Defect detection in CFRP combining SK-means clustering and probability of detection analysis

In this work, an improved K-means clustering algorithm, termed subpart K-means, has been proposed. Carbon fibre reinforced polymer specimens with artificial flat bottom holes were tested using pulsed thermography. The subpart K-means algorithm was used to detect subsurface defects within the specimens. A comparative study was conducted against conventional image processing techniques. Finally, the probability of detection analysis was performed to evaluate the accuracy of the subpart K-means algorithm in detecting defects. The results indicate that the subpart K-means algorithm demonstrates superior capability in defect detection. The subpart K-means method, integrated with probability of detection analysis, offers a more effective approach for pulsed-thermography-based defect detection in carbon fibre reinforced polymer specimens.