-custom de-sign. Similar results can be reported for some of thedesign systems mentioned in Section 1, i.e., PRIDE andDEEP.

3.2. The KBES "consultant" paradigm

Another factor retarding the adoption of expert systemsis the basic consultation paradigm employed by these sys-tems. That is, rule-based KBES design systems are meantto be advisors or consultants, although in this context theyprovide advice to the engineers and designers. In this par-adigm the engineer is then placed in the role of acting asa "sensor for an intelligent machine." Thus, such consul-tative systems are likely to be no more popular with en-gineering designers than earlier medical diagnostic KBESswere with physicians.

3.3. The fragmentation of the AEC industry

And, finally, from the industrial perspective, and espe-cially in domains such as civil engineering and construc-

tion, the fabric of the industry is fragmented. Most AECcompanies are in fact quite small, and they are labor in-tensive rather than capital intensive. Thus, they have lim-ited resources for their own infrastructure development,and they have limited interest in sharing the developmentof appropriate tools with their competitors — unless theyare forced into such investment by the competition. In ad-dition, mitigating against cooperation and data sharingin such a fragmented industry are concerns about liabil-ity and about the ownership of intellectual property.Thus, industry as a whole has not been especially support-ive of academic research in this area—which is in this caseespecially ironic because civil engineering departments areamong the most active academic disciplines in CAE/CAE(S).

3.4. The lack of programmatic financial support

On the academic side, there may be several reasons thatprograms such as the ones at CMU and Stanford Univer-sity have not flourished at many other places. One is, ofcourse, that these ideas emerged at the same time as thebeginning of a slide in the availability of external researchsupport aimed at individual Pi's and projects and whileAmerican colleges and universities were feeling sharp fi-nancial pressures. Thus, the inclination to invest heavilyin new ideas and programs, which was never very strongin the rather conservative engineering education establish-ment, was further reduced by the lack of resources to hirenew faculty and create new laboratories. The successfulprograms flourished because they obtained large-scaleprogrammatic support, such as CMU's Engineering De-sign Research Center (EDRC) and Stanford's Center forIntegrated Facility Engineering (CIFE).

3.5. The perceived lack of domain expertiseof CAE researchers

Further, those departments that were willing to considerhiring new faculty in this area wondered about the depthof the engineering training that the new Ph.D.s were re-ceiving in their engineering fundamentals. There was aconcern that the depth of training in computer and soft-ware engineering was achieved at the expense of depth intechnical areas such as structural or geotechnical engineer-ing. Thus, prospective academic employers were some-what skeptical that graduates of such programs wouldhave the necessary engineering expertise. On the otherhand, and somewhat ironically, for those industrial firmsthat wanted to experiment with KBES technology, suchgraduates were nearly ideal because they could talk bothto in-house computing personnel and to the relevant en-gineers and designers.

The issue of the depth of domain knowledge of CAEresearchers is also important in the context of research as-sessment and evaluation. It was noted above that the most


successful outcomes — at least as measured by the first cri-terion identified in Section 1 —were generated in activecollaborations between academic researchers and domainexperts from industry—and this dichotomy often overlapswith the distinction between computer scientists (the "ac-ademic researchers") and engineers (the "domain ex-perts"). It is, in fact, part and parcel of the same problem.An essential ingredient of a successful collaboration,whether aimed at doing successful research or at build-ing a useful product, is true collaboration between thosewho are really expert in the domain being modeled andthose whose focus is the representation and programmingtechnology. And, to the extent that academic "domain ex-perts" lack real-world experience at dealing with seem-ingly ill-structured problems (i.e., problems not readilyformulated in analytical or numerical terms) or that theirtraining has emphasized the CA part of CAE, the prog-nosis for success is not good. That is, projects fail becausethe domain expertise is not sufficient to deal with the realproblems, whatever they may be.

In saying this, one may in fact be repeating a criticismoften voiced by engineers and managers in industry, thattypical research-minded academics do not have a goodgrasp of real engineering problems. However, it is alsooften the case that academic "domain experts" are notespecially aware of just how brittle their own expertiseand knowledge is. The misuse of FEM packages, for ex-ample, is not limited to novice "black box" users in indus-try. In part this brittleness is a result of increasedspecialization, and in part it is likely a result of a signifi-cant diminution in the depth that doctoral candidates areexpected to show before completing their academic train-ing. This lack of depth will obviously be felt downstreamas new faculty become responsible for training succeed-ing generations of researchers and practitioners. And, itmust be noted, this problem is not limited to students inand graduates of CAE programs, wherein there is an ex-plicit trade-off of depth for a certain kind of breadth.Rather, it is part of a general change in educational val-ues and standards that is moving toward producinggeneralists.

4. LOOKING TOWARD THE FUTURE

What is to be learned from the past that will enable bet-ter preparation for the future, in terms of undergraduateand graduate education, research, and engineering prac-tice? The main points are as follows. First, the age of in-tegrated engineering computing environments is not faroff. Graphics packages are already being converted fromusing layering for data storage and sorting to using object-based representations that facilitate the inclusion of ob-ject attributes and calculation methods with graphical andvisual information (Dym, 1994). And while the KBES ap-plications domain is not as large as anticipated a decade

ago, there are still many researchers and practitioners whoare developing KBESs for engineering design synthesisand as front ends to other engineering tools. And indus-try, in particular, is showing increasing interest in such in-tegrated engineering environments in order to foster moreintelligent design interactions that encompass all thephases of the life cycle of a designed artifact, their moti-vations being to increase both quality and efficiency (byeliminating apparently redundant staff).

Indeed, now that the cost of shells and their platformshas become so low, and their availability so high, manymore companies are coming out of the recession of thelast six years thinking about investing in this technologyfor the future. Higher level tools for design synthesis, basedon AI concepts, are now available (e.g., Design++™,ICAD™). Such tools can be used to reduce both the costof developing designs and the time it takes to introducethem. Further, as noted earlier, many design-intensiveorganizations—such as utilities and large AEC firms—aretrying to reduce their costs by restructuring and shrinkingtheir professional staffs. Thus, the combination of eco-nomic forces, the desire to reinvent the design organiza-tion, and the technical developments in the field may nowlead to the long-delayed boom in the deployment ofKBES design systems. And, as will be noted in the nextparagraph, one of the consequences of the deployment ofKBES technology will be some major shifts in the respon-sibilities and numbers of engineers who will be workingin increasingly computer-intensive environments.

There are several implications for engineering educa-tion. The first is that as more engineering analysis andsynthesis tasks are automated, there is less need for en-gineers who will simply run such programs as black boxes,oblivious to the built-in assumptions and ranges of valid-ity of application. That is, the engineers will truly haveto understand the physical fundamentals of their do-mains, as well as the expected behavior of the artifactsand devices they are analyzing and synthesizing. In designin particular, they will have to recognize the multiplicityof languages and representations of designed objects, sothat they can be conversant with the ways their automated"designer's assistants" are manipulating synthesis infor-mation (Dym, 1993, 1994). Particularly at the undergrad-uate level, one consequence should be an end to theperennial debate about which programming language toteach engineering students. Except for a very small num-ber of engineers who might actually be involved in writ-ing software, most engineers will be users of increasinglysophisticated and accessible software. Thus, studentsshould learn to become discerning and sophisticated usersof packages (whether for spreadsheets, FEM, graphics,etc.) rather than amateur programmers.

At the graduate level, similar cautions would pertain.There ought to be renewed emphasis on fundamentals, onmodeling, on behavior, and on the representations usedto portray the different kinds of engineering knowledge


that are applied in expert problem solving. While the needfor continued research in Al-oriented representation andcomputational integration will persist, the emphasis willlikely shift toward modeling issues that are more organi-zational than technical, as with current work on concur-rent engineering and on modeling design teams (Levittetal., 1994).

Note that these assessments of how undergraduate andgraduate engineering education will change reflect, in ef-fect, the success of CAE(S) as a field when measuredagainst all four criteria set out in Section 1. That is to say,the changes in engineering education that are anticipatedabove mirror the application of new computational par-adigms, a refinement of both the meaning and the struc-ture of engineering knowledge, and continuing change inthe roles that computers play in engineering practice.

Furthermore, it would not be surprising if the spreadof automation to design synthesis for routine and semi-custom work would lead to a significant reduction in thedemand for engineering graduates, particularly those whostop their engineering education at the baccalaureate. Anundergraduate engineering education is still perhaps thebest approach to a "liberal education for a technologicalage," but it will no longer provide an immediate entreeto a well-paid professional life, as has been the case forthe last five decades. Rather, the engineering baccalaure-ate will more likely become an indicator of educationalachievement, much as current B.A. and B.S. degrees areviewed. Thus, the number and kinds of undergradu-ates in engineering might also shift dramatically in thefuture —and the numbers likely in a downward direc-tion — with obvious consequences for engineering schoolsand colleges.

As a natural consequence of some of the previous pos-sibilities, the roles of engineering schools and the natureof their faculties might well be pushed toward a profes-sional model with the characteristics of medical and lawschools. In such a model.the professional degrees wouldclearly be awarded only at the graduate level, and the fac-ulty will become more diverse in the sense that seasonedpractitioners, who may do little or no research, but whowill have experience and contacts in the "real world,"would become mainstays of engineering programs. As ac-complished practitioners and designers become part of thefaculty mix, the reward structure —especially the expec-tations underlying tenure reviews — will have to change,as will perhaps the view of which faculty are truly theelite.

Thus, finally, it would seem that the second and thirdcriteria may be met by the field of CAE(S), if not neces-sarily by individual programs and their creators, becausethis reinvention of engineering education reflects a fun-damental shift in the engineering paradigm. This shift willmove the profession away from the analytically oriented,engineering science paradigm toward one in which knowl-edge is applied through many structures and representa-

tions, often through computational means. The encomiumsand rewards will be earned by those who can decide whatneeds to be done to solve an engineering problem and howthat solution should be implemented, and can then artic-ulate why the approach taken is the correct one.

REFERENCES

Danviche, A., Levitt, R.E., & Hayes-Roth, B. (1989). OARPLAN: Gen-erating project plans by reasoning about objects, actions and re-sources. Artificial Intelligence for Engineering Design, Analysis andManufacturing 2(3), 169-181.

Demel, C.T., Dym, C.L., Summers, M.D., & Wong, C.S. (1992). DEEP:A Knowledge-Based (Expert) System for Electrical Plat Design. Tech-nical Report, Southern California Edison — Harvey Mudd CollegeCenter for Excellence in Electrical Systems, December.

Dym, C.L. (1993). The role of symbolic representation in engineeringdesign education. IEEE Trans. Education 36(1), 187-193.

Dym, C.L. (1994). Engineering Design: A Synthesis of Views. Cam-bridge University Press, New York.

Dym, C.L., Henchey, R.P., Delis, E.A., & Gonick, S. (1988). Aknowledge-based system for automated architectural code-checking.Computer-Aided Design 20(3), 137-145.

Dym, C.L., & Levitt, R.E. (1991a). Knowledge-Based Systems in En-gineering. McGraw-Hill, New York.

Dym, C.L., & Levitt, R.E. (19916). Toward the integration of knowl-edge for engineering modeling and computation. Engineering withComputers 7(4), 209-224.

Dym, C.L., & Levitt, R.E. (1994). Whatever happened to CAE? InBridging the Generations: Proc. Int. Workshop on the Future Di-rections of Computer-Aided Engineering, (Rehak, D.R., Ed.). Car-negie Mellon University, Pittsburgh, Pennsylvania.

Feigenbaum, E.A. (1993). Tiger in a cage: The applications of knowl-edge-based systems. Keynote Lecture, AAAl Symposium, Washing-ton, DC.

Fenves, S.J. (1993). Personal communication, June.Levitt, R.E. (1990). Merging artificial intelligence with CAD: Intelligent,

model-based engineering. Henry M. Shaw Memorial Lecture, NorthCarolina State University, Raleigh, North Carolina; reprinted asWorking Paper No. 28, Center for Integrated Facility Engineering,Stanford University, Stanford, California.

Levitt, R.E., Cohen, G.P., Kunz, J.C., Nass, C.I., Christiansen, T., &Jin, Y. (1994). "The Virtual Design Team": Simulating how orga-nization structure and information processing tools affect team per-formance. In Computational Organization Theory, (Carly, K. &Prietula, M., Eds.). Lawrence Erlbaum Associates, Hillsdale, NewJersey.

Mishelevich, D.J., Katajamaki, M., Karras, T., Lehtimaki, H., Axwor-thy, A., Riitahuhta, A., & Levitt, R.E. (1991). An open-architectureapproach to knowledge-based CAD. In A rtificial Intelligence in En-gineering Design, (Tong, C. & Sriram, D., Eds.), Vol. Ill, AcademicPress, New York.

Mittal, S., Dym, C.L., & Morjaria, M. (1986). PRIDE: An expert sys-tem for the design of paper handling systems. Computer 19(7),102-114.

Oralkan, G.A., Staudinger, J., Levitt, R.E., & Roberts, P.V. (1994). TheHAZTIMATOR: Knowledge-based time and cost estimating of haz-ardous waste remediation projects. Working Paper No. 23, Centerfor Integrated Facility Engineering, Stanford University, Stanford,California.

Reich, Y. (1994). What is wrong with CAE and can it be fixed? In Bridg-ing the Generations: Proc. Int. Workshop on the Future Directionsof Computer-Aided Engineering, (Rehak, D.R., Ed.). Carnegie Mel-lon University, Pittsburgh, Pennsylvania.

Riitahuhta, A. (1988). Systematic engineering design and use of an ex-pert system in boiler plant design. Proc. ICED Int. Conf. Engineer-ing Design, Budapest, Hungary.

Texedeau, J., & Souchet, R. (1991). Innovative architecture realizedusing the finite element method. Structural Engineering Rev. 3(3),181-185.


Clive L. Dym is the Fletcher Jones Professor of Engineer-ing Design at Harvey Mudd College. He has previouslyheld appointments at the University of Massachusetts atAmherst (1977-1991), Bolt Beranek and Newman (1974-1977), State University of New York at Buffalo (1966-1969), and Carnegie Mellon University (1970-1974). Hehas also held visiting appointments at the Technion —Israel Institute of Technology (1971), the Institute forSound and Vibration Research at the University of South-ampton (1973), Stanford University and Xerox PARC(1983-1984), and Carnegie Mellon (1990). After complet-ing the Ph.D. degree at Stanford University (1967), Dr.Dym did research on a variety of problems in applied me-chanics and acoustics. Since the early 1980s his researchhas focused on the development of knowledge-based (ex-pert) systems for engineering design and analysis. He haspublished more than 100 archival journal articles, miscel-laneous papers, and technical reports. Dr. Dym has alsoedited two and written seven books, the latest being En-gineering Design: A Synthesis of Views, published byCambridge University Press (May 1994).

Raymond E. Levitt is Professor of Civil Engineering atStanford University, where he also serves as Associate Di-rector of the Center for Integrated Facility Engineering.After completing the Ph.D. degree at Stanford in 1975,he served on the faculty at MIT for five years before re-turning to Stanford. Dr. Levitt's early research employedorganization theory and risk analysis to model decisionmaking in facility engineering organizations. Since 1980he has explored the use of artificial intelligence andknowledge-based systems as nonnumerical tools for sitelayout, design synthesis, project planning and control,and other facility engineering applications. Dr. Levitt's in-dustrial experience includes design, estimating, and on-site management positions in marine and buildingconstruction, and numerous consulting assignments onorganizational design, safety management, and expertsystem applications. He is also a director of DesignPower, Inc., and serves as a scientific or technical advi-sor to several companies in the United States and Europe.Dr. Levitt is also the co-author, with Professor CliveDym, of the book, Knowledge-Based Systems in Engi-neering, published in 1991 by McGraw-Hill.

Documents

On the Evolution of CAE Research