Generative manifold learning thermography for non-destructive evaluation of defects in composite materials

Abstract

In the non-destructive evaluation of infrared thermography, the thermographic data modeling and analysis steps play an important role in improving the inspection results. However, thermal image analysis still faces challenges such as a small number of informative images and difficulty in extracting defect features. In this work, a novel generative manifold learning thermography (GMLT) method for defect detection of composite materials is proposed. In detail, the spectral normalization generative adversarial network is used as a data augmentation strategy to generate more informative thermal images. Sequentially, the MLT-based thermographic data analysis method is adopted to extract and visualize defects in the thermal images. Experiments on carbon fiber reinforced polymers verify the effectiveness and advantages of the proposed method. Key Words: non-destructive evaluation, generative adversarial network, manifold learning, thermographic data analysis, carbon fiber reinforced polymer.