Using model-based fault detection to differentiate transients and loss of coolant accidents

In the nuclear power industry, one important class of accidents is the loss of coolant accident (LOCA). This paper presents methods to detect a LOCA that is initiated: (i) while the plant is going through a small transient, and (ii) with a time-varying leak magnitude. The accident is simulated using a generic pressurized water reactor (GPWR) simulator. The fault is detected using a model-based approach with models identified using GPWR data. The model-based approach is multiple-model adaptive estimation (MMAE), which uses multiple system models representing both normal and faulted operating conditions. During operation, these models simulate the potential operating conditions, incorporating measurement feedback in a Kalman filter state-estimation structure. Faults are detected by selecting the model that most closely matches the system according to statistical characteristics. For a LOCA, data-driven models of the pressurizer liquid level are derived using first-principles and system identification. In system identification, a physics-based model form is derived that contains unknown parameters. System identification is then used to estimate the parameter values based on measurement data, providing plant-specific pressurizer models. For the accident scenario described above, the proposed methods differentiate between the transient and the accident, and provide real-time estimates of the leak magnitude after it has been initiated.