discussions, stats, and author profiles for this publication at: http://www.researchgate.net/publication/234154482
CITATION
DOWNLOADS
VIEWS
86
146
4 AUTHORS, INCLUDING:
Samuel Gomes
Varret Antoine
3 PUBLICATIONS 4 CITATIONS
SEE PROFILE
SEE PROFILE
Jean-Claude Sagot
Universit de Technologie de Belfort-Mont
248 PUBLICATIONS 355 CITATIONS
SEE PROFILE
60
61
Introduction
The engineering design of complex systems (Rechting, 1991) with a systematic approach
(Pahl and Beitz, 1995) is a decision-making process with the purpose of choosing from
among a set of options that leads to an irrevocable allocation of resources. It is inherently
a multiobjective process. As products become increasingly complex, their design is
usually on a large scale, typically with a significant number of design variables,
parameters, requirements, constraints and objectives. Consequently, multiobjective
optimisation is being used more often to provide one optimal solution.
The main trend, particularly in industrial companies, is to propose complex products
whose design spans several engineering contexts and disciplines. At the same time,
although companies have grown in complexity, they have also reduced the number of
areas of competence, in order to be specialised in one (or several) discipline(s); thus,
the use of subcontractors is now very common. In addition, besides the traditional
cost considerations, more recent industrial requirements, such as robustness, reliability
and design performance and also marketing criteria, have been identified and have
quickly become important characteristics of the design and of the optimisation process.
Nowadays, actual real-world engineering design problems involve simultaneous
optimisation to meet several objectives and to ensure compliance with various constraints
determined by the design team.
The main purpose of this work is to develop a design methodology and a direct
multiobjective optimisation approach, integrating functional design and knowledge-based
engineering features, such as expert rule definition or design experience feedback, in
order to reduce costs, lead time and also improve product qualities and values. This
methodology helps the designer to take parametric Computer-aided Design (CAD)
models of an optimised product through functional requirements, design rules and
design objectives that can be verified and reached using optimisation loops. This
methodology is applied in a collaborative design process (Kvan, 2000) using ACSP
(in French: Atelier Coopratif de Suivi de Projets) (Gomes and Sagot, 2002), a
self-developed knowledge-based Product Lifecycle Management (PLM) environment
(Shen, 2003; Gomes et al., 2005) based on internet technologies (Liu and Xu, 2001;
Zhuang et al., 2000).
62
As a first step, using the requirements specification and the technical characteristics
of the product, integrated by the project members into our ACSP PLM system (functional
and product architecture parameters), we are able to generate and automatically produce a
parametric product architecture and its geometric skeleton, in a commercial CAD
software program. CAD designers can then complete this 3D skeleton by using generic
templates of parts stored in a shared database (Gomes et al., 2006).
In the second step, a knowledge management approach (Grundstein, 2000) including
a knowledge-based engineering software program, using constraint propagation with a
first-order inference engine, is used as an interface between the PLM system and the
CAD software (Peltonen, 2000).
The last step is to develop an application for the optimisation of the product by using
specific algorithms to minimise weight or number of parts or to achieve other objective
functions with respect to various functional parameters, geometrical variables and design
rules. As the objective functions are considered nonlinear functions, depending on the
functional parameters, it is necessary to use a heuristic search method, such as Genetic
Algorithms, in order to find the best solution in the design context.
Our methodology of coupling the Genetic Algorithm optimisation approach,
constraint propagation, with the inference engine and the PLM environment will
be presented. To validate our research hypotheses, an experimental case study is
chosen: the ground-link suspension system of a racing car design and manufacturing
project, including conceptual, embodiment and detailed design phases as well as
manufacturing phases.
Before describing our functional, knowledge-based engineering (Sferro, 1999;
Whitney et al., 1999) and optimisation design methodology, we will first present the
main principles of multidisciplinary and multiobjective optimisation.
Companies need to optimise their products and processes daily, hence optimisation plays
a significant role in todays design cycle. Optimising a system means selecting the best
available option from a wide range of possible choices. This specific approach can be a
complex task as, potentially, a huge number of options should be tested. There are several
sources of complexity, such as the computational difficulties in modelling physics
problems, the potentially high number of free variables, or a high number of objectives
and constraints/constraining factors.
A single optimisation approach is not sufficient to deal with real-life problems.
Therefore, engineers are frequently asked to solve problems with several conflicting
objective functions.
Multidisciplinary optimisation consists in finding the optimal design of complex
engineering systems, which requires analyses that take into account interactions amongst
the disciplines (or parts of the system) and which seek to synergistically exploit
these interactions.
In order to help engineers and decision-makers, old and new optimisation techniques
are studied and widely used in industry. Each optimisation technique is qualified by its
search strategy, which determines the robustness, the reliability and/or the accuracy of the
63
method. Robustness means that the objective functions are met even when starting far
from the final solution. On the other hand, accuracy measures the capability of the
optimisation algorithm to get as close as possible to the functional desired limits.
There are hundreds or thousands of optimisation methods in scientific literature; each
numerical method can solve a specific or more generic problem. Some methods are more
appropriate for constrained optimisation, others for unconstrained continuous problems,
or for solving discrete problems.
Many classical optimisation methods exist; these methods can be used provided that
certain mathematical conditions are satisfied. Thus, for example, linear programming
efficiently solves problems where both the objective and the constraints are linear with
respect to all the decision variables. Other specific numerical methods can be useful for
solving quadratic programming, nonlinear problems, nonlinear least squares, nonlinear
equations and multiobjective optimisation.
Unfortunately, real-world applications often include one or more difficulties which
make these methods inapplicable. Most of the time, objective functions are highly
nonlinear or may not have an analytic expression in terms of the parameters.
From a mathematical point of view, a multiobjective optimisation problem can be
written as follows:
min [f1(X), f2(X), ... , fi(X)],
where:
X = (x1, ..., xn+u) is the vector of the variable generally integrating the n customers
requirements and u technical design parameters of the product. These variables are
considered as input data of the optimisation problem. These input parameters are the
quantities that the designer can vary. It is by modifying these values that the search
for an optimum is performed. The variables can be continuous or discrete. The
problem may even contain a mixture of continuous and discrete variables.
f1,...,fi are the objective functions, the response parameters. When i > 1 and the
different and conflicting functional requirements are observed, we speak about
multiobjective optimisation. These are the quantities that the designer wishes to
maximise or minimise. For example, the designer can maximise the efficiency and
the performance, or can minimise the cost and the weight.
Gi(x1, ..., xn+u) 0, and G1(x1, ..., xn+u) = 0 are the constraints. Equality and
inequality constraints directly connected to the requirements are imposed on the
project. These constraints correspond to the limits that the designer must meet due to
the need to comply with standards, or due to the particular characteristics of the
environment, the functionalities, the physical limitations, etc. These constraints must
be satisfied in order to be able to consider a certain solution as acceptable. All the
constraints define a feasible region in a multidimensional solution space. For
example, designers can impose some general constraints, such as the maximum
admissible stress, the maximum deformation, or the minimum performance. The
designer can even impose some special constraints on the variables such as the total
volume, the thickness range and so on.
With a multiobjective problem, the notion of what is optimum changes as the aim is to
find good compromises rather than a single solution. So, a multiobjective optimisation
does not produce a unique solution but a set of solutions. These solutions are called
64
Pareto solutions. The set of solutions can be called trade-off surface or Pareto frontier.
In the Pareto frontier, none of the components can be improved without the deterioration
of at least one of the other components.
The design activity involves many contributors and experts throughout the product
lifecycle, which starts with project tasks, proceeds to functional definition, product
modelling, manufacturing and ends with its destruction or recycling. Whether the design
solution is a tangible product, service, software program, process, or something else,
designers typically follow these main steps, applying a concurrent (Sohlenius, 1992) and
distributed (Brissaud and Garro, 1996) engineering process:
define the problem they must solve, and also the functions and subfunctions they
have to develop in order to satisfy these needs (Functions vector Fu = [fu1,, fum])
create and select a product solution including one or several parts (Parts vector
P = [p1,, pk])
analyse and optimise the proposed solution by finding optimal sets of technical
characteristics or parameters for each part, subproduct or product assembly
(Technical parameters vector T = [t1,, tu]), while maximising or minimising
objective functions and verifying design constraints and rules (Rules vector
Ru = [F1(X),, Fi(X)]). In this context, X is defined by the concatenation of
both R (R = [r1,, rn]) and T (T = [t1,, tu]) vectors: X = [x1, ..., xn+u]
= [r1,, rn, t1,, tu]
integrate the customers requirements during the entire design process, applying the
House of Quality principles (Mocquo, 2007)
store their design information during the design process in a PLM system, using the
Multi-Domains and Multi-Viewpoints data model, based upon the Axiomatic design
(Suh, 1998) and the Multi-Viewpoints Product Model (Tichkiewitch, 1996)
65
can combine the design parameters and make each of them evolve within the
limits of the customer requirements while following all expert rules, using an
inference engine in a knowledge-based engineering system (Whitney et al., 1999;
Serrafero, 1998)
make the best possible design decisions due to multidisciplinary and multiobjective
optimisation loops, involving metaheuristic algorithms such as genetic algorithms,
for minimising and/or maximising objective functions.
All designers can generate Project, Product, Process and Usability information in our
PLM system, by integrating knowledge engineering features. At the same time, a
continuous and automatic process of knowledge capitalisation, using the KATRAS
multiagent society, is carried out (Monticolo et al., 2006).
The product architect creates a parametric product architecture and skeleton in the
PLM environment, and the same features can be automatically generated in the CAD
software, using a Visual Basic script file, exported directly from the PLM system. At
the same time, he/she generates a table of parameters (X = [x1, ..., xn+u]) with various
acceptable domains for continuous variables (for example : x1 [8,10.5]) or
discrete variables (x2 {steel, aluminium, carbon}).
Then, the CAD designers can begin to build solid features directly on the previous
parametric product architecture and skeleton, with a limited access control on
the parameters.
66
Finally, product architects and expert designers can choose various vectors of
optimal solutions (input/output parameters Xi = [x1,, xn+u]i) situated on the
Pareto frontier of the optimisation problem. After a last checking loop in the
knowledge-based engineering platform, these optimal solutions, defining various
optimal design alternatives, are then archived in the CAD model table of parameters,
in order to verify the compatibility with the other nonalgebraic expert rules.
CAD designers, expert designers and product architects can finally visualise the
optimal design alternatives by opening the CAD model synchronised with the table
of parameters.
Figure 1
Designers
PLM environment
2. Generation of a parametric product architecture and skeleton (*.catvbs script file)
PRODUCT DOMAIN
Requirements (R)
Functions (Fu)
Parts (P)
Technical
characteristics (T)
(T),
KBE application
with an inference
engine for
constraints
propagation
PROCESS DOMAIN
Plants,
Assembly facilities
Manufacturing
means ,
EXPERT
PROCESS
EXPERT
EXPERIENCE
7. Design choices
definition
Dynamic
table of
parameters
PROJECT DOMAIN
Tasks
Resources
Milestones
costs,
Multi-Disciplinary
Multidisciplinary
optimisation
optimization
processor
7. Vectors of optimal
input/output
parameters
(Xi = [x 1,, xn+u ]i on
Pareto Frontier )
8. CAD model
synchronisation
Parametric
CAD modeler
3. Model solid
features
EXPERT
RULES (Ru)
EXPERT
VOCABULARY
4. Validate expert
knowledge
KM environment
Expert
Designers
CAD
Designer
67
reduction of the possible fields of solutions, starting from a numerical interval (for
example: x1[8,10.5]) or a list of discrete values (x2{steel, aluminium, carbon}).
In order to illustrate the constraint propagation in the KADVISER inference engine, the
following example is given, using numerical fields:
x1, x3, x4 are real: x1 [8.0, 10.5], x3 [22.0, 41.3] and x4 is unknown
then
x4 [11.0; 30.3]
then
x4 [11.0; 16.5]
then
x4 [12.0; 16.5].
68
In order to illustrate our proposals, an experimental design case is chosen. Every year, the
Mechanical Engineering and Design Department at Belfort-Montbeliard University of
Technology has to develop and prototype an entire new racing vehicle. To simplify the
demonstration, we choose to limit the experimental case study to a subproduct of the
racing car: the ground-link suspension system. This subproduct of the racing car includes
many mechanical parts linking the wheel to the chassis. The design and optimisation
process will be focused on the suspension triangles (from the diagrams, this component
looks like the one called the wishbone in English, an A-frame component linking the
chassis to the wheel-hub) of the ground-link system. The main steps of our previously
presented methodology are applied in this experimental case. For the optimisation study,
the technical characteristics of external systems in interaction with the ground-link
suspension (chassis, hubs, brakes, wheels) have been considered as requirements (R), as
described in Table 2.
To complete the requirements, the mechanical characteristics of the ground-link
suspension are considered as technical parameters (T), the continuous or discrete
variables of our design and optimisation case study (Table 3).
As presented in Table 1 and Table 4, describing objective functions, constraints and
expert rules, the equations for the optimisation problem are defined in two steps. First, we
apply traditional mechanical analysis of nondeformable solids, allowing the definition of
the maximum deceleration acceptable to the vehicle and loads generated at the various
link points of the suspension triangles (wishbones). Then we define a standard material
strength calculation in order to identify the stresses and deformations corresponding to
these loads.
The first step allows the highest load applied to the lower triangle at point A to
be identified (the respective values of the load vectors in X and Y directions: 1792 N;
1482 N, in a constant deceleration phase, value = 10 m/s corresponding to an
emergency braking condition). However, this calculation identifies the front tie-rod of the
lower triangle as the most highly stressed component.
69
{wheel/hub/brake}
set geometry
0
25.5
A 86
0
108
B 96
0
0
124
46.5
C 22
RS 262.5
1600 mm
m = 15 kg
M = 460 kg
front: Mv = 160 kg
h = 0,35 m
e = 2,495 m
Suspension inclination
angle in relation to
the ground
am = 45
0,85
Note:
rear: Mr = 300 kg
Lower triangles included in horizontal planes. For each triangle, bisectrix of the
angle formed by the two tie-rods and the vehicle axis are perpendicular.
Table 2
Continuous variables
Angle between the tie rods
[20; 120]
Thickness
e [1 mm; 3 mm]
Tie-rod length
L = 300 mm
L
E
70
Table 2
Discrete variables
Section type
square shape: B = H
or
round shape: H = 2R
Material
Table 3
item
material
Re (MPa)
steel
700
7850
210
235
aluminium
4000
2900
75
180
stainless
4500
8700
203
185
carbon
5000
1530
50
555
1.5
Objective functions and constraints defined for our design and optimisation case study
Objective functions (Fi(X))
m = .S.L
max = tc + f
f =
f + dl
Constraints (Gj)
Maximum stress allowed in normal section
Notes:
Table 4
max
Re
n
Tix: 1792 N
Tiy: 1482 N
1
1
Dn = Tix.cos ( 2 ) Tiy.sin( 2)
2
2
1
1
Dt = sin ( 2 ) +
Tix + Tiy.cos( 2)
2sin( 2)
2
2
f =
Dn.L3
3EI
71
Expert rules defined for our design and optimisation case study
Rules (Ru)
f =
Dn.L.H
2I
round section: I =
square section: I =
dl =
64
( H 4 ( H 2e) 4 )
1
( H 4 ( H 2e) 4 )
12
Dt .L
E .S
tc =
Dt
S
round section: S =
( H 2 ( H 2e)2 )
During the second step, we determine the normal and tangential load components in the
front tie-rod and the resulting strain and stress.
As defined previously, during the project, these requirements, technical parameters,
objective functions, constraints and expert rules are integrated by the expert designers
into the ACSP PLM environment, directly associated to the product part list (Figure 2).
The data are then extracted and structured in order to create:
Figure 2
Example of an ACSP interface describing parameters and rules associated with the
product part list (see online version for colours)
72
The KADVISER software, linked with the ACSP PLM environment, enables the expert
designer to quickly generate specific knowledge-based engineering applications, as
shown in Figure 3.
Figure 3
73
Step 2
Step 3
elimination of the vectors outside the Pareto frontier Figure 4. On the left-hand
diagram, these vectors are all points on the right of the curves (we can identify
four curves, which represent the Pareto frontiers for the four materials)
Step 4
Figure 4
Design ID
section
material
stress
Re/n
strain
mass
cost
142
1.0
30.0
120
34.6
370.0
0.36
0.042
0.209
214
1.4
34.0
120
21.5
370.0
0.19
0.066
0.329
252
1.0
16.0
120
58.7
58.8
0.34
0.141
0.099
298
1.0
28.0
120
27.9
45.0
0.18
0.094
0.376
318
1.0
24.0
120
46.5
58.8
0.16
0.170
0.119
367
1.2
40.0
120
15.2
45.0
0.07
0.162
0.648
1351
2.6
34.0
120
12.2
370.0
0.11
0.118
0.589
Notes:
The last step is to visualise the results obtained directly, by opening the CAD model and
selecting the desired optimal design. As explained previously, the MS Excel spreadsheet
containing the optimal Xi vectors is also declared as a parameter table for the parametric
CAD model.
Figure 5 presents three different optimal solutions computed from the same
parametric CAD model, when selecting the optimal design number 142, 298 and 318
from the parameter table resulting from the application of our methodology in a racing
car experimental design project.
74
Figure 5
Conclusion
ensuring that the product design remains within the limits set out by the customers
requirements, in so far as a true functional analysis and design is applied
ensuring that the design developed is verified using all the expert rules, which must
be frequently updated and validated by various expert designers
archiving the data and information from the project, and also accumulating
knowledge (expert rules, expert vocabulary, expert experiences, etc.)
defining several design parameter sets corresponding to optimal solutions and good
compromises between different contradictory objectives. The multiobjective genetic
algorithm allows us to eliminate classical analysis-synthesis-evaluation loops of a
traditional design process (Yannou et al., 2003), and to go farther than a just
acceptable solution. This approach is performed without reconsidering the
performance of the optimisation algorithm used
75
Our methodology shows a few lacks that must be addressed in the future. First, the
application field of this methodology is limited to routine design. Product architecture
and technical principles must be determined before the beginning of the optimisation
loop, as an input of the illustrated process.
In our example, we use explicit mathematical functions, making the application of
optimisation algorithms faster and easier, considering the very simple shapes used for the
product (simple solids with constant section tubes) and also well-known mechanical
analysis models. Future work will consist in considering more complex shapes and solid
features and integrating data coming from other fields. These data can be measurements
resulting from numerical simulations and analyses performed in Computer Aided
Engineering (stress, strain, Young Modulus, etc.) or CAD (volume, centre of gravity,
material density, etc.) systems.
Another issue we intend to address in further work consists in reducing the number of
steps of our methodology, which can introduce a lack of traceability of the design choices
during the feedback loops.
During the whole design process, the traceability between the different choices made
about the product will be implemented through a matrix-based modelling and system
analysis method, describing product design information: requirements (Rn vector of n
requirements), functions (Fum vector of m functions), parts (Pk vector of k parts) and
technical parameters (Tu vector of u technical parameters). Our traceability method will
use several matrices:
76
Acknowledgements
The authors would like to thank the Metropolitan Community of the region of
Montbliard, the Franche-Comt Region Council, OSEO Innovation, the French
Ministry of Industry and the Automotive of the Future cluster, for their funding of this
research activity.
References
Bluntzer, J-B., Gomes, S. and Sagot, J.C. (2006) Application de la modlisation CAO base de
connaissances la conception fonctionnelle, collaborative et multi sites de produits
modulaires, 13me Sminaire CONFERE (Collge d'Etudes et de Recherches en Design et
Conception de Produits) sur l'Innovation et la Conception, Marrakech, 67 July.
Brissaud, D. and Garro, O. (1996) An approach to concurrent engineering using distributed
design methodology, Concurrent Engineering: Research and Applications, Vol. 4, No. 3,
pp.303311.
Browning, T.R. (2001) Applying the design structure matrix to system decomposition and
integration problems: a review and new directions, IEEE Transactions on Engineering
Management, Vol. 48, No. 3.
Eggers, J., Feillet, D., Kehl, S., Wagner, M.O. and Yannou, B. (2002) Optimization of the
keyboard arrangement problem using an ant colony algorithm, European Journal of
Operational Research, Elsevier.
Gomes, S., Bluntzer, J.B. and Sagot, J-C. (2006) Functional design through a PLM system for
fastening routine definition of CAD models, PLEDM 2006, Playa Del Carmen, Mexico,
26 November1 December.
Gomes, S. and Sagot, J.C. (2002) A concurrent engineering experience based on a cooperative and
object oriented design methodology, Integrated Design and Manufacturing in Mechanical
Engineering, Dordrecht: Kluwer Academic Publishers.
Gomes, S., Serrafero, P., Monticolo, D. and Eynard, B. (2005) Extracting knowledge from PLM
systems, an experimental approach, International Conference on Product Lifecycle
Management PLM05, Lyon, 1113 July.
Grundstein, M. (2000) From capitalizing on company knowledge to knowledge management,
in D. Morey, M. Maybury and B. Thuraisingham (Eds.) Knowledge Management, Classic and
Contemporary Works, Cambridge, MA: MIT Press, pp.261287.
Jin, Y. and Lu, S.C.Y. (1998) An agent-supported approach to collaborative design, Annals of the
CIRP, Vol. 47, No. 1, pp.107110.
Kvan, T. (2000) Collaborative design: what is it?, Automation in Construction, Vol. 9, No. 4,
pp.409415.
Liu, D.T. and Xu, X.W. (2001) A review of web-based product data management systems,
Computers in Industry, Vol. 44, No. 3, pp.251262.
Mocquo, A. (2007) Modelling scheme for capturing and analysing multi-domain design
information a hair dryer design example, International Conference on Engineering Design,
ICED07, Paris, ISBN: 1-904670-01-6.
Monticolo, D., Gomes, S., Hilaire, V. and Serrafero, P. (2006) A multi-agent architecture to
synthesize industrial knowledge from a PLM system, PLEDM 2006, Playa Del Carmen,
Mexico, 26 November1 December.
Pahl, G. and Beitz, W. (1995) Engineering Design, A Systematic Approach, in K. Wallace (Ed.),
K. Wallace, L. Blessing and F. Bauert (Trans.), 2nd ed., New York: Springer-Verlag.
77
Peltonen, H. (2000) Concepts and an implementation for product data management, Acta
Polytechnica Scandinavica, Mathematics and Computing Series No. 105, Espoo, The Finnish
Academies of Technology, ISBN: 951-666-538-1, ISSN: 1456-9418.
Rechting, E. (1991) Systems Architecting, New York, NY: Prentice Hall.
Serrafero, P. (1998) From CAD/CAM to KAD/KAM for the design of congruent products of the
extended enterprise, UNIMEP Seminar, 3e Seminario Internacional de Alta Tecnologia,
San Paolo, pp.90104.
Serrafero, P. (2002) Vers la mesure de la quantit de connaissance et de comptence industrielle:
le modle Knova, confrence invite, 1er Sminaire International en Gestion des
Connaissances et Comptences, Nantes.
Sferro, P.R. (1999) Direct Engineering: Toward Intelligent Manufacturing, in K. Kamrani (Ed.),
Boston, MA: Kluwer Academic Publishers.
Shen, W. (2003) Knowledge sharing in collaborative design environment, Computers in Industry,
Vol. 52, No. 1, pp.13.
Sohlenius, G. (1992) Concurrent engineering, Annals of the CIRP, Vol. 41, No. 2, pp.645655.
Suh, N.P. (1990) The Principles of Design, New York, NY: Oxford University Press.
Suh, N.P. (1998) Axiomatic design theory for systems, Research in Engineering Design, Vol. 10,
pp.189209.
Tichkiewitch, S. (1996) Specifications on integrated design methodology using a multi-view
product model, System Design and Analysis Conference, Montpellier, France, pp.101108.
Whitney, D.E., Dong, Q., Judson, J. and Mascoli, C. (1999) Introducing knowledge-based
engineering into an interconnected product development process, Proceedings of the 1999
ASME Design Engineering Technical Conference, Las Vegas, Nevada, September,
DET99-DTM8741.
Yannou, B., Simpson, T.W. and Barton, R.R. (2003) Towards a Conceptual Design Explorer Using
Metamodeling Approaches and Constraint Programming, DETC2003/DAC48766.
Zhuang, Y., Chen, L. and Venter, R. (2000) CyberEye: an internet-enabled environment to support
collaborative design, Concurrent Engineering Research and Applications, Vol. 8, No. 3,
pp.213229.