A Research on State Estimation Based on Causal Inference

Fault diagnosis is to judge whether the system or equipment is faulty, and it is often necessary to consider the current specific status. In this paper, several inference modes based on causal network are analyzed in detail. Both discrete and continuous variables inference are considered for status estimation. Based on actual monitoring data of the nuclear main pump bearing system in a Pressurized-Water Reactor (PWR) nuclear power plant, a causal graph (i.e. Directed Acyclic Graph, DAG) is constructed. And, by using the proposed inference method, the operating state of a concrete bearing is inferred to diagnose the fault. Fault diagnosis and prediction based on causal network is feasible, and the inference process and results are flexible and interpretable, which has great application value and guiding significance.