Abstract
CRUD depositions on fuel cladding are generated as a result of coolant chemistry and corrosion of primary system components during the normal operation of a nuclear reactor. These depositions can affect both reactor power peaking and fuel element structural integrity through CRUD Induced Power Shifts (CIPS) and CRUD Induced Local Corrosion (CILC) phenomena, respectively. The CIPS/CILC phenomena are typically assessed using the Boron-induced Offset Anomaly (BOA) computer code, which predicts the characterization of the distribution of CRUD within a reactor and the risk for occurrence of CIPS/CILC within a given cycle.
Machine learning (ML) algorithms have often been applied towards understanding large data sets and acquiring useable information in many industries, including nuclear. In the nuclear engineering field, ML is increasingly being used as a method to improve nuclear power plant efficiency and reduce engineering costs. The application of machine learning to enhance the prediction of CIPS/CILC increases the possibility of reducing the cost associated with CRUD mitigation. In Westinghouse, exploratory analysis was performed to identify ways in which ML could be utilized to improve CIPS risk assessments. There were two paths identified, the first one was ML applications for CIPS with the standalone BOA code, and the second path was ML applications for CIPS calibration with coupled ANC/FUELUDTYDRV/VIPRE-W/BOA code. The exploratory study demonstrated that ML was capable of accurately predicting the boron mass in the cores of selected nuclear power plants as surrogate models of the BOA code.
