Infrared thermography (IRT) is an economical nondestructive testing technique for structural health monitoring of composite materials. However, the nonlinear nature of the thermographic data and the adverse effects of noise and inhomogeneous backgrounds prevent it from achieving satisfactory results. Most of the existing thermographc data analysis methods are supervised and/or linear, which therefore are not favorable for nonlinear feature extraction of unlabeled thermograms. In this work, a deep autoencoder thermography (DAT) method is proposed for detecting subsurface defects in composite materials. The multi-layer network structure of DAT can handle nonlinear temperature profiles, and the output of the intermediate hidden layer is visualized to highlight defects. The layer-by-layer feature visualization reveals how the model extracts defect features. A loss inflection point scheme is utilized to determine a suitable depth of the model. Moreover, a new quantitative index is proposed to compare the defect detectability of different methods.

The subspace learning (dimensionality reduction) algorithms have played an important role in the analysis of thermographic data: a key step in infrared thermography-based nondestructive testing of subsurface defects in composite materials. However, one of its branches, manifold learning, with excellent ability to preserve local data structure, is rarely applied. Here, a spatial-neighborhood manifold learning (SNML) framework is proposed for thermographic data analysis. Different from traditional manifold learning methods, SNML uses the spatial-neighborhood information instead of the traditional k-nearest neighbors, or epsilon-neighborhood, to construct the adjacency graph. This overcomes the difficulty of parameter selection and extracts local features in images in a more reasonable way. Additionally, the data preprocessing step and the means of thermographic data normalization in the proposed framework are discussed. For performance comparison, three traditional manifold learning methods are also implemented. The experiments on carbon fiber reinforced polymer specimens demonstrate the validity and feasibility of SNML.

Thermographic data processing and analysis is critical for effective infrared thermography non-destructive testing of defects in composite materials. In this research work, the concept of blind source separation is introduced into this field for facilitating subsurface defect detection in polymer composites. In detail, a thermal image generated by infrared thermography is viewed as a linear mixture of several independent and non-Gaussian sources, which can be decomposed by adopting the proposed independent component thermography (ICT) technique. After the decomposition, the non-Gaussian sources are extracted and plotted as a number of images with the same size as the original thermograms. Eventually, the defect signals can be separated from the non-uniform backgrounds which are usually caused by uneven heating. As a result, the location and shape information of the subsurface defective regions is highlighted. The feasibility of the ICT method is illustrated with its application to a carbon fiber reinforced polymer specimen.

Increasing machine learning methods are being applied to infrared non-destructive assessment for internal defects assessment of composite materials. However, most of them extract only linear features, which is not in accord with the nonlinear characteristics of infrared data. Moreover, limited infrared images tend to restrict the data analysis capabilities of machine learning methods. In this work, a novel generative kernel principal component thermography (GKPCT) method is proposed for defect detection of carbon fiber reinforced polymer (CFRP) composites. Specifically, the spectral normalization generative adversarial network is proposed to augment the thermograms for model construction. Sequentially, the KPCT method is used by feature mapping of all thermogram data using kernel principal component analysis, which allows for differentiation of defects and background in the dimensionality-reduced data. Additionally, a defect-background separation metric is designed to help the performance evaluation of data analysis methods. Experimental results on CFRP demonstrate the feasibility and advantages of the proposed GKPCT method.