Automated delamination detection in CFRP using flash infrared thermography and deep learning method

Abstract: Carbon fiber reinforced polymer (CFRP) composites have become an important material for many industry applications due to their high specific stiffness and specific strength. Inevitably, defects generated during manufacturing and/or inservice process could compromise the structural health of the CFRP components. Flash infrared thermography(IRT) is a promising NDT technique, which has already been successfully applied for the detection of various defects in a range of materials. In order to bring this technique to the next level, this study aims to accomplish automatic detection and localization of delaminations in CFRP using flash IRT experiments and deep learning-based object detection method. A virtual dataset has been generated by means of a custom developed fast numerical simulator (programmed in Fortran) of flash IRT. Then, a pre-trained object detection framework from literature, i.e. Faster-RCNN, is applied to the virtual dataset using the concept of transfer learning. A comparison with classical post-processing methodologies, e.g. thermographic signal reconstruction, and principal component thermography, is presented. Finally, a CFRP slab including twelve artificial rectangular delaminations was inspected using flash IRT and evaluated through trained Faster-RCNN.