Impact damage characterization in CFRP samples with self-organizing maps applied to lock-in thermography and square-pulse shearography images

This paper proposes an approach based on self-organizing maps (SOMs) to characterize defects in optical lock-in thermography (LT) and square-pulse shearography (SPS) images of forty samples made of carbon fiber reinforced epoxy. Low-energy impact damages from 1 J to 12 J were inflicted on the samples in a controlled way. The intersection over union metric was used to evaluate the segmentation results, determining which of the configurations were the best ones for characterizing damages in the LT and SPS images. The outcomes were compared to techniques from the literature, such as principal component analysis and absolute thermal contrast-based segmentation procedures. The best configuration of the proposed approach is at least equivalent to the ones of optimal state-of-the-art tools, and provided clean and reliable segmentation results. Besides, since SOM is a single tool and has potentially a higher capability of adaptation to the image due to the training process, this shows that such technique can be useful for damage detection and description in other industrial scenarios.