BAYESIAN-NETWORK-BASED HYDRO-POWER FAULT DIAGNOSIS SYSTEM DEVELOPMENT BY FAULT TREE TRANSFORMATION

Abstract

Currently, fault diagnosis of reservoir facilities relies mostly on check-list evaluation. The results and qualities of evaluation are limited by experience and ability of the evaluators, which may not achieve the goal of systematic assessment in a consistent manner. To overcome the limitation of the traditional approach, this research develops a fault diagnosis and evaluation system for reservoir facility by utilizing multi-state Fault-Tree Analysis (FTA) technique, in conjunction with Bayesian Networks (BN) which incorporate expert experiences through lateral linkages among BN nodes and weighting factors. The system has been used to analyze and verify against three hydro-power systems currently in operation. It was found that through BN analysis the fault trend is consistent to that from historical data analysis via Weibull distribution. This indicates that the transformation of a multi-state Fault-Tree (FT) and BN is reasonable and practical. Based upon the analysis of BN by inputting prior information of the hydro-power systems, the probabilities of fault occurrences are effectively computed based on which proper preventive maintenance strategies can be established.