Framework for fault diagnosis with multi-source sensor nodes in nuclear power plants based on a Bayesian network

Abstract

Fault detection and diagnosis (FDD) provides safety alarms and diagnostic functions for a nuclear power plant (NPP), which comprises large and complex systems. Here, a technical framework based on a Bayesian network (BN) for FDD is introduced because of its advantages of easy visualization, expression of parameter uncertainties, and ability to perform diagnosis with incomplete data. However, a BN raises a new problem when it is applied to NPPs; i.e., how to cope with parameter or node information from multiple sensors. Sensor data must be consolidated because creating a single node for each sensor in the network would lead to information overload. This paper proposes a possible solution to this issue and then constructs an FDD system framework with a BN as the backbone. Within this framework, principal component analysis is used to remove information from malfunctioning sensors, and fuzzy theory and data fusion are combined to further improve data accuracy and combine data from multiple sensors into one node. On this basis, a BN inference junction tree algorithm is used in FDD because it can deal with incomplete data. A BN model for a pressurized water reactor is created to validate the method framework. Simulation experiments indicate the suitability of the proposed method for online FDD in NPPs using multi-sensor information. It is thus concluded that the proposed method is a feasible scheme for the FDD of NPPs.