Automated detection of defects in eddy current inspection data using machine learning methods

Abstract

In this study, automated detection of defects in eddy current (EC) impedance signals is explored using Machine Learning (ML) methods. A feature based classification has been proposed with three features which are derived based on similarity and closeness. XGBoost, Random Forest and Radial Basis Function Neural Network models were implemented and applied on a heat exchanger tube EC inspection data. The performance of the ML models were assessed based on the accuracy, precision and recall metrics. The ML models are able to correctly classify the occurrence of defect events in EC impedance data with an accuracy of 99%. This work lays a foundation for creating more robust and effective systems for automated detection of defects.