A multi-modal particle filtering method for fault detection and diagnosis of digital instrumentation and control systems

This paper introduces a novel multi-modal Particle Filtering (PF)-based method, enhanced by integrating sliding window detection and a likelihood ratio test of time series observation data, for fault detection and diagnosis in digital instrumentation and control systems. The method is applied to a digital U-shaped Tube Steam Generator (UTSG) water level control system in pressurized water reactor nuclear power plants. The system simulation and testing incorporate eight fault modes affecting various digital components such as sensors, control valves, PI controllers, and CPU modules. Comparative analyses between the multi-modal PF-based method and the Long Short-Term Memory (LSTM) network are conducted using simulation data as both training and validation sets in real-time diagnostic applications. The results demonstrate that the multi-modal PF-based method, with its precise mathematical and simulation model integration, achieves robust diagnosis coverage of transient faults and their generation mechanisms. The diagnostic accuracy and timeliness provided by the sliding window particle filtering algorithm offer significant insights into the operational efficiency and safety enhancements of digital process control systems.