pi^r;V

HD28
.M414
no.

WORKING PAPER
ALFRED P. SLOAN SCHOOL OF MANAGEMENT

Evaluating Prototyping Technologies

for Product Design

by
Matthew B. Wall
Karl T. Ubich
Woodie C. Flowers

WP #3334-91-MSA September 1991

MASSACHUSETTS
INSTITUTE OF TECHNOLOGY
50 MEMORIAL DRIVE
CAMBRIDGE, MASSACHUSETTS 02139
 Evaluating Prototyping Technologies
for Product Design

by
Matthew B. Wall
Karl T. Ulrich
Woodie C. Flowers

WP #3334-91-MSA September 1991

 '
OCT 8 199i
To appear in Research in Engineering Design, Volume 3, 1991.

Evaluating Prototyping Technologies for Product Design

Matthew B. Wall
Karl T. Ulrich*
Woodie C. Flowers

Massachusetts Institute of Technology

Cambridge, Massachusetts USA

Abstract

Firms that design mechanical and electromechanical products confront a variety of difficult issues in
their prototyping activities. For a given part, how can a choice among fabrication technologies be
made? Where should investments in new prototyping technology be focused? How can new and existing
prototyping technologies he evaluated? Our primary goal has been to develop a systematic method of
evaluating prototyping processes in order to determine the best process for a given situation. A
secondary goal has been to map the "space" of prototyping processes in order to determine future process
development needs. Using data from a field study at the Kodak Apparatus Division, we have
developed a systematic method for evaluating and selecting prototyping processes. Our data is drawn
from (Da user survey of prototyping perceptions and needs, (2) a survey to determine the importance of
various prototype part performance attributes, and (3) estimates of the fabrication time, cost, and part
performance for 104 parts and four prototyping processes.

Acknowledgements

The research described in this paper was supported by the MIT Leaders for Manufacturing Program, an
educational and research partnership between MIT and eleven major U.S. corporations. We are
indebted to the Apparatus Division of the Eastman Kodak Company for providing data and insights
for this research. A version of this paper appeared in the proceedings of the Third International ASME
Conference on Design Theory and Methodology, 1991.

Address all correspondence to this author at: Massachusetts Institute of Technology, Sloan School of
Management, Room E53-390, Cambridge, MA 02142-1347 U.S.A.
1. INTRODUCTION
Prototyping is one of the most critical activities in new product development. Firms
that design mechanical and electro-mechanical products coiifront a vciriety of
difficult issues in their prototyping activities. Some of these issues are tactical: For a
given part, how can a choice among fabrication technologies be made? Some of the
issues are strategic: Where should investments in new prototyping technology be
focused? How can new and existing prototyping technologies be evaluated? We
believe that there are useful conceptual frameworks and fundamental relationships
that can be developed and used to support prototyping decisions, even though broad
generalizations are likely to fail in some Such a framework
particular situations. is

particularly important today as lead times in product development become more

critical [Fitzgerald87, Ulrich91] and as novel prototyping technologies emerge
[Dickerson88, Kutay90, Lindsay90].

This paper describes research addressing these issues. Our primary goal has been to
develop a systematic method of evaluating prototyping processes in order to
determine the best process for a given situation. A secondary goal has been to map
the "space" of prototyping processes in order to determine future process
development needs.

Our basicmethodology has been to study prototyping in the context of an industrial

manufacturer. By using a data-directed approach based on existing parts and
processes, we have been able to provide empirical support for a systematic method
of evaluating and selecting processes. Our data is drawn from (1) a user survey of
prototyping perceptions and needs, (2) a survey to determine the importance of
various prototype part performance attributes, and (3) estimates of the fabrication
time, cost, and part performance for over 104 parts and four prototyping processes.
We focus on parts intended to be injection molded; the prototyping processes on
which we concentrate are computer-aided design solid modeling (CAD),
stereolithography (STL), computer numerical controlled machining (CNC), and
rubber molding (RM).

The paper has four sections. In the balance of the introduction we define
prototyping, present a conceptual framework for evaluating prototypes and
prototyping processes, and describe our research focus. In the second section we
present our approach and methodology. The third section details our results. The
final section lists our conclusions and suggests areas for future research.
What is Prototyping?

Technically, a prototype is the first thing of its kind. But "prototype" has come to
mean many different things in the context of product development. For the
developer of commercial satellites, the prototype may be the final product. At the
other extreme, the development of a new ballpoint pen may involve more than 10

Page 2
prototypes before the desigr. is finalized. Some prototypes are used to assess styling,
some are used to perform life testing, some are used to verify fit. Other words
frequently used to describe the function of a prototype include breadboard, model,
and mock up. Each is an example of a prototype; each emphasizes a different aspect
of what a prototype does. In our definition of a prototype we include both electroruc
and physical representations of the part or product.

In most cases prototypes are built to answer questions. In the engineering context, a
prototype proves a technology or its application. In the marketing context, the
prototype is a vehicle through which the marketer simulates customer response or
finalizes design requirements. In the manufacturing context, a prototyp>e proves a
process or procedure. Effective prototyping requires understanding which questions
to ask, when to ask them, and how to answer them.

Different types of prototypes are used in many different ways to address different
types of questions. Electronic prototypes are often used for simulation. They may be
subjected to finite element analysis, mass properties calculations, tolerance analysis,
assembly analysis, or motion analysis. Some prototypes are used for verification;
often analysis and simulation have been used make many design decisions, and
to
the prototype used to detect unanticipated
is problems. Some prototypes are used to
perform tests; they may be used for functional testing, customer perception testing,
life testing, assembly planning, etc. Other prototypes serve as crystal balls to
anticipate future problems; they may be used to prepare for tooling design, or to
compare the evolving product with customer needs.
Framework

Product development professionals have highly varied prototyping needs.

Designers concerned with styling have needs different from those of mechanism
designers. Prototypes to support conceptual design have requirements different
from those used in pre-production. The design of some parts may demand
prototypes with very tight tolerances, while other parts may require special material
properties. The required quantity is important in evaluating processes; some
situations may require only one part, some may require 100 parts. Finally, in every
case the user of a prototype cares to some extent about the cost and time required to
procure a prototype.

To capture some of these issues, we

evaluate prototyping processes along three
dimensions: part performance, unit cost, and lead time. We view part performance
as an aggregate measure of the fidelity of the properties of a prototype part, made
with a particular process, with respect to the properties the part would have if it
were made with the intended final production process. This performance measure
is composed of attributes such as appearance, material properties, and dimensional

accuracy. Using an importance weighting for each attribute, derived from a

particular user's needs, the performance of a process with respect to each attribute
can be combined into a single scalar part performance. Part cost and lead time could
also be considered to be dimensions of performance, but since designers often think

Page 3
of them independently of physical part attributes, we chose to separate them from
other performance measures.

More formally, associated with each part, i, fabricated with a particular process, there
is an associated cost, c,, lead time, f,, and performance vector A,-. The performance
vector for the part and process is composed of the performance of the process with
respect to each of the attributes, ay , or

ai
A,=

an

The set of attributes used to define the vector may

vary depending upon the
application or organizational context. Typical attributes might be strength, stiffness,
density, color, surface finish, etc. Each Uj is assigned a value from to 1 representing
the degree of fidelity of the prototyping process to the final production process for
the particular part. A score of is a poor match and 1 is a iperieci match.

For a given part, the attribute importance weighting vector, W, contains the
importance weightings for each of the attributes, or

Each Wj is assigned a value from to 1 where is unimportant and 1 is very

important. For example, for a part to be used in a structural test, stiffness, strength,
and dimensional accuracy may be very important while non-mechanical material
properties like electrical conductivity may be unimportant.

The dot product of the performance vector and the transpose of the importance
vector yields the scalar part performance, pi

A,- WT = Pi
The determination of the appropriate weights in the importance vector is thus
crucial to the use of thismethod. Each department in an organization could
conceivably use a different weighting vector to determine jjerformance of a process
for its applications. A weighting vector might even be different for different parts
within the same department (e.g. a duct versus a bracket). The weighting vector
may vary with time as well; the vector used at the concept stage of development
would differ from that used during the pre-production phase. For example, a

Page 4
"breadboard" prototype intended only to verify a kinematic function might have an
associated weighting vector that emphasizes dimensional accuracy, while a pre-
production prototyp>e intended to test consumer response might have an associated
weighting vector that emphasizes surface finish.

In our study, we
determined the relevant performance attributes and the
importance weightings for a particular class of parts in a particular company. Then,
based on estimates of the time, cost, and p>erformance of four processes on over one
hundred parts, we were able to determine the average performances, costs, and
times for the processes. This information allows us to prescribe a prototyping
decision procedure for this particular set of attributes and importance weightings.

Focus

We focus on the prototyping of plastic piece parts. All of our data were gathered at
the Apparatus Division of the Eastman Kodak Company (KAD) in Rochester, New
York. Although product design involves a diverse set of prototyping applications,
processes, and needs, we focus on a particular class of prototyping problems in a
specific industrial setting in order to generate meaningful While the
results.
specific results are valid only for this setting, the methodology can be applied to
other types of parts and processes. Evaluating prototyping technologies for printed
circuit boards, metal shafts, or food products would require performance vectors
substantially different from those of plastic parts, but we believe the basic framework
would remain valid and useful.

The Parts

All of the parts used in our study were intended to be plastic in final production.
For over 85% of the parts, the final production process had been determined to be
injection molding. For the balance of the parts, the final process had not been
determined, or, in a few cases, other processes such as vacuum forming were to be
used. The parts were obtained from a database of parts at the stereolithography
laboratory at KAD, and were drawn from several types of products including
photocopiers, medical imaging equipment, and printers. The parts varied in
complexity, size, and shape. We believe that the part sample represents a uniform
distribution of part types for injection molded parts in medium volume products.
Figure 1 shows two typical parts from the sample.

The Processes

The study was based on four prototyping processes: computer-aided design solid
modeling (CAD), stereolithography (STL), computer numerically controlled
machining (CNC), and room temperature vulcanizing rubber molding (RM). It
should be noted that the processes are not entirely independent: STL and CNC
depend on a CAD model, and RM requires a master (often generated by STL or
CNC). The processes are described briefly below.

Page 5
Figure 1: Sample parts.

Page 6
CAD solid modeling. For our study, we assume that a part design exists; we base the
prototype cost and time performance for CAD on the resources required to translate
a defined design into a solid model. In many firms, solid modeling is becoming the
standard means of creating, representing, and storing design data [Hurt89, Stix91]; in
these settings, there would be no additional cost or time required to create a CAD
solid model for prototyping purposes. As a baseline case, we include only the
visualization and geometric reasoning capabilities of solid modeling. With enough
time and money, simulation and analysis packages can be used to understand
almost every performance aspect of a part without building a physical prototype. At
KAD these tools are typically used on only the most critical parts in a product. For
this reason, and because there is so much variability in both the available software
and in user skill, we confine our definition of CAD as a prototyping process to the
basic functionality of a solid modeler. In the final section of the paper we discuss a
set of dear opportunities for CAD to supplant physical prototyping.

Stereolithography. Stereolithography is a process by which solid plastic parts are

created by building a stack of cross sections. Each cross section is fabricated by using a
laser to cure a thin layer of a liquid ultraviolet-sensitive photo-polymer. Estimates
of STL parts were made based on the performance of a 3D Systems, Inc. SLA500
machine. There are currently only a few polymers that can be used in the
stereolithography process [Lindsay 90]. Fabricating an STL part requires a CAD
model of the part.

Computer Numerically Controlled Machining. Evaluation of CNC parts assumed

fabrication on Although many parts could be
a lathe or five-axis milling machine.
built on a three- or four-axis machine, five axes were assumed in order to minimize
the number of fixtures and setups. Some of the parts required setup on both a lathe
and a mill [Ford90].

Rubber Molding. Many plastic parts can be made by casting a polymer such as
polyurethane in a silicone rubber mold. This process involves first casting the
rubber mold around a part master and then filling the resulting mold cavity with
the desired part material. Molding compounds are available that approximate the
properties of manythermoplastics used in injection molding. At KAD, both CNC
and STI. are used to create the part master.

2. APPROACH AND METHODOLOGY

We used estimates and surveys to answer the following questions for our focused
domain:
What attributes are important for prototype parts (i.e. What attributes
make up A)?

Page 7
 What are the relative importances of these properties (i.e. What is

W)?
How well do the different processes perform with respect to these
importances (i.e. What are the values for the ay 's in A?)?
What are the lead times and costs for the processes (i.e. what are c and
0?

A User Survey of Prototyping Perceptions and Needs was used to identify critical
attributes. An Attribute Importance Survey was developed to establish the
importance of each attribute at various phases of development (the values of the
elements of W). A Process/Part Evaluation was designed to determine cost, lead
time, and process performance for each attribute (the values for c, t, and the
elements of A).

User Survey of Prototyping Perceptions and Needs

In order to gather general information about prototyping perceptions and needs, we

conducted an informal written survey of 24 designers, engineers, and managers at 6
of the lines of business within Eastman Kodak. The survey asked for such things as
a definition of "prototype", the purpose of prototypes within the organization,
material and process needs, standard procedures for procuring prototypes, and the
characteristics of the ideal prototyping facility.

Based on the responses to this survey we identified the performance attributes given
column of Table 1. In order to simplify our presentation of the results,
in the right
we have aggregated the attributes into four higher-level categories.
Table 1: Performance Attributes Identified in the User Survey

aggregate attributes
Attribute Importance Survey

A written survey was sent to over 900 designers, engineers, tool-and instrument-
makers, and project managers at Eastman Kodak. 171 (18%) of the surveys were
returned. The survey asked each respondent to rate the importance of each of the
attributes in Table 1 at the concept, engineering, and pre-production phases of
development. At each phase, each attribute received a ranking from "unimportant"
to "very important." The survey also requested information about the respondents'
primary function (design, engineering, or management) and discipline (mechanical,
Of the 171 respondents, 143 identified themselves as
electrical, optical, etc.).
involved in a "mechanical" discipline, 12 specified "electrical," 7 specified "optical,"
and 1 1 specified "other." Of the 143 in the mechaiucal discipline, 52 specified that
"design" was their primary fimction, 63 specified "engineering," 18 specified
"management," and 9 indicated "other." The respondents were selected from a
mailing list of users of the Unigraphics CAD system at Eastman Kodak and from
personal contacts made by the authors. A copy of the survey is shown in Appendix
A. These data were used to establish the imp>ortance weighting vector for the study.

Process/Part Evaluation

Parts were evaluated along the three dimensions

performance, time, and cost
for each of four processes.Estimates of time and cost were made by experienced
employees at KAD. In order to reduce variability due to communication issues, the
majority of the estimates were made with the part in hand. Table 2 lists the
assumptions upon which the estimates were based and the components of the time
and cost estimates. The performance of each part was calculated using the attribute
rating system described in Section 1, Framework, and the attributes listed in Table 1.

Table 2: Components of Estimates and Assumptions.

txDmponents of cost
components of time
translation of generation of definition of tool design of mold
drawing or support path fabrication of
defined concept structures debugging of inserts, supports,
to CAD solid translation and program etc.
model manipulation of setup time fabrication of
build files runtime mold
build time hand finishing mold cure time
cure time and/or post pKJuring of part
hand finishing machining part cure time
where needed

A number of factors make estimating processing time and cost difficult. First, the
reliability and consistency of the estimator had to be established. In many
organizations estimating is still more art than science; often the estimators would
produce one estimate then start over as they realized a more elegant solution to the
fabrication problem. Second, we had to define where one process finished and
another started, and we had to define the "average" target level of performance for
each process. For example, a stereolithography part destined for the paper path of a
copier required more hand finishing time than a similar stereolithography part
designed to hold rolls of film. Third, the skill levels of the operators were also an
issue. Fourth, we often found variations within a given process; a given part can
often be fabricated by a given process a number of different ways. Finally, the
capacity utilization of a process determines the time a part will spend waiting for the
process. For the time estimates, we assumed that capacity utilization was low
enough that the queueing effects were negligible. This will only be true in
organizations that allocate enough capacity to prototyping that jobs move through
the shop without contending for resources.

Rather than spend a great deal of time calculating an exhaustive estimate with a
very tight margin of error, the estimators were instructed to maintain consistency
between parts; the accuracy of one part relative to another was more important than
the estimate for any given part. It should also be noted that the estimators were
explicitly told that they would not be held to their estimates (we tried to decouple
the estimating process from the organizational incentives within KAD). Also, the
estimators were familiar with many of the parts; in some cases they had fabricated
the parts within the past year.

For each part, one of us (Wall) rated performance in each attribute category on a
scale of to 1 . The basis for the rating was the answer to the question, "How well
does the prototype part approximate the production part?" In some cases, the
question was, "How much information does the prototype part tell about the
production part?" For some of the attributes [and for the CAD process] one question
makes more sense than the other. A score of 1 meant that the prototype part
matched the production part for that attribute (or that the prototype part provided as
much information as the production part). For example. Tables 3a & 3b show the
estimates for and evaluation of the two sample parts in Figure 1.

Page 10
 Table 3a: Example of Estimates
(times in hoursrminutes)

process pnxess component "bracket" "duct"

CAD time to create solid model

Once these evaluations were performed for all 104 parts, the part performance
ratings, the cost estimates, and the time estimates were averaged to rank the
processes in terms of performance, cost, aind time.

3. RESULTS
Summary of Results

The results of the study are summarized in Figures 2-10. Figure 2 shows the
importance of each aggregate prototype attribute based on replies to the Attribute
Importance Survey. (The individual attributes that make up the aggregate attributes
shown in Table 2 are listed in Table 1.) Note that the importance of each attribute
increases as the project progresses and that time was rated most important at every
phase of product development. As expected, there are differences in the
importances indicated by the respondents. There is also some difference in the
mean responses among respondents who classified themselves as "engineering" or
"design" and those who claimed "management." Depending on the goals of the
prototyping evaluation process, a weighting vector could be defined for a single
individual or for a particular group of individuals. For presentation purposes, we
created a weighting vector by averaging all of the responses. The resulting vector is
W^ = [o.73 0.74 0.53 0.49] for the attributes strength /stiffness, dimensional accuracy,
appearance, and material properties other than strength and stiffness.

The results of the lead timeestimates from the Process /Part Evaluation are shown
in Figure 3 for a lot size of 1 and
in Figure 4 for a lot size of 25. The results of the
cost estimates from the Part/Process Evaluation are shown in Figures 5 for a lot size
of 1 and in Figure 6 for a lot size of 25. Note that the horizontal axes in figures 3
through 7 do not represent continuous variables, rather they are part numbers 1
through 104 evenly spaced along the horizontal axis. The values for each process
are connected together by lines to make the data legible and not to suggest a
continuous function. The data in figures 3 through 7 were all sorted in order of
increasing lead time for the CNC process and a lot size of 1. Lead time to fabricate a
single part for each of the processes depended loosely upon part complexity (where
complexity was related to the number of geometric primitives needed to define the
part) and so the parts are roughly ordered in terms of increasing complexity.

The performances for each part and process are plotted in Figure 7. The
scalar part
scalar performance of each part was calculated using a performance vector (A), based
on tiie attributes in Table 1, and the weighting vector, W^ = [0.73 0.74 0.53 0.49],
derived from the Attribute Importance Survey. The data were sorted as in Figures
3-6.

Averaging of the part performances for each process yields the average process
performance results shown in Figure 8. For each process, the time and cost scores
are normalized values relative to the other processes, where 1 is fastest /least
expensive and is slowest /most expensive. Time and cost ratings are for a lot size

ofl.

Page 12
 1

C) C)
C)
.[]
0.8 ()
[] []

0.6
[]
[] []

0.4
O pre-prod
D engineering
0.2
O concept

strength/ form
lumi appearance naterial
m--'^'"" time cost
stiffness accuracy properties
(other than
strength &

Figvire 2: Mean and Standard Deviation of the Importance

of Performance Attributes, Time, & Cost at Various Phases of Development

Page 13
 RM(CNC)

i
i
; ,, RM(STL)
'
I 1 '.

part

Figure 3: Lead Time for Each Part (lot size of 1)

500 M(CNC)

,,^
/ RM(STL)

400 -

, STL

300
V , CNC
Of

E
^
TO
200
01

100

part

Figure 4: Lead Time for Each Part (lot size of 25)

Page 14
10000

RM(CNC)

CNC

RM(STL)

I
V /: (. ,- STL

CAD

part

Figure 5: Cost of Each Part (lot size of 1)

1000

800

^600

400
u

RM(CNC)
200

RM(STL)

part

Figure 6: Cost of Each Part (lot size of 25)

Page 15
 1 -I

0.9

0.8

60
c 0.7

at
- 0.6 H

0.5

a. / fv\--~^, .-. Aa,^

0.4
-
v^
"^
M''\'\rr-^^^^^^
'
J - / /
/, A
\\ ^.\
.*. ;\ CAD
0.3 - V vv
0.2
part

Figure 7: Aggregate Performance of Parts

O ^
[] ^ ^
.[]
i
k
M
.[]
0.8
f'
u
^b
0.6

0.4 O RM
D CNC
<^
O STL
0.2
X CAD *
-L^
strength/ form appearance material time cost
stiffness accuracy properties
(other than
strength k. stiffness)

Figure 8: Mean and Standard Deviation

of Performance, Time, & Cost Ratings for Different Processes

?age 16
Finally, Figures 9 and 10 illustrate the quantitative relative time- and cost-
p>erformance of each process based on lot sizes of 1 and 25. Note that becavise the
relative ordering of the cost and time values for each of the processes is roughly the
same regardless of the parts (indicated by infrequent crossing of the process lines in
figures 3 through 6), the large standard deviations in figures 9 and 10 do not imply
an inability to predict relative overall process performance. Rather, a part costing
substantially less than the mean to produce by STL v^ll likely also cost substantially
less than the mean to produce by CNC.

Interpretation of Results

A decision among processes can be made based on

for prototyping a particular part
the data in Figures 8-10. The first step is to determine, using Figure 8, w^hich
processes w^ill provide the information required of the prototype. Then Figures 9 &
10 are used to determine which of these processes is best in terms of time or cost for
a given lot size. For example, the bracket shown in Figure 1 must be made of a
conductive material. Based on this fact and the process ratings in Figure 8, the
alternative processes are RM(CNC), RM(STL), and CNC. If we assume that 25 parts
are required and that time is the driving factor in the process decision. Figure 9
shows that CNC is the preferred process. If we extend the example and assume
instead that cost were the driving factor. Figure 10 shows that rubber molding from
a stereolithography master is preferred.

Table 4 summarizes the process decisions for a small set of decision drivers. The
table uses time, cost, the four aggregate attributes, and lot size as the factors that
determine which process to use. Each entry in the table was made by (1) narrowing
the set of possible processes to those that will perform well with respect to a critical
attribute (using Figure 8), and (2) determining the best time or cost performance for
the appropriate lot size (using Figures 9 & 10). The parenthesized entries in the
"form accuracy" row indicate the preferred process if a physical part is required. In
some cases either STL or CNC would be the "preferred" process; whereas the
average values indicated one process, the standard deviations were large enough to
suggest that the other might be a viable alternative. In these cases alternatives are
provided in parentheses.

Page 17
 500 3500
RM(CNC)

RM(CNC)
RM(STL) 3000

400
()

2500
CNC ()

RM(STL)
>^
i;? 300
3 2000
O
Vi

STL
(A
O
T3
TO
1500
200 STL

CNC
\
1000
RM(CNC)
RM(CNC)
RM(STL)
100
() RM(STL)
C)
CAD(|) CNC
C)
CNC 500

STL STL C) C)
CAD

lot size: 1 lot size: 25 lot size: 1 lot size: 25

Figitre 9: Mean and Standard Deviation of Figure 10: Mean and Standard Deviation of
Process Lead Times Process Costs

Page 18
 Table 4: Preferred Prototyping Processes Based on Results.

critical attribute
labor cost issues in design decision making. The time suggests that
criticality of
firms should maintain enough prototyping capacity that queueing delays do not
influence the lead time of the product. Justifying the associated capital outlay is
difficult without some explicit estimate of the value of prototyping lead time to the
organization.

Conventional CNC machining is among the best prototyping alternatives. For

plastic parts, process performance depends on the details of the application.
Nevertheless, it is interesting that the best choice of prototyping technology (Table 4)
is dominated by CNC machining for the time-driven case. We
have observed that
many designers are surprised by this result because the trade literature has given a
lot of attention to so-called rapid prototyping processes.Although we see
tremendous opportunities for rapid prototyping technologies like stereolithography,
machining is often the best available prototyping technology. In our subsequent
discussion of process development we will articulate the process developments that
may change the relative strengths of these processes.

Rapid is a relative term for prototyping. The promise of instant models from CAD
data is still largely unfulfilled. The average lead time to acquire a physical prototype
from stereolithography was 9.3 hours for our part sample, assuming that there was
no queue for the stereolithography machine. However, in general STL parts can not
be used in functional tests because of the relatively poor material properties of the
stereolithography resins. The average time to acquire a part with good stiength and
stiffness properties was 24.2 hours (CNC machining). TTiis can be as long as three
calendar days for a one-shift prototype shop.

Processes tend to perform better on simple parts than complex parts. This result is
intuitive but because we have not displayed the geometry of each of the 104 parts, it
is not clear from the presentation of our data. In general, the differences between

processes were greater than the differences in time, cost, and performance between
parts. (In fact, this is what allows us to make the part-independent
recommendations in table 4.) Nevertheless, simpler parts (measured by the
number of geometric primitives in the CAD file) in general obtained higher
performance ratings. One implication of this observation is that part designers
might consider simplifying or tailoring part geometry in anticipation of prototyping
requirements. A related idea was suggested for analysis tools in [Suri89] and seems
equally appropriate for prototyping.

There is still no substitute for a skilled operator. Many views of "desktop

manufactUi-ing" neglect the knowledge and skills of machine operators by trying to
make the designer omniscient. We found that many ideas are born as designers
interact with fabricators.

Supporting Design Decision Making

A complex electromechanical product may contain thousands of parts with diverse

intended functionality and diverse intended final production processes. When

Page 20
integrated into the design process, a rough-cut prototyping process evaluation
system could be a powerful tool for the designer, engineer, and manager. A
computerized implementation may be imagined as follows. As the project is
irutiated, attributes and importances are defined and agreed upon for different
classes of parts. Decisions about the number of prototypes that will be required at
various stages of the process are established. This information can be used to
inform the designer what the recommended prototyping process is and what the
associated time, cost, and performance measures for the current part or assembly are.
The designer could then add this information to the information available for other
evaluation criteria in order to make better detailed part design decisions. As the
needs of the project evolve, the project team might change the desired lot sizes or
call for re-evaluation of attribute importances. Such a tool would fit naturally into a
group design environment utilizing comprehensive, interactive design software.
Improving Prototyping Processes

Figures 11 and 12 and Table 5 summarize the process cost and time components for
CNC, STL, and RM. Along with the average process p>erformance results shown in
Figure 8, these results help to focus prototyping process development. Note that in
this case the STL and CNC figures do not include the time or cost to create a CAD
model and the RM figures do not include the time or cost to fabricate a master.
Note that the lead times for 25 parts could be reduced substantially in some cases by
parallel processing. For example, if the parts are small enough, more than one can
be built simultaneously on the SLA500 platform. Or, if more than one master is
available, multiple parts can be molded simultaneously.

Table 5: Average Lead Times (hours)

STL CNC RM
1 part 9.3 24.2 59.2

25 parts 130.9 88.8 370.9

Each of the processes has weaknesses and could benefit immensely from research
and development. The figures illuminate important areas for process development:
Nearly 50% of the time to fabricate one CNC part is programming
time. If that is reduced (by automated generation of cutter paths, for
example) then CNC becomes even more useful as a prototyping tool.
STL performs very where material properties
well on average except
are critical; An improvement in
the available resins are too limited.
resins would immediately make STL the dominant prototyping
technology for plastic parts.
CAD is very fast and has high geometrical accuracy. But, it falls short
in evaluating strength, stiffness, and other material properties. We
did not assess finite element analysis methods of prototyping which
address some of these weaknesses. However, the state-of-the-art in

Page 21
 simulation an inadequate substitute for a physical model.
is still

Advances in non-linear,dynamic simulation of multiple parts would

enable CAD to emerge as an even more important prototyping
technology.
Rubber molding an attractive process because it can be used for
is

small scale production. But the process is slow. There is a dear need
for either fast-curing polymers or for technologies that allow
thermoplastics to be molded from quickly-procured tooling.
Developments in spray metal tooling may eventually enable the
fabrication of batches of 100 plastic parts relatively quickly [Weiss90].

We hope from our study soon will be obsolete because of

that the specific results
rapid advances in process development. We believe, however, that the
methodology we have developed can be used to make intelligent and rational
prototyping decisions as processes change and in differing product development
contexts. The methodology supports the tactical decisions surrounding part design
and choice of prototyping technologies as well as the strategic decisions surrounding
investment in process development.

Page 22
 generate support
tructures
finish part
prepare files for build
generate build files

20001

B labor cure part

build part

M machine

1500
D materials finish part

build fixture

program
cost
per
part 1000
($)

setup/debug

fabricate inserts

500 cure part buOd mold

pz:q

pour part

qty: 1 25 1 25 1 25

STL CNC RM
cure mold

Figure 11: Components of Cost Figiuel2: Process Components

5 a Fraction of Time (lot size of 1)

Page 23
5. REFERENCES
"Ford Finds New Formula For Prototypes." Mechanical Engineering, April 1990, pp
32-33.

Deitz, Daniel, "Stereolithography Automates Prototyping." Mechanical

Engineering, February 1990, pp 34-39.

Dickerson, K. R, Benton, R. H.; Atyeo, M. A. "Rapid prototyping tools and

techniques." British Telecommunications Technology Journal, Vol 6, No 4,
October 1988, pp 65-72.

Fitzgerald, Karen, "Compressing the Design Cycle." IEEE Spectrum, October 1987,
pp 39-42.

Hurt, James, "A Taxonomy of CAD/CAE Systems." Manufacturing Review, vol 2,

no 3, September 1989, pp 170-178.

Kutay, Aydan, "Strategic Benefits of Rapid Prototyping Technology." The National

Conference on Rapid Prototyping, 5 June 1990.

Lindsay, Karen F., "Rapid prototyping shapes up as low-cost modeling alternative."

Modern Plastics, August 1990, pp 40-43.

Stix, Gary, "Plane Geometry," Scientific American, March 1991.

Suri, Rajan and Masami Shimizu, 'TDesign for Analysis: A New Strategy to
Improve the Design Process." Research in Engineering Design, 1989, 1:105-
120.

Ulrich, Karl et al, "A Framework for Including the Value of Time in Design-for-
Manufacturing Decision Making," MIT Sloan School of Management
Working Paper #3243-91-MSA, February 1991.

Weiss, L. E. et al, "A Rapid Tool Manufacturing System Based on Stereolithography

and Thermal Spraying," Manufacturing Review, Vol. 3, No. 1, March 1990.

Page 24
6. APPENDIX A - THE ATTRIBUTE IMPORTANCE SURVEY
Name: ^^^1^
'^"V.Tn"^
Matthew Wall
I^ate: 2-4-EP mail code: 35326
LoB:
Primary Function (choose one): Primary Discipline (choose one):
O engineering O mechanical
O design O electrical
O management O optical
O other O other

At what stage(s) of the product development cycle are you involved with prototype
design, fabrication, or assembly?

O concept development
O product engineering
O pre-production

Given that prototypes are built for concept engineering pre-prod

different purposes, how important c
(on average) are the following c s
attributes in your prototype parts t
during the phases of product g
development listed to the right? '|
 Date Duef-^'^'u

APR. 25 1c 35

^ Gta'ff

Lib-26-67
 MIT LIBRORIES

3 TD6Q 00721EEE S