Proximity based automatic defect detection in quadratic frequency modulated thermal wave imaging

Abstract

Automatic and reliable anomaly detection without human intervention is a challenging task in thermal wave imaging. For this purpose, various processing algorithms proposed earlier are outperforming one another to exhibit smaller and deeper anomalies. Machine learning-based defect detection approaches are attracting the research community due to their reliable performance on employing over the stimulated thermal response in active thermography. The present article explores the deployment of local outlier factor (LOF), one-class support vector machine (OCSVM), and isolation forest (IF) algorithms over an experimental carbon fiber reinforced polymer specimen with artificially drilled flat bottom holes of different sizes at different depths to verify the detection capability of these approaches in Quadratic frequency modulated thermal wave imaging even with noisy synthetic data. In addition, a quantitative study has been performed using various thermographic and machine learning based performance metrics.