Test Selection Design with Unreliable Tests Based on Bayesian Network and SA-GA

Testability is a crucial issue in the field of fault detection and diagnosis (FDD). Testing every parameter of a system is an unacceptable option in practice, because it often generates an overwhelming amount of data with sparse value density and costs a lot of time and resource consumption. Optimal test selection design is essential for effectively performing FDD strategies. This paper focuses on the unreliable test problem in testability to minimize the test cost subjected to lower constraints on fault detection, and fault isolation. Expectation-maximization and Bayesian network are applied to solve the missed detection issue. To avoid the local optimum phenomenon, a combination of simulated annealing algorithm and genetic algorithm method is adopted to search the optimal test sequence. The proposed optimal test selection design scheme under unreliable tests is implemented on the Automatic Gauge Control (AGC) simulation platform in Baosteel Co., Ltd., Shanghai, China. The results show that the obtained optimal test sequence can greatly reduce the test cost, and ensure good fault detection and isolation performance.