A BAYESIAN METHODOLOGY USING COMBINED DUANE GROWTH MODEL PREDICTION AND WEIBULL MODEL PREDICTION TO PLAN AND EVALUATE SYSTEM RELIABILITY DEMONSTRATION TESTING

摘要

This paper develops a Bayesian methodology for deriving a reasonable estimate of reliability in a system reliability demonstration testing (or in terms of a "Marathon" in SEMI I300I Standard) based on Duane model prediction and Weibull analysis model prediction. The method uses a prior distribution for Bayesian inference process in which the Dane model and Weibull model are used by means of a heuristic combine the prediction results to generate a posterior distribution (the estimator of the true failure distribution). The effects of different modeling results to the precision of generated posterior distribution are discussed and compared to generate criteria to the confidence of posterior distribution. Our discussion is supported by simulation results.