Machine learning based eddy current testing: A review

Abstract

• Provided detailed description of data sources and nature of investigations made using machine learning for Eddy current testing (ECT) systems. • Identified various types of probes/sensing elements used to collect ECT data. • Examined characteristics of ECT datasets used to train the machine learning models. • Analyzed various feature extraction approaches. • Reviewed various types of machine learning models used in different areas of investigation within ECT systems. • Evaluated performance measures, hyperparameter optimizations, cross validations, uncertainty quantification and verification and validation of machine learning models for ECT systems. • Discussed research gaps and proposed future directions. Eddy current testing (ECT) is an established non-destructive evaluation (NDE) technique to evaluate materials. In last decade, machine learning (ML) has revolutionized many areas and ECT is not an exception. The focus of ML in ECT system is to automate some of its analyses for the possible in-situ monitoring of the process and to alleviate the interpretation burden on the operator. The fusion of ML and ECT is not new, however, due to recent advancements in machine learning, there is a need to assess the current potential of ML for ECT systems and identify any gaps and shortcomings for automated data analysis. Thus, this article discusses the findings of a literature survey about the contemporary methods of using machine learning for the automatic analysis of ECT data. The application of machine learning for the ECT system is described in a general workflow manner that begins with data collection and ends with the verification and validation of the performance of ML models. Findings on potential areas of application of the fusion of ML and ECT along with limitations and potential gaps are discussed. This study also identifies the need for common datasets, sample size determination and uncertainty quantification of ML models.

Topics & Concepts