The effectiveness of pulsed thermography (PT) for detecting delamination in carbon fiber-reinforced polymer (CFRP) plates has been widely verified. However, delaminations are usually characterized by weak visibility due to the influences of inspection factors and the delaminations with weak visibility are easily missed in real inspections. In this study, by introducing a deep learning algorithm-stacked autoencoder (SAE)-to PT, we propose a novel approach (SAE-PT) to enhance the visibility of delaminations. Based on the ability of SAE to learn unsupervised features from data, the thermal features of delaminations are extracted from the raw thermograms. The extracted features are then employed to construct SAE images, in which the visibility of delaminations is expected to be enhanced. To test the performance of SAE-PT, we inspected CFRP plates with prefabricated delaminations. By implementing SAE-PT on the raw inspection data, the delaminations were more clearly indicated in the constructed SAE images. We also compare SAE-PT to the widely used principal component thermography (PCT) method to further verify the validity of the proposed approach. The results reveal that compared to PCT, SAE-PT can show delaminations in CFRP with higher contrast. By effectively enhancing the delamination visibility, SAE-PT thus has potential for improving the inspection accuracy of PT for non-destructive testing (NDT) of CFRP.
Keywords: carbon fiber-reinforced polymer; delamination detection; pulsed thermography (PE); stacked autoencoder (SAE).