An ensemble data-driven method for fault detection and diagnosis of digital control systems in nuclear power plants

Fault detection and diagnosis (FDD) is essential for maintaining safety and preventing hazardous situations in industrial process control. Effective fault diagnosis allows for the timely detection and correction of anomalies, preventing potential disruptions and maintaining optimal performance. In the paper, we present a unified framework for fault detection and diagnosis by combining the real-time sensitivity of the moving window particle filtering (PF) with the diagnostic precision of the generalized likelihood ratio test (GLRT). Within the framework, the particle filtering is integrated to provide accurate real-time state monitoring and prediction in scenarios with nonlinear digital control system dynamics and non-Gaussian noise. The moving window (MW) is adopted to identify anomalous patterns within a stream of data by focusing on a fixed-size segment that moves across the data. The GLRT is then used to isolate the specific type of fault that has occurred based on the observed data and the different fault hypotheses and models. The method is demonstrated with a digital U-shaped tube steam generator water level control system in pressurized water reactor nuclear power plants. Comparative studies have also been conducted with LSTM network to demonstrate the effectiveness and superiority of the proposed PF-based MW-GLRT method. The demonstration results show that the proposed PF-based MW-GLRT framework can provide a robust and efficient solution for identifying and characterizing faults in complex digital control systems.