A comparison of Bayesian analysis methods for reliability parameter estimation in PSA

Parameter estimation is an important constituent of data analysis in Probabilistic Safety Analysis (PSA) for nuclear power plants. Bayesian analysis techniques have become industry standard tools for uncertainty analysis and for updating reliability parameter estimates. This paper considers the relative accuracy of different Bayesian methods for reliability parameter estimates, and two established methods are examined. "One stage" Bayesian analyses use data observed from a single plant, in conjunction with a prior distribution. "Two stage" Bayesian analysis has a more complex underlying model, and allows for the explicit incorporation of observed data from diverse sources in order to estimate reliability parameters. One and two stage Bayesian analyses are reviewed and their application to data is demonstrated. To evaluate the accuracy of the two Bayesian methods of parameter estimation, Monte Carlo simulation is used to create data on theoretical plant component failures. One stage and two stage Bayesian analyses are then performed on the simulated data allowing the accuracy of the two analysis methods to be compared. The results are analysed in order to evaluate the applicability of the different Bayesian methods to estimation of plant reliability parameters.