Chapter 8

PRELIMINARY VEHICLE STRUCTURE

DESIGN APPLICATION

Michael J. Scott and Erik K. Antonsson

Proceedings of the
10th International Conference on Design Theory and Methodology
ASME, Paper Number DETC98/DTM-5646, (September, 1998)

Abstract The Method of Imprecision, or MoI, is a formal method for incorporating im-
precise information into a design process. This methodology has been exercised
on a problem in preliminary vehicle structure design in collaboration with VW
Wolfsburg. Results show that the method is useful in trading off multiple con-
flicting attributes, including styling preferences and engineering requirements.

Keywords:
Industrial Applications of DTM; Vehicle Structure Design; Design Meth-
ods and Models; Design Representations; Computational Methods of Design;
Fuzzy Sets

Introduction
Preliminary design is inherently imprecise [3, 4, 12, 40], and many prelim-
inary design decisions are made informally. Preliminary design has enormous
economic importance, as much of the cost of a design is determined by these
(often informal) preliminary decisions [37]. A further complication is the dif-
ficulty of communicating imprecise information between different members or
groups involved in the design process. Many “interface” decisions are made
after design analysis is complete; these post hoc decisions can result in costly
redesigns.
The Method of Imprecision, or MoI [39, 1] has been developed to formally
incorporate imprecise information into the engineering design process. In the
summer of 1997, an application of the MoI to preliminary vehicle structure
183
184 IMPRECISION IN ENGINEERING DESIGN

design was demonstrated for Volkswagen Wolfsburg. The application serves

both to demonstrate the capabilities of the method, and as an introduction to
some of the underlying concepts.
A brief introduction to the Method of Imprecision is followed by a descrip-
tion of the demonstration project. The application of the MoI to the problem is
then described in detail, and the implications of the results are discussed.

The Method of Imprecision

This introduction is necessarily brief; the application serves as a further
tutorial to explicate the ideas reviewed here.
The original work on the MoI [38, 39] formulated the design problem as
a decision problem: given a set of candidate designs, identified by vectors d~
of design variables in a Design Variable Space (DVS), a set of performances,
described by vectors p~ of performance variables in a Performance Variable
Space (PVS), and a mapping f : d~ 7→ p~, choose the candidate design d~∗ which
maps to the “best” possible performance ~p∗ = f (d~∗ ). So stated, there is in-
sufficient information to determine what constitutes a “best” performance, and
hence a “best” design. On the one hand, requirements are imprecise, while
on the other hand, there is no obvious way to compare different performance
variables which are usually not expressed in the same units.
The need to include imprecision in engineering design can be illustrated by a
simple example. Figure 8.1 shows a specification for one performance variable
(pj ), with the performance preference µp on the vertical axis. As specifications
are commonly written, pj ≥ 500 km would be represented by the dashed line
(the sharp-edged rectangular step), where µp = 1 in the acceptable region,
and µp = 0 for unacceptable values. However, this crisp specification (or
requirement) indicates that two different designs, one with dj = 500 − , and
another with dj = 500 + , would have completely different acceptabilities,
no matter how small  becomes. Thus two designs, indistinguishably different
in dj (as  → 0), have completely different preferences: one is completely
acceptable and one is unacceptable. This situation makes no sense.
Alternatively, the solid line shown in Figure 8.1 indicates a smooth transi-
tion of acceptability of performances from unacceptable (µp = 0) to accept-
able (µp = 1), and thus reflects a more realistic specification. The range over
which the transition from unacceptable performance to most desired perfor-
mance takes place will depend on the particular design problem, and may be
more or less steep, and smooth or faceted.
Thus the MoI introduces the notion of preferences, denoted by µ ∈ [0, 1],
both to represent the imprecision inherent in the preliminary design problem,
and to provide a basis for comparison between different attributes. Perfor-
mance preferences µp on the PVS express the requirements more completely
 Preliminary Vehicle Structure Design Application 185

µp
1
crisp
fuzzy

0
500 km range

Figure 8.1 Example Imprecise Specification.

than crisp targets. In addition, engineers express design preferences µd on de-

sign variables, allowing the incorporation of performance aspects that are not
explicitly calculated by f . Preferences are naturally represented and manipu-
lated using the mathematics of fuzzy sets [42].
~ which are specified on the DVS, can be mapped
The design preferences µd (d),
onto the PVS by use of the extension principle [41]. The various prefer-
ences are then combined with an aggregation function P; at first, the MoI
made use of two different aggregation operators [24], the non-compensating
Pmin (µ1 , µ2 ) = min(µ1 , µ2 ) for situations where the overall performance is
dictated by the lowest-performing attribute, and the compensating PΠ (µ1 , µ2 ) =
√
µ1 µ2 , when high performance on one attribute is deemed to partly compen-
sate for lower performance on another. Each candidate design d~ thus has an
associated overall preference:
 
~ = P µd (d),
µo (d) ~ µp (f (d))
~

~ and µp (~
(where µd (d) p) are themselves aggregations of their constituent pref-
erences), and candidate designs can be compared on the basis of this overall
preference.
Further research on the MoI developed techniques for including noise [26]
and adjustments, or tuning parameters [25], in the imprecision calculations,
and placed an axiomatic framework on the calculations [23]. Implementa-
186 IMPRECISION IN ENGINEERING DESIGN

tion of the MoI continued with the development of a computational tool [19].
The applicability of the method was seen to be limited by large computational
requirements, so the inclusion of Design of Experiments (DOE) approxima-
tions [21] and other computational innovations [18] followed. The need for
more than two aggregation functions to model different trade-off levels was
recognized, and a family of aggregations introduced [32, 33]. The MoI is re-
viewed in more detail, and compared to other methods, in [1].
Other researchers have applied fuzzy methods to design optimization prob-
lems [7, 8, 28, 29, 6]. Related research includes: chemical process synthesis
[9]; fuzzy constraint propagation applied to manufacturing [10]; fuzzy schedul-
ing [11]; application of fuzzy methods to windturbine design [13]; multiob-
jective scheduling [14]; management of uncertain knowledge in engineering
design [15]; engineering design optimization [16]; imprecise calculations in
engineering design [5]; evaluation of design alternatives [17]; fuzzy MADM
methods in system design [22]; fuzzy evaluations [27]; multiobjective fuzzy
optimization [30, 34]; fuzzy ratings and utility analysis in preliminary design
evaluation of multiple attributes [35]; scheduling system design [36]; and fuzzy
multi-criteria decision making [44, 45, 43]. In addition, there is increasing in-
terest in related work in evolutionary algorithms [46], including the combina-
tion of evolutionary algorithms with the MoI [31].

Preliminary Vehicle Structure Design

The general vehicle structure design problem is the engineering of a body-
in-white, which consists of the (usually metal) frame to which components
and exterior panels are fastened. While there are interesting alternative so-
lutions such as space frames and monocoques, this paper is concerned with
the welded metal structure typical of passenger automobiles of the present day
(see Figures 8.2 and 8.3). The vehicle structure engineers must design a body-
in-white that meets certain measurable engineering targets such as stiffnesses,
stress levels under load, and weight. In addition, they must satisfy many per-
formance targets associated with less easily measured concepts such as style,
manufacturability, and requirements of other engineering groups involved in
the design process. These unmeasured peformances are handled informally,
often by negotiation between groups working on the same vehicle. The MoI
was developed to allow for a formal approach to the incorporation of this im-
precise information.
In order to avoid any difficulties involving confidential information, it was
decided that an older model vehicle would provide an effective demonstration
of the method. To this end a 1980 VW Rabbit (see Figure 8.2) was acquired.
The vehicle was stripped to the structural body-in-white, and torsional and
bending stiffnesses were measured. The intact body-in-white was found to
 Preliminary Vehicle Structure Design Application 187

Figure 8.2 1980 VW Rabbit in Stiffness Testing

have a torsional stiffness of approximately 4900 N-m/degree and a bending

stiffness of approximately 2500 N/mm. Tables of data from some of the load
tests are shown in the Appendix. In addition, geometric data were gathered and
used to create a solid model (Figure 8.3). The solid model and the structural
stiffness information together were used to create and calibrate a finite element
model (Figure 8.4).
The finite element model was parameterized with five1 design variables:
1. A-pillar thickness (mm)
2. B-pillar thickness (mm)
3. floor pan thickness (mm)
4. floor rail thickness (mm)
5. B-pillar location (mm aft of a nominal point chosen by stylists)
and the performance was assessed with three measures:

1 The demonstration here was conducted using a subset of the design variables; the method can be applied

directly to a larger set of variables.

188 IMPRECISION IN ENGINEERING DESIGN

Figure 8.3 Geometric Model of Body-In-White in SDRC I-Deas

 Preliminary Vehicle Structure Design Application 189

Figure 8.4 Finite Element Model of Body-In-White

1. Bending stiffness (N/mm)

2. Torsional stiffness (N-m/deg)

3. Weight (kg)

The stated design problem was to achieve 10% improvements over the refer-
ence model in the three measured performances. In addition, it was understood
that that the design must not be difficult to manufacture, and that this year’s
model should have a somewhat longer and sleeker look.

Applying The MoI To Include Imprecise Information

While standard optimization methods could be used to determine the highest
achievable bending stiffness, the highest achievable torsional stiffness, or the
lowest achievable weight for this analysis model, such an optimization would
not tell the designer which designs are the most promising when other relevant
considerations are taken into account. On the one hand, there is a necessary
trade-off between the stiffnesses and the weight; it is impossible to optimize
both simultaneously. Additionally, there is other (imprecise) information to
consider when making the decision, such as manufacturing and styling con-
cerns. The application of the MoI to this problem involves constructing a differ-
190 IMPRECISION IN ENGINEERING DESIGN

1.0 1.0

0.8 0.8

0.6 0.6
µ µ
0.4 0.4

0.2 0.2

0.0 0.0
2000 3000 4000 5000 6000 7000 8000
Bending Stiffness (N/mm) Torsional Stiffness (N-m/deg)

1.0

0.8

0.6
µ
0.4

0.2

0.0
120 140 160 180
Weight (kg)

Figure 8.5 Imprecise Performance Requirements

ent “optimization” problem that includes the imprecise information that would
be left to the negotiation stage in traditional design.
The calculated performance requirements on bending stiffness, torsional
stiffness, and weight were originally expressed as targets of 10% improve-
ments over the reference model. As was discussed above, this is unrealistically,
and indeed unproductively, precise. In place of these hard targets, imprecise
performance requirements were specified with a linear interpolation between
two points. In the implementation of the MoI, it is common to name the cus-
tomer as the source of the performance preferences; in fact, it is more likely to
be a manager, perhaps informed by market research, serving as the customer’s
proxy. To specify these imprecise requirements, the manager must answer two
simple questions: “What is the lowest performance you can live with (where is
µ = 0)? What performance would satisfy you completely (where is µ = 1)?”
These bounds are clearly dependent on a number of factors, including the target
market and the performance of competitors’ products; we have found that en-
gineering managers can answer these two questions with little more effort than
is needed to settle on the initial crisp target. Figure 8.5 shows the imprecise
requirements on stiffnesses and weight.
 Preliminary Vehicle Structure Design Application 191

To include requirements on manufacturing, availability, style, and other

things which are not calculated in the finite element analysis, designer prefer-
ences are specified on the design variables. As with the imprecise performance
requirements, they range from µ = 0 at the unacceptable limit to µ = 1 at
the most preferred. A preference is defined on each of the five design vari-
ables, as shown in Figure 8.6. Each preference is representative of imprecise
information that can be incorporated using the MoI:

1. The sheet steel for stamping the A-pillar is only available in certain in-
crements, so this plot is discrete rather than continuous. The manufac-
turing engineer has a higher preference for thinner sheets, since they are
easier to form; this is a design preference for manufacturability.

2. The B-pillar thickness is continuous and more complicated than the lin-
ear performance preferences. This preference does not indicate that the
physical B-pillar might be 1.113 or 1.114 mm thick; rather it means that
the designer knows that the finite element model is simplified, and that a
high number for B-pillar thickness means that more reinforcing features
will need to be added to the B-pillar. The designer would like to keep
the B-pillar as simple as possible.

3. The floor pan thickness is preferred thicker by the designer for ease of
attachments and for durability.

4. The floor rail thickness preference is an example of a sourcing, or avail-

ability preference; it states that some thicknesses are more easily ob-
tained than others.

5. The design preference for B-pillar location comes from the stylists, and
captures the directive for a longer, sleeker look for this year’s model. It
has been specified differently from the other design preferences, using
α-cuts [1], so that the stylists have given a range of perfectly accept-
able values, a range of barely acceptable values, and a range of values
that fall in the middle. This method of specifying preferences can have
computational advantages.

In addition to these preferences, each attribute is assigned a weight indicat-

ing its relative importance, and the way in which attributes trade-off against
each other must also be specified. In this test example, it was determined that
bending and torsional stiffness traded-off in a non-compensating manner —
the lowest preference is maximized. Together they traded-off with weight in a
compensating manner, so that high performance on stiffness could partly make
up for low performance on weight, and vice versa. The designer preferences
all traded-off in a compensating manner as well, with the styling preference
192 IMPRECISION IN ENGINEERING DESIGN

1.0 1.0

0.8 0.8

0.6 0.6
µ µ
0.4 0.4

0.2 0.2

0.0 0.0
0.6 0.7 0.8 0.9 1.0 1.1 0.8 1.0 1.2 1.4 1.6
A-pillar thickness (mm) B-pillar thickness (mm)

1.0 1.0

0.8 0.8

0.6 0.6
µ µ
0.4 0.4

0.2 0.2

0.0 0.0
0.9 1.0 1.1 1.2 1.3 1.4 0.7 0.8 0.9 1.0 1.1 1.2 1.3
Floor thickness (mm) Floor rail thickness (mm)

1.0

0.8

0.6
µ
0.4

0.2

0.0
-50 0 50 100 150 200
B-pillar location (mm aft of nominal)

Figure 8.6 Designer Preferences

 Preliminary Vehicle Structure Design Application 193

assigned a relatively high weight to reflect the importance of styling consider-

ations in automobile design. The preferences for the computed performance
variables are also weighted heavily. The correct determination of weights and
trade-off strategies is crucial to the method, and a full range of strategies [33],
of which the original compensating and non-compensating strategies are only
two examples, is available.
Through the application of the MoI, the design problem has been reformu-
lated to be the maximization of the overall preference:
h i
~ = P µd (d),
µo (d) ~ µp (f (d))
~
h  h i
~ Pmin µp (f1 (d)),
= PΠ PΠ µp (f3 (d)), ~ µp (f2 (d))
~ ,
PΠ (PΠ [µd (d1 ), µd (d2 ), µd (d3 ), µd (d4 )] , µd (d5 )) ]

The computation of µo (d) ~ for a single design point d~ is limited by the finite
element stiffness calculation, which takes about a minute on a Sun Ultra1-
170MHz workstation; the calculations of weight and preference aggregation
are of negligible cost, regardless of the trade-off strategies employed. Even in
this relatively modest problem, where there are only five design dimensions,
an exhaustive calculation of preferences over the design space is prohibitively
expensive. The MoI exploits the structure of the problem to speed the search
for preferred solutions: the internal calculations linearize where possible, ef-
fectively reducing the dimension of the search space, and employ Powell’s
method to locate internal extrema [20].

Results
The design problem, including all imprecise information, was solved in
two different ways. First, in order to demonstrate the method, the finite el-
ement analysis was run 3125 (= 55 ) times to provide a coarse but complete
check of the entire design space. The point of peak overall preference of
µo = 0.50 was found at d~ = (1.0, 0.9, 0.9, 1.0, 50), where the design pref-
erences µd are (0.6, 1.0, 1.0, 0.5, 1.0); the stiffnesses and weight at this point
were p~ = (2832, 5836, 147), with preferences (0.23, 0.14, 0.62). The max-
imum achievable stiffnesses are 3365 N/mm (µp = 0.77) in bending and
6029 N-m/degree (µp = 0.25) in torsion, but the corresponding weight of
170 kg is unacceptable. Similarly, a weight of 144 kg (µp = 0.78) is achiev-
able, but stiffnesses drop to 2803 N/mm (µp = 0.20) and 5730 N-m/degree
(µp = 0.08). The combined overall preference µo also takes into account the
design preferences on style, manufacturability, and the like.
The power of the method lies not in an ability to find a single overall “best”
point, but in the information it contains of how the total combined preference
µo varies with each of the design variables. Although it is impossible to dis-
194 IMPRECISION IN ENGINEERING DESIGN

play all five dimensions varying at once, a tool was written that uses a com-
mercial package (Matlab) to display results interactively. Using the tool, the
designer can see the change in preference that would occur by varying each
design variable independently from a chosen beginning point. Results can be
seen on five simultaneous plots in two dimensions (see Figure 8.7), or on a
three-dimensional surface plot (see Figure 8.8) with the remaining design vari-
ables set to nominal values. The interpretation of these graphical results is
discussed in greater detail below.

Approximations
Naturally, the exhaustive evaluation of points in the design space would not
be performed on a real design problem. It was performed here only for com-
parison purposes. An approach that utilizes Design of Experiments (DOE) [2]
to approximate the finite element calculations for bending and torsional stiff-
nesses reached substantially similar results in only 21 runs (or approximately
20 minutes). The average difference (from the exhaustive evaluation) in bend-
ing stiffness was approximately 1%, with a maximum difference of less than
4%, while the average and maximum differences for torsional stiffness were
both less than 1%.
In some cases, the nonlinearities of the analysis function f will defeat a
linear or even polynomial approximation, but in many cases, such as the ex-
ample presented here, these simple approximations can drastically reduce the
required computation. Since precise answers are not required for preliminary
design, it is sensible to exploit approximation tools when possible. If more
computation can be justified, a more thorough calculation can be made.

Discussion
Engineering analysis usually requires some judgement on the part of the
designer. Unless a full-scale exact prototype is to be built and tested, the ac-
curacy of any calculated performance measure depends on the fidelity of the
model employed. Even when exact data are available for some attributes, fi-
nal decisions about a design incorporate other, unmodelled concerns, such as
manufacturing and styling.
The MoI constructs a model of the entire decision process, expressing the
calculated overall performance µo (d)~ as a function of the design variables. It
depends on many factors: the function f for calculating measurable perfor-
mances, the specification of design preferences µd and performance prefer-
ences µp , the weighting of these preferences, and the specification of trade-off
aggregations between attributes. A change in any of these will affect the shape
of the function µo in design space, and thus affect the decision. The analysis
f is here relatively expensive to compute, and changes in the finite element
 Preliminary Vehicle Structure Design Application 195

µmax achievable = 0.501 µo selected = 0.43787 design preferences

1 1 1

0.8 0.8 0.8

0.6 0.6 0.6

µo µo µo
0.4 0.4 0.4

0.2 0.2 0.2

0 0 0
0.7 0.8 0.9 1 1.1 0.9 1 1.1 1.2 0.8 0.9 1 1.1
A pillar thickness B pillar thickness Floor sill thickness

1 1

0.8 0.8

0.6 0.6
µo µo
0.4 0.4

0.2 0.2

0 0
1 1.2 1.4 −50 0 50 100 150
Floor pan thickness B pillar location

Figure 8.7 Graphical User Interface for Preference Display

196 IMPRECISION IN ENGINEERING DESIGN

µ*o = 0.40159
1

0.8

0.6
µo
0.4

0.2

0
1.4
1.3 150
1.2 100
50
1.1
0
1 −50
Floor Thickness B pillar location

Figure 8.8 3-D Graphical User Interface for Preference Display

 Preliminary Vehicle Structure Design Application 197

model are costly to propagate to overall preference. Changes in the other fac-
~
tors, on the other hand, are easily incorporated, as finite element results f (d)
are stored so that the same design point need never be analyzed more than once.
This allows the MoI to support an iterative decision process, when the informa-
tion from the first round of calculations inspires a change in the preference
structure.
In the example shown in this paper, the shape of the function µo (d)~ is sensi-
tive to changes in the styling preference µd (d5 ), which is not surprising, since
this preference is accorded a large weight. This and other features of the de-
sign problem can be seen in the advanced interface shown in Figure 8.7. The
vertical dashed (red) lines indicate the selected values of the design variables.
In this figure, d~ = (0.8, 1, 0.9, 1.2, 50), a point which is representative, not
optimal. The solid (blue) lines indicate how the overall preference µo would
change by varying that design variable while holding the other four fixed at
their current values. For instance, decreasing the floor pan thickness d4 will
result in a more preferred design. The dashed (black) lines show the maximum
achievable µo for each value of each design variable. Finally, the solid (red)
lines joining the circles are the specified preferences on design variables.
In this example, the desired improvements in performance were achieved
by small changes to the sheet-metal thicknesses and B-pillar location. Visually
the improved structure would appear quite similar to that shown in Figure 8.3.
No change in vehicle structure configuration was required here.
Figure 8.7 shows that the overall preference µo varies qualitatively, though
not quantitatively, with four of the five design variables; the exception is d4 ,
floor pan thickness. The variation of the overall preference µo with respect to
d4 shows a conflict between the calculated stiffness and weight requirements
µp and the provisions for attachment and durability captured by the designer
preference µd (d4 ). This indicates d4 as a likely candidate for change in a
potential redesign. The resolution of the conflict would be achieved by a choice
of d4 that provides the best overall trade-off between the competing attributes.
Designers can interact with the preference display to examine trends in the
structure of the overall preference.

Conclusions
In preliminary vehicle structure design, as in preliminary engineering design
in general, many important decisions are made informally on the basis of im-
precise information. Concerns of styling and manufacturability, for instance,
can carry great weight in the design process although they are not modelled by
any formal analysis. The MoI is a tool to formally incorporate such imprecise
information into the design process, and thus to make decisions on a sound
basis. In a demonstration of the MoI prepared for VW Wolfsburg, concerns of
198 IMPRECISION IN ENGINEERING DESIGN

manufacturing, styling, parts availability, and design were incorporated with

the engineering analysis of the structural stiffness of a VW Rabbit. The results
show the usefulness of the method in trading off these conflicting attributes.
Any analysis involving more than two design variables must contend with
two difficulties, the exploding need for computation, and the problems of dis-
playing results in several dimensions. The MoI uses approximations, when
feasible, to address the first difficulty, and an interactive graphical tool for pref-
erence display was developed and applied here to address the second.

Acknowledgements
Volkswagen Wolfsburg, and particularly Mr. W. Lange, and Dr. O. Tegel
(presently Chief Engineer of the Institut für Maschinenkonstruktion at the Tech-
nical University in Berlin), were indispensable in the development of this project.
The authors would like express their appreciation to Zee Khoo and Juan
Nuño for their many hours of work stripping and load-testing the Rabbit, and
building the solid model, and to Mike Gerfen, manager of Caltech’s Central
Engineering Shop, for his generous help in providing space and equipment.
This material is based upon work supported, in part, by the National Science
Foundation under NSF Grant Number DMI-9523232. Any opinions, findings,
conclusions, or recommendations expressed in this publication are those of the
authors and do not necessarily reflect the views of the sponsors.
 Preliminary Vehicle Structure Design Application 199

References
[1] A NTONSSON , E. K., AND OTTO , K. N. Imprecision in Engineering De-
sign. ASME Journal of Mechanical Design 117(B) (Special Combined
Issue of the Transactions of the ASME commemorating the 50th anniver-
sary of the Design Engineering Division of the ASME.) (June 1995), 25–
32. Invited paper.
[2] BARKER , T. Quality by Experimental Design. Marcel Dekker, Inc., New
York, 1985.
[3] B ECKER , J. M. A Structural Design Process Philosophy and Methodol-
ogy. PhD thesis, University of California, Berkeley, 1973.
[4] B LOCKLEY, D. I. The Nature of Structural Design and Safety. Ellis
Norwood Limited, Chichester, 1980.
[5] C HEN , J. E., AND OTTO , K. N. A tool for imprecise calculations in
engineering design. In Proceedings of the Fourth IEEE International
Conference on Fuzzy Systems (FUZZ-IEEE/IFES’95) (Mar. 1995), vol. 1,
IEEE, pp. 389–394.
[6] D HINGRA , A. K. Multiple objective design optimization. In Fuzzy Sets
in Engineering Design and Configuration, H.-J. Sebastian and E. K. An-
tonsson, Eds. Kluwer, 1996, pp. 53–88.
[7] D IAZ , A. R. Fuzzy set based models in design optimization. In Advances
in Design Automation - 1988 (New York, NY, Sept. 1988), S. S. Rao, Ed.,
vol. DE-14, ASME, pp. 477–485.
[8] D IAZ , A. R. A strategy for optimal design of hierarchical systems us-
ing fuzzy sets. In The 1989 NSF Engineering Design Research Con-
ference (College of Engineering, University of Massachusetts, Amherst,
June 1989), J. R. Dixon, Ed., NSF, pp. 537–547.
[9] D JOUAD , S., F LOQUET, P., D OMENECH , S., AND P IBOULEAU , L.
Fuzzy set methods in separation process synthesis. In Fuzzy Informa-
tion Engineering: A Guided Tour of Applications, D. Dubois, H. Prade,
and R. R. Yager, Eds. John Wiley & Sons, 1996.
[10] D UBOIS , D. Tuning the cutting edge conditions on a machining transfer
line using fuzzy constraint propagation. In Fuzzy Information Engineer-
ing: A Guided Tour of Applications, D. Dubois, H. Prade, and R. R. Yager,
Eds. John Wiley & Sons, 1996.
[11] FARGIER , H. Fuzzy scheduling: Principles and experiments. In Fuzzy
Information Engineering: A Guided Tour of Applications, D. Dubois,
H. Prade, and R. R. Yager, Eds. John Wiley & Sons, 1996.
[12] G AVIN , J R ., J. G. Fly me to the Moon: An interview with Joseph G.
Gavin, Jr. Technology Review 97, 5 (July 1994), 61–68.
200 IMPRECISION IN ENGINEERING DESIGN

[13] G ERHART, V. Uncertainty within the design of a wind turbine. Working

paper or, RWTH Aachen, Templergraben 64, 52062 Aachen, Germany,
1997.
[14] G RABOT, B., G ENESTE , L., AND D UPEUX , A. Tuning of fuzzy rules for
multiobjective scheduling. In Fuzzy Information Engineering: A Guided
Tour of Applications, D. Dubois, H. Prade, and R. R. Yager, Eds. John
Wiley & Sons, 1996.
[15] H AMBURG , I., AND H AMBURG , P. Fuzzy logic, A user friendly technol-
ogy for the management of uncertain knowledge in engineering design
process. In Modelling Uncertain Data (Berlin, Mar. 1993), H. Bande-
mer, Ed., Akademie Verlag, pp. 153–157. Mathematical Research series,
Volume 68, Lectures from the GAMM-Workshop: Modelling Uncertain
Data, held in Freiberg from March 21-24, 1992.
[16] H SU , Y.-L., L IN , Y.-F., AND S UN , T.-L. Engineering design optimiza-
tion as a fuzzy control process. In Proceedings of the Fourth IEEE In-
ternational Conference on Fuzzy Systems (FUZZ-IEEE/IFES’95) (Mar.
1995), vol. 4, IEEE, pp. 2001–2008.
[17] K NOSALA , R., AND P EDRYCZ , W. Evaluation of design alternatives in
mechanical engineering. Fuzzy Sets and Systems 47, 3 (1992), 269–280.
[18] L AW, W. S. Evaluating Imprecision in Engineering Design. PhD thesis,
California Institute of Technology, Pasadena, CA, June 1996.
[19] L AW, W. S., AND A NTONSSON , E. K. Including Imprecision in En-
gineering Design Calculations. In Design Theory and Methodology –
DTM ’94 (Sept. 1994), vol. DE-68, ASME, pp. 109–114.
[20] L AW, W. S., AND A NTONSSON , E. K. Optimization Methods for Cal-
culating Design Imprecision. In Advances in Design Automation - 1995
(Sept. 1995), vol. 1, ASME, pp. 471–476.
[21] L AW, W. S., AND A NTONSSON , E. K. Multi-dimensional Mapping of
Design Imprecision. In 8th International Conference on Design Theory
and Methodology (Aug. 1996), ASME.
[22] M ÜLLER , K., AND T H ÄRIGEN , M. Applications of fuzzy hierarchies
and fuzzy MADM methods to innovative system design. In Proceed-
ings of the Third IEEE International Conference on Fuzzy Systems (June
1994), vol. 1, IEEE, pp. 364–367.
[23] OTTO , K. N. A Formal Representational Theory for Engineering Design.
PhD thesis, California Institute of Technology, Pasadena, CA, June 1992.
[24] OTTO , K. N., AND A NTONSSON , E. K. Trade-Off Strategies in Engi-
neering Design. Research in Engineering Design 3, 2 (1991), 87–104.
 Preliminary Vehicle Structure Design Application 201

[25] OTTO , K. N., AND A NTONSSON , E. K. Tuning Parameters in Engineer-

ing Design. ASME Journal of Mechanical Design 115, 1 (Mar. 1993),
14–19.
[26] OTTO , K. N., AND A NTONSSON , E. K. Design Parameter Selection
in the Presence of Noise. Research in Engineering Design 6, 4 (1994),
234–246.
[27] P OSTHOFF , C., AND S CHLOSSER , M. Learning from examples for the
construction of fuzzy evaluations. In Proceedings of the Third IEEE
International Conference on Fuzzy Systems (June 1994), vol. 1, IEEE,
pp. 368–371.
[28] R AO , S. S., AND D HINGRA , A. K. Applications of fuzzy theories to
multi-objective system optimization. NASA Contractor Report 177573,
NASA Ames Research Center, 1991.
[29] R AO , S. S., S UNDARARAJU , K., P RAKASH , B., AND BALAKRISHNA ,
C. Multiobjective fuzzy optimization techniques for engineering design.
Computers and Structures 42, 1 (1992), 37–44.
[30] S AKAWA , M., AND K ATO , K. An interactive fuzzy satisficing methods
for large-scale multiobjective linear programs with fuzzy numbers. In
Proceedings of the Fourth IEEE International Conference on Fuzzy Sys-
tems (FUZZ-IEEE/IFES’95) (Mar. 1995), vol. 3, IEEE, pp. 1155–1162.
[31] S CHLEIFFER , R. Fuzzy engineering design. RWTH Aachen, 1998.
Diplomarbeit.
[32] S COTT, M. J., AND A NTONSSON , E. K. Aggregation Functions for En-
gineering Design Trade-offs. In 9th International Conference on Design
Theory and Methodology (Sept. 1995), vol. 2, ASME, pp. 389–396.
[33] S COTT, M. J., AND A NTONSSON , E. K. Aggregation Functions for
Engineering Design Trade-offs. Fuzzy Sets and Systems 99, 3 (1998),
253–264.
[34] S HIH , C. J., AND WANGSAWIDJAJA , R. A. S. Multiobjective fuzzy
optimization with random variables in a mix of fuzzy and probabilistic
environment. In Proceedings of the Fourth IEEE International Confer-
ence on Fuzzy Systems (FUZZ-IEEE/IFES’95) (Mar. 1995), vol. 3, IEEE,
pp. 1163–1170.
[35] T HURSTON , D. L., AND C ARNAHAN , J. Fuzzy ratings and utility analy-
sis in preliminary design evaluation of multiple attributes. ASME Journal
of Mechanical Design 114, 4 (Dec. 1992), 648–658.
[36] T ÜRKŞEN , I. B. Scheduling system design: Three fuzzy theory ap-
proaches. In Fuzzy Information Engineering: A Guided Tour of Appli-
cations, D. Dubois, H. Prade, and R. R. Yager, Eds. John Wiley & Sons,
1996.
202 IMPRECISION IN ENGINEERING DESIGN

[37] W HITNEY, D. E. Manufacturing by design. Harvard Business Review

66, 4 (July 1988), 83–91.
[38] W OOD , K. L. A Method for Representing and Manipulating Uncertain-
ties in Preliminary Engineering Design. PhD thesis, California Institute
of Technology, Pasadena, CA, 1989.
[39] W OOD , K. L., AND A NTONSSON , E. K. Computations with Impre-
cise Parameters in Engineering Design: Background and Theory. ASME
Journal of Mechanisms, Transmissions, and Automation in Design 111,
4 (Dec. 1989), 616–625.
[40] YAO , J. T. P., AND F URUTA , H. Probabilistic treatment of fuzzy events
in civil engineering. Probabilistic Engineering Mechanics 1, 1 (1986),
58–64.
[41] Z ADEH , L. A. Fuzzy logic and approximate reasoning. Synthese 30
(1975), 407–428.
[42] Z IMMERMANN , H.-J. Fuzzy Set Theory - and Its Applications. Manage-
ment Science/Operations Research. Kluwer-Nijhoff Publishing, Boston,
MA, 1985.
[43] Z IMMERMANN , H.-J. Media selection and other applications of fuzzy
linear programming. In Fuzzy Information Engineering: A Guided Tour
of Applications, D. Dubois, H. Prade, and R. R. Yager, Eds. John Wiley
& Sons, 1996.
[44] Z IMMERMANN , H.-J., AND S EBASTIAN , H.-J. Optimization and fuzzi-
ness in problems of design and configuration. In Proceedings of the Sec-
ond IEEE International Conference on Fuzzy Systems (San Francisco,
CA, June 1993), IEEE, Pergamon Press, pp. 1237–1240.
[45] Z IMMERMANN , H.-J., AND S EBASTIAN , H.-J. Fuzzy design - Integra-
tion of fuzzy theory with knowledge-based system-design. In Proceed-
ings of the Third IEEE International Conference on Fuzzy Systems (June
1994), vol. 1, IEEE, pp. 352–357.
[46] Z IMMERMANN , H.-J., AND S EBASTIAN , H.-J. Intelligent system de-
sign support by fuzzy-multi-criteria decision making and/or evolutionary
algorithms. In Proceedings of the Fourth IEEE International Confer-
ence on Fuzzy Systems (FUZZ-IEEE/IFES’95) (Mar. 1995), vol. 1, IEEE,
pp. 367–374.
 Preliminary Vehicle Structure Design Application 203

A Appendix: Stiffness Test Results

Torsion
Load (N) Moment (N-m) Deflection (mm) Twist (deg)
0.00 0.00 0.00 0.00000
126.99 212.84 1.09 0.04407
275.78 455.03 2.41 0.09736
404.77 667.87 3.48 0.14041
551.55 910.06 4.47 0.18038
680.54 1122.90 5.72 0.23060
845.12 1394.45 6.99 0.28184
974.11 1607.28 8.08 0.32591

Fitting to y = mx + c:

m = 4960.74 N-m/deg
c = −10.16 N-m

Fitting to y = mx + 0:

m = 4917.04N-m/deg

moment (N-m) vs. twist (deg)

1800.00

1600.00

1400.00

1200.00
moment (N-m)

1000.00

800.00

600.00 data points

fitted line
400.00

200.00

0.00
0.00000 0.05000 0.10000 0.15000 0.20000 0.25000 0.30000 0.35000
twist (deg)

Figure A.1 Load Test, Torsional Stiffness

204 IMPRECISION IN ENGINEERING DESIGN

Bending
Load (N) Deflection (mm)
0.00 0.00000
284.67 0.101060
551.55 0.17780
836.22 0.33020
2001.60 0.78740
2286.27 0.91440
2837.82 1.16840

Fitting to y = mx + c:

m = 2426.16N/mm
c = 51.79N

Fitting to y = mx + 0:
m = 2484.80N/mm

load (N) vs. deflection (mm)

3500.00

3000.00

2500.00

2000.00
load (N)

1500.00
data points
1000.00 fitted line

500.00

0.00
0.00000 0.20000 0.40000 0.60000 0.80000 1.00000 1.20000 1.40000
deflection (mm)

Figure A.2 Load Test, Torsional Stiffness