Data-based modeling for monitoring and fault detection in small modular reactors

The University of Tennessee is currently engaged in research and development projects regarding instrumentation and controls for small modular reactors (SMRs). SMRs with integral layouts present new challenges to existing instrumentation used in large reactors due to their reduced size and introduction of components such as steam generators and control rod drives to the reactor vessel. In order to overcome these challenges, alternative monitoring methods could be required as there will be less space available for sensor placement. An experimental flow control loop was developed with a submersible pump and full data acquisition system to test the effectiveness of a data-based modeling approach for monitoring process parameters. An auto-associative kernel regression (AAKR) model was trained with normal operational data from the flow loop and then faulty, user-generated data were used as a query to the AAKR model to determine if the model could identify the fault using the sequential probability ratio test (SPRT). The model uses several process variables and the current drawn from the pump motor as its input signals. Faults were applied to a flow rate sensor and to the tank water level controller to test the model's fault detection capability. It was found that the model was capable of detecting each fault with multiple predictor variables, providing a robust system for monitoring system parameters.