Automatic IRNDT inspection applying sparse PCA-based clustering

Recent progress in Thermal and infrared Non-Destructive Testing (IRNDT) in different fields have provided interesting defect detection solutions. Principal Component Analysis (PCA) based K-means clustering have been successfully introduced and used in many clustering applications. However, PCA suffers from being relatively more sensitive to the noise due to having a linear transformation. On the other hand, Sparse Principal Component Analysis (SPCA) has a superior performance in relation to noise because of l1 and l2 norm additional terms which increase its robustness. As such, we propose SPCA based K-means clustering for defect segmentation. Principal Component Thermography (PCT) and Candid Covariance-Free Incremental Principal Component Thermography (CCIPCT) are also used as a pretreatment of data in order to reduce the number of components. Three types of specimens (CFRP, Plexiglass and Aluminum) have been used for comparative and quantitative benchmarking even in the case of adding Gaussian noise (0%– 35%). The results conclusively indicate the promising performance and confirmed the outlined properties.