- MIYAZAWA Yuto, CHIBA Ryosuke, OTA Yutaro
Structural materials undergo various types of damage during use, which can lead to deterioration and failure. To prevent failure or recurrence, it is necessary to identify the causes of damage. However, in conventional damage investigations, estimations have been mostly based on the knowledge and experience of evaluators, leading to subjective and unstable results. It is expected to make generalized estimations, that are not dependent on knowledge or experience, by using machine learning image classification methods for microstructure images of damaged materials. This paper presents the prediction results of machine learning model, which accuracy was 89 % by using EBSD (Electron Backscatter Diffraction) images of three types of damaged materials (creep, creep-fatigue, fatigue). This result shows the potential for accurately estimating damage patterns through this method.