Bayesian Inference of Fatigue Life Estimated by Inspection Data
Shinsuke SAKAI, Satoshi OKAJIMA, Satoshi IZUMI and Atsushi IWASAKI
Abstract:Recently, importance of risk assessment has greatly increased. However, data of failure probability do not exist sufficiently in our country. Therefore, the method to obtain the probability of failure from inspection data is required. This paper investigates the method to estimate the mother distribution of failure life of structures by inspection data using Bayesian inference. Though Bayesian inference is widely said to be effective for the estimation from small samples, the effectiveness greatly depends on the initial setting of a prior distribution. In this paper, the applicability of Bayesian inference is examined first. For this purpose, Bayesian inference is applied for estimating fatigue life by the inspection data of boiler parts. Next, influence of the initial prior distribution on the results is studied systematically by giving variation to mother parameters. Weibull distribution is assumed for the fatigue life distribution and the investigation is made on scale parameter and shape one. In order to assist the inspection planning, the estimations of MTBF and hazard rate are also investigated. As the result, the relation between the initial prior distribution and the effective range by Bayesian inference is clarified. Key Words:Bayesian inference, Weibull distribution, Fatigue life estimation, Mean time between failure, Hazard rate, Inspection planning, Risk assessment