Thermographic clustering analysis for defect detection in CFRP structures

Abstract Owing to the high cost of carbon fiber, pulsed thermography (PT) has been adopted for defect detection in carbon fiber-reinforced plastic (CFRP) structures. In previous studies on this topic, there were two major problems. First, analyzing thermal images by visual inspection alone is laborious and time-consuming. Second, a few thermal images selected from the thermographic dataset may not contain all necessary defect information. To solve these problems, this study proposes a thermographic cluster analysis (TCA) method for automatic defect detection based on three-dimensional image segmentation. Specifically, the minimum spanning tree (MST) clustering algorithm is adopted to take both temperature differences and spatial distances between pixels into consideration. The proposed method makes use of all thermal images in the thermographic dataset, with no prior knowledge needed for image selection. Compared with conventional methods, TCA more accurately identifies the shape of each defective region in an automatic way.