THE DESIGN OF EXPERIMENTS IN A MULTIRESPONSE FRAMEWORK: A REVIEW
Abdelatif OUALOUANE, Zohra CHERFI, Nassim BOUDAOUD
University of Technology of Compiègne JE ODIC
Centre de recherches Pierre Guillaumat
BP 60319 - Compiègne cedex. FRANCE
E-mail: {zohra.cherfi ; nassim.boudoud}@utc.fr
Abstract: In this paper we present a review of the different approaches in a multi
response robust design framework. The main objective of robust design is to improve the
quality of a product by minimizing the effects of variation without eliminating the causes
(since they are too difficult or too expensive to control). Different criteria are reviewed
and discussed. An illustration of the possible uses of the multi-responses optimization
techniques is presented on engine calibration problem.
Keywords: Multiresponse, Optimization, Lost Function, Desirability, Robust design,
design of experiments , Taguchi
1. INTRODUCTION
Quality and productivity improvement are most
effective when they are integrated in the product and
process development cycle. The introduction of the
statistical experimental design methodology at the
earliest stage of the product design or process
optimization is often the key to over all product
success. The purpose of its use in industrial
framework is often to improve the robustness of a
product.
The purpose of robust design is to improve the quality
of a product by minimizing the effects of variation
without eliminating the causes (since they are too
difficult or too expensive to control). The robust
design method is instituted at both the product and
process design stage to improve product
manufacturability and reliability by making products
insensitive to environmental conditions and
component variations.
In a general way, three stages are identified when
designing a new product: the system design, the
parameter design and the tolerance design.
The first step is to define the architecture of the
system such as the technological choices. The second
step, the parameter design, is the base of the robust of
design as defined by Taguchi (interested by this
work). In this step the parameters are tuned at a level
making the product less sensitive to noise. The last
step, determination of the tolerances, is performed to
provide tolerances if the robustness of the system is
insufficient.
Generally, the optimization of a product/process
performance is provided by taking into account
several objective criteria. The optimization of several
responses at the same time with the same experiments
represents a great interest, for obvious reasons of cost
of implementation.
Several criteria were developed by different authors
in order to combine the several responses:
- Loss functions (Leon.N.A 1997) ;
- Desirability functions (Castillio, 1996);
- Multi-response Signal Noise Ratio
(Logothetis,1998) ;
- Total performance index of production
(Duarte et al. 2000) ;
- Artificial Neural Network approach (David,
1996) and (Tong , 2000)
In the next sections, we will present in further details
the different approaches and focus on the main
advantages and drawbacks of their use in industrial
framework..
2. THE DIFFERENT APPROACHES
Like that was mentioned higher, the performance of a
product is often characterized by a group of responses
which are measurements of one or several
characteristics of quality. These responses are
generally correlated and can be expressed in different
measuring units. In parameter design stage, the
problem is to find the optimal levels of the parameters
by combining the different responses. Several criteria
were developed by different authors the main
difference is the choice of the optimization criteria
(loss function, desirability function, signal noise ratio,
performance index of production).
.....