Automated crack detection on metallic materials with flying-spot thermography using deep learning and progressive training

ABSTRACTIn non-destructive testing for metallic materials, ‘Flying-spot’ thermography allows the detection of cracks thanks to the scanning of samples by a local laser heat source observed in the infrared spectrum. However, distinguishing a crack from other surface structures such as air ducts or non-planar shapes on the material surface can be challenging in an automation perspective. To address this, we propose to use deep learning techniques, which can exploit contextual information but require a significant amount of labelled data. This study presents a training method based on curriculum learning and recent denoising diffusion models to generate synthetic images. The protocol progressively increases the complexity of training images, using successively simulated data from a multi-physics finite-element software, synthetically generated data with diffusion process, and finally real data. Several detection scores are measured for various machine learning and deep learning architectures, demonstrating the benefits of the proposed approach for regular application cases and degraded experimental conditions, consisting of limited thermal enlightenment recordings.KEYWORDS: Non-destructive testingflying-spot thermographydeep learningcurriculum learningdenoising diffusion models Disclosure statementNo potential conflict of interest was reported by the author(s).Additional informationFundingThe work was supported by the Agence de l’innovation de Défense.Notes on contributorsKevin HelvigKevin Helvig is a Ph.D. student currently doing research at ONERA Palaiseau in France. His work is dedicated to the application of computer vision techniques to laser thermography for non-destructive materials testing, in particular exploring the coupling between active IR and visible spectrum examinations. He is graduated with an engineering degree from IMT Mines Albi, specializing in non-destructive testing and materials.Pauline Trouvé-PelouxPauline Trouvé-Peloux after completing her engineering training in optics at the Institut d'Optique Graduate School, Pauline Trouvé-Peloux obtained her doctorate in Information and Mathematics Science and Technologies from the Ecole Centrale de Nantes in 2012, specializing in signal and image processing. Since 2012, she has held the position of research engineer at ONERA, within the Information Processing and Systems Department (DTIS). Her research activities focus on the joint design, or co-design, of an imager through joint optimization approaches of its optics and processing parameters. The application areas of her work particularly concern compact 3D sensors for robotics or industrial inspection.Ludovic GaverinaLudovic Gaverina graduated with an engineering degree from Telecom Saint-Etienne and a master of research in optic, image, and computer vision from Jean Monnet University in 2013. In 2017, he received the PhD degree (title of his thesis: “Thermal characterization of heterogeneous material by flying spot laser and infrared thermography”) in heat transfer from Bordeaux University under the supervision of Christophe Pradère. He currently works at ONERA, within the Materials and Structures Departement, focusing on automated multiphysics non-destructive testing (NDT) techniques.Baptiste AbeloosBaptiste Abeloos is a Research Scientist at The French Aerospace Lab ONERA, within the Information Processing and Systems Departement. His PhD thesis, titled ”Searches for Supersymmetry in the Fully Hadronic Channel and Jet Calibration with the ATLAS Detector at the LHC,” focused on enhancing jet energy measurement accuracy and propelling the search for supersymmetry, aiding in extending the known limitations on squark and gluino masses. Currently, he works on deep learning techniques for explainability, vision-language models, and non-destructive testing.Jean-Michel RocheJean Michel Roche is a senior research scientist in ONERA, leader of the R&D team dedicated to non-destructive testing and structural health monitoring. He graduated from Ecole Centrale de Lyon in 2007, specializing in aeroacoustics, and successfully defended his PhD thesis in the field of the absorption of resonant liners, in 2011. Since then, he has been working on thermal-based approaches to detect defects in aeronautic structures.