Machine Learning in Nondestructive Testing

Abstract At nuclear power plants (NPPs), critical structures, systems, and components must comply with national codes and standards. To ensure this compliance, an In-Service Inspection (ISI) program is mandatory. The ISI program relies on Nondestructive Testing (NDT), which uses various methods and techniques to verify the quality and integrity of critical structures, systems, and components. One of these methods is phased array ultrasonic testing (PAUT), which can sometimes produce many indications that require evaluation. The final assessment of these indications is crucial for the continued safe operation of the NPP. This process can be time-consuming and demanding for NDT personnel, especially during plant shutdowns. To support this task, artificial intelligence can be used to assist NDT personnel in the evaluation of PAUT data. This article presents an approach for applying machine learning (ML) techniques to train a neural network for this purpose.