Analytical model approximation for defect classification in fiberglass composites inspected by long-pulse thermography

This paper presents a complete pipeline for automatic detection and classification of defects within composite laminates inspected by active IR thermography. Specifically, long-pulse thermography is proposed for nondestructive evaluation of samples made of Glass Fiber Reinforced Polymer (GFRP). A model approximation based on exponential functions is used to achieve an efficient representation of temperature decays at the surface of the samples. At the end of the pipeline, several decision forests are implemented to process input features and label corresponding areas among three classes of interest: sound regions, surface defects, and in-depth discontinuities. Results prove that the proposed methodology performs with good accuracy also in case of inspection of GFRP samples tested by long-pulse thermography.