A Utility-Driven Bayesian Design: A New Framework for Extracting Optimal Experiments from Observational Reliability Data
Open Access
Abstract
In this study, a procedure to build Bayesian optimal designs using utility functions and exploiting existing data is proposed. The procedure is illustrated through a case study in the field of reliability, by applying a hierarchical Bayesian model and performing Markov Chain Monte Carlo simulations. Two innovative contributions are introduced: (i) the definition of specific utility functions that involve several key issues and (ii) the use of observational data. The use of observational data makes it possible to build the optimal design without additional costs for the company, while the definition of the utility functions accounts for the specific characteristics of the reliability study. Features like model residuals, i.e., discrepancies between observed and predicted response values, and the costs of the electronic component are addressed. Costs are also weighted considering the environmental impact. Satisfactory results are obtained and subsequently validated through an in-depth sensitivity analysis.Keywords
This publication has 18 references indexed in Scilit:
- Optimal Bayesian experimental design for subsurface flow problemsComputer Methods in Applied Mechanics and Engineering, 2020
- A Comparison of Alloy-Surface Finish Combinations Considering Different Component Package Types and Their Impact on Soldering ReliabilityIEEE Transactions on Reliability, 2015
- T‐optimality and neural networks: a comparison of approaches for building experimental designsApplied Stochastic Models in Business and Industry, 2012
- An Optimal Experimental Design Criterion for Discriminating Between Non-Normal ModelsJournal of the Royal Statistical Society Series B: Statistical Methodology, 2007
- A new metric for probability distributionsIEEE Transactions on Information Theory, 2003
- Divergence measures based on the Shannon entropyIEEE Transactions on Information Theory, 1991
- The design of experiments for discriminating between two rival modelsBiometrika, 1975
- Bayes Estimates for the Linear ModelJournal of the Royal Statistical Society Series B: Statistical Methodology, 1972
- Optimum Experimental DesignsJournal of the Royal Statistical Society Series B: Statistical Methodology, 1959
- A Statistical Distribution Function of Wide ApplicabilityJournal of Applied Mechanics, 1951