A digital twin approach to system-level fault detection and diagnosis for improved equipment health monitoring

Abstract

Automating the task of fault detection and diagnosis is crucial in the effort to reduce the operation and maintenance cost in the nuclear industry. This paper describes a physics-based approach for system-level diagnosis in thermal–hydraulic systems in nuclear power plants. The inclusion of physics information allows for the creation of virtual sensors, which provide improved fault diagnosis capability. The physics information also serves to better constrain diagnostic solutions to the physical domain. As a demonstration, various test cases for fault diagnosis in a high-pressure feedwater system were considered. The use of virtual sensors allows constructing performance models for two first-point feedwater heaters which would not have been possible otherwise due to the limited sensor set. Real-time plant data provided by a utility partner were used to assess the diagnostic approach. The detection of an abnormal event immediate after a plant startup pointed to faulty behaviors in the two first-point feedwater heaters. This double-blind fault diagnosis was subsequently confirmed by the plant operator. In addition, several simulated sensor fault events demonstrated the capability of our algorithms in detecting and discriminating sensor faults.