|
Items in cart [0] |
Questions? Call 888-624-8373 |
| ||||||||||||
| Copyright © 2009. National Academy of Sciences. All rights reserved. Terms of Use and Privacy Statement |
Below are the first 10 and last 10 pages of uncorrected machine-read text (when available) of this chapter, followed by the top 30 algorithmically extracted key phrases from the chapter as a whole.
Intended to provide our own search engines and external engines with highly rich, chapter-representative searchable text on the opening pages of each chapter.
Because it is UNCORRECTED material, please consider the following text as a useful but insufficient proxy for the authoritative book pages.
Do not use for reproduction, copying, pasting, or reading; exclusively for search engines.
OCR for page 17
4
METHODS, THEORIES, AND TOOLS
Various tools are commonly used to aid designers, and several additional theories offer more
analytically rigorous support to engineering designers. Concurrent engineering may be the most
practical method to improve the design process, and other common tools are used to obtain input
from stakeholders in the design process (the Pugh Method, Quality Function Deployment,
Decision Matrix techniques, and the Analytical Hierarchy Process). These tools incorporate
relatively high levels of subjective judgment. An additional set of tools address variability,
quality, and uncertainty in the design process (Projected Latent Structure, the Taguchi method,
and Six Sigma). These tools are more analytical and are typically coupled to the processes used to
produce products. Still other tools are used to generate alternatives for designers (artificial
intelligence and TRIZ). Design theories also exist (Dym's, Suh's Axiomatic, Yoshikawa's, and a
Mathematical Framework) that are less widely used but offer more rigorous analytical bases.
Finally, certain other tools are used primarily in the fields of management science and economics,
and are being explored in current research for applicability to decision making in engineering
design.
This review is illustrative rather than exhaustive given the voluminous existing literature
about each tool and theory. Series of individuals and groups are devoted to research and
application for each separate tool or theory mentioned here. Subjective comments on the
applicability or limitations of the tools and theories are of course based only on the authors'
knowledge of the tools and of the relevant literature.
CONCURRENT ENGINEERING
Concurrent engineering is defined as "a systematic approach to the integrated, simultaneous
design of products and their related processes, including manufacture and support. This approach
is intended to cause the developers, from the outset, to consider all elements of the product life
cycle from conception through disposal, including quality, cost, schedule, and user requirements"
(Winner et al., 1988~. According to Dean and Unal (1992) concurrent engineering consists of
getting the right people together at the right time to identify and resolve design problems.
Concurrent engineering includes designing for assembly, availability, cost, customer satisfaction,
maintainability, manageability, manufacturability, operability, performance, quality, risk, safety,
schedule, social acceptability, and all other attributes of the product. Table 4-1 highlights the
contrast between concurrent engineering and conventional engineering design.
The concurrent engineering environment has the following characteristics:
Reduced cycle time
Overlapping of functional activities
Collaboration in functional decisions
5
17
OCR for page 18
18
APPROACHES TO IMPROVE ENGINEERING DESIGN
Table 4-1 Comparison Between Concurrent Versus Linear (Serial) Engineering
Concurrent Engineering
Parallel design of products and processes
Multifunctional team
Linear (Serial) Engineering
Sequential design
Independent designer
Concurrent consideration of product life cycle
Total quality management tools
All stakeholder inputs
Sequential consideration of product life cycle
Conventional engineering tools
Customer and supplier not involved
Concurrent evolution of system and component decisions
. Critical sequencing
Concurrent engineering strives to meet the need for continually shorter product development
cycles and the need to represent the inputs of all stakeholders. Decision-making tools and
processes continue to evolve to support making the best decisions in this environment. In the
concurrent environment functional activities such as engineering and manufacturing are done
simultaneously. This is contrasted with a traditional serial process in which the design team
finishes its work before the drafting department prepares and releases the drawings, at which time
manufacturing starts. Because of the overlapping of functional activities, cycle times may be
greatly reduced and decisions become highly interdependent. This environment causes many
decisions to be made without complete information, which has led to the development of methods
to share product requirements, to assess risks effectively, and to develop abatement plans. System
and component trades must be made concurrently, which requires effective methods for flow-
down and tracking of requirements, as well as flow up and tracking of status. The tight schedules
resulting from reduced cycle times require disciplined scheduling and monitoring of key
decisions as well as management of the required sequencing and the downstream impact of
design decisions. Cross-functional decision making requires integrated, highly reliable, and
readily accessible databases.
The tools and methods being developed to support this environment, most of which are
computer based, can be categorized as follows:
t
.
LAN- and Web-based management tools. Often an individual decision (say,
material selection for a component) is dependent on a higher level system
requirement (say, operation in a corrosive environment). Without good
management tools the individual decision maker may be unaware of the
requirement. These management tools usually include a function for informing
the individual decision maker of multi-level critical decision dates and status. An
informative interconnected data management system is essential.
Tools and processes for assessing risks and developing risk abatement plans.
Because of compressed cycles and overlapping functional activities, decisions
must be made with incomplete information or data. For example, manufacturing
of components may begin before the final product drawing is complete and
issued. There is inherent risk in starting manufacturing prior to drawing release
as a change in definition may result in rework or scrap. Tools need to be
s
?
OCR for page 19
s
IMPLICA TIONS FOR EDUCA TION AND RESEARCH
developed to aid the decision maker to assess the risks and develop suitable
abatement plans. Work with design structure matrices is a good start in this area.
When these situations are routine occurrences, as with the above example,
standard abatement plans may be developed.
.
19
Automated analysis tools for rapid evaluation of alternatives. The accuracy
of decisions and reduction of risk can be greatly improved by the quantification
of alternatives. Many tools exist or are under development to automate or codify
analysis to improve cycle time and efficiency. Parametric Modeling, Design of
Experiments, and Automatic Finite Element Meshing are examples.
Data-sharing tools. In a cross-functional concurrent decision making
environment with compressed cycle times, data must be readily available to all
decision makers. Data must be accurate, timely, complete, easily accessed, and in
many cases, configuration controlled. The data must be informative, not merely
numerous. An excellent example of a shared database is the use of three-
dimensional solid models for distributing geometric information for drawing,
tooling, manufacturing, and customer appraisal analysis.
As decisions are made in the concurrent environment, the impact of each decision on the
product is evaluated. This makes the concurrent engineering environment not only contextual but
also a fundamental part of the decision process. As a process rather than a decision tool, it
requires a wide variety of supporting tools as described above and in the following sections.
Through continual feedback, the concurrent environment decision process facilitates evaluation
and modification of decisions whenever undesirable or unexpected consequences result.
TOOLS TO OBTAIN STAKF.HOLDER INPUT
THE PUGH METHOD
Stuart Pugh, University of Strathclyde, is the author of two books discussing the problems of
engineering design (Pugh, 1990, 1996~. The second book (Pugh, 1996) is largely a compendium
of his many papers.
Pugh (1990) developed the product development process and a set of discipline-independent
methodologies to carry out the process, such as customer surveys, the product design
specification document, and the method of controlled convergence. In the Pugh method a
decision matrix is prepared with columns to identify design concepts (variant) and the rows to
represent criteria. A design team chooses both concepts and criteria. One of the column concepts
is chosen as a datum against which all others are to be judged. In the matrix cells for each row
criteria a plus (+), zero (0), or minus (-) sign is then used to indicate whether the concept is
better, equivalent, or less than that of the datum. For each concept the number of plus and minus
signs is noted and the best concept is selected. Omitted concepts with unique plus cells are
especially studied to provide insight. New concepts (variants or designs) are now formed, criteria
modified and added, a new datum selected, and the process repeated. The process requires
continuous elaboration until the datum column becomes uniquely best. This variant initiates the
final detailed design process.
Pugh argues against ranking or weighting of either concepts or criteria beyond the simple
plus (+), zero (0), and minus (-I. In defense of his matrix approach to engineering design Pugh
states, "The matrix does not make the decisions: it is simply a procedure for controlled
convergence onto the best possible concept and is not composed for absolutes in the mathematical
l
2
OCR for page 20
q
20
APPROACHES TO IMPROVE ENGINEERING DESIGN
sense; the decisions remain with the user" (Pugh, 1990~. The authors agree with the method's
creator that his method should not be used to make decisions.
QUALIltY FUNCTION DEPLOYMENT
Quality Function Deployment (QFD) has its origins in Japan in the late 1960s, during an era
when Japanese industries broke from their post-World War II mode of product development
through imitation and moved on to product development based on originality (Akao, 1997~. It
provides a synthesis of the concepts and tools required to translate customer aspirations into the
final technical and manufacturing activities needed to produce a quality product. QFD is a matrix
management tool meant to translate the factor termed the "voice of the customer" into items that
can be measured, assessed, and improved. It is particularly useful in finding gaps in a developing
program and offering opportunities for new ideas to enter a design activity.
Don Clausing of the Massachusetts Institute of Technology is largely responsible for the
introduction of QFD methodology to U.S. industry in the early 1 980s. The blending of the ideas
of QFD with those of Pugh's Total Design theory is often called EQFD (Enhanced QFD)
(Clausing and Pugh, 1991).
Quality Function Deployment is used to identify critical customer attributes and to create a
specific link between customer attributes and design parameters. Matrices are used to organize
information to help marketers and design engineers visualize and answer three primary questions:
What attributes are critical to our customers?
What design parameters are important in meeting those customer attributes?
. What should the design parameter targets be for the new design?
QFD constructs, in a fashion similar
to Pugh, an evaluation matrix at each
stage of product development both to
display relevant states of knowledge and
to provide a mechanism for decisions.
This matrix correlates the identified
customer needs called "the whets"
(rows) to the engineering specifications
called "the bows" (columns). Ideally, a
cross-functional team made up of
members from the core functions in
product development should develop the
matrix. The matrix rows are usually
descriptive verbalizations of the
customer's wants. The matrix columns
are more likely to be quantitative
measures of engineering requirements.
The engineering measures are frequently
correlated, or represent incompatible
desiderata, and an additional triangular
array is often placed atop the matrix
displaying these associations. The
schema, often called the "House of
Quality," is illustrated in Figure 4-1.
C n ~ n a ~ n s u c s ~
R~UI!~mCDb Sl S] S] S4 SS S6 S7 S8 S9 S10 S]l SIt
N.
N2
N.
N5
N6
Hi,
~8
Ng
~ _ _ _ -
8~eir`6l221191 1 1~110135118121 _ 4) 12
Manic of
Associz~on
Figure 4-1 The House of Quality. Source: Carriere and
Finster (1989).
OCR for page 21
IMPLI CA TI ONS FOR ED UCA TI ON AND RESEA R CH
A
A
Key proceSS 1\ 1 ,
operatlor~s ~ \ .
A
/ \
Production I
requ~romonN |
. ,
/ ~ , / \
l inginefftng ~ Ports 1\
63~L (: - '1
~1
_ _
IY]
(1~1
IV
HOUSE PARTS PROCESS PRODUCTION
OF QWWU DEPLOYMENT PLANNING PLANNING
Figure 4-2 A cascade of evaluation matrices. Procedure adapted from Hauser and Clausing (19884.
The first step in building a House of Quality is to construct an evaluation matrix indicating
how each engineering specification affects each customer need. Many cells of the matrix may be
empty (blank), while others will contain a score testifying to the importance of the row-column
association. Information on technical difficulties and cost can also be displayed. An additional
step is to construct the "roof" for the house, a triangular matrix displaying relationships among
engineering specifications. For instance, specifications for take-off weight for a commercial
aircraft will strongly affect the required take-off distance. The roof of the House of Quality also
provides a good indicator of design trade-offs to consider in the future. Additional information
matrices may be attached to the "house." For example, information on customer perceptions of
competing products and competitive benchmark data can provide "wings" and "cellars."
The essential motivation for constructing a House of Quality schema is to provide a viewer,
new or old to the design problem, a quick and thorough appraisal of its scope. It is an aid to the
decision-making process.
The House of Quality can be used as a stand-alone tool to generate answers to a particular
development problem. Alternatively, it can be applied in a more complex system in which a
series of decision tools are used. The procedure is illustrated in Figure 4-2.
The use of evaluation matrices is only a part of a larger philosophy implied by QFD. QFD
also emphasizes the importance of teamwork by diverse experts. Moreover, customers are not
viewed solely as those who will use the final product but also as those at the receiving end of
each stage of the creation of the final product.
DECISION MATRIX TECHNIQUES
Decision matrix techniques are used to define attributes, weight them, and appropriately sum
the weighted attributes to give a relative ranking among designs. An example of the framework
for such a process is shown in Figure 4-3.
The advantages of this method are as follows:
.
q
The method encourages team interaction (causes the design team to consider
attributes of a variety of potential solutions and their relative importance and thus
a good way to help calibrate the team).
Analysis can be performed relatively quickly.
The method can identify non-viable design variant options and remove them
from further consideration.
OCR for page 22
22
APPROACHES TO IMPROVE ENGINEERING DESIGN
Disadvantages of this method are the following:
Criteria may have interdependencies.
Risk must be overtly addressed as an additional criterion.
Subjective weighting often reflects the design team's opinion rather than the
customer's view.
It is not a stand-alone decision tool.
. Sometimes it is used merely to rationalize decisions.
The decision matrix is prepared using technical or economic criteria (or both). When all row
criteria are not be considered equally important they are weighted individually. For every column
concept (variant) each criterion is assigned a value, typically on a scale of 1 to 10. The final
weighted value for all matrix cells is found by multiplying the criteria weights by their
corresponding design variant value. The sums of both the values and the weighted values are then
found for each concept variant.
The "Weighted Sum of Attributes" decision matrix shown in Figure 4-4 is an example typical
of frequently encountered design decisions in which a variety of concepts are viable but vary
considerably in their ability to meet conflicting requirements. In the example, for instance, all the
concepts will provide attachment, but only one has loose parts. To help reach a decision, as only
one concept will be used, the decision matrix shown in Figure 4-4 was prepared. As is often the
case with this tool, (1) the weighted sums are not highly differentiated; (2) the weighting changes
the decision relative to the unweighted sums, thus demonstrating the need for careful selection of
the weights; and (3) the values often have to be normalized to prevent the dominance of one key
requirement due to its high values. In the current example, if the parts count were in the hundreds
of parts, its value would dwarf other considerations, unless of course, it was assigned a very low
weight.
This example does not indicate a clear winner, merely that one choice (Quick-Nuts) can be
eliminated, provided the weightings and assigned values are reasonable. The matrix evaluation
nevertheless can still be a usefi~1 tool. Further consideration of the weightings and playing "what
if'' will at least allow the decision maker to understand which requirements have been emphasized
in the selection of a concept and which requirements may need careful attention in detail design.
.~
Concept Variants
Conceptl | Concept2 | Concept3
Criteria Weights Value I WV I Value T wv I Value
Criterion 1 l l l l
Criterion 2 T ~ I
Criterion 3 T I I
Criterion 4 _
.
...
:
Sum: ~ ~
Weighted Sum: .
WV
Figure 4-3 General format of the decision matrix.
l
OCR for page 23
IMPLI CA TI ONS FOR ED UCA TI ON A ND RESEA R CH
23
Concept Variants
Kit ~ Requirements Cam-Lock Screws Quick-Nuts
.
Weighting Value Weighted Value Weighted Value
Factors ~ Value Value
2 Part count # 5 10 12 24 12 .
10 Loose parts # O O 12 120 12_
.
8 Weight in Ibs _ 10 80 1 8 2 .
10 Reliability MBTF 10 100 10 10
7 Assemblytime, min 5 35 10 70 .
Sum 30 - 36 43
Weighted Sum 225 232
Recommendation Eliminate option
Weighted
Value
24
120
16
; 100
49
309
Note: Lower scores indicate superior choices. MTBF is mean time between failures.
Figure 4-4 Decision matrix for access door attachment.
As part of playing "what if," sensitivity studies can also be conducted by perturbing the values
with varying ranges or probability distribution functions. The "Weighted Sum of Attributes"
decision matrix is an important decision tool, but its limitations need to be well understood by the
decision maker.
ANALYTIC HIERARCHY PROCESS
The Analytic Hierarchy Process (AMP) is a methodology for multi-criteria analysis and
decision making developed by Thomas L. Saaty (1980, 1987~. It can help decision makers to:
examine a complex problem with a number of possible solutions,
evaluate and prioritize alternatives, and
organize the information and judgments used in decision making.
s
The analytic hierarchy process allows the relative independent judgments made by people to
be used in a more formalized decision making process. The basic idea assumes it is much easier
for a person to say a car is much bigger than a motorcycle than it is to say how many times it is
larger, or even exactly what is meant by "larger." It is similar in approach to the Pugh method of
employing plus (+), zero (0), and minus (-) judgments. The word "hierarchy" in the name reflects
the notion that these judgments may be made at several levels and then combined to provide an
overall evaluation. While proponents of the process claim it has many uses even beyond decision
making, the focus of this discussion is on its use as a decision aid.
At each level of the hierarchy the process proceeds by having the user establish rough pair-
wise comparisons between attributes by using such comparatives as "more important" and "much
more important." These pair-wise comparisons, often redundant, are used to produce a relative
weighting of the various attributes. This process is repeated at the next level in a manner
demonstrated by example to provide an overall evaluation.
A general form of a hierarchical model of a decision problem is a pyramid with a broad
overall objective at the highest level. Lower levels list the criteria and respective subcriteria used
to choose among alternatives. At the lowest level are the alternatives to be evaluated.
Consider the example of selecting an automobile to purchase. Presume the attributes of
interest are comfort, performance, and safety. At the highest level of the hierarchy the user would
?
OCR for page 24
24
APPR OA CHES TO IMPR O HE ENGINEERING DESI ON
be asked to judge, for example, whether comfort is more important than performance, equivalent
to performance, or less important than performance. Then a similar set of comparisons would be
made between performance and safety, and finally, another set of comparisons between comfort
and safety. These would be manipulated to provide the relative weights, usually in the form of
percentages, to be assigned to comfort, performance, and safety in purchasing the car.
The next step is to address which car best meets the criteria. Suppose three cars are under
consideration: A, B. and C. The user would be asked to compare car A with car B with respect to
comfort; then A to C with respect to comfort; and finally B to C with respect to comfort. After the
averaging process, these comparisons would result in the relative percentage weightings of the
three cars with respect to comfort. The same judgments would then be made with respect to
performance. Is A much better than B with respect to performance, equivalent to B in
performance, etc. Then A would be compared to C, and B would be compared to C with respect
to the same performance attribute. After the averaging process, the relative weights of the three
cars would be obtained with respect to performance. Finally, the same procedure would be carried
out for safety. Now it can be said for car A what percentage of the values for comfort,
performance, and safety it should receive; and the same computation can be done for cars B and
C. For each car, multiplying the car's percentage of each value attribute by the percentage the
value represents with respect to a desirable car and summing overall attributes will provide for
each car the percentage of overall value attributable to it. In one case, car A could have 55 percent
of total value, car B could have 30 percent of total value, and car C could have 15 percent of total
value. Proponents of the process would say this is an argument for purchasing car A.
The procedure, as can be seen, is very easy to perform because it uses only simple judgments.
The problem is determining the strength of the recommendation of car A because it is highest on
this scale. Additional difficulties occur when a large number of alternatives require paired
comparisons, because k alternatives require k (k-1~12 pairs. Moreover, care should be taken to
ensure that some of the paired comparisons do not contradict each other.
The axiomatic structure of this process does not guarantee the alternative with the highest
rating will be the most preferred alternative. Unfortunately, it can be shown that the addition of a
new alternative may change the ranking of existing alternatives, a property seen as undesirable in
a decision process. The analytic hierarchy process has difficulty with uncertainty, which it can
handle only in an approximate way. The process therefore provides no basis for valuing the
elimination or reduction of uncertainty.
The main advantage of the analytic hierarchy process is ease of understanding and
application. It may have real value in making decisions with robust influence factors, where there
is no possibility of a major loss and where the complete set of alternatives is known a priori.
The difficulty with the analytic hierarchy process, in addition to the theoretical features
mentioned above, is that it cannot answer the questions necessary to build confidence in the
selection of an alternative. The very simplicity of the process limits its ability to answer hard
questions.
METHODS AND TOOLS TO ADDRESS
VARIABILITY, QUALITY, AND UNCERTAINTY
Once the decisions have been made and product design concept finalized, the next steps are
to translate the concept to reality. This section deals with decision-making tools, which are
methods to address the quality of the design process, to address the variability in the process, and
to convert the concept to final product. The general process of making decisions is greatly
s
p
OCR for page 25
IMPLI CA TI ONS FOR ED UCA TI ON AND RESEA R CH
25
affected by the context (see Figure 2-1) in which the decisions are made. Design decision making
in the context of variation can be conceptualized as shown in Figure 4-5.
The context of variation in Figure 4-5, similarly to Figure 2-1, has been segmented by the
categories of input, output, controllable design parameters, and uncontrollable (noise) parameters.
In Figure 4-5, the context is related to variation; therefore, the above four categories provide a
context for decisions in which the variation needs to be considered in decision making.
While there may be variation in the input requirements, the primary variation to be
considered is in the design, environmental, and manufacturing parameters. An example of
variation in a design parameter is the seal clearance in a shaft. Examples of variation in
environmental and manufacturing parameters are ranges in the line voltage a product will see in
use or differences in the ability of machines to meld tolerances. As a result of such variations, the
performance of individual product units will vary with respect to the design target. If the output
variation is too great or the mean is not appropriately centered near the design target, then some
of the units will not perform acceptably. The decision process must adequately consider variation
in design, manufacturing, and environmental parameters to ensure products delivered to the user
will perform within specified limits of design intent.
The consideration given to variation in the design process differs depending on whether the
variation is in a design-controlled parameter or in manufacturing- and environmental-
uncontrolled parameters. In the context of design decision making for products, the design
parameters in Figure 4-5 are controllable whereas the environmental and manufacturing
parameters are for the most part uncontrollable or at least contain an element of random variation
(noise). The strategy in the design decision process for controllable design parameters is to use
analysis, including statistical tools such as "Design of Experiments," to select values (settings) for
these parameters such that the product performance is within acceptable limits. The noise
variation of environmental and manufacturing parameters cannot be changed or controlled by
selection of parameter values as can be done with design parameters. The variation in
environmental and manufacturing parameters either is known or can be measured and included in
sensitivity analysis of design parameters. Experience has shown that inclusion of environmental
and manufacturing noise variation in design decisions is crucial for products to consistently meet
the design intent.
In summary, design decision making in the context of variation can significantly contribute to
the success of a product from the standpoint of customer satisfaction (market share) and
economic viability (profit to business). Employing "Design of Experiments" or similar methods
enables one to rapidly quantify product performance and economic viability across a broad design
space. Including variability or noise parameters in the design and decision process, as illustrated
in Figure 4-5, enables the designer to quantify the sensitivity of the product to variation and
determine the probability of success for achieving objectives relative to design limits.
Additionally, for those controllable noise variables, product performance and cost trade-offs can
be quantified in terms of design intent and probability of success. In total then, the process
conceptualized in Figure 4-5 enables design decision making based not only on deterministic
assessment but also on the inherent, real-world characteristics of product design.
.~
5
OCR for page 26
26
PROJECTED LATENT STRUCTURE
APPROACHES TO IMPROVE ENGINEERING DESIGN
Variation in
Design Control
Variables
Input:
Requirements
& Constraints
Material Properties
Dimensions/ Tolerances
· Fits and Clearances
· etc.
'I K~~ -
~ ~ ~ D~:is~io~n~:-: ~ ~~)~
An, ~~ ~~ ; Sprout. HISS- ~~ ~~:~?
1 _1 I:—
~ 1
Variation In ~ · Corrosiveness
Environmental ~ · Machine Capability
anc menu acturing 1~ · Actual Use Profile
Variables ~~
L_ · etc.
.0' .0. ,., ,35 =
Uncontrollec! Environmental anc! Manufacturing Variables
Figure 4-5 Decision making in the context of variation.
Output Distribution
Output: \
Design Intent / Design
/ Limits
~ / 1
' 1' · '
~ ~-
.
\ ~
"w . . ~ , ...., an.........
/ Design Rang
/
Design ~ ~ ,
Acceptable
~ Performance
Unacceptable Performance
Also called Partial Least Squares or PLS, Projected Latent Structure is a method for
constructing predictive models when controllable variables are many and highly redundant. The
emphasis is on predicting the responses and not necessarily on trying to understand the
underlying relationships among the variables. PLS is not usually appropriate for screening out
factors with a negligible effect on the response. However, when prediction is the goal ant ehre is
no practical need to limit the number of measured factors, PLS can be a useful tool (Tobias,
1995~. The original mathematics of PLS is the work of Herman Wold (Word, 1985~. Svante
Wold and B. Kowalski (Word, 1978; Beebe and Kowalski, 1987) are given credit for establishing
the field of "chemometrics," based largely upon PLS methodology. In chemometrics the X factors
(controllable variables) may include the many spectroscopic measures taken on samples drawn
from a chemical process, along with associated measures of temperatures, pressures,
concentrations, and flow rates. The Y responses (the behavior of other variables) in turn may
represent the mass, volume, viscosity, density, flow rates, and other quality measures on
intermediates and final products, once again gathered across many samples. The objective of PLS
is to analyze the data sets X and Y in the hope of discovering one or more signs of structure (low
dimensional linear relations) while recognizing X and Y may both have structural aspects
unrelated to one another. The idea of PLS is to extract latent factors, accounting for as much
variation as possible while modeling the response well. PLS has been successfully applied in the
s
Y
OCR for page 27
IMPLICA TIONS FOR EDUCA TION AND RESEARCH
chemical process industries. The opportunity to explore applications of PLS to the design and
assembly of hardware appears unexploited to date.
TAGUCHI METHOD
In Japan in the early 1 960s, Genichi Taguchi began an introduction of statistical patterns of
experiments (the fractional factorial, orthogonal arrays, and response surface "experimental
design") as aides in the design and manufacture of products (Taguchi, 1988~. Several concepts
were involved.
Quality should be measured by the deviation from a specified target value, rather
than by conformance to preset tolerance limits.
A defined loss Function should be established to provide a financial measure of
customer dissatisfaction with a product's performance as it deviates from a target
value.
Process parameters are not constants and their intrinsic variability will be
reflected in the variability of the product.
The product user's environment adds further variability challenges to quality
performance.
Engineering design must provide a robust product that is on target and
simultaneously insensitive to variability arising from both the process and the
environment.
27
Taguchi argues for the application of statistical methods throughout the entire engineering
design process, from product concept to customer usage. He was among the first to emphasize the
importance of statistical planning and analysis of experiments to identify and measure sources of
variability and sensitivity to assist in resolving the problems of design engineering. Of particular
note is Taguchi's development of parameter design wherein non-linearity in response is used to
decrease the sensitivity of that design to a given level of noise variability in manufacturing and
use. Taguchi's great contribution to the practice of engineering design is his emphasis on the need
to study the sensitivity of product responses to variability in both manufacturing and
environmental factors (Ross, 19964.
SIX SIGMA
In statistics the Greek letter sigma is used to identify the standard deviation, as a measure of
the variability of measurements. When measurements are reasonably approximated by a normal
distribution located on target, then the interval of target plus or minus two sigma will contain
approximately 95 percent of all the measurements. If this interval also defines product
specifications, then 5 percent of the product will be defective. When the specification limits are
set at target plus or minus six sigma, this results in only 3.4 defects per million outputs. However,
the terminology "six sigma" has come to mean far more than a simple counting of defects. It now
identifies an entire quality culture of strategies, statistics, and tools for improving a company's
bottom line (Pyzdek, 2001~.
The Six Sigma concept was introduced by Mikel Harry at Motorola in the 1 980s. Many other
corporate giants, including Texas Instruments and General Electric, have adopted it since then.
The tools utilized in Six Sigma include problem statement development, brainstorming,
histograms, fish-bone diagrams, process mapping, measurement-systems analysis, graphical
analysis, capability analysis, hypothesis testing, regression analysis, analysis of variance, supply-
s
OCR for page 28
OCR for page 29
OCR for page 30
OCR for page 31
OCR for page 32
OCR for page 33
OCR for page 34
OCR for page 35
OCR for page 36
OCR for page 37
OCR for page 38
Representative terms from entire chapter:
design parameters
5
28
APPROACHES TO IMPROVE ENGINEERING DESIGN
chain management, design of experiments, statistical process control, and failure-mode-effects
analysis.
METHODS AND TOOLS FOR GENERATING ALTERNATIVES
Generating alternatives to meet the requirements of a product is an inherent part of the design
process. In the simplest form a single designer can consult design guides for past practices and
select a reasonable option. More creative designers could examine other fields to adopt new ideas
or imagine a novel solution to the problem.
Group practice, in its simplest form, includes brainstorming to generate design alternatives.
Newer tools draw on a range of computational capability to enhance the generation of options.
Generally, the techniques can identify alternatives by employing a very large set of data using
several methods or tools, possibly including artificial intelligence. A few of these techniques are
summarized here.
DESIGN INFORMATION SYSTEMS, SUPPORT SYSTEMS, AND ENVIRONMENTS
Design decision making involves more than methods and tools for addressing decision
events. Decisions usually occur in information-rich environments involving a range of
stakeholders and disciplines. A wide variety of efforts have been developed aimed at supporting
designers to make decisions in such rich contexts.
These efforts usually are pursued under the rubric of decision support, often involving many
of the methods and tools discussed elsewhere in this report, but also frequently involving various
levels of artificial intelligence. Gero and Sudweeks (1996, 1998), in two recent proceedings of the
biennial International Conference on Artificial Intelligence in Design, provide a broad view of the
many efforts in this area, which include design processes (decision-driven process models,
cognitive theories of design decision making), knowledge management (representation,
acquisition, sharing), decision support (reasoning, structure generation, form generation,
component selection, optimization), and design support (interactive tools, collaboration support).
Figure 4-6 indicates the scope of this field, gleaned from an analysis of the papers included in
these volumes. Traditionally, the focus has been on detailed design and automated decision
making. However, current research is moving toward conceptual design and problem formulation,
with necessary emphasis on decision support and, consequently, provision of information and
knowledge needed for making design decisions. Gero and Sudweeks assert, "Computer-aided
decision making will produce better designs."
Numerous efforts have been undertaken to integrate much of the above into design
environments. The long series of studies by Cody, Rouse, and Boff (1995) is a good example of
work in this area. Analyses of the results of studying hundreds of designers yielded the
requirements for the architecture of an interactive design environment, supported by an
artificially intelligent Designer's Associate.
Evaluation of a series of prototypes led to an important insight. Specifically, the cost of
developing and maintaining the knowledge bases associated with what the Designer's Associates
envisioned would be prohibitive. Simply, there would be too much data and information to
compile and maintain. This led to the recommendation to develop more focused support systems
where the embedded knowledge required would be far less comprehensive.
One direct result of this conclusion is the Product Planning Advisor (
IMPLICATIONS FOR EDUCATION AND RESEARCH
29
evaluating alternative functions/features of product concepts relative to market models, as well as
both current and likely competitive offerings. This much more focused design support tool
combines intelligent support with multi-stakeholder, multi-attribute market models and quality
function deployment representations of relationships among product functions/features and
stakeholders' attributes. The Product Planning Advisor has been employed in hundreds of product
planning efforts, many involving top high-technology companies.
Artificial intelligence (AI) developers have, at various stages in its history, promised
comprehensive automated decision making. AI as a category includes such approaches as neural
networks, expert systems, genetic algorithms, intelligent agents, and fuzzy logic. More
realistically, however, for all but routine design decisions, AI can best provide powerful
knowledge-based support of design decision making, rather than automated decision making. It
does this in part by building on an understanding of human decision making in design contexts
and tailoring support to enhance human abilities (e.g., pattern recognition) and overcome human
limitations (e.g., errors). Thus, AI is not used to automate human decision making; rather the
theory determines how AI can best assist decision making. Some recent attempts to exploit AI for
the purpose of generating new design alternatives show promise.
TRIZ
TRIZ, a Russian acronym that translates to Theory of Inventive Problem Solving, describes a
process to encourage creativity. The theory is based on the presumption that a conflict between
two objectives creates a need for creativity. For example, the design has to be light and strong.
The developer of the process reviewed hundreds of thousands of patents to see what ideas were
used to reconcile these conflicting objectives. He developed a matrix that showed, for the
intersection of conflicting objectives, alternatives used in existing patents. Some of the
intersections are relatively sparse; others have several suggestions. The intent is that one of the
suggestions might spark the creativity of the person facing the design problem. TRIZ was created
by Genrich S. Altshuller in Russia (Altshuller, 19884. He and his coworkers analyzed about 1.5
million patents and worked out a methodology to resolve the technological problems.
The salient TRIZ principles are as follows:
All engineering systems have uniform evolution in the direction of increased
ideality. Many others (such as economic and educational systems) have the same
evolutionary trends.
Any innovative problem represents a conflict between new requirements and an
old system.
TRIZ is composed of various systematic techniques including inventive principles;
psychological inertia overpass system; physical, chemical, and geometric effects; substrate-field
Support/Focus | Detailed Design | ConceptualDesign | Problem Formulation
Information/Knowledge
Decision Support
Decision Making l l l
9
Figure 4-6 Scope of artificial intelligence in design.
30
A PPR OA CHES TO IMPR O VE ENGINEERING DESI ON
and functional analysis; technological ideality concept; and technology forecasting. It helps find a
quasi-icleal solution to innovative problems through the resolution of hidden conflict based on
inferred knowledge of system evolution.
TRIZ is more than a methodology. It represents a unique way to enhance creativity by getting
individuals to think far beyond their own experience, to reach across other disciplines, and to
resolve problems using knowledge extracted from other areas of business, science, or technology.
FORMAL METHODS FOR REPRESENTING DESIGN PROBLEMS
This section covers some limited formal methods and theories for representing design
problems. They are selected as representatives from different schools of thought in approaching
design problems: traditional engineering, decision theory, and artificial intelligence.
ENGINEERING DESIGN: A SYNTHESIS OF VIEWS
Dym (1994) attempts to articulate what is meant by engineering design, how it can be
discussed both for designed artifacts and for the process of design, and what areas are amenable
to, and perhaps most require, formal research approaches. The central thesis elevates
representation as the key element in design. In addition, recent research in AI aimed at rendering
design computable is purported to provide new techniques for design representation to enable
improved understanding of design concepts and processes.
Design activities encompass a spectrum from routine design of familiar parts and devices,
through variant design requiring some modification in form or function, to truly creative design
of new artifacts. The spectrum of design concerns includes some processes susceptible to
thoughtful analysis in other words, cognitive processes. One principal consequence of this
recognition is the need to focus on the languages of engineering design. On the other hand,
creative designs are almost certainly not susceptible to encapsulation with current representation
technologies and, therefore, cannot be modeled.
For the last half of the twentieth century mathematics was the language of engineering, owing
in part to the central role of mathematics in physical sciences modeling. Engineers understand
that much of what they really know cannot be expressed in mathematics alone. Engineers use
graphics and pictures, as well as words, although often in a very structured way (e.g., in
specifications and codes, in heuristics or rules of thumb). Designers and modelers of cognitive
design processes understand that many languages of design have been devised and used.
Design knowledge includes information about such features as design procedures and
shortcuts, as well as about designed artifacts and objects and their attributes. Designers think
about design processes when they begin to create sketches and drawings to represent the objects
they are designing. A complete representation of designed objects and their attributes requires a
complete representation of design concepts that may be less easy to describe or represent than
physical objects (e.g., design intentions, plans, behavior).
The languages or representations used in design include:
.
.
5
verbal or textual statements to articulate design projects, describe objects,
describe constraints or limits, communicate between different members of design
and manufacturing teams, and document completed designs;
graphical representations to provide pictorial descriptions of designed artifacts
such as sketches, renderings, and engineering drawings;
IMPLICA TIONS FOR ED UCA TION AND RESEAR CH
31
mathematical or analytical models to express some aspect of an artifact's function
or behavior, which is in turn often derived from some physical principles that
can be expressed mathematically;
numbers to represent discrete-valued design information (e.g., part dimensions);
and
continuously varied parameters in design calculations or within algorithms
representing a mathematical model.
Different languages are used to represent engineering and design knowledge at different
times, and the same knowledge is often cast in different languages in order to serve different
purposes. For example, fundamental structural-mechanics knowledge can be expressed
analytically, as in formulas for the vibration frequencies of structural columns, numerically, as in
discrete minimum values of structural dimensions or in finite element meshing algorithms for
calculating stresses and displacements; and in terms of heuristics or rules of thumb, as in the
knowledge that the f~rst-order earthquake response of a tall, slender building can be modeled as a
cantilever beam whose foundation is excited. Therefore, designers must realize that what they
need to know is not just a set of formulas; they must know how to apply knowledge in different
forms to serve different purposes. This means designers must master the languages of engineering
design.
SUM'S AXIOMATIC DESIGN
t
This theory (Suh, 1990) describes design as a mapping between what designers want to
achieve and how they achieve it. The framework of axiomatic design views design as a collection
of mappings between four domains: the customer domain, the functional domain, the physical
domain, and the process domain. In each domain the design is specified using different elements:
customer attributes (CAs), functional requirements (FRs), design parameters (DPs), and process
variables (PVs). In addition there are constraints (Cs). The design process starts with the
identification of customer needs and attributes, and formulates them as FRs and constraints.
These FRs are then mapped onto the physical domain by conceiving a design embodiment and
identifying the DPs. There may be more than one solution to this mapping. Each DP is then
mapped onto a set of PVs to define it. Each DP typically introduces new Fits, DPs, and PVs, and
so the mapping process iterates by zigzagging between domains, until the design can be
implemented without further composition. In principle this approach takes a broad view of
design, but the axioms and methods are almost entirely about mapping from the functional to the
physical domain, so they do not address all aspects of design. The principles of this theory
potentially apply to a variety of design problems, including mechanisms, software, and
organizations.
Based on Nam Suh's experience and observations about existing designs and his assessment
of successful and unsuccessful designs, he proposes two axioms:
.
.
5
Axiom 1, the Independence Axiom: Maintain the independence of functional
requirements. This means that when two or more functional requirements exist,
the design solution must be such that each one of the FRs can be satisfied without
affecting the other Fits. That in turn means a designer must create a design (or
design parameters) able to satisfy each FR independently of the other Fits. Thus,
this first axiom establishes a requirement about the design to guide the creative
process and any decisions among alternatives.
Axiom 2, the Information Axiom: Minimize the information content. Sub
defines information as the logarithm of inverse probability of meeting the
32
APPROACHES TO IMPROVE ENGINEERING DESIGN
functional requirement. Among designs satisfying Axiom 1, Suh's axiomatic
design approach states that the best design is the one with the minimum
information or maximum probability of meeting the functional requirements,
considering tolerances and nominal requirements.
In addition to these two axioms (and numerous associated theories and corollaries), Sub
through case studies and examples essentially outlines a design methodology consistent with his
axioms. The methodology involves techniques for zigzagging between functional requirements
and design parameters and the use of matrix algebra to assess independence. The theoretical
framework has some appeal to experienced designers who recognize that achieving conflicting
functional requirements with one design parameter (independence axiom violation) is the source
of some badly compromised designs and that the information content embedded in the functional
requirements might be a valid assessment of such complexity. Suh's axiomatic approach
represents a substantial and potentially useful addition to design methods, but the technique has
not shown significant practical application, as is discussed below. Moreover, the theoretical basis
has some apparent limitations. It is not clear that Suh's assertion is correct that an ideal design
always has an equal number of functional requirements and design parameters.
On the one hand, although we can agree that independence is desirable, design constraints
such as manufacturability, low cost, and ease of use may at times conflict with independence or
for objective reasons override independence. The best design, therefore, may have more
functional requirements than design parameters. On the other hand, there are cases where
decreased sensitivity to variations in use or manufacturing may be particularly important and can
be improved by having more design parameters than functional requirements. Thus the
independence axiom can result in a useful assessment tool but is not a requirement for all good
designs. Further development of the information definition may also be needed to best meet
customer needs rather than simply meeting the given tolerances.
In summary, the axioms while useful do not appear to constitute a complete and optimal
design method. This could be why the best practical applications to date use axiomatic design in
combination with other design methods. One can use the independence axiom in combination
with robust Taguchi methods to examine which design parameters to use in achieving a robust
design. One could also use axiomatic principles to assess concepts created by TRIZ methodology
(Mann, 1999~.
YOSHIKAWA'S GENERAL DESIGN THEORY
t
Yoshikawa and his associates began in the late 1970s to publish papers on a general theory of
design. This work attempts to address design in a rather complete fashion by defining design as
"creating artificial things ~that] have not existed in the real world previously." Some of these
papers have focused on how the approach might be applied to extending computer-aided design
(CAD) systems to include engineering and simulation information. Although a formal
mathematical basis is sketched, this approach remains largely philosophical with some interesting
general observations about the nature of design. This approach has resulted in no new design
methods or engineering design tools, nor (as yet) has it seemed to directly add new tools to the
intelligent CAD area.
A MATHEMATICAL FRAMEWORK FOR ENGINEERING DESIGN
5
The decision-based design view of engineering design states that much of design consists of
decision-making activities, and that decision support methods used in engineering design should
reliably produce good advice. This is a non-trivial condition that demands that design methods
IMPLICA TIONS FOR ED UCA TION AND RESEARCH
33
adhere to the mathematics of decision theory. Only in this way can paradoxical results, in which a
design method might recommend even the worst design alternative, be prevented. Rigorous
decision theory has developed over the last 300 years and has its roots in the work of many
mathematicians and economists, including Daniel Bernoulli, Charles Lutwidge Dodgson (Lewis
Carroll), John von Neumann, Oskar Morgenstern, and Kenneth Arrow. Hazelrigg (1998, 1999)
uses the results of these mathematicians and economists to lay out a decision theory-based
framework for engineering design, thereby extending the earlier work of such people as Myron
Tribus (1969), Richard de Neufville (1990), and Andrew Sage (1977~.
The purpose of Hazelrigg's framework is to provide a self-consistent method for rank
ordering alternatives in the context of engineering design. The framework recognizes some key
aspects of design, in the context of decision theory, that other methods fail to consider: (1) that all
design decisions are made under conditions of significant uncertainty and risk; (2) that the
preferences of key importance in a decision are those of the decision maker (not those of the
customers or stakeholders); and (3) that alternatives must be ranked on a valid and validated real
scalar measure. Thus, Hazelrigg uses the preference of the company CEO or other decision
authority as the basis for a valid scalar measure (typically net present value of cash flow
generated by a design), together with von Neumann-Morgenstern utility theory to assure validity
of the measure under uncertainty and risk.
Hazelrigg adds two axioms to those of von Neumann and Morgenstern, although the first can
be derived from the von Neumann-Morgenstern axioms and is presented only for convenience.
The two additional axioms are as follows:
.
Axiom 1, the Axiom of Deterministic Decision Making: Given a defined set of
alternatives from which to choose, each with a known and deterministic outcome,
the decision maker's preferred choice is the alternative whose outcome is most
preferred.
Axiom 8, the Axiom of Reality in Engineering Design: All engineering designs
are selected from among the set of explicitly considered potential designs.
The von Neumann-Morgenstern axioms provide a basis for comparison of known
alternatives. They do not provide a basis for comparison of a known alternative with an unknown
alternative, and some engineers thus argued that Hazelrigg's framework is not valid. The addition
of Axiom 8, which states that any chosen engineering design is a known option included in the set
of options under consideration for selection (this should be intuitively obvious, as we do not
produce products we never imagined), assures that the von Neumann-Morgenstern results apply
to engineering design.
The von Neumann-Morgenstern axioms provide two results of consequence:
.
The Expected Utility Theorem. Given a pair of alternatives, each with a range
of possible outcomes and associated probabilities of occurrence, the preferred
choice is the alternative with the highest expected utility.
The Substitution Theorem. A decision maker is indifferent between a lottery
and a certainty outcome whose utility is equal to the expected utility of the
lottery, and for purposes of analysis, the two may be substituted one for the other.
Early applications of these ideas to engineering design include Greenberg and Hazelrigg
(1974). Recent applications of utility theory exist, for example, Thurston et al. (1994); and many
other recent papers apply decision theory to engineering design, but they largely fail to consider
uncertainty and the decision maker's attitudes toward risk. Marston and Mistree (1997) use the
von Neumann and Morgenstern axioms, but advocate additional areas (such as subjectivity in
q
34
APPROACHES TO IMPROVE ENGINEERING DESIGN
options and in designer preferences) to be included in design theory. A recent paper by Thurston
(2001) assesses the appropriateness and usefulness of decision-based design. Of course, the
desirability of a design to customers, as expressed in willingness to pay, is an important
consideration in formulating an objective function usually profits for the designer's company.
The Nobel laureate Kenneth Arrow in 1951 provided results of extreme importance to
engineering design. Arrow states six conditions that should be satisfied by a selection method,
such as the following: If, under all conditions and by every measure, alternative A is preferred to
alternative B. then the selection method should not choose B over A. He goes on to prove that, in
the case of three or more alternatives and three or more selection criteria (voters, for example), no
selection method can be assured of giving a valid result. Arrow's Impossibility Theorem points to
the dangers of naive multi-criteria decision methods that comprise many engineering design
selection methods. Based on Arrow's result, Haunsperger and Saari (1991) have provided
numerous paradoxical examples illustrating how naive decision support methods fail.
An early application of these ideas to engineering design can be found in Dyer and Miles
(1976~. Recent work by Allen (2001) points out, in the context of the von Neumann and
Morgenstern setting for decision making under uncertainty, that a weakening of Arrow's axioms
permits a possibility result for group decisions with risk aversion. Scott and Antonsson (1999)
argue that despite the common participation of many individuals in the design process,
engineering design is closer to multiple criteria decision making than to social choice theory, so
Arrow's theorem need not apply.
DECISION MAKING IN MANAGEMENT SCIENCE AND ECONOMIC
FIELDS
The fields of management science, game theory, and economics commonly use decision-
making techniques, and some of these may have application to engineering design. In fact, many
opportunities exist for cross-application of decision-making tools among unrelated fields.
Operations research has developed numerical methods useful in computational economics and
game theory. Constrained optimization packages for linear programming, integer programming,
and non-linear programming are well known. Fixed-point algorithms can be used to find
. . . . .
equ1 1 aria in economies ant games.
DECISION MAILING IN ECONOMICS
..~
5
The academic discipline of economics lies behind the business side of engineering design and
technology management. Using economics terminology, engineering design consists of product
selection and technology choice. Economists use techniques from constrained optimization,
decision theory, game theory, and microeconomics (the study of resource allocation) in general to
solve such problems. The methodology used in economics also typically differs somewhat from
that used in engineering.
Economics tends to focus on the development of an overriding general framework for
analysis of a wide variety of applied problems and policy issues. Such a framework involves the
elucidation of a fully consistent general model from fundamental principles, starting with
constrained optimization. A goal is to rigorously examine the implications of appropriate
assumptions. Numerical work should be preceded by a precise statement of the equilibrium
concept or constrained optimization problem (i.e., the objective function and feasible set) and by
a careful examination of the conditions under which the model has a solution. Note the term
"model" refers to the general and rigorous framework. The model includes the abstract exogenous
information, assumptions, and definitions.
IMPLICA TIONS FOR ED UCA TION AND RESEAR CH
35
The basic product selection problem can be cast as a constrained optimization problem along
the following lines. Once a firm has made its product selection decision, the firm's profits
(revenues minus costs, where revenues depend on demand) depend on its costs, the price at which
the product is sold, and the number of units of the product produced and sold. Of course, profits
also depend on the selection of products manufactured and sold by all other firms and their
respective prices. Given the products of all other firms and their prices, one could construct the
profit function for each feasible product choice and find its maximum value, subject to price and
quantity being consistent with market demand. Then one could pick the product for which
maximum profits are greatest. (This assumes other firms do not strategically alter their decisions
in response to the product choice of the firm in question.) This extension can be analyzed with
game theory. Engineering considerations enter through the feasibility constraints faced by firms
and through their cost functions (which depend on the production technology, the quantity
manufactured, and input prices) for each possible product. The basic question reduces to a (highly
non-trivial) constrained optimization problem. (The case of a multi-product firm is more difficult
to analyze, but the same principles can be applied.)
Decision making in economics, whether for consumers or firms, is based on constrained
optimization; however, the objective functions and constraints faced by consumers are different
from those for firms.
The consumer's decision problem starts from preferences essentially, data in the form of a
yes or no answer to the question of whether some combination of items to be consumed is at least
as good as some other combination of items. Standard rationality assumptions on preferences are
well known and give rise to utilities. The following axioms are typically used to study individual
choice behavior in economics:
Symmetry: X is at least as good as X.
Consistency (transitivity): If X is at least as good as Y and Y is at least as good as
Z. then X is at least as good as Z.
Decisiveness (completeness): Either X is at least as good as Y or Y is at least as
good as X (or both).
No drastic changes (continuity): If X is strictly better than Y. then X~ is strictly
preferred to Ye whether X~ is sufficiently close to X and Ye is sufficiently close
to Y.
More is better (monotoricity): If X is greater than Y. then X is as least as good as
Y.
Frequently an additional axiom stating that variety is strictly desirable (strict convexity) is
added in order to guarantee that optimal choices are unique. Utility functions summarize the
preference relation such that the utility associated with one combination exceeds the utility of
another combination if and only if the first combination is strictly better according to the
consumer's preference. The consumer's constraints reflect affordability (given all prices and
income) and survivability (minimum quantities of food, shelter, and the like may be required).
If uncertainty is involved, preferences over lotteries (randomizations over combinations of
items to be consumed) are the appropriate fundamental concept. Rationality axioms lead to
cardinal utility representation (see von Neumann and Morgenstern, 1980~. Cardinal utilities
reflect attitudes toward risk and are unique (given the preferences over lotteries) up to
multiplication by a positive constant and addition of a constant.
s
For firms, profits are the appropriate objective functions in simple cases. A sole entrepreneur
should maximize expected utility of profits when uncertainty is present. In intertemporal settings,
36
APPROACHES TO IMPROVE ENGINEERING DESIGN
the firm should maximize the present discounted value of profits or the expected utility of profits.
If shares of the firm are traded, the basic goal is to maximize the firm's market value, although
complications (such as shareholder purchase of significant amounts of the firm's output) can
cause the objective to change. The firm's constraints are derived from its available production
technology, which when combined with all input prices, determines costs and from that
aggregate, demand for its products.
The interactions and interdependencies among individuals, firms, and products can be
captured by market equilibrium. When strategic aspects of individual and firm behavior matter,
game theory provides a rigorous analytical tool.
GAME THEORY
Game theory is the study of formal models of strategic behavior in which the payoff or utility
received by a player (individual or firm) can depend not only on the player's own decisions but
possibly also on the decisions of all other players. Game theoretic models are classified into
cooperative and non-cooperative games.
A non-cooperative game is specified by a set of players, a strategy set for each player, and a
payoff function (or utility depending on the strategies chosen by all the players) for each player.
A Nash equilibrium is defined by the principle that each player chooses a strategy to optimize his
or her own payoff given the decisions of all other players. Well-known conditions guarantee that
a game has a Nash equilibrium, which may require randomized strategies. When there are
multiple Nash equilibria, refinement concepts can narrow the equilibrium set. Non-cooperative
games were introduced into the engineering design literature by Vincent (1983) for the study of
collaboration within teams.
In a cooperative game, players can communicate and make binding agreements within
coalitions (non-empty subsets of players), but cooperative games do not analyze the formation of
coalitions. Many solution concepts are available, some of which are defined axiomatically.
SUMMARY OF METHODS, THEORIES, AND TOOLS
.`
Table 4-2 provides a cursory rating of several tools with respect to potential values in current
use, concept creation, concept development, selection among alternative concepts, and ease of
use. Some decision analysis and applied decision theories are also included in this comparison.
Concurrent engineering is included here as a tool, but it is more of an operating philosophy. It has
its primary basis in the economics of product and process development, whereas the other
approaches with a primary basis in economics build on theories of preferences (e.g., von
Neumann-Morgenstern axioms).
The "ease of use" criterion is used in two ways to describe the tools and methods,
encompassing both conceptual difficulty and execution difficulty. For example, QFD is
overwhelming to execute if pursued fully, not because it is conceptually difficult but because
filling in the huge number of cells in the matrices becomes daunting. Sub, however, is
conceptually difficult. Mathematically rigorous approaches can often be conceptually difficult to
employ (e.g., in terms of understanding and justifying assumptions and perhaps overly narrow
problem definition). On the other hand, broadly applicable methods, including Concurrent
Engineering and QFD, face difficulty in some applications because of the very breadth they
address.
5
IMPLICA TIONS FOR EDUCA TION AND RESEARCH
Table 4-2 Summary of Tools and Applications Examined
37
Primary Basis
Ratingsa - Potential Value for:
'A a., an
it,, ° E ,,, ~ ~ E ~ :
Y _ ~ ~ an ~ ~ Cay ~ cry ~ us
Practical Concurrent X 4 2 4 4 1
Engineering
Qualitative Decision Matrix X X 4 1 2 4 5
Pugh Method X 3 4 5 1 2
QFD X 2 2 4 2 1
AMP X 3 1 2 4
Product Plan X X X 3 2 3 4 3
Advisor
Statistical PLS X X 1 3 3 2 1
TaguchiMethod X X 4 1 4 4 2
Six Sigma X X 3 3 3 3 2
Creative AlSupport X 2 4 2 2 2
TRIZ X 3 3 1 1 3
Axiomatic Suh's Theory X X 2 2 3 5 1
Yoshikawa Theory X 1 1 1 1 1
Math Framework X X X X 1 1 1 5 3
Validating Game Theory X X 1 1 1 3 2
Decision Analysis X X X 3 1 4 5 3
aRating by several authors: 1 = low; 5= high.
The "selection among alternative concepts" criterion has multiple interpretations. The multi-
attribute matrix-oriented techniques are often used to select among overall product concepts,
while the statistical methods are more typically used to select among process alternatives and
among more detailed design differences. This criterion demonstrates the main strength for some
of the more mathematically rigorous approaches. Not surprisingly, such approaches have little
basis for generating alternatives. In contrast, knowledge-based approaches such as AI and TRIZ
are much better for generating alternatives than for choosing among them. Approaches such as
QFD and Pugh attempt to aid in both.
p
38
APPROACHES TO IMPROVE ENGINEERING DESIGN
Cooper et al. (2000) state, "The choice of tool may not be that critical; indeed, the best
performers use an average of 2.4 tools each—no one tool can do it all!" One could use Table 4-2
and the discussion in this chapter as a guide to choosing approaches for design application. For
example, effective choices include:
Concurrent engineering as an overall framework for decreasing costs and time to
market;
TRIZ for generating alternatives;
Some form of Decision Matrix Technique for initial screening of ideas;
Six Sigma for process design and evaluation with emphasis on quality control;
Decision Analysis for making major investment decisions and for selection
among viable concepts; and
. Taguchi and axiomatic methods for reliable, robust design development.
Design is an intellectual activity of the human brain and is therefore difficult to understand or
even to describe using mathematical theories. Because all human intellectual activity includes
decision making, decision making is an integral and inherent part of any design process. A variety
of tools, methods, and theories have been developed over time to help describe and facilitate
decision making processes, and some of these have been applied to various aspects of design, but
none approach a general and useful theory fundamental to all areas of design.
In current practice each of these formal approaches to representing design processes is
valuable yet individualistic. Specific theories do not currently acknowledge or make reference to
others, making it difficult to determine the compatibility or contradictions among them. This lack
of coordination impedes the teaching and general reduction to practice of potentially usefi~1 tools.
Although some work is being conducted to understand the connections among these theories (din
and Lu, 1998), much more is needed.
The validation of individual theories is anecdotal and difficult to justify across the wide range
of design processes. Each of the tools described here must take into account the uncertainty of the
data input. This data may come from actual measurements, from analysis of historical data, from
solicited expert opinions, or from moderated opinions of potential users. Much of this input data
depends on the ability of humans to judge attributes, and on how their judgment is affected when
the number of attributes becomes large or complex. This variable makes validation of designs and
design tools difficult.
.
Each methodology can clearly be an intellectual activity of value provided its potential
applicability and limitations are well understood. However, comparison and contrast of the results
of each tool can provide additional insight to the designer, and tools used in tandem may result in
incalculable synergies. While the value of a single tool with applicability across every design
query might be desirable, today's designers must use what is available.
In summary, as in all human activities, the tools and methods used in design are the ones that
have the most utility given the constraints imposed by time and other resources available. Box
(1979) has stated that "all models are wrong, some are useful," to which the committee adds the
codicil, "all tools are useful, some are appropriate."
5
