Case Study on Bayesian Reliability Estimation of Software Design of Motor Protection Relay

Abstract

A case study on the reliability estimation of software design of a motor protection relay is presented. The case study is part of a long-term research effort to develop methodology and support for the reliability estimation of computer-based systems to be used in the safety-critical applications of nuclear industry. The estimation method is based on Bayesian inference and the case study is a follow-up to previous case study presented in SAFECOMP 2003.

In the case study reliability estimate of the protection functions of the relay is built in a sophisticated expert judgement process. The expert judgement process consists of two phases including several sessions where the relay designers from different development stages participated. The sessions are named according to the phases as qualitative and quantitative sessions. The qualitative sessions are used to identify and record possible uncertainty and unpunctuality in the planning and documentation of the software design. The quantitative sessions are used to analyse the recordings and to generate a prior reliability estimate. Finally, the prior estimate is updated to a posterior estimate using the operating data of the relay.

The estimation demonstrates the excellence of Bayesian modelling in the reliability estimation of computer-based systems. The reliability estimation typically involves evidence of different kind and with Bayesian modelling the evidence can be combined coherently and transparently together. The estimation method is particularly attractive for probabilistic safety assessment (PSA) of nuclear industry. The method provides informative posterior probability distributions on the failure rates of the protection functions to be used in the PSA models.