Automatic defect detection based on segmentation of pulsed thermographic images

Abstract

Pulsed thermography, widely used as a nondestructive testing method, offers many advantages for material defect detection. However, most existing methods for pulsed thermographic data processing aim to enhance the defect signals in each single thermal image, whereas automatic defect detection is not achieved. Instead, laborious and time-consuming visual inspection of the processed images is required to draw final conclusions. It is usually impossible to visually inspect all images. Therefore, manual selection of a few informative images is often a required step before thermal image processing, probably resulting in the oversight of necessary defect information. To overcome the drawbacks of the existing methods, a hyper-image segmentation method is proposed in this study, which analyzes all thermographic data simultaneously to achieve automatic defect detection and avoid the risk of losing information. Specifically, an iterative defect detection procedure is designed on the basis of the Laplacian eigenmap algorithm. The results of a case study on the carbon-fiber-reinforced plastic (CFRP) materials show the feasibility of the proposed method.