Srikanth Doraiswamy and S. Krishnamurty, Bayesian Analysis in Engineering Model Assessment, Design EngineeringTechnical Conference 2000 ,Design Theory and Methodology-14546
This paper addresses issues relating to the study of model uncertainty
and its impact in the design decision-making process. A structured decision-based
methodology is presented for the assessment of engineering models, leading
to the identification of the best model from a design decision perspective.
Engineering design decisions often need to be made when complete knowledge of the system is not available. Getting such complete deterministic information may be impossible or impractical as the resources required to get the information may be unaffordable. The challenge then is to identify and construct an information-gathering strategy, one that can be expected to yield most reliable results in an efficient manner during design. In this context, an engineering model can be viewed as an information-gathering strategy, using which predictive information regarding the performance of a design can be acquired by means of mathematical simulation studies.Focusing on this issue, this paper presents the development of a Bayesian analysis based model selection strategy to understand and deal with the uncertainty inherent in engineering models. Specifically, this work offers a basis for assessing engineering models under conditions of uncertainty through a methodical generation of the required probabilities in a Bayesian tree format, and subsequently using it in the selection of the best analysis model from a design decision perspective. This work also extends the use of decision trees to handle multiple attributes effectively. The resulting tree structure can be construed to be a graphical representation of utility-based decision model building. A model selected by using this methodology will be an optimal trade-off between the value of information and the cost of gathering that information under conditions of uncertainty.Modeling and analysis of a windshield-wiper arm is used to illustrate the application of the proposed methodology and the results are discussed.
Keywords: Engineering Design, Bayesian Analysis, Finite Element Analysis, Expected Utility Theory