Res Eng Des (1989) 1:51-67 Research in

Engineering
Design
. t - . 1 9 5 ' ) S p r i n v o - V e d i T N c w Y o d , I nc

A Review of Research in Mechanical Engineering Design. Part I:

Descriptive, Prescriptive, and Computer-Based Models
of Design Processes
Susan Finger," John R. Dixon'
'Robotics Institute, Carnegie Mellon University, Pittsburgh, Pennsylvania, USA; 'Department or Mechanical
Engineering, University of Massachusetts, Amherst, Massachusetts, USA

Abstract. This is the first of a two part paper summariz- researchers to place their work in context and thus
ing and reviewing research in mechanical engineering guide continuing work, The series of papers is als6
design theory and methodology. Part 1 includes: 1) de- intended to be an efficient starting place for those
scriptive models; 2) prescriptive models; and 3) com- who wish to become familiar with the engineering
puter based models of design procgsses.
-
design literature relevant to their interests.
which will appear in the next issue of this journal, will
include: 4) languages, representations, and environments 1.1 Scope
for design; 5) analysis in support of design; and 6) design
for manufacture and the life-cycle. For each major area, There are, of necessity, limits to the nature and
we discuss the current topics of research and the state of scope of this review. First, the review is not in -
the art, emphasizing recent significant advances. We also tended to be a substitute for reading complete pa -
discuss the important open research issues in each area. pers; it is intended only as a brief summary of, and
The six categories are .certainly not mutually exclusive guide to, the literature. Although we have made
nor even collectively exhaustive; however, some organi- every reasonable effort to be complete, omissions
zation is necessary, and these categories have been effec-
are inevitable. There can also be errors of commis -
tive in making sense of a body of research that is expand-
ing rapidly in many exciting and promising directions.
sion caused by misinterpretation or lack of full Un-
The mechanical engineering design research community derstanding on our part of papers included in the
has made major advances over the last few .years. The review. We apologize to both readers and research-
research community in mechanical engineering design ers for these errors.
has made significant progress not only in advancing our
knowledge of design, but also in clarifying the research
The scope is limited in several ways. We intend
methods necessary to study design. Great progress is be- only to include research in engineering design, and
ing made toward a better understanding of design, and then only that portion of en ,'ineerin desi rn broad
hence toward better design tools. called "mechanical," which includes roducts ma -
chines, structures, and the like. Research in geo-
metric modeling, architectural design, manufactur-
Introduction
ing, expert systems, and Optimization are included
only when the research is directly relevant to design
This paper, the first in a series of reviews that will of mechanical systems. We have also not attempted
be published in Research in Engineering Design, to cover the many new, commercial computer -
summarizes and reviews the state of research in aided design (CAD) systems which have begun to
engineering design theory and methodology, con- incorporate the research ideas discussed in this re-
centrating on mechanical engineering design. Sub- view.
sequent reviews will concentrate on other areas of
engineering design or on special sub -topics. The The Research discussed in this review a er has
goal of the series is to inform the community at b een co nd ucted p rimaril in t he United State
large of advances in the developments in engineer- Work outside the U.S. has not been excluded, but is
ing design research. We also hope that it will enable riot covered systematically. Finally, research on
mechanical design in very specific technical do -
* Reprint requests: Robotics Institute, Carnegie Mellon Uni - mains (e.g., mechanisms and heat exchangers) is
versity, Pittsburgh, PA 15213, USA
not covered unless it is clearly extendible to other
mechanical design domains.
 52 Finger & Dixon: Research in Mechanical Engineering Design

1.2 Organization classical scientific method used for testing theories

This review is organized into six sections based' on . with the much less well-understOod research
our current view of the active design theory and approaches needed for generating theories. He
methodology research areas. These six areas are: asserts that scientific theory is the ultimate goal of
design research and discusses how pre -theory
I. Descriptive models of design processes
research methodology in engineering design can
2. Prescriptive models for-design
3. Computer-based models of design processes advance toward that goal. His view is that
4. Languages, representations, and environments cognitive studies involve far too many ill-defined
for design variables to support a theory and that prescriptive
5. Analysis to support design decisions models are premature until they can be based on
6. Design for manufacturing and other life cycle is- validated theory; however, he also believes that
sues such as reliability, serviceability, etc. computer-based studies, if used appropriately to
discover and explain the knowledge and strategies
These six categories are certainly• not mutually ex-
needed for design, could lead to the desired
clusive, and some research overlaps two or more
areas. In such cases, we have done our best to in- theoretical foundations.
form readers where research projects have been Yoshikawa [151] in a paper entitled "Automation
placed. of Thinking in Design" has looked at the results of
design research by asking the question: What in de-
sign do we understand well enough to be able to
1.3 Design Methodology
automate? Whether one agrees that the goal of de -
A common thread in this•paper is our discussion of sign research is to automate design, answering the
the research methodologies used to approach de - question makes clear those aspects of design which
sign problems. In a mature field, the research com- are well understood and those which arc not. While
munity will share a common view of what are ap- his paper does not address methodology per se, it
propriate research methodologies, wha t are the defines a criterion for measuring the success of de -
difficult, unanswered questions, and what consti - sign research.
tutes high-quality research. In the emerging field of
In some areas of design research, especial ly
design research, no such consensus exists, This
those that draw from the behavioral sciences, the
lack of consensus causes some chaos, but it also
scientific paradigm of hypothesis testing through
makes the design field exciting and gives promise
experimentation is clearly relevant. Newsome
that revolutionary, new paradigms will emerge; it
[73], Perlman [84], and Adelson [2] have stressed
also fragments the design community and makes
the need for well-designed experiments in design
writing this review difficuIL The following para-
research. They point out that there are well-known
graphs discuss briefly several papers that have ad-
guidelines and methodologies for studying human
dressed the issue of the methods to be used in de -
behavior. Much is known both about the types of
sign research,
evidence necessary to draw conclusions and about
Cross et al. [2 1], in "Design Method and Scientific the types of inferences that can be drawn from ob-
Method," asks whether the scientific method can be servable behaviors. In addition, there is always the
applied to design. After giving a brief history of need for multiple experiments that replicate conclu-
design and the scientific method, they argue that sions from a single experiment or situation, so that
design method and scientific method are not the conclusions can be validated by different experi -
same and that design is a technological activity.' ments studying the same phenomena.
Their arguments reflect the earlier work of Simon
11111 who argues that a theory of design must be a
"science of the artificial." In this phrase, artificial 2 Descriptive Models of Design Processes
retains its original sense of pertaining to an artifact,
that is, something constructed by man as opposed
Many researchers from different fields have studied
to something- existing in nature.
the question of how humans create designs; that is,
Dixon [23] in his paper entitled "On Research they have studied what processes, strategies, and
Methodology Towards a Scientific Theory of Engi- problem solving methods designers user. Most of
the research is based on techniques from artificial
"Fhe book in which this article appears, Design: Science:
intelligence such as protocol analysis in which data
Method [421 is an interesting collection of articles that give is systematically gathered from human subjects.
differing opinions on the possibility of a science of design. The spirit of this research is in sharp contrast to
neering Design," notes that design research is in a earlier work that focused on the development of
pre-theory stage. He compares the well-established
 Finger & Dixon: Zesearch in Mechanical Engineering Design 53

techniques such as brainstorming, inversion, anal - The implication for research on tools to support
ogy, etc., that were designed to enhance the clea - software design is that designers will be aided by a
tivity of a designer, rather than to categorize,.study, whole range of tools. When a tool will be helpful will
or model the cognitive processes themselves. depend on the level of the designer's experience with
The work discussed in this section can be divided the object being designed and the domain being de-
into two categories: one that gathers data on how signed in. For example, when the designer has de-
designers design and the other that builds models of signed an object previously he will need tools that
the cognitive process. (See Adelson Hp In this help retrieve the previous designs. When a designer
has experience with elements of the design but not
cont en t, a co gni ti ve model is a mod el th at de-
with the design as a whole, he needs tools that will
scribes, simulates, or emulates the mental pro -
help to retrieve and assemble the elements in a simu-
cesses used by a designer while creating a design. lation. When the designer has not had experience in
the domain, he will need tools that help him infer the
constraints on the design."
2.1 Protocol Studies of Indiuidual Designers
Much of the design process is a. mental process; the
.itches and drawings that form The visible record Ullman and Dietterich [132, 133] have performed
of designs do not disclose the underlying processes a similar study of mechanical designers looking at
by which they were created. In a design protocol, novice and expert designers designing mass -pro-
the actions of a person performing a design task are dneed and one-off products. The motivations for
recorded as the design evolves, Usually, the de - their study were to understand the mechanical de -
signer is encouraged to think aloud and is ques - sign process better, to improve the efficiency of the
tioned when information seems to be missing or in- mechanical design process, and to support the de-.
velopment of intelligent CAD systems through Al
complete. Because there is no single design process
tools. Of particular interest is their conclusion that
or design strategy, most protocol studies are set up
designers pursue a single design concept, and that
to study a few well-defined questions. For example,
one of the studies discussed below [ compares the they will patch and repair their original idea rather
strategies used by expert designers performing a fa- than generate new alternatives. This single-concept
miliar design task with their strategies performing design strategy, which has also been observed in
an unfamiliar design task. software designers [I], does not conform to the tra-
ditional view of what the design process ought to be
Most of the protocol studies have been done dur- as discussed in Section 3.1.
ing the preliminary design stage. While there is a
consensus that during all phases of design from pre- Waldron et al. [141] study the differences in vi-
liminary to final design, designers exhibit the sual recall by expert and naive mechanical design -
range of design strategies, this assumption has never ers. They conclude that experienced designers han-
been ,steel. In addition, few formal protocol dle visual data more efficiently and that they use
studies have been done on design teams. information at a higher symbolic level. Esterline et
al. [30] initiated a study on the process of problem
One of the major criticisms of design protocols is formulation in the domain of mechanism design.
that a designer's words cannot reveal those pro - However, they found that the process of problem
cesses that are inherently nonverbal, for example, formulation could not be separated from the design
geometric reasoning. Moreover, the requirement to process. Their paper ends with a general formula -
verbalize may interfere with the design process it - tion of the design process for mechanism design
self. Finally, all protocol studies must address the based on the results of their protocol study.
problem that even though subjects may not have
any reason to withhold information, they may do so Scholl [102], based on protocol studies of design-
unconsciously. All of these factors must be taken ers, concludes that designers work from underlying
into account when studying the results of the design types, such as pavilion or .New England farmhouse,
protocols. which are general and yet concretely particular. He
discusses how designers transform the types and
Adelson and Soloway [1] and Adelson [2] have
studied the knowledge and skills used by software how the types transform the design. He discusses
designers. Using protocol analysis [28, 29], they the implications for computer tools that could pro-
have compared novice and expert designers work- vide repertoires of functional types as well as spa -
tial gestalts or backgrounds for the designer. Maher
ing in familiar and unfamiliar domains designing fa-
and Gero [63] have created a system that repre -
miliar and unfamiliar objects. Their focus was on
sents design knowledge as prototypes, which are
identifying the general behaviors used by all the de-
signers and on identifying the specialized behaviors equivalent to what Schou calls types. In their sys -
that arise from knowledge of the domain or prior
experience in designing an object. They conclude:
54 Finger & Dixon: Research in Mechanical Engineering Design

tern, prototypes can be generated, refined and both sociological and technical aspects of the design
adapted to create novel designs. process. One of his more interesting conclusions is
Newsome and Spillers [73] discuss research that designers reuse familiar solutions and will not
methodology in psychology, focusing on methods explore alternatives or innovative ideas unless their
and results that are directly relevant to conceptual new design fails badly and cannot he salvaged. This
design in engineering. They conclude with a -discus- conclusion is supported by some of the protocol
sion of_the implications of this research for support studies cited above.
tools in CAD. Wallace and I-hales [142] advocate participant ob-
servation of engineering design projects and analy-
sis of collected field data. They report on the obser-
2.2 Cognitive Models
vation of a design project that lasted for 2.8 years
The goal of much of the research in cognitive sci - and involved 37 people. They emphasize the impor-
ence is to build computer-based models that..de- tance of distinguishing between the design process
scribe, simulate, or emulate the skills that humans and the context within which the process takes
use as they solve problems [55]. A cognitive model place.
describes the processes that underlie the set of be- Bucciarelli [12] sees design as a fundamentally
haviors that constitute a skill. For example, a cogni- social process. He has studied, through participant
tive model could be created to describe the skill of observation, the discourse of design and how the
remembering a name. The model is specified as a language and multiple representations of design
set of mechanisms with defined functionality; each shape the description of an artifact. Allen [7] has
mechanism is described as a process that can trans- studied the interactions between the designer and
form classes of input into classes of output. The the design, and the environment in which both are
model also specifies the interactions among the embedded. Calling the process situated design, she
mechanisms. Because the model describes a cogni- focuses on how the design itself changes and influ-
tive system at the level of its functional mecha - ences the environment within which it evolves.
nisms, it generates explanations and predictions
about the skill being studied. Stults [116] recorded on video the individual par-
ticipants in a group design project and enabled them
Gero [34] asserts that design systems must be to use the recordings as reference material and as
based on human design processe. Fie argues: `f)e,. 7 documentation for the design. Stults views design
sign paradigms based on mathematical models .in- as a fractal-like process in which the stages of de-
herit the properties of the mathematical models on sign repeat continuously at different times and at
which they are based. Thus, it is possible to prove different levels of detail.
such characteristics as feasibility and optimality
about a resulting design. However, such a design
paradigm has limitations in two major areas: the 2.4 Other Descriptive Models
processes used to achieve designs are far removed Many descriptive models exist which are not based
from the way humans carry out this process; and on formal observation of the design process, but
much of design can only be represented symboli- which make intuitive sense to many designers.
cally but not mathematically." Schools of design have been formed around many
Developing cognitive models to support the pro- of these descriptive systems, for example those of
cess of design is a relatively recent research topic, Ostrofsky 1771, Rzevski [101], Beitz [78], and
so there are few papers to cite. In another paper in Hubka [40].
this issue of Research in Engineering Design, Rveski [101] gives five axioms for design that are
Adelson [3] discusses the methodology used to con- "derived from philosophy, psychology, and the
struct. a model of the skill used in simulating an systems approach." One of his axioms is "Every
incomplete design. design solution (that is, every artefact produced as a
result of design) will inevitably change equilibrium
relationships within its environment and thus create
2.3 Case Studies of the Design Process unforeseen problems."
Most of the case studies of large design projects Warfield [145] presents a Generic Design The -
have been performed in Europe. Two examples of ory, which is intended to be both prescriptive and
this type of research are given by Marples [64] and descriptive. The goal of Generic Design Theory is
Wallace and Hales [142]. Marples describes two de- to capture what is common to all human design pro-
sign projects that we observed. From his observa- cesses no matter what the particular domain. Ge -
tions, he describes a design process that includes neric Design Theory is based on four postulates
Finger & Dixon: Research in Mechanical Engineering 55
Design

which imply the three Laws of Design: the law of other, and most express fundamental, incontrovert-
requisite variety, the law of requisite parsimony, ible philosophies of design.
and the law of requisite saliency. A large body of research has been published in
Another descriptive model of quite a different German on the design process, only a small fraction
character is that of Fitzhorn [32]. Fitzhorn asserts of which has been translated into English. The two
that the design process can be defined to be a Turing books that are readily available in English are Prin-
Machine, that the design artifact is an enumerated ciples of Engineering Design by Hubka [40] and En-
string from a formal grammar, and that the rules of gineering Design by Phal and Beitz [78]. Pahl and
design are formal state changes that govern the Beitz include an extensive German bibliography.
string enumeration. Fitzhorn's work is discussed in Much of the work that is discussed below draws
more detail in Part 11, Section 5.1.2. from the work done in the seventies by Hubka 140],
Koller [52], Rodenacker (97], and Roth [100].

2.5 Summary 3.1 The Canonical Design Process

-it of the research on the design process in me- An interesting controversy is discussed by Broad -
chanical engineering is motivated by the realization bent in the Book Design: Science: Method [42].
that better computer-aided design tools are required He describes an earlier confrontation of three
and that to create better tools requires knowledge of schools of thought within the British design commu-
how designers design. Several authors have dis - nity in which one group believed that the design
cussed possible improvements in the experimental process should be chaotic and creative, another
techniques used by the researchers doing protocol group believed that it should be organized and disci-
studies in this field. To date, much of the work has plined, while the third believed that no process
focused on generating hypotheses based on obser- should be imposed on a designer. This controversy
vations of designers without designing experiments is interesting because most of the recent work on
to test these hypotheses. the human design process in the United States has
In comparison with studies of individual design- focused on what the design process is without dis-
ers, little research has been done on groups of de- cussing what it should be. That is, the focus of the
signers. Some additional work is cited in the section protocol studies is on the design process as prac -
on design environments (Part II, Section 5.5). Also, ticed by individuals, whereas the research dis -
there is a growing body of work on computer-sup- cussed in this section focuses on the process the
ported cooperative work [19] much of which is rele- designers ought to be following. An implicit (and
vant to group design. occasionally explicit) assumption of this research is
that if designers follow the prescribed process, bet -
The field of cognitive modeling is relatively ter designs will result. The authors of this article are
ig, particularly when compared to the field _of unaware of any research in which this assumption is
mechanical engineering. As the body of research on tested scientifically.
cognitive design processes grows, and as the field of
cognitive modeling matures, a better understanding A standard description of what the design pro -
of the skills and strategies used by designers should cess should be recurs in much of the design litera -
emerge. From this understanding, guides to better ture, especially in design textbooks. In an unpub-
computer-based tools for designers should also lished review paper, Eisler [47] summarizes the
emerge. common elements from a number of authors [9, 13,
44, 64, 78] with apparently conflicting views. He
gives the following areas of agreement on the design
process:
3 Prescriptive Models for Design
 The creativity and effort a designer puts into a
Prescriptive models can be divided into two catego - design varies depending on the type of design
ries: those that prescribe how the design process problem. Designs problems are divided into three
ought to proceed and those that prescribe the attrib- types:
utes that the design artifact ought to have. Unlike  Original or new designs
the other sections in this paper in which areas of  Transitional or adaptive designs
general agreement can be covered with minor ex-
 Extensional or variant designs
cursions for differences of opinion, in this section
each prescriptive system stands alone. Although  The design process is an iterative progression
some have attributes in common, few build on each through the following stages:
56 Finger & Dixon: Research in Mechanical Engineering Design

 recognition of need scribes a highly structured process in which, at each

 specification of requirements step, the designer is told what ought to happen
 co ncep t fo r mulatio n . next. Pugh [89] discusses the role of systematic
 concept selection design and concept selection for both conventional
 embodiment of design detail and non-conventional product concepts.
 production,. sales, and maintenance
 In design there are three stages of thought: 3.3 Prescriptive Models of the Design Artifact
 divergence: In this stage emphasis is on ex- The two axiomatic systems discussed in this section
tending the design boundary. The design is each describe the attributes that the designed arti -
unstable, ill-defined, and no evaluation is per- fact should have as opposed to describing the pro -
formed. cess by winch the design should be generated. Inter-
 transformation: In this stage, the problem be- estingly, both systems are particularly concerned
comes bounded, judgments are made, the with the relationship between design and manufac-
problem is decomposed, and subgoals are turing. This link between the creation of the design
modified. and its efficient transformation into a useful artifact
 convergence: In this stage, there is a progres- is rarely a concern in the prescriptive models of the
sive reduction of secondary uncertainties un- design process. Another axiomatic system, that of
til a single design emerges. Yoshikawa et al. [130, 131, 150, 152] gives axioms
for design knowledge and is the topic of another
 A design is strongly influenced by the lifestyle, paper in this issue.
training, and experience of the designer.
Juster cites several authors [41, 128] who studied 3.3.1 Axiomatic design. In Suh's axiomatic de-
design in practice and found that the process in sign system [117-119], there are two axioms for de-
practice did not follow the prescribed process—the sign: 1) Maintain independence of functional re -
variance being attributed both to the designers who quirements, and 2) Minimize the information
are not systematic enough and to the theory that is necessary to meet the functional requirements. Or
unrealistic in its assumptions about the orderliness stated alternatively: a good design meets its various
of the process. functional requirements independently and simply.
Within this axiomatic system, it is important to
make the distinction between functionally indepen-
3.2 Morphological Analysis
dent and physically independent. Two functions
Detailed design methodologies have- been devel- may be performed independently while being em-
oped for some of the stages in the design process. bodied in the same part. For example, a hammer
Morphological analysis [6, 78, 154, 155] is a highly can be used both to drive nails and to pull nails. The
evolved methodology prescribed to generate and two functions are independent even though they are
select alternatives. Morphological analysis is based both performed by the hammer head. Suh has used
on the following assumptions: the axiomatic system to study the relationship be-
tween design and manufacture. Many complete ex-
1. Any complex engineering problem can be di- amples are given in his textbook, Principles of De-
vided into a finite number of subproblems sign [119], as well as in [68] and [92].
2. Each subproblem can be considered separately
and its relations with other subproblems tempo-
rarily suspended 3 .3 .2 Qua lity lo ss. T aguchi [1 2 1 -1 2 5] states
3. All subproblems and their solutions can be pre- that a good design minimizes the quality loss over
sented in a morphological table the life of the design, where quality loss is defined to
4. A global solution to any complex engineering be the deviation from desired performance. The
problem can be found as a combination of solu- system for design developed by Taguchi, often re-
tions to individual subproblems ferred to as "the Taguchi method," uses statistical
5. A global solution can be round in an unbiased techniques, especially design of experiments, for
way through a .random generation of combina- parametric design and tolerance specification.
tions of solutions to subproblems from the mor- Much has been written about the statistical aspects
phological table of this approach, so they will not be discussed here.
Of more interest is Taguchi's definition of the good-
Morphological analysis is most suited for configura-
ness of a design. Whereas the various approaches
tion design as discussed in Section 4.2; however, it
described above assume that a good design meets a
is discussed here because the methodology pre -
set of well-defined functional, technical perfor -
 Finger & Dixon: Research in Mechanical Engineering Design 57

mance, and cost goals, Taguchi is concerned with contrast, expresses a method by which a computer
the sensitivity of a design to uncontrollable factors may accomplish a specified task. A computer-based
that may be encountered in both manufacturing and model may in part be derived from observation of
use. The undesirable, uncontrollable, and costly how humans think about the task, but such a con-
factors that cause a functional criteria to deviate nection is not necessary. Computer -based models,
from target values are called "noise factors." De- if they are successful, may in turn suggest prescrip-
signs that are less sensitive to noise are robust. So, tions for human processes, though this is not a nec-
while other methods are concerned with function essary connection. (For an interesting discussion of
and cost, Taguchi is concerned with robustness. It the relationship between human and machine be -
should be noted that Taguchi is concerned with the havior relevant to manufacturing, see Seering
design of products and processes, not with tradi - [103].) Broadly, cognitive models are concerned
tional quality control as used to control manufactur- with how humans do design whereas computer -
ing processes and meet specifications in a manufac- based models are concerned with how computer
turing environment. can design or assist in designing, A more complete
review, of research in computer -based design is
.given by. Nevi!! [72].
3 .4 S u mma ry
We distinguish between computer processes that
The canonical design process as given by [9, .33,
design (i.e., make design decisions) and those that
44, 64, 78] is not followed precisely by designers
analyze - (i.e., provide information on which design
in practice. This conclusion is supported both by
evaluations and decisions may be based). An analy-
informal observation and by the protocol studies
sis is a procedure which, given a design, provides
covered in Section 2.1. However, no research.that
information on the performance of that design alo ng
we are aware of attempts to verify that, better de-
one or more dimensions. Thus, research in analysis
signs would result if the prescribed process were to
per se is not included in this review of research on
be followed. The implications for future research
design; however, design-oriented interfaces to anal-
are important; if designers do not generate alterna-
yses are included in Part II, Section 6.1.
tives, and generating alternatives results in measur-
ably better designs, then efforts should be focused Optimization methods are closed -form proce-
on methods, such as morphological analysis, that dures that can produce, in certain situations, de -
help designers generate alternatives. The results signs from specifications. We have included se -
from protocol studies should be used to enrich and lected references to optimization in this section,
modify the behaviorally-based descriptive models. although we believe that optimization is a tool to be
Both Pugh [88] and Papalambros et al. [80] have used in a design model when appropriate. Thus,
used college-level design classes to -experiment with some design-oriented interfaces to optimization
different design methods. Pugh concludes: ". . . in methods are included in Part II, Section 6.1.
broad terms, qualitative methods seem to be of lim-
A computer-based model for design generally is
ited use and effect whilst quantitative methods may
specific to a well-defined class of design problems,
act as a positive restriction upon the emergence of
in this review, we refer to the following basic design
the best design solutions."
problem types: parametric, configuration, and con-
ceptual. (or preliminary). Each of these types is de-
hi the methods that prescribe the attributes of the fined below. For a more detailed taxonomy of de -
designed artifact, difficulty arises in measuring and sign problem types, see Dixon [24]. Discussions of
accounting for all the sources of quality loss or in- classifications of design problem and process types
formation gain over the life of a product. The idea of are given by Chandrasekaran [15] and -Ullman 1.1341.
quantifying what is a "better" design is appealing to
engineers, but little work has been done to relate the
life-time performance of designs either to the pro - 4.1 Parametric Design
cess by which it was created or to quantifiable mea-
sures. In parametric design, the structure or attributes of
the artifact are known at the outset of the design
process. Parametric design is then the process of
4 Computer-Based Models of Design Processes assigning values to attributes which are called the
parametric design variables. It should be noted that
The word model is used in two different ways in this the values to be assigned are not always numeric,
review. As used in Section 2, a cognitive model, but may also be a type or class designation (e.g., a
whether given in words or in computer code, is one material choice or a motor type).
that attempts to describe, replicate, or simulate hu- If the goodness of a design can be stated in terms
man mental processes. A computer-based model, in of a single criterion function of the design variables,

A
 10.40040,,,WAVW

58 Finger & Dixon: Research in Mechanical Engineering Design

and if all the associated constraints are known, has been demonstrated in a number of design prob -
and if the criterion function and the constraints lem domains.
can all be expressed quantitatively, then
DPMED PO, 106, 1131 is another hill-climbing
optimization models often can be used. There
algorithm for parametric design, It evaluates its per-
exist a large number of optimization methods that
formance within a design domain based on informa-
can assign values to design variable which maximi-
tion about the goals and the design criteria, along
ze or minimize the criterion function 179, 81, 91,
with a depenAency graph based on analysis equa-
147]. The range of realistic mechanical design
tions. The system has been demonstrated in the de -
problems for which optimization methods are
sign of gear pairs, v-belts, bearings, and shafts.
appropriate has been relatively small, but new
Engenious [7 1] is another parametric design model
methods are appearing that extend their usefulness.
that uses a Comain-specific, rule-based system to
Optimization methods are powerful, and it is
guide redesign. It has some embedded search con-
useful for designers to know when the methods
trol knowledt c to avoid search difficulties that arise
apply and how to apply them. This review will not
from local optima.
cover this specialized, extensive body of
knowledge in detail. For review articles on Qualitative analysis often can provide insight,
optimization research, see Arora 181 and Vanderplaats simplify, or e -fen find solutions in a parametric de-
[1391. sign problem, once decisions have been made about
design varialles, evaluation criteria, and con -
Two optimization methods especially relevant to straints. Agor ino [4, 5] has applied monotonicity
design should be noted. Many design problems in- analysis SUCC , -ssfully for this purpose. In another
volve discrete variables; however, optimizing dis- approach, Ril:derle [941 reduces the coupling [92]
crete variables is computationally intensive, and among the design variables and performance pa -
most optimization models are limited in their ability rameters by ystematie transformations of the de -
to handle such variables. One proposed method for sign variables . An explanation of the approach and
including discrete design variables can be'foundin examples are ncluded in [951.
[14]. Another optimization method of special rele-
vance to design is mixed-integer; non-linear pro- Rossignac A al. [99] have developed a system
gramming. This technique enables configuration de- for capturing t designers' specifications and rela -
sign to be done simultaneously with parametric tional aspects . )1' a design using a sequence of para-
optimization because the presence or absence of meterized tray sformations. Designers can specify
whole components can be included as design vari- their intention., by interacting with'a representative
ables. See, For example, Duran [261, Grossman [371, example of t he design. And, although covered else-
and Kocis 1511. where in this re view, the research of Antonsson and
Wood [148, 14 )] is also relevant to computer-based
models for pat ametric designs.
A review of selected design-oriented interfaces
to optimization methods is included in Part 11, Sec- In a variatic n on the usual concept of parametric
tion 6.1. Most of these interfaces are knowledge - design, Ward rid Scoring [143, 1441 instantiate me-
based; that is, they u s e d e s i gn kn o w l ed ge wi t h i n chanical desig ,s that are given as schematic dia-
a d o m a i n to assist in Formulating an optimiza - grams; their mechanical compiler" then selects
tion problem. A number of knowledge-based ap- components fo)m catalog information embedded in
proaches have been developed for parametric de - their program. The schematic diagrams are at the
sign that do not use existing optimization methods. level of motors pumps, valves, and the like. Thus it
These methods have an interface to their own for- appears that tb program performs both the configu-
mulations an d methods for finding solutions, opti- rations and par metricdesign steps when it selects
mal or not. One of these approaches is Dominic [22, catalog artifact .
39] that uses a method based on iterative redesign.
Brown and 1 Thandrasekaran [111 have created a
Dominic is a hill-climbing algorithm that guides re-
task-level lane wage called the Design Specialists
design by using explicit, but domain-independent,
and Plans Law:nage (DSPL) to model routine de -
knowledge of dependencies between multiple,
sign behavior. -hey assert that routine design is top
fuzzy performance parameters and design vari -
down, and the have created a computer model in -

ables. Dominic 11. [75, 76] has a meta-control system

which specialise:-, choose from existing plans, make
that enables the program to monitor its progress in
commitments, and then instruct lower level special-
order to select more productive strategies from a
ists to refine the plans. They conclude that an ex-
library of strategies. Dominic is limited to paramet-
pert system usir g a hierarchical approach following
ric design of components and standard modules; it
the same procedures humans do shows that com-
cannot handle systems or assemblies in which the
puters can do mntine design work."
parts are decomposed into sub-systems. Dominic
 Finger & Dixon: Research in Mechanical Engineering Design 59

4.2 Configuration Design of abstraction, though he does not apply his theory
to physical artifacts.
In configuration design, or structure design, a phys-
Another important approach to design of assem-
ical concept is transformed into a configuration
blies from components is taken by Rinderle [94]. In
With a defined set of attributes, but with no
this work, the configuration is represented by a net-
particular values assigned. For example, a beam is a
work of parameters (nodes) and constraints (links).
phySical concept with the possible cross-section
Parameters include both design variables and be -
configuration of an I-beam. Research in
havioral, or performance, characteristics. Con -
configuration design can be placed in two
straints express physical principles, spatial rela -
categories. The first is development of an assembly
tions, specified requirements, and material
from a set of standard components (e.g., gears,
limitations. The network can be used to determine
shafts, bearings, and motors); the other is
the important form-function relationships implicit in
development of a non-standard form (e.g., an
the configuration, and thus to evaluate the configu-
extrusion, bracket,. and truss) by redesign or
ration. Additional references to this work include
directly from the functional requirements.
[17, 18]. In related work, Rinderle et al. [82, 83]
describe an experimental system, called Mechani -
cal Engineering Design Assistant, in which compo-
nents are modeled in terms of their behavior and
t.2.1 Configuration of parts into assemblies. A
,

can be aggregated in functional perspectives to cre-

general review of research related to representation
ate trial configurations. The system is implemented
issues in the design of mechanical assemblies is
for electro-mechanical design where the component
contained in Libardi [60]. One of the crucial issues
connections are kinematic relations, and the behav-
in configuring assemblies is the representation of
ioral primitives are stored in a component library.
geometry and spatial relations among the partS.
This subject is treated in a number of paperS includ -
ing [27, 49, 50, 56, 57, 61, 126, 146]. These
papers address the question of how the relative lo- 4.2.2 Configuring non-standard forms. In addi-
cation of parts can be expressed and represented tion to . the configuration of assemblies from compo-
during the design and modification of an assembly nents, configuration design also encompasses the
of parts. Boundary representations and T -matrices development of non-standard forms. To date, re-
are used in various ways, and approaches such as search on the design of non-standard forms has fo-
location graphs and virtual links are developed to cused on relatively simple domains (e.g., beam
represent assemblies. Methods for expressing the cross-sections and structural frames). The auto -
spatial relationships are quite well developed. mated design of structural configurations has been
Methods for propagating change are also reasona - explored by Nevill and Paul in a computer program
bly well understood, though some computational .celled MOSAIC [69-71]. This program produces
is• 3 remain. Less well developed is the ability to a set of possible two-dimensional mechanical frame
design in the presence of abstract or incomplete ge- configurations which connect loads to support
ometry. Some work in this direction has _been done points while avoiding forbidden areas. The system
by Struss [115] and by Shubert [109]. - uses a hierarchical problem decomposition, heuris -
tic knowledge, and the strategy of iterative rede -
sign. Nevill and Paul have begun to extend their
Some research addresses the representation of work to include a high-level representation of the
the connectivity and organization of assemblies design using features in order to reason effectively
without the representation of geometry_or spatial about spatial relations,
relationships. Struss' work [115] is directed at diag-
nosis of machine problems, but his representation A discussion of the synthesis of structural config-
points to ways of providing designers with levels of urations and their selection for numerical optimiza-
abstraction. Even though he does not address geo- tion is presented by Shah [104, 105, 107]. He dis -
metric . and spatial relationships, his system pro - cusses the need for formal synthesis of structural
vides for connectivity among components. In addi- shapes rather than arbitrary selection at the concep -
tion, he discusses the concept of different views of, tual stages of design. He presents a methodology
or ways to partition, an assembly. Shubert [110] for generating structural configurations of slender
also addresses the issues of viewpoints, but in a and polygonal elements using a shape algebra 1108].
different way. He develops what he calls a P -graph T h e s e s h ap e s ar e t h e n s c r e e n ed b a se d o n a l -
to assist in the formal analysis of relationships in an gorithmic and heuristic knowledge such as the load
assembly; that is, he enables the designer to answer paths and stress patterns within a configuration.
such questions as ' ' Is a a part of b?" Hobbs [38] Duffey [25] has produced a system for the auto-
introduces the idea of granularity to describe levels
 f?
4

60 Finger & Dixon: Research in Mechanical Engineering Design

mated design of non-standard extruded shapes is, a rule-based system, is used to fill in the non-
which function as beams. Although this program critical configuration attributes based on manufac-
has a limited domain, it starts from a problem speci- turability. Suzuki points out that top-level design
fication and automatically develops a configuration constraints are inadequate to guide the generation
and instantiation. Currently, the function of the of good configurations; other concerns, such as
beam is only to support bending; work is now M manufacturability, must be taken into consideration
progress to include torsion and manufacturing con- to complete the design.
siderations as well, The parametric instantiation is
done by the Dominic program [75, 76].• In this 4.2.3 Managing constraints in configuration de -
system, Duffey uses configurational operators as sign. Another crucial issue in the configuration of
rules in a production system that transform configu- designs is how to deal with inconsistencies that in-
rations. Fenves [31] and Murthy [67] have also cre- e v i t a b l y a r i s e . I n a s ys t e m c a l l e d R O S A LI E ,
ated design systems that based on transformational Cholvy and Foisseau [16] use a relational data -base
operators. and object-oriented programming to allow designers
An expert system for .configuration .design of to proceed , after a warning, even tho ugh co n -
three-dimensional high rise building structures has straints have been violated. Light and Gossard [61]
been developed by Maher [62] in a program called have developed a system that notifies designers
HI-RISE, The program generates different _struc- when a two-dimensional system is over- or under-
tural configurations, made up of standard structural constrained. In HIMADES-1 [49], created by Ki-
subsystems, based on user defined constraints of tajima and Yoshikawa, a graph structure is used to
size and load. The rules guiding the generation are enable the identification of lOops in connectivity.
based both on heuristic knowledge and knowledge Once a loop is identified, equations are formed and
about basic physical principles. solved using Gauss iteration or Newton -Raphson,
as appropriate; then the design is re-displayed in its
For assistance in the configuration design of inte- new configuration. Popplestone [85 -87] has de-
grated circuits, a program called VEXED has been veloped a system called E.DS (Edinburgh Designer
developed by Steinberg et al. [114]. The program System) that enables a designer to omit details at
uses an interactive rule-based systern to help the the outset and then to selectively add detail as the/
human designer develop the layout, or configura - design progresses. An ATMS (Automatic Truth
tion, or an integrated circuit. An important feature Maintenance System) is used to manage constraints
that this system has is the ability to backtrack to any as they are entered by the designer, and spatial rela-
previous point in the design. Backtracking is useful tionships are computed by a system called RAPT as
to a designer when a line of reasoning leads to unde- the design is modified or extended. Gross et al. [36]
sirable consequences. MEET [90], a program have also created a system that enables the designer
based on the VEXED methodology, has been devel- to navigate the constraints of a design. These con-
oped for the design of mechanical power transmis- straint-based systems represent considerable pro -
sion systems. It uses the input/output requirements gress in handling inconsistencies; however, much
to select and configure gear-pairs, V:belts, and work remains to be done before systems that keep
bearing-shaft assemblies. MEET uses the DPMED track of inconsistencies but allow designers to con-
[106] program to perform parametric instantiation. tinue working, can be realized in practice.
Mittal et al. [66] have developed a computer
program called PRIDE for the design of paper trans-
port mechanisms in copiers. The program uses a
knowledge-base to generate, evaluate, and redesign 4.3 Conceptual Design
configurations of rollers to guide the paper along a
In conceptual design, functional requirements are
smooth path based on design constraints. The sys-
transformed into a physical embodiment or configu-
tem uses a search algorithm with dependency-di-
ration. For example, if the desired function is to
rected backtracking and the added feature of advice
support a load over an open space, then possible
to guide possible redesign paths. This advice is de-
embodiments are a beam, a truss, or a suspension
rived from the evaluation information and is very
structure. Some of the papers cited in this section
important in reducing the amount of design space
discuss configuration design as described in the pre-
searched in order to achieve an acceptable design.
vious section, but the emphasis here is different. In
Suzuki and Kimura [120] are working on optimiz- the papers referenced in Sections 4.2.1 and 4.2.2,
ing the configuration of fiat -.sheet metal parts. The
research involves classifying the parts of a configu
ration to be either critical or not critical to the func -
tion of the component. A production system, that
 Finger & Dixon: Research in Mechanical Engineering Design 61

function is usually implicit and is used primarily to system that selects the best linkage
evaluate designs; in this section, function is explicit configuration for a given set of functional
and is also used to generate designs. requirements. Again, the embodiment, in this case, a
Ulrich and Seering define conceptual design as linkage, is given. This system is based on a
the transformation from functional or behavioral re- system of generalized linkage classification
quirements to structural descriptions; that is, to together with an extensive heuristic knowledge
configurations. In [135, 136], they describe a pro- base.
gram which, given a functional description, creates The subject of innovation or creativity often
new mechanical fasteners from novel combinations arises in connection with conceptual design. Coyne
of existing designs. In this work, the embodiment is [20] argues the view that creativity is knowledge -
given; that is, the system can design only-something based and uses examples of automated design sys-
with a drive, head, body, tail, and tip. The system, tems to study the issue. Both knowledge acquisition
in effect, instantiates the attribute types of the con- and-knowledge control at different levels of abstrac-
figuration (e.g., specifies the body is to be tion are shown to be characteristics of highly crea -
reacted). This system is similar to the morphologi- tive systems.
cal systems described. in Section. 3.2, except that Goel and Chandrasekaran [35] propose case -
here the generation of new designs is driven auto- based reasoning, that is, reasoning from acquired,
matically by the functional requirements. organized, experiential knowledge, as an approach
They have extended this approach to dynamic to reasoning about function in engineering designs.
systems [137, 138] and have created a system that They find that causal understanding of the structure
generates a schematic description of . functional of a design is required to attain the required func -
components which meet the behavioral specifica - tion.
tions. The schematic is transformed into an initial
physical system by substituting one devices for
each function component. Finally, they use debug- 4.4 Models for Distributed Design Problems
ging to improve the physical characteristics of the
Most product designs involve first the design and
design. Their representation of the functional com-
then the integration of components and sub-assem-
ponents is discussed in Part II, Section 5.2.
blies into larger assemblies (e.g., bicycles, automo-
Physical principles are intimately involved in biles, and air conditioners). In this process, the sub-
conceptual design; that is, in the development of assemblies and their components are often designed
physical form embodiments or configurations from independently to keep size and complexity at man-
function. Rinderle [93, 96], for example, con - ageable levels; however, the subcomponents are de-
cludes that "form-function characteristics of me - pendent on each other in many ways. One goal in
anical components and devices exist and can be decomposing design problems into subproblems is
identified from physical principles," and he reports to minimize the inter-dependence, but complete in-
on some initial qualitative progress at relating form dependence is seldom, if ever, possible. The sub -
with function via physical principles. assemblies must share resources such as space,
Cagan and Agogino [13] also relate physical prin- weight, or allowed cost; the assembled components
ciples to function and form. They describe a design may create natural frequency problems; tolerances
methodology called I stPRINCE based on the asser- may stack-up, and so on.
tion that innovative designs require reasoning from
Problems of this type are called distributed. One
the first principles. Analysis of analyses is used to
way to solve distributed problems is to have several
uncover functional relationships among parame -
communicating problem solving agents work on dif-
ters. This work is another exploration of the idea
ferent aspects of the decomposed problem. Re -
that function and form can be operationally con-
searchers in artificial intelligence have worked ex-
nected via operations with expressions of physical
tensively on distributed problems, but their models,
principles.
derived mostly from military applications, do not
Some work has been done in conceptual designs transfer easily to mechanical design. See, for exam-
related specifically to kinematic linkages. Josko - ple, Lesser and Corkill [58, 59]. Although some
wicz [45, 46] describes an algorithm that designs work in this area ultimately may prove useful in
kinematic pairs from a functional specification , mechanical design, we review only those recent pa-
even when the specification is qualitative or incom- pers that apply most directly.
plete. A representation is proposed that uses the A program called PAGES, described by Zarefar
concept of configuration space, an intermediate
step between function and the final design.
Erdman et al. [53, 34, 98, 129] have developed a
 g S ;e , 741-
; ; I; ' 4 5 4

,
Vi4 'e l fb

60: '146VY

62 Finger & Dixon: Research in Mechanical Engineering Design

[153], designs parallel axis gear drive systems that Computer-based models have been developed and
are divided into a gear design module, an axis de- studies for parametric design, configuration design,
sign module, and a lubricant selection module, and for conceptual or preliminary design. Work at
These three subsystems are controlled by a system the parametric level is the most mature, though no
coordinator. The information used as the basis for single theoretical approach has yet evolved. In ad-
control is not reported; however, it appears that the dition to optimization, various knowledge-based ap-
system coordinator uses domain specific knowledge proaches (e.g., Dominic, DPMED, Engenious, and
on how the three subordinate modules design and DSPI.,) have different methods of evaluating the
select. goodness of trial designs and different methods of
Meunier [65] and Verrilli [140] describe a com- using the evaluation results to guide redesign. How-
puter program for hierarchical distributed problem ever, the conceptual similarity among Dominic's
solving that is based on iterative respecification. In dependencies, the sensitivities in Agogino's mono-
this model, system and subsystem nodes, called tonicity analysis, and Rinderle's coupling coeffi -
managers, each formulate a subsystem design that cients is interesting. All relate performance to the
meets the specifications passed down from a higher . design variables. Much research remains before
manager. Managers meet their specifications by these models can handle non-numeric parameters; a.
writing specifications for _subsystem or component material choice, for, example, is a selection of a
design nodes below them, and then integrate the whole set of design variables including yield
subordinates' solutions into a complete sub-system. strength, elastic modules, cost, etc. Also, evalua -
The resulting design is evaluated. If it is not accept- tions and tolerances of the type proposed by Ta -
able, then the manager must change the. subordi- guchi (see Section 3.3.2) are not yet included in the
nate's specifications, hence the term "iterative re- models, Also, except for Engenious, which is used
specification," to obtain subsystem • solutions that inside General Electric, the models have not been
result in a better integrated design. It should be transferred to engineering practice.
noted that this model -allOws for no direct communi-
cation among managers that are at the same level in Research in models for configuration design is
the hierarchy. beginning to appear. Here the paramount question
is the evaluation of a configuration without first as-
Sriram [112] has proposed a system to provide signing values to the attributes. Identifying the ap -
cooperation among multiple designers working in propriate features on which to base evaluation is
different disciplines, especially those connected crucial and difficult. Currently, uninstantiated de-
with architecture and construction. It consists of signs can only be evaluated subjectively and implic-
knowledge modules, control mechanisms, and a ity. As the work in computer-based models for con-
blackboard and is heavily influenced by related figuration , design proceeds, the evaluation will
work in AI. Computer Aided Simultaneous Engi-
neering, or CASE [127], is a proposed system for explicit xplicit based on associated features and
distributed problem solving based on human-com- dependencies. Better evaluation of incomplete de-
puter interaction. Here, however, distributed im- signs will not only make computer -based design
plies the use of multiple representations and agents, more practical, but will also enable designers to ex-
or critics, to provide design or redesign suggestions. plore more configuration alternatives.
Computer-based models for conceptual design
Johnson and Benson [43] has developed a basic are in an even more embryonic state than those for -
two-stage decomposition strategy for design optimi- configuration design. At the conceptual level,
zation. The strategy assumes that all the subprob- rune_ tion becomes an explicit issue, but the link
lems can be solved by optimization techniques and between function and form is not yet fully
that all the subproblems are independent of each understood. Progress is being made, but representing
other. Azarrn [10] applies optimization methods to a and reasoning about function is still an open
decomposed problem using monotonicity concepts. research area.
In both of these models, however, the subproblems
are assumed to be independent, a situation seldom
realized in engineering practice. Kinoglu et al. [48] References
describe a system that designs by decomposing ob- 1. Adelson, 13. and Soloway, E., "A Cognitive Model of Soft-
jects into their smallest parts in tree-like form. De- ‘rare Design," Report 342, Department of Computer Sci -
sign is performed by propagating constraints and ence, Yale, 1984
design requirements throughout the tree. This sys - 2. Adelson, 13., "Modeling Software Design within a Problem-
Space Architecture," Design Theory '88: Proceedings of
tem also handles redesign. 4 . 5 S u m m a r y
 Finger Sc. Dixon: Research in Mechanical Engineering Design 63

the 1988 NSF Grantee Workshop on Design Theory and ence on Expert .Systems in Computer-Aided Design, 1111),
Methodology, Springer-Verlag, New York, 1989 Sydney, Australia, 16-20 February 1987
3. Adelson, 13., "Cognitive Research: Uncovering 11°0 De- 21. Cross, N., Naughton, J., and Walker, 0., "Design Method
signers Design; Cognitive Modeling: Explaining and Pre- and Scientific Method,” Design: Science.. Method, West-
dicting How Designers Design," Research in Engineering bury House, Guilford, England, 1980, pp. 18-27
Design, Vol. 1, No. 1, 1989 22. Dixon, J,R., Howe, A., Cohen, P.R. and Simmons, M.K.,
4. Agogino, A.M. and Alrngren, A.S., "Symbolic Computa - "Dominic 1: Progress Towards Domain Independence in
tion in Computer-Aided Optimal Design," Expert .Systems Design By Iterative Redesign," Proceedings, AS/rIP 1987
in Computer-Aided Design, Gero, J.S., ed., North-Hol- International Computers in Engineering, Conference,
land, Amsterdam, 1987, pp. 267-284 American Society of Mechanical Engineers, Chicago, 11.„
July 24-26, 1987
5. Agogino, A.M. and Almgren, A.S., "Techniques for Inte-
grating Qualitative Reasoning and Symbolic Computation 23. Dixon, J.R., "On Research Methodology Towards A Sci-
in Engineering Optimization," Engineering Optimization, entific Theory of Engineering Design," Artificial Intelli-
Vol. 12, No, 2, Sept/Oct 1987, pp. 117-135 gence for Engineering, Design, Analysis and Mannfactur-
6. Allen, M.S., Morphological Creativity, Prentice-Hall, ing (Al-EDAM), Vol. I, No. 3, 1988
Englewood Cliffs, NJ, 1952 24. Dixon, J.R., Duffey, M.R., Irani, ILK., Meunier, K.L. and
7. Allen, C., "Situated Design," Master's thesis, Design Orelup, M.F., "A Proposed taxonomy of Mechanical De-
Studies, Carnegie-Mellon University, 1989_ sign Problems," Proceedings of the ASME Computers in
8. Arora, J.S. and l3aenzinger, G., "Uses of Artificial Intelli- Engineering Conference, American Society of Mechanical
gence in Design Optimization," Computer Methods in Ap- Engineers, San Francisco, CA, July 31–August 3, 1988
plied Mechanics and Engineering, Vol. 54, 1986, pp. 303- 25. Duffey, M.R. and Dixon, J.R., "Automating the Design of
323 Extrusions: A Case Study in Geometric and Topological
9. Asimow, M., introduction to Design, Prentice-Hall, New Reasoning for Mechanical Design," Proceedings of the
York, 1962 ASME Computers in Engineering Conference, American
10. Azarm, S. and Li, W.C., "Optimal Design Using a Two- Society of Mechanical Engineers, San Francisco, CA, July
Level Monolonicity-Based Decomposition Method," 31–August 3, 1988
Technical report, Department of Mechanical Engineering 26, Duran M.A. and Grossman, I.E., "A Mixed-Integer Non-
and Systems Research Center, University of Maryland, linear Programming Algorithm for Process Systems Synthe-
1986 sis," AIChE Journal, Vol. 32, No. 4, April 1986
11. Brown, D.C. and Chandrasekaran, B., "Expert Systems 27. Eastman, C.M., "The Design of Assemblies," Technical
for a Class of Mechanical Design Activity," in Proceedings Paper 810197, SAE, 1981
of the First International Conference on AI Applications in 28. Ericsson, K.A. and Simon, H.A., "Verbal Reports as
Engineering, Sriram D. and Adey B., eds., Computational Data," Psychological Review, Vol. 87, No. 3, 1980,
Mechanics, U.K., 1986 pp. 215-251
12. Bucciarelli, L.L., "An Ethnographic Perspective on Engi- 29. Ericsson, K. and Simon, H., Protocol A nalysis, MIT Press,
neering Design.," Design Studies, Vol. 9, No. 3; July . Cambridge, MA, 1984
1988 30. Esterline, A., Rosen, D., Otto, K., Nelson, L., Ilessburg,
13. Cagan, J. and Agogino, A.M., "IstPRINCE: Innovative T., Riley, D.R. and Erdman, A.G., "A Methodology for
Design from First Principles," 7th National Conference on Capturing Mechanical Design Expertise," Proceedings of
Artificial Intelligence, AAAI-88, Minneapolis, MN, August the A SME Computers in Engineering Conference, Ameri-
21-26, 1988 can Society of Mechanical Engineers, San Francisco, CA,
•-14. Cita, J.Z. and Mayne, R.W., "Optimization with Discrete July 3I–August 3, 1988
Variables via recursive Quadratic Programming,". Ad - 31. Fenves, S.J. and Baker, N.C., "Spatial and Functional
vance.s• in Design Automaticm-1987, American Society of Representation Language for Structural Design," Ex -pert
Mechanical Engineers, September 1987 . Systems in Computer-Aided Design, Elsevier Science
15. Chandrasekaran, B., "Design: An Information Processing- (North-Holland), IFIP 5.2 1987
Level Analysis," Technical Report, Laboratory for Artifi- 32. Fitzhorn, P., "A Computational Theory of Design," De-
cial Intelligence Research, Ohio State University, 1988 sign Computing, Vol. 2, No. 2, 1988
16. Cholvy, L. and Foisseau, J., "Rosalie: A CAD Object- 33. Fren ch, Engineering Design, The Conceptual Stage,
Oriented and Rule-Based System," InfOrmation Process- Heinemann, London, 1971
ing 83, Mason, R.E.A., ed., North-Holland: Elsevier Sci- 34. Gero, J.S. and Coyne, R.D., "Knowledge-Based Planning
ence, 1983, pp. 501-505 as a Design Paradigm," Design 'Theory in Computer-Aided
17. Colburn, E.R. and Rinderle, J.R., "Identifying Design Re- Design, Proceedings of the IFIP WG 5.2 Working Confer-
lationships," Internal Lab Report CMU-MEDL-88-25, De- ence 1985, Tokyo, Yoshikawa, H. and Warman, E.A., eds.,
partment of Mechanical Engineering, Carnegie Mellon Uni- North Holland, Amsterdam, 1985, pp. 261-295
versity, 1988 35. Goei, A. and Chandrasekaran, B., "Integrating Model -
I8. Colburn, E.R. and Rinderle, J.R., "Supporting Design Rea - Based Reasoning and Case-Based Reasoning for Design
soning with a Constraint-Based Representation," Internal Problem Solving," Technical Research Report, Laboratory
Lab Report CMU-MEDL-88-22, Department of Mechani- for Artificial Intelligence Research, Ohio State University,
cal Engineering, Carnegie Mellon University, 1988 1988
19. Proceedings of the Conference on Computer-Supported 36.. Gross, M., Ervin, S., Anderson, J. and Fleisher, A., "Con -
Cooperative Work, ACM, Portland, OR, September, 1988 straints: Knowledge Representation in Design," pesign
20, Coyne, R.D., Rosenman, M.A., Radford, A.D. and Gero, Studies, Vol. 9, No- 3, July 1988
J.S., "Innovation and Creativity in Knowledge -based 37. Grossman, I.E., "Mixed-Integer Programming Approach
CAD," Proceedings of the IFIP W.G.5.2 Working Confer- for the Synthesis of Integrated Process Flowsheets," Corn-

„10141,4-gby Oggi ,,
r s c r4I01.104,1fiv.W.-.F.4,4
 tiffiV.?

64 Fi n ger & Di xon : R es ea rch in M ech anica l En gin eerin g Desi gn

paters and Chemical Engineering, Vol. 9, No, 1, 1985, 57. Lee, K. and Gossard, D.C., "A Hierarchical Data Struc -
pp. 463-482 ture for Representing Assemblies: Part 2," CAD, Vol. 17,
38. Hobbs, R.J., "Granularity," Proceedings, of the Ninth No. 1, 1985, pp. 20-24
EICAI, International Joint Conference on Artificial Intelli- 58. Lesser, V. and Cotkill, D., "Functionally Accurate Coop-
gence, Los Angeles, CA, 1985,. pp. 1-4 erative Distributed Systems," IEEE Transactions on Sys-
39. Howe, A.E., Cohen, P.R., _Dixon, J.I. and Simmons, tems, Man, and Cybernetics, Vol. 11, No. 1, 1981, pp. 81--
M.K., "Dominic; A Domain-Independent Program for Me- 96
chanical Engineering Design," Artificial Intelligence, Vol. 59. Lesser, V. and Corkill, 13., "The Distributed Vehicle Moni-
1, No_ 1, 1986 toring Test Bed: A. Tool for Investigating Distributed Prob-
40. Ilubka, V., Principles of Engineering Design, Butterworth lem Solving Networks," Al - Magazine, Vol. 4, No. 3, 1983,
Scientific, London, 1980 pp. 15-33
41. Ilykin, D.H.W. and Lansing, L.C.. "Design Case Histo- 60. Libardi, E.C., Dixon, J.R., and Simmons, M.K., "Com -
ries: Report of a Field Study of Design in the United King- puter Environments for the Design of Mechanical Assem-
dom Engineering Industry," Proceedings of the Institute blies: A Research Review," Engineering with Computers,
fur Mechanical Engineers, Vol. 189, No. 23, 1975 1988, pp. 121-136
42. Jacques, R. and Powell, J.A., Design: Science: Method, 61. Light, R. and Gossard, D., "Modification of Geometric
Westbury Efouse, Guilford, England, 1980 Models Through Variational Geometry," CAD—Com-
43. Johnson, R.C. and Benson, "A Basic Two-Stage De- puter-Aided Design, Vol. 14, No. 4, July 1982, pp. 209-214
composition Strategy for Design Optimization," ASME 62. Maher, M.L., "III RISE and Beyond: Directions for Ex-
-

Journal of Mechanisms, Transmissions, and Automation in pert Systems in Design," Computer-Aided Design, Vol. 17,
Design, Vol. 106, 1984, pp. 380--386 1985, pp. 420-427
44. Jones, J.C., Design Methods--7 -he Seeds of Human Fu- 63. Maher, M.L., "A Knowledge-Based Approach to Prelimi-
tures, Wiley—Inter -science, Loudon, 1972 nary Design Synthesis," Engineering Design Research
45. Joskowicz, L. and Addanki, S., "From Kinematics to Center Paper EDRC-12-14-87, Carnegie Mellon University,
Shape: An Approach to Innovative Design" 7th National 1987
64. Marples, D.L., "The Decisions of Engineering Design,"
Conference on Artificial Intelligence, AAAI-88, Minneapo-
IRE Transactions' on Engineering Management, Vol. EM
lis, MN, August 21-26, 1988
46. Joskowicz, L. "Reasoning- about Shape and Kinematic
,
8, 1961, pp. 55-70
Function. in Mechanical Devices," Robotics Research 65, Meunier, K. and Dixon, J.R., "Iterative Respecification: A
Technical Report 402/173, New York University Courant Computational Model for Hierarchical Mechanical System
Institute of Mathematical Sciences, September 1988 Design," Proceedings of the ASME Computers in Engi -
neering Conference, American Society of Mechanical Engi-
47. Juster, N.P., "The Design Process and Design Methodolo- /
gies," Technical report, University of Leeds, May, 1985 ricers, San Francisco, CA, July 31-August 3, 1988
48. Kinoglu, F., Riley, D., and Donath, M., "Knowledge - 66. Mittal, S.M., Dyes, C.L. and Morjaria, M., "PRIDE: An
Based System Model of the Design Process," Computers in Expert System for the Design of Paper Handling Systems,"
Engineering, 1986, Gupta, G., ed., American Society of IEEE Computer, November 1985, pp. 102-114
Mechanical Engineers, New York, 1986, pp. 181-191 67. Murthy, S.S. and Addanki, S., "PROMPT: An Innovative
49. Kitajima, K. and Yoshikawa, H. "HIMADES-1: A Hierar- Design Tool," Proceedings, AAA!. Sixth National Confer-
chical Machine Design System Based on the Structure ence on Artificial Intelligence, uol, 2, AAA', Seattle, WA,
Model for a Machine," CAD, Vol. 16, No. 6, 1984, pp. 299- July 13-17, 1987, pp. 637-642
307 68. Nakazawa, 1I., "Information Integration Method," Design
50. Ko, H. and Lee, K., "Automatic Assembling Procedure and Synthesis, Elsevier Science Publishers B. V (North-
Generation from Mating Conditions," CAD, Vol. 19, No. 1, Holland), 1985
1987, pp. 3-10 69. Nevill, G.E. and Paul, G.H., "Heuristic Evaluation of Par -
51. Kocis, G.R. and Grossman, I.E., "General Optimization of tial Structural Designs," Proceedings of the ASME Com-
Nonconvex Mixed -Integer Nonlinear Programming puters in Engineering Conference, American Society of
(MINLP) Problems in Process Synthesis," 1 and EC Re- Mechanical Engineers, New York, August 1987
search, Vol. 27, 1988 70. Nevill, G.E. and Paul, GIL "Knowledge-Based Spatial
"0, 5 2 Kotler, R., Konstruktionsmethode fur den Mashinen -, Reasoning for Designing Structural Configurations,'' Pro-
Gerate-, and Appariebau, Springer-Verlag, Berlin, 1976 ceedings of the ASME Computers in Engineering Confer -
53. Kota, S., Erdman, A.G., Riley, D.R., Esterline, A. and ence, American Society of Mechanical Engineers, New
Slagle, J.R., "An Expert System for Initial Selection of York, August 1987
Dwell Linkages," Proceedings 1987 ASME Computers in 71. Nevill, G.E., Jackson, L.A. and Clinton, J.H., "Auto-
Engineering Conference, American Society of Mechanical mated Hierarchical Planning for Structural Design," Com-
Engineers, New York, August 1987
puters in Engineering, 1988, American Society of Mechani-
54 Kota, S., Erdman, A.G. and Riley, D.R., "Minn -Dwell- c a l E n gi n e e r s , Sa n Fr a n c i s c o, C A, Au g u s t 1 9 8 8 ,
Computer Aided Design and Analysis of Linkage -Type pp. 395-402
Dwell Mechanisms," Proceedings 1987 ASME Computers 72. Nevill, G.E., "Computational Models of Design Pro -
in Engineering Conference, American Society of Mechani- cesses," Design Theory '88: Proceedings - of the 1988 NSF
cal Engineers, New York, August 1987 Grantee Workshop on Design Theory and Methodology,
55 Laird, I.E., Newell, A. and Rosenbloom, P.S., "SOAR: Springer-Verlag, New York, 1989
An Architecture for General Intelligence," Al Journal, 73. Newsome, S.L. and Spillers, W.R., "Tools for Expert De-
Vol. 33, No. 1, September 1987, pp. 1-64 signers: Supporting Conceptual Design," Design Theory,
56. Lee, K. and Gossard, D.C., "A Hierarchical Data Struc - '88: Proceedings of the 1988 NSF Grantee Workshop on
ture for Representing Assemblies: Part 1," CAD, Vol. 17, Design Theory and Methodology, Springer-Verlag, New
No. 1, 1985, pp. 15-19 York, 1989
 Finger & Dixon: R es earch in Mechanical Engineering Desi gn 65

74. Nicklaus, D.J., Tong, S.S. and Russo, C.J., "Engenious: A puter Aids for Industrial Productivity, Rutgers -State
Knowledge Directed Computer Aided Design Shell," Pro- University of New Jersey, January 1988
ceedings of tine Third Conference on AI Applications, IEEE 91. Reklaitis, C.V., Ravindran, A., and Ragsdell, K.M., Engi-
Computer Society, The Computer Society Press, February neering Design Optiatization, Wiley, New York, 1983
1987 92. Rinderle, J.R. and Sub, N.P., "Measures of Functional
75. Orelup, M.E., Dixon, J.R. and Simmons, M.K., "Dominic Coupling in Design," ASME Journal of Engineering for
II: More Progress Towards Domain Independent Design by Industry, Vol. 104, 1982, pp. 383-388
Iterative Redesign," Proceedings of the ASME Winter An- 93. Rinderle, J.R., "Implications of Function-Form-Fabrica-
nual Meeting, American Society of Mechanical Engineers, tion Relations on Design Decomposition Strategies," Com-
Boston, MA, December 13, 1987 puters in Engineering, 1986, American Society of Mechani-
76. Orelup, Dixon, J.R., Cohen, P.R. and Simmons, cal Engineers, Chicago, 1986, pp. 193-198
M.K., "Dominic 11: Meta-Level Control in Iterative Rede- 94. Rinderle, J.R., "Function and Form Relationships: A Basis
sign," Proceedings, AAA1 Seventh National Conference for Preliminary Design," Proceedings from the NSF Work-
on Artificial Intelligence, AAAI-88, St. Paul, MN, August shop on the Design Process, Waldron, M.B., ed., Ohio
23-26, 1988 State University, Oakland, CA, February 8 -10, 1987,
Ostrofsky, 13., Design, Planning, and Development Meth- pp. 295-312
odology, Prentice-Hall, Englewood Cliffs, New Jersey, 95. Rinderle, J.R. and Watton, J.D., "Automatic Identification
1977 of Critical Design Relationships," ICED 87, International
Paid, G. and Beitz, W., Engineering Design, The Design Conference on Engineering Design, Cambridge, MA, Au-
Council, Springer-Verlag, London, 1984 gust 1987
96. Rinderle, J.R., Colburn, E.R., Hoover, S.P., Paz-Soldan,
79. Papalambros, P.Y., "Integration of Knowledge Forms in
J.P. and Watton, J.D., "Form-Function Characteristics of
Design Optimization," Proceedings from the NSF Work-
Eiectro-Mechanical Designs," Design Theory '88, Pro-.
shop on the Design Process, Waldron, M,13., ed., Ohio
ceedings of the 1988 NSF Grantee Workshop on Design
State University, Oakland, CA, February 8 -10, 1987,
Theory and Methodology, Springer-Verlag, New York,
pp. 417-438
1989
80. Papalambros, P.Y., "The Design Laboratory, Interdiscipli-
nary Research and Education," 1988 NSF Grantee Work- Rodenacker, W.G., Methodisches Konstruieren, Springer-
shop on Design Theory and Methodology, Rensselaer Poly- Verlag, Berlin, Vol. Second Edition, 1970.
technic Institute, Troy, NY, June 2-5, 1988 98. Rosen, D., Erdman, A. and Riley, D., "A General Design
81. Papalambros, P.Y. and Wilde, D.J., Principles of Optimal Knowledge-Based System Shell, with Application to Dwell
Design, Cambridge University Press, Cambridge, England, Mechanism Design," Proceedings 1987 A,S11E. Computers
1988 in Engineering Conference, American Society of Mechani-
cal Engineers, New York, August 1987, pp. 29-36
82. Paz-Soldan, J.P. and Rinderle, J.R., "Abstractions and
99. Rossignac, J. R., Borrel, P. and Nackman, L.R., "Interac -
Structure in Conceptual Design Environments," Internal
tive Design with Sequences of Parameterized Transforma-
Lab Report CMU-MEDL-88-24, Department of Mechani-
tions," Intelligent CAD Systems 2: Implememational Is -
cal Engineering, Carnegie Mellon University, 1988
sues, Springer-Verlag, New York, Eurographics Seminar
83. Paz-Soldan, J.P., "Conceptual Design Environments for
Book 1988
Opportunistic Refinement," Internal Lab Report CMU-
MEDL-88-2I, Department of Mechanical Engineering, /7 1 0 0 . R o t h , K. , Konstruieren m i t K o n s t r u k t i o n s a n l a g e n ,
Carnegie Mellon University, 1988 Springer-Verlag, Berlin, 1982
Perlman, G., "Descriptive Models of Cognitive Aspects of 101. Rzevski, G., "On the Design of a Design Methodology,"
the Engineering Design Process," Design Theory '88: Pro- Design: Science: Method, Westbury House, U.K., 1981
ceedings of the 1988 NSF Grantee Workshop on Design 102. Sefton, D. A., "Designing: Rules, Types, and Worlds,"
Theory and Methodology, Springer-Verlag, New York, Design Studies, Vol. 9, No. 3, July 1988
1989 103. Seering, W.P., "Who Said Robots Should Work Like Peo-
85. Popplestone, R.J., Ambler, A.P. and Baas, I., "An Inter- ple?," Technology Review, April 1985, pp. 58-67
preter for a Language for Describing Assemblies," Artifi- 104. Shah, J.J., "Development of a Knowledge Base for and
cial Intelligence, Vol. 14, No. 1, 1980, pp. 79-107 Expert System for Design of Structural Parts," Proceed-
86. Popplestone, R.J., "An Integrated Design System for ings of the ASME Computers in Engineering Conference,
Engineering," Third ISRR, ISRIZ, Gouvieux, France, American Society of Mechanical Engineers, Boston, MA,
1985 November 1985
87. Popplestone, R.J., "The Edinburgh Designer System as a 105. Shah, J.J. and Pandit, L., "Dezinev—An Expert System
Framework for Robotics," Proceedings of IEEE Interna- for Conceptual Form Design of Structural Parts," Com-
tional Conference on Robotics and Automation, IEEE, puters in Engineering, 1986, American Society of Mechani-
1987, pp. 1972-1977 cal Engineers, Chicago, 1986
88. Pugh, S. and Smith, D.G., "The Dangers of Design Meth- 106. Shah, A., Ramachandran, N. and Langrana, N., "Knowl-
odology," 1st European Design Research Conference: edge Based Design of Primitive Mechanical Components,"
Changing Design, European Design Research Council, Technical Report CA1P-TR-052, Center for Computer Aids
Portsmouth Polytechnic, April 1976 for Productivity, Rutgers University, November 1987
89. Pugh, S., "Further Development of the Hypothesis of 107. Shah, J.J., "Synthesis of Initial Form from Structural
Static/Dynamic Concepts in Product Design," Design and Shape Optimization," ASME Transactions, Journal of Vi-
Synthesis, Elsevier Science Publishers B. V (North-Hol- brations, Acoustics, Stress and Reliability, January 1988
land), 1985 108. Shah, J.J., "Shape Algebra for Structural Configuration
90. Ramachandran, N., Shah, A. and Langrana, N.A., "Ex- Synthesis," Proceedings of the ASME Computers in Engi-
pert System Approach in Design of Mechanical Compo- neering Conference, American Society of Mechanical Engi-
nents," 'Technical Report CAIP-TR-058, Center for neers, San Francisco, CA', July 31-August 3, 1988
Corn-

uw
sq, • V.
,Ir"PAN

66 Finger & Dixon: Research in Mechanical Engineering Design

109, Shubert, L.K., "Representing and Using Knowledge about 129. Thompson, T.R. , Riley, D.R. and Erdman, A.G., "An Ex-
Parts," Computing Science Technical Note, University of pert System Approach to Type Synthesis of Mechanisms,"
Alberta, 1979 Proceedings of the ASAIE Computers in Engineering Con-
110. Shubert, L.K ., "Problems with Parts," Proceedings of the ftrence, American Society of Mechanical Engineers, Bos-
Sixth LICA'', International Joint Conference on Artificial ton, MA, November 1985
Intelligence, Tokyo, 1979, pp. 778-784 130. Tomiyama, T. and Yoshikawa, IL, "Extended General De-
111. Simon, Ii. A., The Sciences of the Artificial, MIT Press, sign Theory," Design Theory in Computer-Aided Design,
Cambridge, MA, 1969 Proceedings of the IFIP WG 5.2 Working Conference 1985,
112. Sriram, D., Logcher, R.D., Grolean, N. and Cherneff, J., Tokyo, North Holland, Amsterdam, 1985, pp. 95 -130
"DICE: An Object Oriented Programming Environment for 131. Tomiyama, T., Kiriyama, T., Takeda, 11., Xue, D. and
Cooperative Engineering Design," Technical report, Intel- Yoshikawa, H., "Metamodei: A Key to Intelligent CAD
ligent Engineering Systems Laboratory, NUT, August 1988 Systems," Research in Engineering Design, Vol. I, No. 1,
113. Steinberg, LI, "Design as Refinement Plus Constraint 1989
Propagation: The VEXED Experience," Proceedings of 132. Ullman, D.G. and Dietterich, T,A., "Mechanical Design
the Sivth National Conference on Artifi cial Intelligence, Nlethodology," Computers in Engineering, 1986, American
AAAI, Seattle, WA, July 1987 Society of Mechanical Engineers, New York, 1986,
114. Steinberg, 1,.1., "Design = Top Down Refinement Plus pp. 173-180
Constraint Propagation Plus What?," Technical report, Al! 133. Ullman, D.G., Stauffer, L.A., and Dietterich, T.G., "Pre-
VLSI Project, Computer Science Department, Rutgers liminary Results of an Experimental Study of the Mechani-
University, August 1987 cal Design Process," Proceedings from the NSF Workshop
115. Stress, P,, "Multiple Representation of Structure and on the Design Process, Waldron, M.B., ed,, Ohio State
Function," Expert Systems in Computer -Aided Design, University, Oakland, CA, February 8 10, 1987, pp. 145-
-

1987 188
116, Stidts, R., "Shoptalk 2", Video Tape, Xerox PARC, Palo 134. Ullman, D-G., "A Taxonomy of the Mechanical Design
Alto, CA Process," Personal Communication, Oregon State Univer -
117. Sub, N.P., Bell , A.C., and Gossard, D.C., "On an Axiom - sity, 1988
tric Approach to Manufacturing and Manufacturing Sys- 135. Ulrich, K. and Seering, W., "A Computational Approach
tems," ASME Journal of Engineering for Industry, Vol. to Conceptual Design," Proceedings of ICED 87, Interna-
100, No. 2, May 1978, pp. 127-130 tional Conference on Engineering Design, Cambridge, Au -
11g. Suh, N.P., "Development of the Science Base for the Man- gust 1987
ufacturing Field through the Axiomatic Approach," Ro- 136. Ulrich, K.T. and Seering, W., "Conceptual Design as
botics and Computer-Integrated Manufacturing, Vol, 1, Novel Combinations of Existing Device Features," Ad-
No. 3/4, 1984, pp. 397-415 van•es in Design Automation-1987, American Society of
119. Sub, N.P., The Principles of Design, Oxford University Mechanical Engineers, Boston, MA, September, 27 -30,
1987
Press, Oxford, UK, 1988
120. Suzuki, H. and Kimura, F., ''Framework for Constructing 137. Ulrich, K. and Setting, W.P., "Conceptual Design: Syn -
intelligent CAD Systems," Proceedings of the IFIP WG thesis of Systems Components," Intelligent and Integrated
5.2 Working Conference on Expert Systedts in Computer - Manufacturing Analysis and Synthesis, American Society
Aided Design, IFIP, Sydney, Australia, February 1987 of Mechanical Engineers, New York, 1987, pp. 57 -66
121. Taguchi, G., "Off-Line and On-Line Quality Control Sys- 138. Ulrich, K.T. and Seering, W.P., "Synthesis of Schematic
tems," International Conference on Quality Control, To- Descriptions in Mechanical Design," Research in Engi-
kyo, 1978 neering Design, Vol. 1, No. 1, 1989
122. Taguchi, G. and Wu, Y., Introduction to Off-Line Quality 139. Vanderplaats, G.N. and Sugimoto, 11. , "Numerical Opti-
Control, Central Japan Quality Control Association, Japan, mization Techniques for Mechanical Design," Design and
1980 Synthesis, Elsevier Science Publishers B. V (North-Hol-
123. Taguchi, G., On-Line Quality Control During Production, land), 1985
Japanese Standards Association, Tokyo, Japan, 1981 140. Verrilli, R.J., Meunier, Kin, Dixon, J.R. and Simmons,
124, Taguchi, G., Introduction to Quality Engineering, Asian M.K., "Iterative Respecification Management: A Model
Productivity Organization, (available from UN(PUB, New for Problem-Solving Networks in Mechanical Design,"
York), 1986 Proceedings, ASME Computers in Engineering Confer -
125. Taguchi, G., System of Experimental Design, vol. 1 and ence, American Society of Mechanical Engineers, New
vol. 2, UNIPUBIKratis International Publications (White York, August 1987
Plains) and American Supplier Institute, Inc. (Dearborn), 141. Waldron, M.B., Jeiinek, W., Owen, D. and Waldron, K.J.,
1987 "A Study of Visual Recall Differences between Expert and
126. Takase, H., Nakajima, N., "A Language for Describing Naive Mechanical Designers," International Conference
Assembled Machines," Design and Synthesis, Elsevier on Engineering Design, ICED, Cambridge, MA, 1987,
Science Publishers B. V (North-Holland), 1985, pp. 471- pp. 86-93
476 142. Wallace, K.N. and Hales, C., "Detailed Analysis of an
127. Talukdar, S., Elles, A., Papanikolopoulos, N., "Concur- Engineering Design Project," Proceedings of the Interna-
rent Design, Simultaneous Engineering and Distributed tional Conference on Engineering Design, ICED, Cam-
Problem Solving," Technical Report, Engineering Design bridge, MA, 1987, pp. 94-101
Research Center, Carnegie Mellon, 1988 143. Ward, A.C. and Seering, W.P., "Quantitative Inference
128. Tebay, R., Atherton, J., and Wearne, S.H., "Mechanical in a Mechanical Design Compiler," Ai Memo 1062,
Engineering Design Decisions: Instances of Practice Com- Department of Mechanical Engineering, MIT, September
pared with Theory," Proceedings of the Institute for Me- 1988
chanical Engineers, Vol. 19813, No. 6, 1984 144. Ward, A.C. and Seering, W.P., "Representing Compo -
 .1! ,11NA 'Pr 14.

Finger & Dixon: R es earch in Mechanical Engineering Desi gn 67

nents Types for Design," Technical Report, Department of 150. Yoshikawa, H., "General Design Theory and a CAD Sys-
Mechanical Engineering, MIT, 1988 tem," Alan-Machine Communication in CAD1CA41, Pro-
145. Warlield, I.N., "Domain of Science Model: Evolution and ceedings of the IFIP WG 5.2/5.3 Working Conference 1980 ;
Design," Proceedings of the Society for General Systems Tokyo, North Holland, Amsterdam, 1981, pp. 35-58
Research, 1986, May 1986
151. Yoshikawa, H., "Automation of Thinking in Design,"
146. Wesley, M.A., Lozano-Perez, T., Lieberman, L.I., Lavin,
Computer Applications in Production and Engineering,
M.A. and Grossman, D.D.; "A Geometric Modeling Sys-
North-Holland, Amsterdam, 1983, pp. 405-417
tem for Automated Mechanical Assembly," HMI Journal
of Research and Development, Vol. 24, No. 1, 1980, 1.52. Yoshikawa, H., "General Design Theory as a Formal The -
pp. 64-74 ory of Design," IFIP WG 5.2 Workshop on Intelligent
147, Wilde, D.J., Globally Optimal Design, Wiley, New York, CAD, IFIP, Cambridge, MA, 1987
1978
153. Zarefar, H., Lawley, Ti. and Eitesami, F., "PAGES: A
148. Wood, K.L. and Antonsson, E.K., "Computations with Parallel Axis Gear Drive Expert System," Computers in
Imprecise Parameters in Engineering Design: Applications Engineering, 1986, American Society of Mechanical Engi-
and Examples," Engineering Design Research Laboratory neers, New York, 1986, pp. 145-149
Report 88-02, California Institute of Technology, June 1988
149. Wood, K.L. and Antonsson, E.K., "Modelling Impreci - 154. Zwicky, F., Morphological Analysis and ConslruCtion, Wi-
sion and Uncertainty in Preliminary Engineering Design," ley Interscience, New York, 1948
Engineering Design Research Laboratory Report 88e, Cali- 155. Zwicky, F., Discovery, Invention, Research through Mor-
fornia Institute of Technology, October 1988 phological Analysis, MacMillan, New York, 1969