Proximity: An Automatic Approach for Defect Detection and Depth Estimation in Infrared Non-destructive Testing

Abstract

AbstractNon-destructive testing (NDT) is a technique used to assess the integrity and reliability of industrial components without impairing their future functionalities. Thermal non-destructive testing (TNDT) is an evolving NDT technique to assess the subsurface details of various test objects. For the past two decades, non-stationary stimulation schemes have been promoting as a prominent stimulation approach by surpassing the shortfalls in conventional stimulation techniques. Nevertheless, the requirement of enhanced defect detection propelled towards fascinating post-processing research with the aid of various signal processing techniques. On the other hand, recent advancements in NDT toward automation recommended various machine learning algorithms as efficient data processing techniques in thermal NDT. However, in the thermographic data point of view, a small defect in the test sample at subsurface layers is covered by significantly fewer thermal profiles than the non-defective region. These fewer thermal profiles become local or global outliers to the thermal response of the non-defective region, depending on the depth and size of the defect. The present work introduces a proximity-based outlier detection algorithm for defect detection or classification in quadratic frequency modulated thermal wave imaging. Further, the thermal profiles of detected defects are fed to K-nearest neighbor regression model to estimate their depths. A carbon fiber-reinforced polymer sample is used to validate the proposed methodology with the aid of local and global outliers created by deeper and shallowest defects, respectively. Furthermore, thermographic and machine learning metrics are concerned to qualify the reliability of the proposed methodology.KeywordsThermal non-destructive testingQuadratic frequency-modulated thermal wave imagingProximityAn outlier