Application of image processing and machine learning for classification of laser-induced damage morphology

Interest in qualitative analysis of damage morphology of laser-induced damage test sites has increased in recent years. Such analysis can potentially provide valuable information about underlying damage mechanisms and can be used for separation of different damage modes. However, morphological analyses are currently performed manually and only on a few test sites at a time. In this work, a novel computational approach to the analysis of damaged test sites is presented. Image processing algorithms were applied to images of test sites in order to identify damaged test sites and extract features of damage morphology. Unsupervised machine learning was performed to automatically cluster damaged test sites. It was shown that ZrO2 single layer's laser-induced damage can be separated into well defined clusters. The clusters were grouped to distinct catastrophic and color-change modes. Characteristic damage curves of different damage modes were investigated to reveal different fatigue behavior.