Crack angle estimation with induction thermography and machine learning

Abstract

Induction thermography is a well-established method for detecting and analysing cracks in metal products, such as rails. However, quantifying defects, particularly those with complex geometries, remains a challenging and intricate task. This paper addresses one critical aspect of defect quantification: the determination of crack inclination angles, which is essential for accurate depth estimation and hazard level assessment. We propose a novel approach that combines induction thermography data analysis with machine learning regression models to estimate crack angles. The regression model is trained on a dataset generated through numerical simulations, ensuring robust and reliable performance. The effectiveness of the proposed method is demonstrated through both numerical and experimental results, showcasing its potential for improving crack characterization in industrial applications. This work advances the field of non-destructive testing by providing a more precise and automated solution for crack inclination angle determination, contributing to enhanced structural integrity assessments