Non-destructive defect evaluation of polymer composites via thermographic data analysis: A manifold learning method

Abstract

Recently, various thermographic data analysis methods have been utilized in the field of non-destructive evaluation (NDE) to process thermal images and enhance the visibility of defects. However, most of them extract only linear features, leading to cumbersome results. In this work, manifold learning is introduced into the thermographic data analysis field. As a nonlinear dimensionality reduction technique, manifold learning can identify an intrinsically low-dimensional manifold in a high-dimensional data space. Specifically, an isometric feature mapping (ISOMAP) based manifold learning thermography (MLT) method is proposed to analyze the thermographic data, which can effectively distinguish the uneven background, noise, and defect characteristics contained in thermal images and make the defect detection easier. The feasibility of MLT is illustrated using a carbon fiber-reinforced polymer (CFRP) specimen. The results show that, comparing to the conventional linear methods, the present method can better determine the defect information, including the positions, sizes, and shapes.