Implementing Component Degradations into a Modelica Model of an iPWR System to Develop Health Monitoring Techniques

Abstract

The economic operation of small modular reactors will partly rely on managing and reducing inspection and maintenance activities while supporting new operational paradigms like load-following. Turbine control valves throttle the steam from the steam generator into the steam turbine while maintaining the pressure within the steam generator at a constant set point. Degradation of these components could impact the ability to manage electrical power production. Utilizing the Idaho National Laboratory Hybrid repository and the Oak Ridge National Laboratory TRANSFORM library developed for multiphysics simulations in Dymola/Modelica, an integral pressurized water reactor system was modeled based on the available specifications of the NuScale power module. The effects of various component degradation modes have been implemented into the model in order to simulate faulted plant data during both steady-state and load-following operations. The fault modes resemble different physical fault modes that may occur at an operating nuclear power plant; a leaking turbine control valve and a valve actuator failure due to loss of hydraulic pressure have been implemented. A neural network autoencoder is employed in conjunction with statistical analysis, namely, simple signal thresholding (SST) or sequential probability ratio testing (SPRT), to identify the presence of a fault. Fuzzy logic is additionally employed in a novel and promising manner to classify the state of the system based on the cumulative sum of the neural network residuals. SST and SPRT are both successfully validated using healthy data and proved capable of identifying both fault types; fuzzy logic identified the false positives and classified the faulted data correctly.