이미지 데이터 증강 및 설명가능 인공지능을 이용한 반사경 구조의 변형 특성 평가

Large-aperture mirrors are the main optical elements of artificial satellite telescopes, and they require high optical performance. However, their optical performance is determined by wavefront error (WFE), which deteriorates rapidly even under a small load. The current evaluation process of WFE is complicated, where early performance evaluation is required. In this study, early evaluation of large-aperture mirror was performed using explainable artificial intelligence, gradient-weighted class activation mapping (Grad-CAM). A small amount of displacement-based image data was collected and classified by its design variables and performance. Then, the dataset was augmented and applied to Grad-CAM. Visual results and model accuracy for early performance evaluation were derived by comparing the results with analysis of means (ANOM). The results showed that it is effective to adjust the inner pattern thickness. Specifically, a thickness of 4 mm leads to optimal performance.