Generation of Ultrasonic Signals for Aged Materials through Machine Learning

Abstract

This study aims to generate ultrasonic signals of aged nuclear structural materials through machine learning models. Ultrasonic signals were generated using a variational auto-encoder model with a mathematically explainable and reproducible method. Two machine learning models for classification (k-nearest neighbor and multi-layer perceptron) were used to verify the quality of the generated signals. It was confirmed that the generated ultrasonic signals possessed characteristics similar to those of the experimental data, which implies that a large number of new ultrasonic signals can be generated using a small amount of data collected through experiments.