Unsupervised Machine Learning for Automatic Image Segmentation of Impact Damage in CFRP Composites

Abstract

In this work, a novel unsupervised machine learning (ML) method for automatic image segmentation of low velocity impact damage in carbon fiber reinforced polymer (CFRP) composites has been developed. The method relies on the use of non-parametric statistical models in conjunction with the so-called intensity-based segmentation, enabling one to determine the thresholds of image histograms and isolate the damage. Statistical distance metrics, including the Kullback–Leibler divergence, the Helling distance, and the Renyi divergence are used to formulate and solve optimization problems for finding the thresholds. The developed method enabled rigorous and rapid automatic image segmentation of the grayscale images from the micro computed tomography (micro-CT) scans of the impacted CFRP composites. Sensitivity of the segmentation results with respect to the thresholds obtained using different statistical distances has been investigated. Based on the analysis of the segmentation results, it is concluded that the Kullback-Leibler divergence is the most appropriate statistical measure and should be used for automatic image segmentation of impact damage in CFRP composites.