You are on page 1of 19

CIRP Annals - Manufacturing Technology 57 (2008) 697715

Contents lists available at ScienceDirect

CIRP Annals - Manufacturing Technology


journal homepage: http://ees.elsevier.com/cirp/default.asp

Recent advances in engineering design optimisation: Challenges and future trends


Rajkumar Roy (2)a,*, Srichand Hinduja (1)b, Roberto Teti (1)c
a
b
c

Decision Engineering Centre, Department of Manufacturing, Craneld University, Craneld, Bedford MK43 0AL, UK
School of Mechanical, Aerospace and Civil Engineering, University of Manchester, Manchester M60 1QD, UK
Department of Materials and Production Engineering, University of Naples Federico II, 80125 Naples, Italy

A R T I C L E I N F O

A B S T R A C T

Keywords:
Design
Optimisation
Algorithm

Traditional engineering design optimisation which is the process of identifying the right combination of
product parameters is often done manually, time consuming and involves a step by step approach. This
paper identies recent approaches to automating the manual optimisation process and the challenges
that it presents to the engineering community. Engineering design optimisation is classied based on
design evaluation effort and degrees of freedom viewpoints. An overview of different approaches for
design optimisation is presented. The study identies scalability as the major challenge for design
optimisation techniques. Large-scale optimisation requires signicant computing power and efcient
algorithms such as swarm intelligence.
2008 CIRP.

1. Introduction
Every time a product is created or designed to satisfy human
needs, the creator tries to achieve the best solution for the task in
hand and therefore performs optimisation. Engineering design
optimisation (EDO) is not new. This process is often manual, time
consuming and involves a step by step approach to identify the
right combination of the product and associated process parameters for the best solution. Often the manual approach does not
allow a thorough exploration of the solution space to nd the
optimum design, resulting in sub-optimal designs. With increasing
global competition, it is necessary to design products that are able
to satisfy human needs in the most effective manner. This keynote
paper identies recent approaches to automating the manual
optimisation process and the challenges this brings for the
engineering community.
In real life, identication of the optimum design of an industrial
problem is often not possible because of the size of the problem
and lack of knowledge. In this situation, design optimisation is
essentially seen as design improvement. This paper uses design
optimisation as the goal for any design improvement task.
The terms optimise, optimisation, optimal, optimum are
often used in a very loose sense without necessarily referring to the
use of specic optimisation techniques. In this paper, only
optimisation as relating to mechanical design problems or discrete
products is considered. It obviously excludes optimisation
problems in thermouid processes, manufacturing processes as
well as process manufacturing areas [1,2]. However, the optimisation of shapes such as turbines or tools in interaction with the
thermouids or manufacturing processes are considered as well as

* Corresponding author.
E-mail address: r.roy@craneld.ac.uk (R. Roy).
0007-8506/$ see front matter 2008 CIRP.
doi:10.1016/j.cirp.2008.09.007

the constraints arising from these interfaces such as manufacturability. It should be noted that optimisation of the organisation/
management of the design process [35] is not included in the
paper. It is recognised that the design is dependent on the
manufacturing processes to be used and the whole life cycle issues.
This paper considers the manufacturability and life cycle issues as
constraints or one of the objectives for the design optimisation.
Multi-disciplinary [6,7] and civil engineering structural optimisations [8] are not covered in this keynote paper in order to limit the
scope.
1.1. Basics
Performing EDO often requires knowledge about the stage of
design, design variables and their minimum and maximum limits
(independent variables), constraints, measurement of the design
performance (dependent variables), design parameters and
relationships between the independent and dependent variables
(i.e. a design evaluation model). The design variables are
dependent on the level of product denition available at the
different stages of the design optimisation. The design stages can
vary from conceptual or preliminary design, and from conguration to detailed design.
Design variables are expressed in either quantitative or
qualitative terms. In many design optimisation problems, it is
easy to measure the design variables, such as length, weight and
temperature. They are referred to as quantitative design variables.
But in real life design problems, variables such as aesthetics and
manufacturability are difcult to measure and they are referred to
as qualitative variables [9]. The nature of the model development
will vary depending on the types of the variables. In order to reduce
the possible choices for a design, each variable is given a minimum
and maximum value, called the bounds of the variable. Some
design parameters remain constant for a given design problem.

698

R. Roy et al. / CIRP Annals - Manufacturing Technology 57 (2008) 697715

A constraint on a design, on the other hand, denes physical


or functional limits of a design. The constraint is dened as a
function of the design variables and can be of two types: nonequality and equality type. For example, the temperature on a
turbine blade has to be less than the melting point of the
material, i.e. T(x) < Tmax or the cost of a design must be less than
or equal to a budgeted amount, i.e. C(x)  Cmax, where x is the
vector of design variables: (x1, x2, . . ., xi)T. In case of mechanical
design a dimensional constraint [10] introduces tolerance on the
dimension and becomes an in-equality constraint as
Lmax  tolerance  L(x)  Lmax + tolerance, where L stands for
the length of a component. On the other hand, equality
constraints are those that must be satised, examples include
physics-based constraints, such as F = ma. Satisfying constraints
is a challenge for any optimisation task, especially if it is an
equality constraint [11]. It is advisable to convert an equality
constraint to a suitable in-equality constraint where possible.
Development of the design evaluation model (or objective
function or tness function) is dependent on the nature of the
variables and constraints. The model assumes that certain aspects
of the design remain constant, and they are called design
parameters. The design parameters are often used to reduce the
total number of design choices (or design space). The model can
also be either quantitative or qualitative in nature. Quantitative
models can be either simulation based, analytical or empirical,
whereas the qualitative models are generally knowledge-based.
Considering the above three aspects, the optimisation problem can
be mathematically represented as
optimiseminimize or maximizeF j x; p;
min

j 1; 2; . . . ; J

(1)

max

 xi  xi
; where i = 1, . . ., n, p is a
with variable bounds as xi
vector of design parameters, and subject to constraints:
g k x  0 and hm x 0

(2)

where k = 1, 2, . . ., K and m = K + 1, . . ., M.
Fj(x, p) are the objective functions for the design problem.
Minimisation of a function F(x, p) is equivalent to maximisation of
F(x, p). It should be noted that the multiple objective functions
represent different measures of performance for the same design,
such as weight, cost and quality [12].
This paper starts with classication of EDO problems. Then the
major mathematical challenges faced in design optimisation are
highlighted with a special focus on mechanical and real life design
problems. EDO approaches are then grouped and mapped against
the types of optimisation problems in Section 4. Section 5 presents
an analysis of trends in different optimisation techniques in the
last 10 years. This section identies the most popular and
promising techniques for design optimisation based on the
literature. Section 6 is about current practice in industry. The
next three sections discuss in detail the three major approaches
and present a critical analysis of the research with examples.
Sections 10 and 11 discuss the future challenges and potential
techniques to address large-scale design optimisation.
2. Classication of engineering design optimisation problems
A classication of the EDO problem is necessary to select the
right approach for a given problem. An enhanced version of the
classication proposed in an earlier publication [13] for EDO
problems is presented in Table 1. The classication is developed
based on ve basic schemes and two view points. The basic
schemes are: design variables, constraints, objective functions,
problem domains and the environment for the design. The two
viewpoints are design evaluation effort and the degrees of freedom
of the design problem. The ve major classication schemes and
their categories as discussed below:
 Design variables play a major role in EDO. The number of design
variables, their natures, permissible values and mutual depen-

dencies can affect the overall complexity of the optimisation


task. Most of real life or industrial design optimisation problems
are likely to be multi-dimensional [14]. The complexity is
dened as the level of effort required to formulate the
optimisation problem and identify the optimum solution(s). In
static or parameter optimisation problems, the design variables
are independent of each other whereas in trajectory or dynamic
optimisation problems, the design variables are all continuous
functions of some other variable(s). Another perspective of this
classication is provided by other researchers [15], based on
time-dependence of optimisation problems. Depending on the
values permitted for design variables, EDO problems can be
categorised as integer-valued, real-valued and mixed-integer
(that involve both integer and real variables). Variable dependence occurs when the variables are functions of each other. It is
often observed that there are variable dependencies among real
life design problems. This has an effect of constraining the search
space [13].
 Existence of constraints in an EDO problem affects the optimisation approach to be used. The constraints can be linear or
nonlinear in nature. A mixed-integer programming (MIP)
problem is one where some of the design variables are
constrained to have only integer values at the optimal solution.
Constraint programming denes higher-level constraints that
apply to integer variables. The most common and useful higherlevel constraint is the all-different constraint, which applies to a
set of variables, say x1, x2, x3, x4 and x5. This constraint assumes
that the variables can have only a nite number of possible
values (say 1 through 7), and species that the variables must be
all different at the optimal solution. One of the possible values for
the optimimum could be (7, 6, 4, 3, 2). The travelling salesman
problem is an example of constraint programming. The last two
types of constraints make optimisation much more difcult. The
number of constraints, constraint development and evaluation
time are factors that signicantly affect the optimisation process
[16,17].
 Objective functions are used to evaluate a design solution within
the optimisation context. The number of objective functions,
their nature and whether they are separable determines the
complexity of the optimisation task. In real life most of the
optimisation problems are multi-objective. Multi-objective
optimisation becomes more complex with more than about 10
objectives for a problem [18]. Quantitative objective functions
can be further classied as simulation based (e.g. FEA, CFD)
[19], analytical (e.g. mathematical models created from rst
principles and with domain knowledge) [20] and empirical
[21]. Qualitative objective functions involve issues like
manufacturability and aesthetics [14]. One of the major
challenges in EDO is to deal with computationally expensive
objective functions. Typically, simulation-based models take a
long time to evaluate. The nature of a search space also
classies the EDO problems as uni-modal or multi-modal
based on the number of optimal solutions that the problem has.
Multi-modal problems can also be categorised as sensitive and
robust. The former has mostly very sensitive optima, whereas
the latter has at least one robust optimum. The nature of the
search space can also be classied as linear and nonlinear
based on the nature of underlying equations in the objective
function.
Based on this criterion, EDO problems can also be classied as
continuous, discontinuous depending on whether the equations
involved in the problem have any discontinuities and not-dened
outside the feasible space when the objective function cannot be
evaluated if any constraint is violated. A function is said to be
separable if it can be decomposed into functions that involve
groups of variables rather than just a single variable. Inseparability
manifests itself as cross-product terms, and makes the effect of a
variable on the function dependent on the values of other variables
in the function.

R. Roy et al. / CIRP Annals - Manufacturing Technology 57 (2008) 697715

699

Table 1
Classication of EDO problems.

 Problem domain brings different physics consideration within the


optimisation. Multiple domains require a multi-disciplinary
approach to the optimisation [6,22]. Establishing interdependence between the domains for real life design problems and
optimising them simultaneously, such as aircraft design, requires
signicant effort and makes the optimisation more complex than
single domain optimisation.
 The optimisation environment involves considerations like
uncertainties in the design, level of knowledge available about
the design solutions, importance of designer involvement and

nally the nature of the environment. Lately there has been


signicant interest in design optimisation with uncertainties
[23]. The uncertainties can be associated with the design variable
denition as well as in the model development [24]. Knowledge
about the design environment is often lacking for real life
problems [14]. Not knowing about the design space and location
of the optimum makes the optimisation task more challenging.
Some design tasks require designer involvement to improve their
condence and also to involve them in qualitative design
evaluation; this is called interactive optimisation [25]. The

700

R. Roy et al. / CIRP Annals - Manufacturing Technology 57 (2008) 697715

involvement increases the degrees of freedom of the optimisation due to non-uniform behaviour of human experts and also
involves signicant effort from the expert designer. And nally
the nature of the environment could be static or dynamic. The
dynamic nature of the environment will impact the design
variables as well as the design evaluation. If it is dynamic, the
optimisation will require more effort and involve more degrees
of freedom than a static environment.
The design evaluation effort viewpoint is related to the
computational effort required to develop and evaluate a design
model due to the nature of the design variables, constraints,
objective functions, problem domains and the environment.
Whereas, degrees of freedom of an EDO problem includes the
number and number of types of design variables, constraints,
objective functions, problem domains and environmental factors
like uncertainty, designer condence required and dynamic
behaviour involved in a design.
Table 1 presents two categories of design evaluation effort:
inexpensive and expensive, and another two categories for the
degrees of freedom: small and large. Based on these two view
points, an EDO problem can be classied as a small-scale, expert
dependent, algorithm dependent and large-scale problem (Fig. 1).
Examples of a small-scale problem are connecting plate design
[26], turbine blade cooling system design [20] and automotive
magnetorheological brake design [27]. Examples of expert
dependent design optimisation includes fast axis feed drive design
[28], microend drill design [29], product and assembly design for a
bre reinforced plastic track wheel [30], parallel kinematic
mechanisms design for ve-axis milling operations [31], machine
tool spindle design [32], bearing design for ultrasonic machines
[33], power transmission system design [34], sintered product
shape optimisation [35], use of rapid prototyping for faster product
optimisation [36], product redesign using value oriented life cycle
costing [37], cutting tool design [3840] and mould design [41].
Examples of algorithm dependent design optimisation include
reliability-based cantilever beam and a simplied car crashworthiness based design [42] and multi-disciplinary aircraft concept
sizing problem with uncertainty [43]. In real life, sometimes we
want to optimise a system (rather than just a component) with
interaction. Technological constraints are used to re-evaluate the
limit boundaries and to formulate a large-scale design optimisation problem [an early example in 44]. There is a lack of research in
large-scale design optimisation area. Most of the component
design optimisations do not address design of an assembly or use
of computationally expensive but more realistic simulation-based
design evaluations (i.e. no further approximation using surrogate
models). The problem of size in design optimisation is also
discussed by other researchers [45]. Multi-objective and multidisciplinary design of blended wing body unmanned aerial vehicle
(UAV) presents a large-scale design optimisation application [46].
3. Mathematical challenges in design optimisation

complex. This section identies the major mathematical challenges that are relevant for EDO, especially for real life problems.
3.1. Global and robust optimisation
Mathematical optimisation aims to determine the globally best
solution for a problem for a given objective. In engineering design,
it is often not possible to even identify if a global optimum is
reached during the optimisation process. In the design context,
feasible solutions against multiple nonlinear constraints that are
signicantly better than the current solutions are often acceptable
considering the computational resources required. Global optimisation becomes very time consuming with a large number of
design variables (typically much more than 30), equality constraints and noisy objective functions (using simulation models).
EDO also applies local optimisation using gradient information and
Hessian matrices (second derivatives) in the local region. In cases
where simulation approaches, such as nite element analysis (FEA)
are employed, even local optimisation becomes difcult because of
lack of knowledge about the design space based on the models
[47]. The gradient-based approach is often not suitable for real life
design optimisation as the models in such real life problems are
often non-differentiable. Due to the presence of uncertainty, in real
life optimisation, it is often required to determine less sensitive
solutions as robust designs (Fig. 2). Robust solutions are areas in
the search space where signicant changes in design variables
produce only insignicant changes in the performance of the
design [24]. The challenge is to identify robust regions in the design
space.
3.2. Multi-modal and multi-objective optimisation
Some designs involve multiple good solutions, such as antenna
design [48] and turbine blade design [14].
This type of optimisation is called multi-modal optimisation
(Fig. 3). The challenge is to identify as many optima as possible to
provide a choice of good solutions to the designer. The task
becomes more difcult with an increase in the number of design
variables. Real life engineering designs often have more than one
conicting objective functions thus requiring a multi-objective
optimisation approach.
The optimisation identies several solutions that are good
considering the objective functions, they are called Pareto
solutions. Fig. 4 shows a Pareto front dening the solutions for a
two objective (F1 and F2) problem. The multi-objective optimisation becomes more difcult with increasing number of objectives
and it has been shown in [18] that existing multi-objective
optimisation algorithms do not perform well with more than ve
objectives.
3.3. Dealing with design variable interactions
Many real-life design optimisation problems also involve
interaction among decision variables. Ideally, the optimum design

Mixed nature of design variables, nonlinearity of the objective


functions and constraints make a design optimisation more

Fig. 1. Classication of EDO problems.

Fig. 2. An illustration of global and local optima and robust solutions.

R. Roy et al. / CIRP Annals - Manufacturing Technology 57 (2008) 697715

Fig. 3. An illustration of multi-modal design optimisation problem.

701

obtained, it is difcult to maintain them since any change in one


variable must be accompanied by related changes in others [12].
There is a lack of research in developing effective design
optimisation algorithms for inseparable interactions.
Generalised regression genetic algorithm (GRGA) is developed
to deal with mixed-integer (i.e. with integer and real variables)
multi-objective design optimisation problems with inseparable
function optimisation [13]. The GRGA works with the relationships
among design objective functions. It is observed that in case of a
multi-objective design optimisation, any continuous part of a
Pareto front will represent a relationship between the objectives.
Consider a two-objective optimisation problem having f1 and f2 as
the two objective functions. For any continuous portion of the
Pareto front, there exists a function F involving f1 and f2. Suppose
the problem has two decision variables x1 and x2 that dene the
functions f1 and f2, i.e., f1 and f2 can be expressed as f1(x1, x2) and
f2(x1, x2), leading to F1:
F f 1 ; f 2 0;
F f 1 x1 ; x2 ; f 2 x1 ; x2 0
) F 1 x1 ; x2 0:

(3)

The above equations show existence of relationship(s) among


the variables of the solutions belonging to any continuous portion
of the Pareto front. GRGA aims to explore this relationship using
nonlinear multi-variable regression analysis [49]. It uses the
relationship thus obtained to

Fig. 4. Pareto front identied using multi-objective optimisation.

could be obtained by varying the design variables of a given


problem in a random fashion independent of each other. This is not
possible for many applications, such as turbine blade design. In this
type of design, if the value of a given variable changes, the values of
others should be changed in a unique way to get the required
results. There are two types of interactions among the design
variables: inseparable function interaction and variable dependence [13]. In this section inseparable function interaction is
discussed as it is more relevant for EDO problems and it is difcult
to handle. Inseparable interaction occurs when the effect that a
variable has on the objective function(s) depends on the values of
other variables in the function (Fig. 5). For example, in Fig. 5(a) the
nature of the effect of A on y does not change due to variation in the
value of B, but in Fig. 5(b) it does (the slope of the line is changed).
The impact of the interaction becomes more signicant with
multiple objectives.
Traditional techniques such as weighted sum approach and goal
programming suffer from serious limitations in dealing with the
complexities introduced by inseparable function interaction [13].
The interaction also poses a signicant challenge to GA-based
optimisation approaches by making it more difcult for it to build
the building blocks used in a GA. Furthermore, in its presence, a
multi-objective optimisation problem cannot be decomposed into
simpler parts. Furthermore, even if a set of optimal solutions are

 perform periodic and nal re-distribution of solutions for aiding


their spread;
 use history of change of regression coefcients for guiding the
search towards the Pareto front;
 use rate of change of regression coefcients as the termination
condition of the algorithm.
3.4. Dealing with uncertainty
Real life design often involves uncertainties with the design
variables, constraints and objective functions. For example,
uncertainties in design could come from manufacturing variations.
There are several approaches to deal with uncertainty: robust
design [50], utility function optimisation [51,52] and reliabilitybased design optimisation [53]. (Eq. (1)) shows formulation of a
design optimisation problem without uncertainty. In practice, both
the design variables and the parameters can introduce uncertainty
in the design making the objective function Fj (x, p) uncertain. The
uncertain objective function can be represented using the mean
(m) and standard deviation (s) (robust design optimisation with
normal distribution):
optimise RmFx; p; s Fx; p

(4)

Or optimise a noisy objective function:


optimise Rx; p Fx; p e

(5)

where e  N(m, s), e.g. normal distribution.


In case of noise in x and F:
optimise Rx; p Fx e1 ; p e2

(6)

If the utility function optimisation approach is used, the


objective function can be represented as
Z
max U
uF pFdF . . .
(7)
To achieve the feasibility of constraints under uncertainty, a
general probabilistic feasibility formulation can be expressed as
follows:
Pg k x; p  0  P ok;k
Fig. 5. An example of inseparable function interaction. (a) No interaction and (b)
inseparable function interaction.

(8)

where k = 1, . . ., L, number of constraints and Pok,k is the desired


probability for satisfying constraint k. Parkinson et al. [54] have

R. Roy et al. / CIRP Annals - Manufacturing Technology 57 (2008) 697715

702

simplied (Eq. (8)) to reduce the computation time as (assuming


normal distribution for the constraints):

mgk  Ik s gk  0
1

(9)
1

where Ik = F Pok,k and F () is the inverse function of the


cumulative density function (CDF) of a standard normal distribution. Locating the optimum design solution is not trivial within an
uncertain environment. Jin et al. [52] have demonstrated that a
meta-modelling approach using Kriging could be used effectively
to locate a solution close to the optimum without much
computational cost. However, this technique depends on the
accuracy of the meta-model.
Reliability-based design optimisation works on the basis of a
concept of reliable design space (RDS), within which any design
satises the reliability requirements [42]. Designing a vehicle with
optimum crashworthiness requires the reliability-based approach.
This type of optimisation problem can be represented as
0

minimise R x1 ; mx2 ; m p

(10)

Such that; probg k x1 ; x2 ; pi0ir x1 ; x1 Lower  x1


k

 x1 Upper ; mx2 Lower  mx2  mx2 Upper


where R0 is the reliability-based objective function, x1 is the vector
of deterministic design variables, and x2 and p are the vectors of
random design variables and design parameters. mx2 and mp are the
mean vectors of x2 and p. gk() is the kth constraint, and there are L
number of constraints. prob() is the probability function which
denotes the probability of satisfying the kth constraint. The symbol
r x1 denotes the design reliability (or desired reliability/probk
ability) of satisfying the kth constraint. Often this optimisation is
considered as an iterative numerical analysis process searching the
most probable point (MPP) [55]. This is a time consuming process
and several researchers have tried to reduce the computational
costs by improving the reliability analysis method [5658]. The
latest of these efforts [59] has demonstrated an improved approach
by (1) analytically converting the feasible design space to the
reliable design space, and (2) performing a deterministic
optimisation constrained by the reliable design space. The major
challenge for these techniques is to address constraint functions of
deep valleys.
3.5. Computationally expensive objective functions and surrogate
models
Design optimisation is limited by the computational cost of the
design evaluation. Use of simulation techniques such as FEA,
computational uid dynamics (CFD) and dynamic simulation of
mechanisms are very expensive. The simulations also bring
unwanted noise in the evaluation [47]. Any optimisation approach
that requires several evaluations of the design is not suitable for
design optimisation. A common approach to address the noise and
computational cost issues is to use the simulation as numerical
experiments (instead of physical experiments), perform a number
of the experiments based on a design of experiment approach and
then develop an approximate model based on the simulation
results. Curve tting and other data modelling techniques are used
for creating this surrogate models or meta-models [60]. A detailed
review of meta-modelling techniques has presented three major
strategies for the modelling: sequential, adaptive and direct
sampling [61]. This review has identied that the meta-modelling
approach becomes less attractive with an increase in the number of
design variables.
3.6. Dealing with qualitative design space
Design optimisation is often performed with quantitative
information. This information is based on a principle of numerical
reckoning and is the ability to express the behaviour of a
phenomenon in numbers using numerical models. There are

Fig. 6. The discontinuous nature of search space for rod shape design, where the
roundness is a qualitative objective function and load quantitative [60]. R-1 stands
for Region 1.

aspects of a physical system that are less understood and


ambiguous or cannot be modelled using a numerical framework;
these are termed as qualitative [60]. The qualitative nature of
design variables, objective function and constraints can contribute
in incorporating human knowledge either through a knowledgebased system framework [60] or through direct interaction [25]. A
knowledge-based approach based on fuzzy logic to deal with the
qualitative design space introduces signicant discontinuities as
shown in Fig. 6 for hot rolled rod shape optimisation. Optimisation
within a highly discontinuous space is difcult. If the quantitative
and qualitative information is treated together within one
optimisation framework, granularity of the qualitative information
should also match the measurement scale of the quantitative
information, and that is not trivial. Another major challenge is to
avoid any individual bias and user inconsistency.
4. An overview of engineering design optimisation approaches
Engineering designs are still optimised mostly through a
manual iterative process where the designer compares a few
designs based on a small number of criteria (such as maximum
stress and weight) and then selects the best design. It is such a
common procedure that it is not often published. Fig. 7 shows an
overview of EDO approaches based on publications over the last 10
years. The designs are initially checked against any constraints
such as maximum cost, and only feasible designs are considered for
optimisation. This manual process is often limited to selecting
designs which are recognised by the designers, and it fails to
identify any unknown but potentially signicantly better designs.
This category of design optimisation is termed here as expertbased optimisation. This expert-based optimisation approach
often uses expert judgement (knowledge based) or simulation
techniques such as FEA or CFD analysis for the design optimisation.
The major advantage of this approach is that designers do not
require any additional skill, it may take less time to select a better
design, and it gives incremental improvement. On the other hand,
the major challenges are in the form of dependency on a few
experts who could evaluate the designs and nd truly novel and
signicantly better designs. Fig. 8 shows how different optimisation approaches are used to address different classes of design
optimisation problems. It is observed that algorithm-dependent
and large-scale optimisation classes require signicant research in
developing optimisation approaches.
A design of experiment (DOE) is a structured, organized method
for determining the relationship between factors (Xs) affecting a
design and the performance of that design (Y) (e.g. maximum
stress, weight or cost).
Once the contributions of the factors (i.e. design variables) on
the performance are identied, the information is used to identify
an ideal set of design variable values that is expected to yield the
best result. The DOE approach to design optimisation can reduce
the design time, and can often nd better performing designs that

R. Roy et al. / CIRP Annals - Manufacturing Technology 57 (2008) 697715

703

Fig. 7. An overview of EDO approaches.

are outside the comfort zone of designers. The approach provides


a structure to the optimisation, but it can still be a very manual
process. The approach works fairly well with design variables that
are independent from each other. In real life situation that is often
not the case [14,62,63].
Use of algorithms for design optimisation is gaining popularity.
This helps to partially automate the optimisation process and
allows a better search for the best design. The algorithmic
approach to EDO can be mostly categorised for ve reasons:
dealing with increasing complexity, real life design requirements,
increasing designer condence, hybrid and other. Roy et al. [64]
identied the above criteria as part of the major challenges in
adoption of algorithmic design optimisation in industry. The
complexity of design optimisation varies from highly constrained
optimisation, multi-disciplinary, multi-objective and multi-modal
optimisation. Multi-disciplinary design optimisation aims to nd a
globally best solution considering more than one discipline, say
structure, weight and aerodynamics [65]. For example, Wang et al.
[66] presents multi-disciplinary optimisation of helicopter air
intake scoop design involving couplings among deicing, aerodynamic, and engine performance. Fig. 9 shows increasing
popularity and demand for multi-disciplinary optimisation compared to other design optimisation approaches. This analysis is
based on searches in Engineering Village database during 1997
2006. Multi-objective design optimisation is also gaining popularity since the last 5 years.

Fig. 8. Engineering design approaches vs. different classes of the optimisation


problem.

Real life engineering design requires designs that are robust,


reliable and can operate with inherent uncertainties associated
with engineering systems. This is a growing area of research in the
last 10 years (Fig. 9). Reliability-based design optimisation looks
for optimal solutions considering probabilistic constraints [42].
This optimisation approach is popular in structural design
optimisation as well. Another real life design requirement is to
obtain robust optimum that is not sensitive to design tolerances,
production parameter drifts during operation, and model sensitivities. Robust optimisation approaches [50] search for solutions
that are in robust regions within a design space and locate the
optimum among the robust solutions. Ciof et al. [67] present a
robust design approach for magnets and identies optimum design
that is not sensitive to design and manufacturing tolerances.
Design optimisation within uncertain environment is a similar
concept to robust and reliability-based optimisation. In this
category approaches to quantify uncertainty and then perform
optimisation is grouped together. Agarwal et al. [43] have
demonstrated the use of evidence theory to quantify epistemic
uncertainty in a multi-disciplinary design problem. Increasing
designer condence on the algorithmic design optimisation is a
key challenge to increase adoption of the approach in industry.
Currently, two main approaches are used to address this issue: by
involving the designers during the optimisation process and by
capturing their knowledge as qualitative objective functions for
the design. Designers often complain about the simplicity of the
approximate models used in algorithmic design optimisation. It is
often not possible to analytically or numerically model every
important aspect of a design mainly due to the level of effort
required.
To address this need, several researchers [6870] have
presented an interactive design optimisation approach where
designers can qualitatively judge designs and give a rating. Designs
are optimised based on this qualitative tness function and other
quantitative functions together. Oduguwa et al. [60] have
developed a knowledge-based model of the qualitative aspects
of steelmaking roll design by involving designers and design team
leaders. The model is then used together with other quantitative
tness measures for a multi-objective optimisation.
The other two categories represent hybrid approaches to
algorithmic optimisation and approximate model based and single

704

R. Roy et al. / CIRP Annals - Manufacturing Technology 57 (2008) 697715

Fig. 9. EDO approaches presented in literature (19972006).

objective optimisations. Rao and Shyju [71] present a hybrid metaheuristic algorithm for combinatorial optimisation of laminate
composite cylindrical skirt which combines the good features of
popular guided local search algorithms like simulated annealing
and tabu search. Qazi and Linshu [72] present a neural network
based data driven approach to identify optimum conceptual design
of spacecraft. Single objective approaches are still used, although
multi-objective optimisation is more close to reality and are
becoming popular. Deb [12] gives a good overview of single
objective design optimisation approaches. The next section of this
paper identies major optimisation techniques used in each of the
optimisation approaches. Fig. 10 shows an overall growth in design
optimisation publications and a comparison with the optimisation
used for mechanical parts. It is observed that adoption of
optimisation approaches in mechanical part design is slower than
the overall growth in all sectors together. Due to complexities
mechanical part design optimisation often uses expert based
iterative process.
5. Trends in design optimisation techniques
There are several optimisation techniques that are used in
design optimisation: human knowledge-based engineering to
intelligent (adaptive) algorithms. Teti and Kumara [73] extensively
investigated the state-of-the-art, technological challenges and
development trends in applications of intelligent computing
methods in production engineering and gave particular attention
to the solution of design problems and issues in the manufacturing
environment. Fig. 11 presents a list of several of these optimisation
techniques against relevant algorithmic design optimisation

Fig. 10. Growth of design optimisation publication and a comparison with the
optimisation for mechanical parts.

approaches. The list attempts to identify all major approaches


used in design optimisation including a few very recent techniques
that are gaining popularity. The purpose of the diagram is to
present the whole spectrum of design optimisation and help in
bringing different approaches together to solve real life design
optimisation problems.
Based on the Engineering Village database (Inspec and
Compendex), it is observed that the top three design optimisation
algorithms used in the last 10 years (19972006) are: Genetic
Algorithms (693 publications), Linear and Quadratic programming
(142 publications) and Simulated Annealing (98 publications).
There are 58,982 publications on genetic algorithms over the last
10 years (19972006) based on the Engineering Village database
search, whereas only 693 of them are related to design
optimisation. Please note, the results are limited to the quality
of the database. EDO needs to exploit more of the growing
algorithmic optimisation techniques.
It is worth noting that while genetic algorithms is the most
popular technique for design optimisation, swarm intelligence [74]
has become popular in the last 5 years and show potential for
further use in design optimisation.
6. Engineering design optimisation in practice
Recently a survey was conducted involving ve UK-based
companies from aerospace, automotive and steelmaking industry
(total nine engineers) to investigate issues related to EDO in
industry. The survey was based on a semi-structured questionnaire
used in face to face interviews and in email response. Results from
the survey were also enhanced through discussions during the
Evolutionary Computing in Practice sessions in the GECCO 2007
conference (http://www.sigevo.org/gecco-2007/). Major remarks
from the survey are reported below (Fig. 12).
 All the organisations had a design development process (at least
informally) and optimisation of the design took place within the
process using mostly expert-based optimisation approach.
 Design optimisation was time consuming and needed at least
50% of design life cycle.
 Most did not use GA for the design optimisation (this is in
contrast to the number of publications where GA is the most
popular).
 Existing expert-based optimisation approaches were relatively
less time consuming than GA.
 Majority admitted that they did not have a process to develop
models for design evaluation.
 40% surveyed followed sequential process like:
FEA and Computer Aided Engineering followed by classical
DOE and regression analysis.

R. Roy et al. / CIRP Annals - Manufacturing Technology 57 (2008) 697715

705

Fig. 11. Major techniques used in relevant algorithmic design optimisation approaches.

 The main criterion to optimise a design was cost.


 The majority measured design efciency manually by comparing:
performance against target;
output with cost;
man-hours spent with other sister centers; or
with previous practice.
 The survey shows the majority of the companies try to achieve
optimised design:
as a balance between cost, quality and time but the design may
not be the best;
40% generally support continuous improvement as a way to
improve the design.
 There is a signicant gap in getting test result feedback on time
for the design optimisation.
 Last minute change in design specication puts a lot of strain on
the resources and therefore inhibits the optimisation.

 Lack of integrated software tools to support the optimisation


process.
 Internationally (through discussions at GECCO) there is an
increasing recognition of algorithmic optimisation for engineering design. GAs are known to designers as a potential technique.
 There are issues related to the recognition of Intellectual
Property rights for designs developed using evolutionary
computing technique (stochastic method) such as GAs.
 Major inhibitors to the use of an algorithmic approach for design
optimisation are: work load and priorities, cost and constraints in
operating especially within global organisations. Global organisations often like to decide the design process and tools to be
used centrally, which could stie local interests in using an
algorithmic approach for optimisation.
Developing novel designs using GAs may fail to describe a
logical product development process involving stakeholders. This
could be difcult when applying for patents or intellectual rights
for a design.
7. Expert based design optimisation

Fig. 12. Industrial survey observation.

Design optimisation of real life problems is often done by


designers using their experience. This process is often iterative,
time consuming and is limited by the knowledge of the designer.
Uhlmann and Schauer [29] have used FEA and empirical load
analysis to optimise the shape of microend mills (see Fig. 13).
Tosatti et al. [31] have presented an approach to optimise parallel
kinematic mechanisms based on kineto-static optimisation criteria
as the only criteria. Zirn et al. [28] have optimised fast axis feed

706

R. Roy et al. / CIRP Annals - Manufacturing Technology 57 (2008) 697715

[82,83] have applied intelligent computing approaches to the


design of multi-step cold forging tools for the fabrication of multidiameter shafts with different congurations and diverse work
materials. Initially, a rule-based expert system was developed for
the generation of cold forging working sequences and identication of the corresponding tool design. Then, supervised learning
neural network techniques were applied to assess the product
technological feasibility and associate the correct forming tool
geometries to be employed.
8. Design of experiment-based design optimisation

Fig. 13. Conventional and optimised microend mills [29].

drives using an analytical approach through nonlinear stability


analysis. The optimisation is based on experimental results which
are limited in numbers, and therefore does not fully explore the
design space. A similar approach is adopted by Lange et al. [30]
where they used FEA simulations and lightweight prototyping for
evaluating a selected set of designs. Other similar researches are
reported by Heisel and Klotz [33]. A trial and error approach to
develop advanced materials for net shape sintering of graded
laminated powder compact is also presented by Mori and Osakada
[35]. FFA was used to evaluate the designed materials. Lutters et al.
and Vaneker et al. [75,76] developed a what-if analysis based
decision support system for design optimisation. The approach
reduces workload of expert designers and provides better insight
about the design space.
But this approach is still limited by the amount of search one
can perform to nd the best solution. Another expert based
approach to optimise design is based on inexpensive and rapid
prototype development. For example, Krause et al. [36] have
presented a methodology to link CAD and rapid prototyping key
areas in a design to support the product development process. It is
argued that for effective use of rapid prototyping in optimisation, it
should be fully integrated as part of the product development. Janz
et al. [37] have developed an approach to optimise designs
manually based on value-oriented life cycle cost analysis. The
approach considers life cycle cost as the only objective, in reality
there are several objectives, such as environmental impact.
The other approach to expert based design optimisation is to
use a computer based system that uses knowledge elicited from
experts and can automate the design. knowledge-based systems
(KBSs) (also called expert systems) are computer programs
embodying knowledge about a narrow domain for solving and/
or searching for the optimal solution to problems related to that
domain [77]. More information on the technology of KBSs can be
found in most standard KBS textbooks [78].
Kimura et al. [79] have developed an expert system for injection
moulding tool design. This is resource intensive during the expert
system development and is dependent on the participation of the
designers. The approach is more suitable for variant type designs.
Yang et al. [80] have described an intelligent design system by
integrating an expert system, FEA and a CAD system for forging tool
design. The integrated approach has reduced the design optimisation time through automated link, but the nal selection is still
driven by the designers. Ismail et al. [81] used an expert system
shell called kappa for press tool design optimisation. This
approach is also suitable for variant type design. Teti et al.

It is often difcult to mathematically establish a relationship


between product performances and its design variables. Design of
experiment (DOE) is a technique to empirically understand the
impacts of design variables on the design performance and
therefore aid in identifying optimum variable values [14]. This
experimental method of optimisation is based on a fractional
factorial experiment which allows an experiment to be conducted
with only a fraction of all the possible experimental combinations
of design variable or design factor values. Standard orthogonal
arrays are used to design the experiments. The orthogonal array
denes the values of the parameters for each experiment.
Frequently, two orthogonal arrays are used, i.e. design and noise
factor matrices. A potential weakness of the approach is that it may
lead to a large number of experiments. To overcome this problem,
some researchers [8487] have proposed the use of a combined
array where control and noise factors are combined in a single
array. Each factor is divided into levels, typically 25. Experiments
are designed as a combination of the factor levels. Signal to noise
ratio is used as a metric to summarise the contribution of each
factor level and the amount of variability observed during multiple
experiments. A mean signal to noise ratio value is calculated for
each design factor level. This data is statistically analysed using
analysis of variation (ANOVA) techniques. To maximise the design
performance and minimise the noise, factor levels with the highest
impact are selected as the optimum design variable values. DOE
has two types of use in design optimisation, one is direct
optimisation and the other is a supporting role, where DOE is
often used to dene the data points for surrogate model
development.
8.1. Direct optimisation using DOE
Rout and Mittal [88] have applied the DOE technique to
optimise design variables for a rather complex planar robot
manipulator. The researchers introduced a normally distributed
noise and used simulation-based experiments instead of physical
trials. Selecting the right levels for each design factor and
identifying the real noise involved are the major challenges in
using DOE. It is also observed that this technique is more suitable
for designs with a smaller number of variables (around 10).
Evangelaras et al. [89] have presented a DOE with combined matrix
to reduce the number of experiments required. Yu and Jin [90] have
used DOE to optimise microelectronic packages in two stages: rst
to identify the most important design variables, and then a full
factorial DOE was carried out to study the key factors and their
interactions.
8.2. Supporting modelling and optimisation
DOE is also used to dene the data points to develop surrogate
models in case of expensive objective functions. Oduguwa et al.
[60] have demonstrated the use of the orthogonal matrix for data
point selection for response surface modelling of FEA results for a
steel rolling system design. Hou et al. [91] have presented an
approach to use DOE with D-optimal criteria technique to
construct response surface for the objective of specic energy
absorption and the constraint of maximum peak load, respectively
to perform crashworthiness based design. The research uses an

R. Roy et al. / CIRP Annals - Manufacturing Technology 57 (2008) 697715

707

iterative linear process to approximate the real nonlinearity in the


problem. Shuaeib et al. [92] have designed a motorcycle helmet
liner material with expensive FEA simulation and DOE to understand the effect of different factors on the helmet design. Similar
research is reported by other researchers [9395]. Yeniay et al. [96]
have utilised a dual response surface approach to quantify
variability in critical performance characteristics during the
conceptual design phase of a launch vehicle. In this example both
mean and variability is minimised simultaneously. There are other
design applications where a trade-off between the mean performance and variability is required.
9. Genetic algorithms for design optimisation
Genetic algorithms are stochastic search and optimisation
algorithms that mimic Darwins theory of biological evolution. The
idea behind a GA is to use this power of evolution to solve
optimisation problems. The father of the Genetic Algorithm is John
Holland who invented it in the early 1970s. As seen in Section 5,
GAs are currently the most popular design optimisation algorithms; the popularity has grown rapidly in the last 10 years. GA
works on the composition of genetic traits called chromosomes, in
which successive operations through crossover or mutation give
rise to better performing off-springs (population) due to successive
renement of these hereditary traits. GA works with a population
of design solutions (rather than single solutions as is the case with
many traditional optimisation algorithms) and tries to nd the best
solution. A design solution, composed of the design variables, is
represented as a single chromosome. This chromosome can be
either binary or real valued. The GA starts with a random
population of the chromosomes. The chromosomes are evaluated
using an objective or tness function. GA is not dependent on
gradient-based model for the design evaluation, and this is an
advantage in using the GA. The chromosomes are manipulated by
GA parameters such as crossover and mutation. Crossover
combines characteristics of two parent chromosomes to produce
two children chromosomes and mutation brings a sudden change
in the chromosome. GA is based on passing hereditary traits from
good designs to the next generation through a selection process
similar to natural selection. A good description of GA is presented
in [12]. GA is more popular for multi-objective design optimisation
than classical optimisation algorithms [12]. This section provides
an overview of GAs to solve mechanical EDO problems. GAs are
mostly used for design parameter (or size) optimisation, shape
optimisation or topology optimisation. Nassef and ElMaraghy [97]
have also used GA for type and magnitude optimisation of
geometric tolerances. The methodology can be computationally
expensive, although the authors offer some ideas to reduce the
cost.
9.1. Genetic algorithms for parameter optimisation
Design parameter optimisation problems include automotive
chassis design, turbine blade cooling system design, vehicle
suspension design, wing planform design, bearing design, microuidics, microactuator design, multibody rail vehicle design, speed
reducer design, composite leaf spring and composite drive shaft
design. Most of these applications are multi-objective in nature
with less than ve objectives [98]. It is also observed that some of
the applications have dealt with more than 10 design variables but
when it comes to constraints, it is observed that the number of
constraints is typically less than 5. Most of the applications involve
mixed type variables. There are applications that are only tested on
test functions, 12 out of 30 parameter optimisation problems are
about real life applications. One of the challenges in the parameter
or size optimisation problem is to deal with design variable
interaction, as mentioned in Section 3.3. Roy et al. [20] have
presented a turbine blade cooling system optimisation problem
with inseparable design variable interaction. Roy [14] presents
development of an analytical model (that is the objective function

Fig. 14. Design of a turbine blade cooling system [14].

for the design) to calculate coolant mass ow for radial passage and
metal temperature gas side (Fig. 14) based on twelve design
variables. The problem also has 15 constraints including the design
variable bounds. The model follows an iterative process to
calculate the tness values. This design problem is nonlinear
and has bias in the design space. The tness function is implicit and
multi-layered. Hence, the determination of tness values requires
an iterative procedure in which the effect that a variable has on the
objective functions depends on the values of other variable in the
function. This shows a high degree of inseparable function
interaction [20]. Full details of the model are presented in [14].
For any continuous portion of a Pareto front, there is a unique
relationship involving objective functions. This relationship is
difcult to obtain analytically, and even if it is found, it has limited
usefulness since mapping from function space to variable space is
very complex. However, the existence of a relationship among
tness functions of Pareto solutions necessarily implies that
corresponding relationship(s) exist among the decision variables
of these solutions [13]. An advanced GA called generalised
regression GA (GRGA) is used to explore this relationship within
the turbine blade design optimisation. It is observed that GRGA has
converged to the global Pareto front as shown in Fig. 15 and against
an exhaustive search (10,000 random points). The application has
established that GRGA successfully handles complex inseparable
function interaction to identify a range of optimum feasible
designs from which one could nally be chosen based on designers
preferences.
One of the major problems in parameter optimisation applications is the computational costs of the tness functions. One of the
common ways to deal with this is to use surrogate models based on
a small number of simulation results. The major challenge is to
develop acceptable surrogate models in the case of larger numbers
of variables. It is also observed that parameter optimisation often

Fig. 15. Experimental results (assuming two objectives) (units: Wcr in kg/s, Twg in K)
[14].

708

R. Roy et al. / CIRP Annals - Manufacturing Technology 57 (2008) 697715

uses hybrid genetic algorithms where the algorithm is used to


identify good solutions and then a local search tries to nd the
optimum solution. A trend in the literature is the growing
application of parameter optimisation for multi-objective design
problems [99104]. Multi-objective optimisation mostly deals
with a small number of objectives (less than 5). Kurpati et al.
describes an improved constraint handling for speed reducer
design and a design of a eet of ships with more than ve
constraints [105]. The improvement is made in the tness
assignment stage of a multi-objective GA and are all based upon
a Constraint-First-Objective-Next model. Most of the other
papers investigated had less than ve constraints. Similarly,
majority of the application papers address design optimisation
with less than 10 design variables [106109]. It is observed that
design problems with more than 10 design variables are often
expensive to evaluate. One approach to deal with such problem is to
use meta-models instead of simulation-based models for the design
evaluation [100,101,110]. Duvigneau and Visonneau [111] have
used neural networks to create an inexpensive model of the design,
similar to a meta-model. The model development would require
more example design solutions than fractional factorial designs, like
kriging. In terms of types of GAs used, hybrid algorithms are more
popular. Mian et al. [112] have presented a kriging assisted multiobjective genetic algorithm where the kriging-based meta-model is
used to evaluate some designs. If the kriging-based evaluation
changes the non-dominated solutions within a GA generation, then
those designs are evaluated using simulation. This approach reduces
the overall number of evaluations required and is suitable for
computationally expensive design.
The other type of hybrid GA is developed by integrating local
search techniques at the end of the GA [110]. Depince et al. [109]
present an innovative mix of GA with a collaborative optimisation
algorithm. This hybrid GA is suitable for multi-level design
problems. Luo and Dai [113] have presented a hybrid GA that
incorporates previous knowledge about the design, called Gene
Expression and Handling Genetic Algorithm (GEHGA). The inclusion of engineering design experiences and knowledge signicantly improves the algorithms quality of the initial population
and provides the capability of generating better genetic elements
for the consequent generations. Another important problem for
size optimisation is hierarchical design, such as machine tool
design. Yoshimura and Izui [114] have presented a hierarchical
chromosome structure that can deal with multilevel design
problems. For a complex multilevel design, such as rod rolling
system design a very long chromosome could occur.

design evaluation cost by using distributed parallel computing


environment.
Nash equilibrium is the solution of a non-cooperative strategy
of multi-objective optimisation rst introduced by Nash in 1951
[123]. For a multi-objective optimisation with N objectives, a Nash
game consists of N players, each in charge of one objective and able
to modify their sub-set of variables. GA implements this concept by
dividing the population of design solutions into a number of subpopulations, each sub-population focusing on one design objective
keeping other objectives xed. This approach helps to implement a
parallel GA with reduced interaction time between the subpopulations. When no sub-population can further improve his
objective function, the system has reached a state of equilibrium
named Nash equilibrium. Although this approach has demonstrated a signicant speeding up as compared to sequential GA, it
could be further improved by better synchronisation requirement
between sub populations.
It is observed that shape optimisation with a larger number of
design variables requires numerical simulation to evaluate the
designs, thus making it often computationally expensive. Use of
approximation methods such as surrogate model-based design
evaluation (to reduce computational cost) is often not suitable due
to the large number of design variables. Galantucci et al. [119] have
presented a novel application of GA and neural networks for shape
registration problem. Registration, dened as the process of
matching geometric entities, is performed when multiple scanned
data sets must be aligned or when an existing model must match
digitized point clouds. This process is crucial in several applications such as reverse engineering, CAD-based inspection and
computer vision. Articial neural networks technique is used for
pattern matching, whereas GA is used to minimize error deviations
between the geometric entities. The formulation of the optimisation problem using GA encoding was performed by subdividing the
population into six sub-populations, one for each parameter to be
optimised, i.e. the translation components and Eulers angles.
Fig. 16a and b show the results achieved by using a GA-based
optimisation.
The other major challenge in shape optimisation is to deal with
the qualitative nature of objective functions [25,124129]. There
are two major approaches in using GA for qualitative optimisation.
One approach is to develop a fuzzy expert system to capture the
qualitative nature of the objective function [60,127] and the other

9.2. Genetic algorithms for shape optimisation


Shape optimisation is characterised by a larger number of
variables and expensive evaluation. Application of shape optimisation includes compressor blade prole design [99], haptic device
design [115], pole shape of the synchronous generators [116],
nozzle shape optimisation [117], parallel mechanism design [118]
and automatic registration of free form surfaces [119]. Similar to
the previous paragraph, hybrid type GA is also popular in shape
design problems. Deterministic local search is integrated with GA
in [111]. Shape optimisation also often involves relatively larger
number of design variables, thus increasing their degree of
freedom [99,115,116,120]. There are several attempts to deal
with computationally expensive shape design optimisation
problems. The expensive optimisation is performed either by
parallel implementation of the GA with game theory [117,121,122]
or by more efcient GAs that require less expensive design
evaluation [99,112]. Wang et al. [121] have presented a
hierarchical GA where at the beginning they use very coarse
and approximate models of design as objective functions and after
the search matures they use more precise models. This reduces the
computational cost signicantly but at the same time achieves the
accuracy required for the nal design solutions. Then they
implemented game theory tools with the GA to address the

Fig. 16. (a) A plastic element of the rear part of a motorcycle chassis [119]. (b) Final
alignment of the point clouds using the GA for the plastic part shown above [119].

R. Roy et al. / CIRP Annals - Manufacturing Technology 57 (2008) 697715

is to involve designers directly within the search using an


interactive GA [25]. The major limitation of these approaches is
that not more than three to four objective functions can be
addressed. With an increasing number of objective functions, the
design space becomes more discontinuous, making it difcult for
the designers to interact effectively due to obvious human fatigue.
9.3. Genetic algorithms for topology optimisation
Before using a size or shape optimisation, an initial design
proposal has to be available that represents the structure of the
design solution. Topology optimisation is widely used in structural
optimisation applications, e.g. truss design [130132], and these
researches are not covered in this paper. Week et al. [133]
identied the potential of topology optimisation in product
development and argued that the target should be the integration
of topology and shape optimisation in a CAD-System in order to
produce a tool which supports the complete design process.
Mackerle [134] presents a recent review on topology optimisation.
Chapman et al. [135] presents a cantilever beam topology
optimisation using GAs. They identify that the optimisation often
involves a large number of variables and involve computational
expensive design evaluations (e.g. using FEA). In order to reduce
the computational time, Kim and Weck [136] have proposed a
variable length chromosome GA for progressive renement in
topology optimisation. The chromosome length is increased by an
increase in the resolution of the existing design variables or by the
addition of new design variables during encoding. Bharti et al.
[137] have used a multi-objective GA to optimally place cables and
struts in a bay or a section of an aircraft wing. Sid et al. [138] have
developed a novel topology representation using Bezier curves
with varying thickness material in a nite element model. This new
technique avoids the formation of disconnected elements and
checkerboard patterns in optimal topology design. They use
modied crossover and mutation GA operators to deal with the
new representation. Lu and Wang have shown how topology
optimisation could identify totally new designs [139]. They apply
GAs for developing designs of slider air bearings that meet the
strict performance demands of current hard disk drives. The
topology-optimised design can be better than the one obtained
after shape optimisation from an initial design. Although GA has
signicant potential for topology optimisation, scalability becomes
a major weakness, especially because of computationally expensive design evaluation. There is a lack of GAs application for real life
and large-scale topology optimisation problems. In future grid
computing based approach for multi-objective GAs implementation could address this issue [140].
10. Future challenges in algorithmic engineering design
optimisation
Considering the growth in publications using the algorithmic
approach for EDO, this approach has the best potential to improve a
design. Fig. 10 shows lack of popularity of algorithmic approach for
mechanical systems design optimisation compared to nonmechanical systems. This section identies the major challenges
of algorithmic approaches for real life optimisation and then
comments on the possible reasons for lack of interest in the
mechanical systems design community. The major challenges are
 Real life features: The features of real-life optimisation problems,
especially the presence of interaction among decision variables,
lack of prior problem knowledge and qualitative issues, create
challenges for optimisation algorithms. The lack of robust
optimisers that can effectively deal with these features prevents
their successful application in industry. This is particularly true
for those industries that deal with a wide range of complex
designs.
 Model development: All optimisation algorithms work on
mathematical models of real-life designs. It was observed that

709

since designers prefer maintaining full control on the design


improvement process, they have little faith in the models that are
provided to them. This makes them skeptical about the results
obtained from the optimisation algorithms. This situation is
further worsened by the fact that there is a lack of model
development skills among designers in industry. There is also a
lack of commercial tools required for carrying out the task of
model development. It is also observed in some cases that
designers lack knowledge about how legacy models may work;
often they nd it difcult to understand existing models.
Designer condence: Another inhibitor to the use of optimisation algorithms in industry is the important role of designers
skills and experience in the design improvement process. This
makes the optimisation task very difcult to model and encode in
algorithmic form. The lack of designers knowledge in using these
algorithms also presents a further obstacle to their use in
industry. This could be addressed by involving designers in the
search process either to provide their past knowledge as starting
point of the optimisation for objective function development [9]
or through an interactive optimisation process [25].
Design improvement process: Each company has its own design
improvement process. This process gradually evolves in the
company, and hence its people resist the implementation of any
new optimisation system and the associated organisational
changes. Further, the costs associated with creation, installation
and maintenance of optimisation algorithms discourage their
use in industry.
Computational expense of design evaluation: One approach to deal
with computationally expensive design evaluation models is to
develop surrogate models to replace the expensive design
models. Increasingly there is demand to work with more
accurate models and nd ways to deal with the computational
costs. In line with this there is increasing interest to solve real life
large-scale design optimisation problems. Recently, use of grid
and distributed computing is beginning to address this issue for
large-scale optimisation problems [141,142]. This is discussed in
more detail in Section 12.
Qualitative design space: Another major challenge would be to
extend the algorithmic optimisation approaches to deal with
larger scale qualitative design spaces. It would be ideal to handle
quantitative and qualitative information together within one
framework. Design evaluation time using qualitative objective
functions (e.g. fuzzy expert system-based model) increases
exponentially with the increase in number of variables. Developing a qualitative objective function that represents a broad range of
physical phenomena from a different perspective at a level which
allows useful and veriable inference to be drawn can also be
computationally expensive and non-trivial [60].
Integration with CAD and simulation: The bidirectional interface
between feature-based parametric CAD models and optimisation/analysis models that ensure automatic bidirectional conversions do not exist at present. Several researchers have
identied this deciency [143,144]. Nosner [143] noted that after
the shape or topology optimisation stage, the design engineer
still has to interact with the results to ensure that features are
integrated into the CAD model in an appropriate way.
Researchers [143,145] have noticed that the lack of feature
information prevents the application of meaningful engineering
constraints. Addressing these needs requires high level geometric reasoning such as feature technology/recognition to be
more integrated into the analysis algorithms and the optimisation procedure to achieve what has been termed feature-based
optimisation [145]. When these are addressed, several advantages will make optimisation techniques more attractive to
engineers in industry who are not experts in optimisation
techniques. For example, most loading and boundary conditions
can be automatically extracted from feature-based CAD model of
a product. Also, it would be easier to automatically generate
intuitive visualisation which has been identied by Hernandez
et al. [146] as a key need in order for engineers in industry to be

710

R. Roy et al. / CIRP Annals - Manufacturing Technology 57 (2008) 697715

comfortable with the use of optimisation techniques. The


application of geometric modelling and reasoning will allow
analysis (e.g. FEA), optimisation algorithms and parametric
feature-based CAD systems to be transparently and intuitively
integrated. In the short term, this may mean an integrated
information backbone or infrastructure [147] that is exible to
support changes in geometry, meshes etc and able to dynamically link with FEA or optimisation and CAD systems through
data exchange of native parametric CAD formats. In the longterm, it would require an extendible integrated information
infrastructure for CAD/FEA/Optimisation based on international
interoperability standards such as STEP (ISO 10303).
 Selecting an appropriate optimisation algorithm: Since there are
several optimisation algorithms, how can the user select the best
method without missing out on the most effective technique but
not getting bogged down with all the mathematics and
computational algorithm details of every new development?
Even in commercial integrated optimisation tools/environment
such as VisualDOC or Altair OptiStruct, the user is still required to
select the optimisation technique to be used to solve a problem.
Egorov et al. [148] have made some contributions in this
direction by proposing an evolutionary self-organising algorithm
which adaptively selects algorithm(s) suitable for each particular
problem. The technique is capable of handling multiple objective
functions and constraints which may be non-differentiable,
stochastic, with multiple optima, and mixed variables and users
are not required to have any knowledge of nonlinear programming or optimisation procedures. However, the user is still
required to set up the equations dening the physics of the
problem. The use of a unied taxonomy of problems and solution
techniques may also help this problem of selecting the best
optimisation algorithm(s). The automatic upgrade mechanisms
already present in most software that allows the update through
the Internet may be helpful also. But the upgrade needs to go
beyond the black box method and be transparent as well as
upgrading the unied taxonomy. However, please note that
there is no single best optimisation algorithm for all problems,
this is called no free lunch theorem [149].
 Dynamic optimisation: Recently there is interest in design
optimisation within a dynamic environment. One example is
to optimise rotor-bearing systems with dynamic constraints
[150]. Maalawi [151] presents a wind turbine design optimisation with dynamic loading conditions from wind speeds and
yawning motions of the turbine. One approach to consider a
dynamic environment is presented in the contribution of Chang
and Joo [152] using dynamic simulation of mechanisms. These
researches are still developing and signicant amount of
research would be required to extend the results to complex
multi-objective design optimisations. In the case of a dynamic
multi-objective optimisation the designers could obtain a set of
Pareto optimal design solutions that change with time. The next
challenge is to develop an optimisation approach that is robust
against the changing environment and supports designers to
select the design solution.
 Education: Designers also lack knowledge of optimisation
concepts and an overview of approaches. stochastic optimisation, like genetic algorithms, is contradictory to conventional
deterministic thinking. It is necessary to introduce optimisation
concepts, techniques and model building approaches within the
undergraduate course curriculum.
The following are some potential reasons why designers are still
not using the algorithmic approach in the design of mechanical
systems as effectively as others:
 Complexity from scalability: Scalability is a major challenge for
mechanical systems design optimisation. Large-scale design
optimisation requires approaches to deal with the complexity.
 Lack of understanding of the behaviour of complex mechanical
systems: There is a lack of understanding about the interaction

between mechanical components and their behaviours. This


makes model development very challenging. Numerical simulation, virtual testing, qualitative model development can address
this problem with additional computational and development
costs.
 Inherent uncertainty in performance of mechanical systems:
Uncertainty is another major challenge for mechanical systems
design optimisation. There is a signicant level of interest in this
area of optimisation research. Robust design, design with
uncertainty and reliability-based design optimisations are
addressing the issue and all of them have problem with
scalability (large-scale problem).
11. Future approaches to engineering design optimisation
It is observed form the analysis presented above there are three
major areas of improvement when it comes to use of computing to
address engineering design optimisation: improve efciency and
speed of optimisation and use human knowledge effectively where
necessary. The two following sections will discuss role of Grid and
distributed computing to speed up the optimisation and involve
multiple experts in the design process; and emergent computing
techniques for better efciency and speed in the optimisation.
11.1. Engineering design optimisation using grid and distributed
computing
Large-scale EDO of complex mechanical systems such as
aerodynamic wing design and gas turbines involve complex
processes with multiple iterative steps that require huge data and
computational resources to obtain satisfactory optimum solutions
[153,154]. The use of control theory and parallel distributed
computing has proved to greatly improve the speed of aerodynamic shape optimisation of supersonic aircraft design [155].
Though parallelising and distributing different computational
tasks to different processors in distributed computing architecture
improves speed of design optimisation, the need for multidisciplinary experts to collaborate and share data and knowledge
using heterogeneous platforms (software and hardware) is a big
challenge. Grid computing is distributed computing taken to the
next evolutionary level. The goal is to create a large and powerful
self-managing virtual computing system out of a large collection of
connected heterogeneous computers sharing various combinations of resources. Grid computing provides the middleware for
multi-disciplinary experts to analyse and optimise designs from
different geographical locations [141]. This section presents the
advantages of using high performance computing (HPC) and grid
computing for the optimisation.
Design optimisation involves workow of steps with dependencies. These workow dependencies can be difcult to
accomplish within the time constrain of most projects because
of the number of analysis tools involve in analysing mesh
generation and sending solutions to the next step. An automated
and integrated workow process can be implemented within a
grid-enabled environment [156,157]. Practical design optimisation
of laminated composite structures leads to high dimensional,
multi-modal and non-differentiable optimisation problems which
are difcult to solve [142]. This problem is further complicated as
there is no unique laminate conguration which can give an
optimum solution for all design criteria simultaneously. Using
distributed parallel computing and grid computing can overcome
this problem as each node will perform computations for each
criteria. These are a kind of divide and conquer optimisation
processes [158].
The process of design search and optimisation involves
modelling and analysis of engineering problems to yield technically and economically improved designs. CFD is involved in this
process when it comes to design optimisation of automotive and
aerospace products. Detailed analysis of the properties of these
products using CFD is usually computationally and data intensive.

R. Roy et al. / CIRP Annals - Manufacturing Technology 57 (2008) 697715

Grid tools like Geodise (grid-enabled optimisation design search


for engineers) incorporates distributed data, computational and
analysis tools as services for engineers to access and use efciently
[159]. Sasaki et al. [160] have demonstrated application of a hybrid
genetic algorithmic for multi-objective optimisation of single stage
rotor/stator blades for a multistage compressor to improve
aerodynamic performance in terms of efciency, blockage and
loss, while satisfying four aerodynamic constraints to maintain the
ow similar to a baseline geometry.
Current limitations in parallel distributed and grid computing
in the area of design optimisation are non-inclusion of company
culture in grid platforms, lack of tangible denition of what quality
of service means to different users and lack of economic models
that implement service-oriented architectures based on demand
and supply chain strategies as obtained in conventional global
market. Another limitation is the lack of security guarantee that
some sensitive data such as cost model which is part of design
optimisation may be accessed by competitors. Future trends in the
research aim to tackle the limitations mentioned. Gridbus is a
project that aims at providing an economic model for serviceoriented distributed and grid computing applications. Workow
management models that tackle design optimisation dependencies
are part of future research trend. Building robust security measures
into the Grid Security Infrastructures so that only the provider has
access to sensitive data is the trend in grid security. All these
features of a large-scale design optimisation can be addressed
using a service-oriented grid computing.
11.2. Emergent computing techniques
Swarm Intelligence was identied as a promising new
optimisation technique in Section 5. It is an attempt to develop
algorithms inspired by the collective behaviour of social insects
and other animal societies. There are two major categories of
algorithmic optimisation swarm intelligence approaches: ant
colony optimisation algorithms and particle swarm optimisation
algorithms. Dorigo and Blum [161] have provided a comprehensive
survey of theoretical foundation of ant colony optimisation. The
algorithms constitute mostly algorithms for discrete optimisation
that took inspiration from the observation of the foraging behavior
of ant colonies [73]. These algorithms have more potential to
address qualitative search spaces in design optimisation. Recently
ant colony optimisation research is extended to continuous design
spaces as well [162] using a hybrid approach with chaotic
sequences. On the other hand particle swarm intelligence is
inspired by human social behaviour [163]. Particle swarm
optimisation algorithms are applied to several parameter and
layout optimisation problems, such as recongurable machine tool
design [164], design of composite structures [165], layout of
satellite modules [166], induction motor design [167] and ceramic
tile manufacturing [168]. The optimisation is also becoming
popular for multi-objective optimisation [169173] and for
dynamic environments optimisation [174]. A recent trend is to
integrate features of swarm intelligence with other optimisation
techniques for more efciency [175178].
Simulated annealing is another popular optimisation technique
as identied in Section 5. The technique is a global optimisation
procedure which is inspired by the physical process of annealing
[179,180]. Suman and Kumar have presented a survey of simulated
annealing application for single and multi-objective optimisations
[181]. The technique is used for both parameter [167,182183] and
shape optimisations [184186].
Other recent approaches with potential for application in
design optimisation are continuous estimation of distribution
algorithm for continuous single and multi-objective design
problems [187189], and differential evolution technique for
constrained design optimisation [17,107,190,191]. Zhou and Sun
have presented a survey of estimation of distribution algorithms
[192]. Emergent synthesis approach is used to solve complex
synthesis problem in manufacturing [193195]. Distributed design

711

for a complex product like aircraft has similar synthesis challenge


(at least it is a case II problem). Design is becoming more complex
with new product-service system-based business models [196]. In
this new business model, companies sell functionality rather than
the product itself, and in some industry it is a long-term provision
[197]. Design of product-service systems includes co-design of
physical (e.g. product) and non-physical solutions (e.g. life long
service and support). Emergent synthesis has signicant potential
to contribute to product-service system design.
Quantum computing is a new computing paradigm with
signicant promise. The basic principle of quantum computation
is that the quantum properties (Qubits) can be used to represent
and structure design related information, and that quantum
mechanisms can be devised and built to perform operations with
this information. Quantum computing provides a faster and more
efcient computing framework [74,198]. The recent trend in
optimisation includes a hybrid approach to use quantum
computing for combinatorial, and non-convex optimisation
problems. It is observed that the optimisation performance has
improved by integrating quantum computing with genetic
algorithms [199,200] or swarm intelligence [201,202]. Quantum
computing based optimisation techniques would have a signicant
potential for large-scale design optimisation problems.
12. Summary and concluding remarks
EDO has evolved with time from a totally manual process to
computer-based approaches. This paper proposes a classication
of the optimisation problems based on two view points: design
evaluation effort and degrees of freedom. These view points are
relevant for mechanical engineering problems and show the major
issues in the optimisation. Mathematical challenges in design
optimisation identify current state of the optimisation. The study
then supports classifying the EDO approaches as expert-based,
DOE-based and algorithmic. It is observed that current approaches
are unable to address large-scale problems. The number of
publications recorded in one of the commercial databases is used
as a criterion to judge the relative popularity of different
approaches and optimisation techniques. It is observed that the
genetic algorithm is the most popular algorithmic technique for
optimisation. However, the algorithmic approach for design
optimisation (e.g. GA) is not very popular within the small number
of industries surveyed. Although algorithmic approaches are
increasingly used to solve mechanical component design optimisation problems, but some of them are never published as papers in
order to protect the intellectual property. The study identies the
major inhibitors in the industry acceptance of the optimisation
approach. An overview of expert-based and DOE-based optimisation and application of GAs have identied the current state of the
research. Lack of knowledge and a systematic model development
process inhibits designers to adopt the algorithmic approach.
It is observed that scalability is the biggest challenge.
Algorithmic optimisation is becoming popular for EDO based on
the number of papers published, but it is still not widely used in
industry. GAs are the most popular algorithmic optimisation
approach. Large-scale optimisation will require more research in
topology design, computational power and efcient optimisation
algorithms. Grid computing provides an opportunity to improve
the speed of EDO and involve multiple experts. Emergent
computing techniques such as swarm intelligence and quantum
computing improve efciency and speed of the optimisation.
Future success of EDO is in application of expert knowledge with
existing and emergent algorithmic and computing approaches to
large-scale designs, supported by education on optimisation.
Acknowledgements
Authors are grateful to Professors F.L. Krause, Moshe Shpitalni,
Tetsuo Tomiyama, G. Seliger, Stephen Lu, Peihua Gu, Neil Dufe,
Joost R. Duou, John Corbett, K. Ueda, S. Tichkiewitch, L. Monostori

712

R. Roy et al. / CIRP Annals - Manufacturing Technology 57 (2008) 697715

and Andrew Y. C. Nee for their contributions, comments and


encouragement. The authors are also grateful to Dr. Oladele
Owodunni, Dr. Jorn Mehnen, Dr. Ashutosh Tiwari, Dr. Essam
Shehab, Professor Kazuhiro Saito, Yoseph A. Tafesse and Gokop
Goteng for their kind support during the preparation of the paper.
References
[1] Byrne G, Dornfeld D, Inasaki I, Ketteler G, Konig W, Teti R (1995) Tool
Condition Monitoring (TCM)The Status of Research and Industrial Application. Annals CIRP-Manufacturing Technology 44(2):541567.
[2] Lucchetta G, Bariani PF, Knight WA (2006) A New Approach to the Optimization of Blends Composition in Injection Moulding of Recycled Polymers. CIRP
Annals-Manufacturing Technology 55(1):465468.
[3] Tseng MM, Jiao J, Merchant ME (1996) Design for Mass Customization. CIRP
Annals-Manufacturing Technology 45(1):153156.
[4] Liu F, Luh PB, Moser B, Kegg RL (1998) Scheduling and Coordination of
Distributed Design Projects. CIRP Annals-Manufacturing Technology 47(1):
111114.
[5] Krause F, Golm F (1996) Planning and Multicriteria Optimization of Design
Processes. CIRP Annals-Manufacturing Technology 45(1):145148.
[6] Nielsen JCO, Fredo CR (2006) Multi-disciplinary Optimization of Railway
Wheels. Journal of Sound and Vibration 293(3-5):510521.
[7] Laban M, Herrmann U (2007) Multi-disciplinary Analysis and Optimization
Applied to Supersonic Aircraft. Part 1: Analysis Tools, in: 48th AIAA/ASME/
ASCE/AHS/ASC Structures, Structural Dynamics, and Materials Conference, April
2326, American Institute of Aeronautics and Astronautics Inc., Reston, VA,
USA: pp. 18451861.
[8] Kicinger R, Arciszewski T, Jong KD (2005) Evolutionary Computation and
Structural Design: A Survey of The State-of-the-art. Computers & Structures
83(23-24):19431978.
[9] Oduguwa V (2003) Rolling System Design Optimisation using Soft Computing
Techniques, EngD Thesis, Craneld University.
[10] Rangavajhala S, Messac A (2006) Towards a Better Understanding of Equality
Constraints in Robust Design Optimization, in: 11th AIAA/ISSMO Multidisciplinary Analysis and Optimization Conference, September 68, American Inst of
Aeronautics and Astronautics Inc., Reston, VA, USA: pp. 307324.
[11] Goldberg DE (1989) Genetic Algorithms in Search, Optimization, and Machine
Learning. Addison-Wesley, Reading, MA.
[12] Deb K (2001) Multi-Objective Optimization Using Evolutionary Algorithms.
Wiley.
[13] Tiwari A (2001) Evolutionary Computing Techniques for Handling Variable
Interaction in Engineering Design Optimisation, PhD Thesis, Craneld University.
[14] Roy R (1997) Adaptive Search and the Preliminary Design of Gas Turbine
Blade Cooling System, PhD Thesis, University of Plymouth.
[15] Schutz M, Schwefel H (2000) Evolutionary Approaches to Solve Three Challenging Engineering Tasks. Computer Methods in Applied Mechanics and
Engineering 186(24):141170.
[16] Coello CAC (2002) Theoretical and Numerical Constraint-handling Techniques Used with Evolutionary Algorithms: A Survey of the State of the Art.
Computer Methods in Applied Mechanics and Engineering 191(1112):1245
1287.
[17] Landa Becerra R, Coello CAC (2006/7/1) Cultured Differential Evolution for
Constrained Optimization. Computer Methods in Applied Mechanics & Engineering 195(3336):43034322.
[18] Corne DW, Knowles JD (2007) Techniques for Highly Multiobjective Optimisation: Some Nondominated Points are Better Than Others, in: 9th Annual
Genetic and Evolutionary Computation Conference, GECCO 2007, July 711,
Association for Computing Machinery, New York, NY, USA: pp. 773780.
[19] Zaeh MF, Oertli T, Milberg J (2004) Finite Element Modelling of Ball Screw
Feed Drive Systems. CIRPP Annals-Manufacturing Technology 53(1):289292.
[20] Roy R, Tiwari A, Corbett J (2003) Designing a Turbine Blade Cooling System
Using a Generalised Regression Genetic Algorithm. CIRP Annals-Manufacturing Technology 52(1):415418.
[21] Shkvar EA (2001) Mathematical Modeling of Turbulent Flow Control by Using
Wall Jets and Polymer Additives Poster Paper, in: 2001 ASME Pressure Vessels
and Piping Conference, July 2226, American Society of Mechanical Engineers,
New York, NY, USA: pp. 277283.
[22] Tomiyama T, DAmelio V, Urbanic J, ElMaraghy W (2007) Complexity of
Multi-Disciplinary Design. CIRP Annals-Manufacturing Technology 56(1):185
188.
[23] Yang S, Ong Y, Jin Y. (Eds) (2007) Evolutionary Computation in Dynamic and
Uncertain Environments. 1st ed., Springer.
[24] Beyer H, Sendhoff B (2007) Robust OptimizationA Comprehensive Survey.
Computer Methods in Applied Mechanics and Engineering 196(3334):3190
3218.
[25] Brintrup AM, Ramsden J, Tiwari A (2007) An Interactive Genetic Algorithmbased Framework for Handling Qualitative Criteria in Design Optimization.
Computers in Industry 58(3):279291.
[26] Deb K (2003) Unveiling Innovative Design Principles by Means of Multiple
Conicting Objectives. Engineering Optimization 35(5):445470.
[27] Park EJ, da Luz LF, Suleman A. (2008) Multidisciplinary design optimization of
an automotive magnetorheological brake design. Computers & Structures
86(35):207216.
[28] Zirn O, Weikert S, Rehsteiner F (1996) Design and Optimization of Fast Axis
Feed Drives Using Nonlinear Stability Analysis. CIRP Annals-Manufacturing
Technology 45(1):363366.

[29] Uhlmann E, Schauer K (2005) Dynamic Load and Strain Analysis for the
Optimization of Micro End Mills. CIRP Annals-Manufacturing Technology
54(1):7578.
[30] Lange S, Schmidt H, Seliger G (2000) Product and Assembly Design for a Fibre
Reinforced Plastic Track Wheel. CIRP Annals-Manufacturing Technology
49(1):105108.
[31] Mollnari-Tosatti L, Fassi I, Legnani G, Jovane F (2003) Kineto-static Optimisation of PKMs. CIRP Annals-Manufacturing Technology 52(1):337341.
[32] Altintas Y, Cao Y (2005) Virtual Design and Optimization of Machine Tool
Spindles. CIRP Annals-Manufacturing Technology 54(1):379382.
[33] Heisel U, Klotz D (2003) An Optimized Design of the Bearing in Machines for
Ultrasonic Machining Processes. CIRP Annals-Manufacturing Technology
52(1):307310.
[34] Kaftanoglu B, Ulugul E, Carkoglu N (1995) Computer-Aided Optimal Design
Strategy of Power Transmission Systems. CIRP Annals-Manufacturing Technology 44(1):8386.
[35] Mori K, Osakada K (1999) Net Shape Approach for Sintering Process of Graded
Laminated Powder Materials Using Finite Element Simulation. CIRP AnnalsManufacturing Technology 48(1):239242.
[36] Krause F, Ciesla M, Stiel C, Ulbrich A (1997) Enhanced Rapid Prototyping for
Faster Product Development Processes. CIRP Annals-Manufacturing Technology 46(1):9396.
[37] Janz D, Sihn W, Warnecke H (2005) Product Redesign Using Value-Oriented
Life Cycle Costing. CIRP Annals-Manufacturing Technol 54(1):912.
[38] Enache S (1971) Researches concerning optimum geometry of the nishing
tools with nose radius. Annals of the CIRP 19(2):411416.
[39] Gim J-S, Cho DW, Lee JM. (1990) Optimal design of face milling cutter
geometry. Annals of the CIRP 39(1):391394.
[40] Kaldor S (1986) A Common Denominator for Optimal Cutting Tool Geometry.
Annals of CIRP 35(1):4144.
[41] Fogg B, Jamieson GA (1975) The Inuencing Factors in Optimizing Press Tool Die
Layouts and a Solution Using Computer Aids. Annals of the CIRP 24(1):429434.
[42] Xingtao L, Qing L, Xujing Y, Weigang Z, Wei L (2008) Multiobjective optimization for crash safety design of vehicles using stepwise regression model.
Structural and multidisciplinary optimization Springer Berlin 35(6):561569.
[43] Agarwal H, Renaud JE, Preston EL, Padmanabhan D (2004) Uncertainty
Quantication Using Evidence Theory in Multidisciplinary Design Optimization. Reliability Engineering & System Safety 85(13):281294.
[44] Tichkiewitch S (1988) Computer Aided Design of Forged Parts Taking into
Account Technological Constraints in Optimisation Process, in: CAD/CAM and
FEM in Metal Working, Pergamon Press: pp. 6169.
[45] Koch PN, Simpson TW, Allen JK, Mistree F (1999) Statistical Approximations
for Multidisciplinary Design Optimization: The Problem of Size. Journal of
Aircraft 36(1):275286.
[46] Lee DS, Gonzalez LF, Srinivas K, Auld DJ, Periaux J (2007) Multi-Objective/
Multidisciplinary Design Optimisation of Blended Wing Body UAV via
Advanced Evolutionary Algorithms, in: Proceedings of the 45th AIAA Aerospace
Sciences Meeting and Exhibit, January 811, AIAA, vol. 1: pp. 296316.
[47] Papalambros PY (2002) The Optimization Paradigm in Engineering Design:
Promises and Challenges. Computer-Aided Design 34(12):939951.
[48] Lee YH, Marvin ACaP, SJ (1999) Genetic Algorithm Using Real Parameters for
Array Antenna Design Optimisation, in: Proceedings of the High Frequency
Postgraduate Student Colloquium, 1999 IEEE: pp. 813.
[49] Draper NR, Smith H (1998) Applied Regression Analysis. Wiley, New York, USA.
[50] Beyer H, Sendhoff B (2007/7/1) Robust OptimizationA Comprehensive
Survey. Computer Methods in Applied Mechanics and Engineering 196(33
34):31903218.
[51] Chen W, Wiecek MM, Zhang J (1999) Quality UtilityA Compromise Programming Approach to Robust Design. Journal of Mechanical Design Transactions of the ASME 121(2):179187.
[52] Jin R, Du X, Chen W (2003) The Use of Metamodeling Techniques for
Optimization Under Uncertainty. Structural and Multidisciplinary Optimization
25(2):99116.
[53] Moses F (1997) Problems and Prospects of Reliability-based Optimization.
Engineering Structures 19(4):293301.
[54] Parkinson A, Sorensen C, Pourhassan N (1993) General Approach for Robust
Optimal Design. Journal of Mechanical Design Transactions of the ASME
115(1):7480.
[55] Ang AH-, Tang WH (1984) ProbabilityConcepts in Engineering-Planning and
Design-Decision Risk and Reliability. John Wiley and Sons.
[56] Youn BD, Choi KK, Yang R-, Gu L (2004) Reliability-based Design Optimization
for Crashworthiness of Vehicle Side Impact, in: National Congress on Computational Mechanics, 02 Springer-Verlag: pp. 272283.
[57] Youn BD, Choi KK, Park YH (2003) Hybrid Analysis Method for Reliabilitybased Design Optimization. Transactions of the ASME Journal of Mechanical
Design 125(2):221232.
[58] Mohsine A, Kharmanda G, El-Hami A (2006) Improved hybrid method as a
robust tool for reliability-based design optimization. Structural and Multidisciplinary Optimization 32(3):203213.
[59] Shan S, Wang GG (2008) Reliable design space and complete single-loop
reliability-based design optimisation. Reliability Engineering & System Safety
93(8):12181230.
[60] Oduguwa V, Roy R, Farrugia D (2007) Development of a Soft Computingbased Framework for Engineering Design Optimisation with Quantitative and
Qualitative Search Spaces. Applied Soft Computing 7(1):166188.
[61] Wang G, Gary. Shan S (2007) Review of Metamodeling Techniques in Support
of Engineering Design Optimization. Trans of the ASME Journal of Mechanical
Design 129:370380.
[62] Khoei AR, Masters I, Gethin DT (2002) Design Optimisation of Aluminium
Recycling Processes Using Taguchi Technique. Journal of Materials Processing
Technology 127(1):96106.

R. Roy et al. / CIRP Annals - Manufacturing Technology 57 (2008) 697715


[63] Madu IE, Madu CN (1999) Design Optimization Using Signal-to-noise Ratio.
Simulation Practice and Theory 7(4):349372.
[64] Roy R, Tiwari A, Munaux O, Jared G (2000) Real-life Engineering Design
Optimisation: Features and Techniques, in: CD-Rom Proceedings of the 5th
Online World Conference on Soft Computing Methods in Industrial Applications
(WSC5), September: p. 142.
[65] Sobieszczanski-Sobieski J, Haftka RT (1997) Multidisciplinary Aerospace
Design Optimization: Survey of Recent Developments. Structural Optimization 14(1):123.
[66] Wang D, Wang G, Naterer G.F (2003) Optimization of Helicopter Air
Intake Scoop Design, in: 2003 ASME Design Engineering Technical Conferences
and Computers and Information in Engineering Conference, September
26, American Society of Mechanical Engineers, New York, NY, USA: pp.
935944.
[67] Ciof M, Formisano A, Martone R (2004) Stochastic Handling of Tolerances
in Robust Magnets Design. IEEE Transactions on Magnetics 40(2):1252
1255.
[68] Xu J, Sun S, Tan Z, Shi F (2007) An Interactive Evolutionary Design System
with Feature Extraction. Human-Computer Interaction. HCI Applications and
Services, in: Proceedings of the 12th International Conference, HCI International
2007: pp. 10851094.
[69] Dragan CI, Parmee I (2002) Agent-based Support within an Interactive
Evolutionary Design System. (AI EDAM) Articial Intelligence for Engineering
Design Analysis and Manufacturing 16(5):331342.
[70] MacHwe AT, Parmee IC (2007) Multi-objective Analysis of a Componentbased Representation within an Interactive Evolutionary Design System.
Engineering Optimization 39(5):591613.
[71] Rao ARM, Shyju PP (2008) Development of a Hybrid Meta-heuristic Algorithm for Combinatorial Optimisation and its Application for Optimal Design
of Laminated Composite Cylindrical Skirt. Computers & Structures 86(7
8):796815.
[72] Qazi M, Linshu H (2006) Nearly-orthogonal Sampling and Neural Network
Metamodel Driven Conceptual Design of Multistage Space Launch Vehicle.
Computer-Aided Design 38(6):595607.
[73] Teti R, Kumara SRT (1997) Intelligent Computing Methods for Manufacturing
Systems, STC-O Keynote Paper. Annals of the CIRP 46(2):629652.
[74] de Castro LN (2007) Fundamentals of Natural Computing: An Overview.
Physics of Life Reviews 4(1):136.
[75] Lutters D, Vaneker THJ, van Houten FJAM (2004) What-if Design: A Synthesis
Method in the Design Process. CIRP Annals-Manufacturing Technology
53(1):113116.
[76] Vaneker THJ, van Houten FJAM (2006) What-if Design as a Synthesizing
Working Method in Product Design. CIRP Annals-Manufacturing Technology
55(1):131134.
[77] Pham DT, Pham PTN (1999) Articial Intelligence in Engineering. International Journal of Machine Tools & Manufacture 39(6):937949.
[78] Durkin
.
J (1994) Expert Systems Design and Development. Macmillan, New York
[79] Kimura F, Ariyoshi H, Ishikawa H, Naruko Y, Yamato H (2004) Capturing
Expert Knowledge for Supporting Design and Manufacturing of Injection
Molds. CIRP Annals-Manufacturing Technology 53(1):147150.
[80] Yang DY, Im YT, Yoo YC, Park JJ, Kim JH, Chun MS, Lee CH, Lee YK, Park CH,
Song JH, Kim DY, Hong KK, Lee MC, Kim SI (2000) Development of Integrated
and Intelligent Design and Analysis System for Forging Processes, CIRP
Annals-Manufacturing Technology 49(1): 177180.
[81] Ismail HS, Hon KKB, Huang K (1995) An Intelligent Object-Oriented Approach
to the Design and Assembly of Press Tools. CIRP Annals-Manufacturing Technology 44(1):9196.
[82] Teti R, Langella A, DAddona D (2000) Process Design of Cold Forged Steel
Shafts Based on a NN Paradigm, in: 2000 International CIRP Des. Sem., May 16
18, Haifa, Israel: pp. 139144.
[83] Teti R, Langella A, DAddona D (2000) Neural Network Based Process Design
of Cold Forged Shats with Different Work Materials. Journal of Manufacturing
Science and Production 3(4):6576.
[84] Borkowski JJ, Lucas JM (1997) Designs of Mixed Resolution for Process
Robustness Studies. Technometrics 39(1):6370.
[85] Box G, Jones S (2000) Quality Quandaries: Split Plots for Robust Product and
Process Experimentation. Quality Engineering 13(1):127134.
[86] Lucas JM (1994) How to Achieve a Robust Process Using Response Surface
Methodology. Journal of Quality Technology 26(4):248260.
[87] Simpson JR, Montgomery DC (1996) A Biased-Robust Regression Technique
for the Combined Outlier-Multicollinearity Problem. Journal of Statistical
Computation and Simulation 56(1):122.
[88] Rout BK, Mittal RK (2008) Parametric Design Optimization of 2-DOF RR
planar ManipulatorA Design of Experiment Approach. Robotics and Computer-Integrated Manufacturing 24(2):239248.
[89] Evangelaras H, Koukouvinos C, Stylianou S (2005) Evaluation of Some Nonorthogonal Saturated Designs with Two Levels. Statistics & Probability Letters
74(4):322329.
[90] Yu G, Jin D (2006) Drop Test Simulation and DOE Analysis for Design
Optimization of Microelectronics Packages, in: 56th Electronic Components
& Technology Conference (IEEE Cat. No. 06CH37766C) Publication, 30 May2
June IEEE: pp. 422427.
[91] Hou T, Su C, Liu W (2007) Parameters Optimization of a Nano-particle Wet
Milling Process Using the Taguchi Method, Response Surface Method and
Genetic Algorithm. Powder Technology 173(3):153162.
[92] Shuaeib FM, Hamouda AMS, Wong SV, Umar RSR, Ahmed MMHM (2007) A
New Motorcycle Helmet Liner Material: The Finite Element Simulation and
Design of Experiment Optimization. Materials & Design 28(1):182195.
[93] Rikards R, Abramovich H, Kalnins K, Auzins J (2006) Surrogate Modeling in
Design Optimization of Stiffened Composite Shells. Composite Structures
73(2):244251.

713

[94] Shyy W, Papila N, Vaidyanathan R, Tucker K (2001) Global Design Optimization for Aerodynamics and Rocket Propulsion Components. Progress in Aerospace Sciences 37(1):59118.
[95] Keane AJ (2003) Wing Optimization Using Design of Experiment, Response
Surface, and Data Fusion Methods. Journal of Aircraft 40(4):741750.
[96] Yeniay O, Unal R, Lepsch RA (2006) Using Dual Response Surfaces to Reduce
Variability in Launch Vehicle Design: A Case Study. Reliability Engineering &
System Safety 91(4):407412.
[97] Nassef AO, ElMaraghy HA (1997) Allocation of Geometric Tolerances: New
Criterion and Methodology. CIRP Annals-Manufacturing Technology
46(1):101106.
[98] Coello CCA, Aguirre AH, Zitzler E (2007) Evolutionary Multi-objective Optimization. European Journal of Operational Research 181(3):16171619.
[99] Jin D, Zhan Z, Gui X (2006) Automatic Design Optimization of Compressor
Blades Based on Hybrid Genetic Algorithm. Tuijin Jishu/Journal of Propulsion
Technology 27(4):349353.
[100] Kim S, Chung H (2006) Multiobjective Optimization Using Adjoint Gradient
Enhanced Approximation Models for Genetic Algorithms, in: ICCSA 2006:
International Conf on Computational Sc. and Its Applications, May 811,
Springer Verlag, Heidelberg, Germany: pp. 491502.
[101] Marcelin JL (2006) Genetic Optimization of Vibrating Stiffened Plates. Structural Engineering and Mechanics 24(5):529541.
[102] Zhang J, Zhang Y, Gao R (2006) Genetic Algorithms for Optimal Design of
Vehicle Suspensions, in: 2006 IEEE International Conference on Engineering of
Intelligent Systems, April 2223, IEEE: pp. 16.
[103] Chen J, Zhong Y, Xiao R, Sun J (2005) The Research of the DecompositionCoordination Method of Multidisciplinary Collaboration Design Optimization, Engineering Computations 22(3): 274285.
[104] Mehnen J, Michelitsch T, Lasarczyk C, Bartz-Beielstein T (2007) Multi-objective Evolutionary Design of Mold Temperature Control Using DACE for
Parameter Optimization. International Journal of Applied Electromagnetics
and Mechanics 25(14):661667.
[105] Kurpati A, Azarm S, Wu J (2002) Constraint Handling Improvements for
Multiobjective Genetic Algorithms. Structural and Multidisciplinary Optimization 23(3):204213.
[106] Chen Z, Jiao Y, Xia S, Huang W (2005) Optimum Design of Rotor-bearing
Systems Considering the Nonlinear Effects of Bearing Fluid Forces, IN:
DETC2005: ASME International Design Engineering Technical Conferences and
Computers and Information in Engineering Conference, September 2428,
American Society of Mechanical Engineers, New York, NY, USA: pp. 1107
1112.
[107] Codreanu, I (2005) A Parallel Between Differential Evolution and Genetic
Algorithms with Exemplication in a Microuidics Optimization Problem, in:
International Semiconductor Conference, CAS 2005, October 35, Institute
of Electrical and Electronics Engineers Inc., Piscataway, NJ, USA: pp. 421
424.
[108] Cristello NM, Kim IY (2006) Multidisciplinary Design Optimization of a Zero
Emission Vehicle Chassis Considering Crashworthiness and Hydroformability, in: 11th AIAA/ISSMO Multidisciplinary Analysis and Optimization Conference, September 68, American Inst. of Aeronautics and Astronautics Inc.,
Reston, VA, USA: pp. 912925.
[109] Depince P, Rabeau S, Bennis F (2005) Collaborative Optimization Strategy for
Multi-objective Design, in: DETC2005: ASME International Design Engineering
Technical Conferences and Computers and Information in Engineering Conference, September 2428, American Society of Mechanical Engineers, New York,
NY, USA: pp. 197205.
[110] Kim Y, Gerwell M, Ling H (2005) Application of the Cauchy Method to Genetic
Algorithms for Broadband Antenna Design, in: 2005 IEEE Antennas and
Propagation Society International Symposium, July 38, IEEE: pp. 590593.
[111] Duvigneau R, Visonneau M (2004) Hybrid Genetic Algorithms and Articial
Neural Networks for Complex Design Optimization in CFD. International
Journal for Numerical Methods in Fluids 44(11):12571278.
[112] Mian L, Genzi L, Azarm S (2006) A Kriging Metamodel Assisted Multiobjective Genetic Algorithm for Design Optimization, in: 2006 ASME International Design Engineering Technical Conferences and Computers and Information in Engineering Conference, DETC2006, September 1013, American Society
of Mechanical Engineers, New York, NY, USA: pp. 1020.
[113] Luo L, Dai L (2005) A Newly Developed Genetic Algorithm for Engineering
Design Optimizations. WSEAS Transactions on Mathematics 4(2):102109.
[114] Yoshimura M, Izui K (2002) Smart Optimization of Machine Systems Using
Hierarchical Genotype Representations. Journal of Mechanical Design Transactions of the ASME 124(3):375384.
[115] Lee S, Kim S (2006) Analysis and Optimal Design of a New 6 DOF Parallel Type
Haptic Device, in: 2006 IEEE/RSJ International Conference on Intelligent Robots
and Systems, IROS 2006, October 915, Institute of Electrical and Electronics
Engineers Inc., Piscataway, NJ, USA: pp. 460465.
[116] Susnjic L, Haznadar Z (2001) Combined Genetic Algorithm2D Finite Element Analysis Applied to Synchronous Machines. Electromotion 8(4):197
201.
[117] Periaux J, Chen HQ, Mantel B, Sefrioui M, Sui HT (2001) Combining Game
Theory and Genetic Algorithms with Application to DDM-Nozzle Optimization Problems. Finite Elements in Analysis and Design 37(5):417429.
[118] Zhang D, Wang L (2005) Conceptual Development of an Enhanced Tripod
Mechanism for Machine Tool. Robotics and Computer-Integrated Manufacturing 21(4-5):318327.
[119] Galantucci LM, Percoco G, Spina R (2004) An Articial Intelligence Approach
to Registration of Free-form Shapes. CIRP Annals-Manufacturing Technology
53(1):139142.
[120] Ong Y, Nair PB, Lum KY (2006) MaxMin Surrogate-assisted Evolutionary
Algorithm for Robust Design. IEEE Transactions on Evolutionary Computation
10(4):392404.

714

R. Roy et al. / CIRP Annals - Manufacturing Technology 57 (2008) 697715

[121] Wang JF, Periaux J, Sefrioui M (2002) Parallel Evolutionary Algorithms for
Optimization Problems in Aerospace Engineering, in: 15th Toyota Conference:
Scientic and Engineering Computations for the 21st Century-Methodologies and
Applications, 12/01 Elsevier: pp. 155169.
[122] Galvan B, Greiner D, Periaux J, Sefrioui M, Winter G (2003) Parallel Evolutionary Computation for Solving Complex CFD Optimization Problems: A
Review and Some Nozzle Applications. in Matsuno K, Ecer A, Satofuka N,
Periaux J, Fox P, (Eds.) Parallel Computational Fluid Dynamics 2002. Amsterdam, North-Holland, pp. 573604.
[123] Nash JF (1951) Noncooperative Games. Annals Mathematics 54:286295.
[124] Oduguwa V, Roy R, Farrugia D (2003) An Integrated Design Optimisation
Approach for Quantitative and Qualitative Search Space, in: 2003 ASME
Design Engineering Technical Conferences and Computers and Information in
Engineering Conference, September 26, American Society of Mechanical
Engineers, New York, NY, USA: pp. 13591368.
[125] Machwe AT, Parmee IC (2007) Supporting Free-form Design Using a
Component Based Representation: An Overview, in: 9th Annual
Genetic and Evolutionary Computation Conference, GECCO 2007, July 711,
Association for Computing Machinery, New York, NY, USA: pp. 2821
2824.
[126] Machwe AT, Parmee IC, Miles JC (2005) Overcoming Representation Issues
When Including Aesthetic Criteria in Evolutionary Design, in: 2005 ASCE
International Conference on Computing in Civil Engineering, July 1215, American Society of Civil Engineers, Reston, VA, USA: pp. 583594.
[127] Huber M, Baier H (2006) Qualitative Knowledge and Manufacturing Considerations in Multidisciplinary Structural Optimization of Hybrid Material
Structures, in: Flexible manufacture of lightweight frame structures. Trans Tech
Publications: pp. 143152.
[128] Binder T, Heitzinger C, Selberherr S (2001) A Qualitative Study on Global and
Local Optimization Techniques for TCAD Analysis Tasks, in: 2001 International Conference on Modeling and Simulation of Microsystems-MSM 2001,
March 1921, Computational Publications, Cambridge, MA, USA: pp. 466
469.
[129] Mukerjee A, Agrawal RB, Tiwari N, Hasan N (1997) Qualitative Sketch
Optimization. (AI EDAM) Articial Intelligence for Engineering Design Analysis
and Manufacturing 11(4):311323.
[130] Dominguez A, Stiharu I, Sedaghati R (2006) Practical Design Optimization of
Truss Structures Using the Genetic Algorithms. Research in Engineering Design
17(2):7384.
[131] Stocki R, Kolanek K, Jendo S, Kleiber M (2001/9) Study on Discrete Optimization Techniques in Reliability-based Optimization of Truss Structures. Computers & Structures 79(22-25):22352247.
[132] Jakiela MJ, Chapman C, Duda J, Adewuya A, Saitou K (2000) Continuum
Structural Topology Design with Genetic Algorithms. Computer Methods in
Applied Mechanics and Engineering 186(2-4):339356.
[133] Week M, Asbeck J, Bussenschutt A (1996) Potentials of Structural Optimization Systems in Product Development. CIRP Annals-Manufacturing Technology
45(1):165168.
[134] Mackerle J (2003) Topology and Shape Optimization of Structures Using FEM
and BEM: A Bibliography (19992001). Finite Elements in Analysis and Design
39(3):243253.
[135] Chapman CD, Saitou K, Jakiela MJ (1994) Genetic Algorithms as an Approach
to Conguration and Topology Design. Journal of Mechanical Design Transactions Of the ASME 116(4):10051012.
[136] Kim IY, de Weck OL (2005) Variable Chromosome Length Genetic Algorithm
for Progressive Renement in Topology Optimization. Structural and Multidisciplinary Optimization 29(6):445456.
[137] Bharti S, Frecker M, Lesieutre G (2006) Optimal Structural Design of a
Morphing Aircraft Wing Using Parallel Non-dominated Sorting Genetic
Algorithm II (NSGA II), Smart Structures and Materials 2006: Modeling,
Signal Processing, and Control, February 27March 2, Proceedings of the SPIE
6166: 616602.
[138] Sid B, Domaszewski M, Peyraut F (2005) Topology Optimization Using an
Adaptive Genetic Algorithm and a New Geometric Representation, OPTI IX, May
2325, WIT Press: pp. 127135.
[139] Chung-Jen Lu. Tai-Kuo Wang. (2004) New Designs of HDD Air-lubricated
Sliders via Topology Optimization. Transactions of the ASME Journal of Tribology 126(1):171176.
[140] Chang Y, Cheng W, Liu X, Xie X (2006) Application of Grid Technology in
Multi-objective Aircraft Optimization System, in: Proceedings 2006 10th
International Conference on Computer Supported Cooperative Work in Design,
May 35, IEEE: pp. 15.
[141] Grauer M, Barth T, Thilo F (2004) Grid-based Computing for Multidisciplinary
Analysis and Optimization, in: Collection of Technical Papers-10th AIAA/ISSMO
Multidisciplinary Analysis and Optimization Conference, August 30September
1, American Institute of Aeronautics and Astronautics Inc., Reston, VA, USA:
pp. 17441753.
[142] Kere P, Lento J (2005) Design Optimization of Laminated Composite Structures
Using Distributed Grid Resources. Composite Structures 71(34):435438.
[143] Nosner RH (2008) Structural optimisation with altair optistruct in practice.
Available: http://www.tecosim.com/leadmin/user_upload/Exposes/altair_
opti-struct.pdf.
[144] Langer H, Puhlhofer T, Baier H (2002) An Approach for Shape and Topology
Optimization integrating CAD Parameterization and Evolutionary Algorithms, in: Proceedings 9th AIAA/ISSMO Symposium on Multidisciplinary Analysis and Optimization, September 46, Atlanta, USA.
[145] Shen J, Yoon D (2004) A Freeform Shape Optimization of Complex Structures
Represented by Arbitrary Polygonal or Polyhedral Meshes. International
Journal for Numerical Methods in Engineering 60(15):24412466.
[146] Hernandez S, Hernandez LA, Anton A (2000) Virtual Laboratories for Analysis
and Design of Civil Engineering Structures, in: Proceedings of 6th International

[147]

[148]

[149]
[150]

[151]

[152]

[153]
[154]

[155]

[156]

[157]

[158]

[159]

[160]

[161]
[162]

[163]

[164]

[165]

[166]
[167]

[168]

[169]

[170]

[171]

[172]

Conference on Applications of High-Performance Computers and Engineering,


January 2628, WIT Press: pp. 477491.
Kolonay RM, Burton SA (2004) Object Models for Distributed Multidisciplinary Analysis and Optimization (MAO) Environments that Promote CAE
Interoperability, in: Collection of Technical Papers-10th AIAA/ISSMO Multidisciplinary Analysis and Optimization Conference, August 30September 1,
American Institute of Aeronautics and Astronautics Inc., Reston, VA, USA: pp.
33103325.
Egorov IN, Kretinin GV, Leshchenko IA, Kuptzov SV (2004) The Main Features
of IOSO Technology Usage for Multi-objective Design Optimization, in:
Collection of Technical Papers-10th AIAA/ISSMO Multidisciplinary Analysis
and Optimization Conference, August 30September 1, American Institute
of Aeronautics and Astronautics Inc., Reston, VA, USA: pp. 34373447.
Wolpert DH, Macready WG (1997) No Free Lunch Theorems for Optimization. IEEE Transactions on Evolutionary Computation 1(1):6782.
Choi B-K, Yang B-S (2001) Multiobjective Optimum Design of Rotor-bearing
Systems with Dynamic Constraints Using Immune-genetic Algorithm. Journal of Engineering for Gas Turbines and Power 123(1):7881.
Maalawi KY (2007) A Model for Yawing Dynamic Optimization of a Wind
Turbine Structure. International Journal of Mechanical Sciences 49(10):1130
1138.
Chang K, Joo S (2006) Design Parameterization and Tool Integration for CADbased Mechanism Optimization. Advances in Engineering Software
37(12):779796.
Goel S, Talya SS, Sobolewski M (2007) Service-based P2P Overlay Network for
Collaborative Problem Solving. Decision Support Systems 43(2):547568.
Quoc-Thuan Ho, Yew-Soon Ong, Cai W, Hee-Khiang Ng, Bu-Sung Lee (2004)
GAD KitA Toolkit for Gridifying Applications, in: Proceedings, December 8
10, Springer-Verlag: pp. 868871.
Reuther J, Jameson A (1995) Supersonic Wing and WingBody Shape Optimization Using an Adjoint Formulation, in: Proceedings of the 1995 ASME
International Mechanical Engineering Congress & Exposition, November 1217,
ASME, New York, NY, USA: pp. 4552.
Jiao Z, Wason J, Song W, Xu F, Eres H, Keane AJ, Cox SJ (2004) Databases,
Workows and the Grid in a Service Oriented Environment, in: Proceedings,
31 August3 September, Springer-Verlag: pp. 972979.
Parmee IC, Abraham J, Shackelford M, Rana OF, Shaikhali A (2005) Towards
Autonomous Evolutionary Design Systems via Grid-based Technologies, in:
2005 ASCE International Conference on Computing in Civil Engineering, July 12
15, American Society of Civil Engineers, Reston, VA, USA: pp. 12411252.
Wang N, Chen L (2005) A Divide-and-conquer Parallel Computing Scheme for
the Optimization Analysis of Tribological Systems. Tribology and Lubrication
Technology 61(1):3846.
Pound GE, Eres MH, Wason JL, Jiao Z, Keane AJ, Cox SJ (2003) A Grid-enabled
Problem Solving Environment (PSE) for Design Optimisation within Matlab,
in: International Parallel and Distributed Processing Symposium (IPDPS 2003),
April 2226, IEEE Comput. Soc.: 8 pp.
Sasaki D, Keane AJ, Shahpar S (2006). Multiobjective Evolutionary Optimization of a Compressor Stage Using a Grid-enabled Environment, in: 44th AIAA
Aerospace Sciences Meeting 2006, January 912, American Institute of Aeronautics and Astronautics Inc., Reston, VA, USA: pp. 40714088.
Dorigo M, Blum C (2005) Ant Colony Optimization Theory: A Survey. Theoretical Computer Science 344(2-3):243278.
Coelho L, Mariani dSVC (2008) Use of Chaotic Sequences in a Biologically
Inspired Algorithm for Engineering Design Optimization. Expert Systems with
Applications 34(3):19051913.
Eberhart R, Kennedy J (1995) A New Optimizer Using Particle Swarm Theory,
in: Proceedings of the Sixth International Symposium on Micro Machine and
Human Science, October 46, IEEE: pp. 3943.
Liu W, Liang M (2008) Multi-objective Design Optimization of Recongurable
Machine Tools: A Modied Fuzzy-Chebyshev Programming Approach. International Journal of Production Research 46(6):15871618.
Omkar SN, Mudigere D, Naik GN, Gopalakrishnan S (2008) Vector Evaluated
Particle Swarm Optimization (VEPSO) for Multi-objective Design Optimization of Composite Structures. Computers and Structures 86(1-2):114.
Zhang B, Teng H, Shi Y (2008) Layout Optimization of Satellite Module Using
Soft Computing Techniques. Applied Soft Computing 8(1):507521.
Padma S, Bhuvaneswari R, Subramanian S (2007) Application of Soft Computing Techniques to Induction Motor Design. COMPEL-The International
Journal for Computation and Mathematics in Electrical and Electronic Engineering 26(5):13241345.
Navalertporn T, Afzulpurkar NV (2007) Optimizing Process Parameters for
Ceramic Tile Manufacturing Using an Evolutionary Approach, in: 4th International Symposium on Neural Networks, ISNN 2007, June 37, SpringerVerlag: pp. 252260.
Brits R, Engelbrecht AP, van den Bergh F (2007) Locating Multiple Optima
Using Particle Swarm Optimization. Applied Mathematics and Computation
189(2):18591883.
Nakamura M, Izui K, Nishiwaki S, Yoshimura M (2006) A Multi-objective
Particle Swarm Optimization Incorporating Design Sensitivities, in: 11th
AIAA/ISSMO Multidisciplinary Analysis and Optimization Conference, September
68, American Institute of Aeronautics and Astronautics Inc., Reston, VA,
USA: pp. 112121.
Ochlak E, Forouraghi B (2006) A Particle Swarm Algorithm for Multiobjective
Design Optimization, in: 2006 18th IEEE International Conference on Tools with
Articial Intelligence, November 1315, IEEE Computer Society: pp. 765772.
Santana-Quintero LV, Ramirez-Santiago N, Coello Coello, Carlos A, Luque JM,
Hernandez-Diaz AG (2006) A New Proposal for Multiobjective Optimization
Using Particle Swarm Optimization and Rough Sets Theory, in: 9th International Conference on Parallel Problem Solving from Nature, PPSN IX, September
913, Springer Verlag, Heidelberg, Germany: pp. 483492.

R. Roy et al. / CIRP Annals - Manufacturing Technology 57 (2008) 697715


[173] Hu X, Eberhart RC, Shi Y (2003) Engineering Optimization with Particle
Swarm, in: Proceedings of the 2003 IEEE Swarm Intelligence Symposium, April
2426, IEEE: pp. 5357.
[174] Pan G, Dou Q, Liu X (2006) Performance of Two Improved Particle Swarm
Optimization in Dynamic Optimization Environments, in: ISDA 2006: 6th
International Conference on Intelligent Systems Design and Applications, October 1618, Institute of Electrical and Electronics Engineers Computer Society,
Piscataway, NJ, USA: pp. 10241028.
[175] Santana-Quintero LV, Ramirez-Santiago N, Coello CAC, Luque JM, HernandezDiaz AG (2006) A New Proposal for Multiobjective Optimization Using
Particle Swarm Optimization and Rough Sets Theory, in: Proceedings,
September 913, Springer-Verlag: pp. 483492.
[176] Ge Y, Rubo Z (2005). An Emotional Particle Swarm Optimization Algorithm,
in: Proceedings, Part III, August 2729, Springer-Verlag: pp. 553561.
[177] Grimaldi EA, Grimaccia F, Mussetta M, Pirinoli P, Zich RE (2005) Genetical
Swarm Optimization: A New Hybrid Evolutionary Algorithm for Electromagnetic Applications, in: 18th International Conference on Applied Electromagnetics and Communications and Joint COST 284 Workshop, October 1214,
IEEE: pp. 14.
[178] Shi XH, Wan LM, Lee HP, Yang XW, Wang LM, Liang YC (2003) An Improved
Genetic Algorithm with Variable Population-size and a PSO-GA Based Hybrid
Evolutionary Algorithm, in: Proceedings of the 2003 International Conference
on Machine Learning and Cybernetics, November 25, IEEE: pp. 17351740.
[179] Romeo F, Sangiovanni-Vincentelli A (1991) A Theoretical Framework for
Simulated Annealing. Algorithmica 6(3):302345.
[180] Pannetier J (1990) Simulated Annealing: An Introductory Review, in:
Invited and Contributed Papers from the Conference, March 1416, IOP: pp.
2344.
[181] Suman B, Kumar P (2006) A Survey of Simulated Annealing as a Tool for Single
and Multiobjective Optimization. Journal of the Operational Research Society
57(10):11431160.
[182] Saravanan R, Ramakrishnan K, Surender S, Dwivedi A (2001) Practical Gear
Design Optimization Using Genetic Algorithm and Simulated Annealing
Algorithm. Journal of the Institution of Engineers (India) Mechanical Engineering
Division 82(2):7479.
[183] Mipo J-, Poloujadoff M, Radulescu MM (2005) Simulated Annealing Approach
to the Design Optimization of Two-speed Induction-Motor Windings, Electromotion 12(1): 1924.
[184] Dexin X, Baodong B, Yingying Y, Fengxiang W, Mohammed OA (1995) Shape
Design Optimization in Non-linear Magnetic Problems Using Simulated
Annealing with Complex Strategy, in: IEEE International Magnetics Conference: pp. 35693571.
[185] Shiyou Y, Guangzheng N, Yan L, Renyuan T (1997) Shape Optimization of Pole
Shoes in Harmonic Exciting Synchronous Generators Using a Stochastic
Algorithm, in: 7th Conference on Electromagnetic Field Computation-CEFEC,
03 IEEE: pp. 19201930.
[186] Bai Feng-xian, Shao Yu-huai, Sun Jian-zhong (2003) Optimization Design of
Pole Shape of SRM by Using Intelligent Simulated Annealing Algorithm,
Proceedings of the CSEE, 23(1): 126131.

715

[187] Bosman PAN, Grahl J (2008) Matching Inductive Search Bias and Problem
Structure in Continuous Estimation-of-Distribution Algorithms. European
Journal of Operational Research 185(3):12461264.
[188] Bosman PAN, Thierens D (2007) Adaptive Variance Scaling in Continuous
Multi-objective Estimation-of-distribution Algorithms, in: 9th Annual Genetic
and Evolutionary Computation Conference, GECCO 2007, July 711, Association
for Computing Machinery, New York, NY, USA: pp. 500507.
[189] Zhang Q, Sun J, Tsang E (2007) Combinations of Estimation of Distribution
Algorithms and Other Techniques. International Journal of Automation and
Computing 4(3):273280.
[190] Huang F, Wang L, He Q (2007) An Effective Co-evolutionary Differential
Evolution for Constrained Optimization. Applied Mathematics and Computation 186(1):340356.
[191] Lampinen J (1999) Differential EvolutionNew Naturally Parallel Approach
for Engineering Design Optimization, in: 3rd Euro-Conference on Parallel and
Distributed Computing for Computational Mechanics, March 2025, Civil-Comp
Press: pp. 217228.
[192] Zhou S, Sun Z (2007) Survey on Estimation of Distribution Algorithms.
Zidonghua Xuebao/Acta Automatica Sinica 33(2):113124.
[193] Ueda K, Markus A, Monostori L, Kals HJJ, Arai T (2001) Emergent Synthesis
Methodologies for Manufacturing. CIRP Annals-Manufacturing Technology
50(2):535551.
[194] Teti R, DAddona D (2006) Emergent Synthesis in Supply Network Tool
Management. Advanced Engineering Informatics 20(3):233246.
[195] Ueda K, Fujii N, Inoue R (2007) An Emergent Synthesis Approach to Simultaneous Process Planning and Scheduling. CIRP Annals-Manufacturing Technology 56(1):463466.
[196] Aurich J, Fuchs C, Barbian P (2004) An Approach to the Design of Technical
Product Service Systems. Industries Management 20(5):1316.
[197] Komoto H, Tomiyama T, Nagel M, Silvester S, Brezet H (2005) Life Cycle
Simulation for Analyzing Product Service Systems, in: 4th International
Symposium on Environmentally Conscious Design and Inverse Manufacturing,
Eco Design 2005, December 1214, Inst. of Electrical & Electronics Engineers
Computer Soc., Piscataway, NJ, USA: pp. 386393.
[198] Fortnow L (2003) Introduction to Quantum Computing from the Computer
Science Perspective and Reviewing Activities. NEC Research and Development
44(3):268272.
[199] Wang L, Tang F, Wu H (2005) Hybrid Genetic Algorithm Based on Quantum
Computing for Numerical Optimization and Parameter Estimation. Applied
Mathematics & Computation 171(2):11431158.
[200] Han K, Kim J (2002) Quantum-inspired Evolutionary Algorithm for a Class of
Combinatorial Optimization. IEEE Transactions on Evolutionary Computation
6(6):580593.
[201] Huang Jian-jiang. Xu Wen-bo. Jun S, Dong Hong-wei. (2006) Study on Layout
Problem Using Quantum-behaved Particle Swarm Optimization Algorithm.
Journal of Computer Applications 26(12):30153018.
[202] Wang L, Niu Q, Fei M (2007) A Novel Quantum and Colony Optimization
Algorithm, in: International Conference on Life System Modeling and Simulation,
LSMS 2007, September 1417, Springer-Verlag: pp. 277286.

You might also like