低線量X線CTの機械学習デノイズによる炭素繊維の界面き裂抽出

抄録

Synchrotron radiation X-ray computed tomography (CT) is an effective tool for visualizing individual carbon fibers and interfacial cracks in them. The propagation behavior of matrix cracking under cyclic loading can be monitored by repeatedly conducting CT imaging on the same sample. However, the potential deterioration of mechanical properties due to X-ray irradiation is a concern. This study aims to evaluate low-dose X-ray CT. Thus, we investigated the effectiveness of the image-denoising capabilities of Gaussian filters and deep-learning algorithms. A comparative analysis was conducted on reconstructed CT images before and after denoising under several conditions, each with a different exposure time and projection number. Maintaining the number of projections and reducing the exposure time yielded a clear reconstructed image. Additionally, binarized images of carbon fibers and matrix cracking were analyzed, which reveal that utilizing deep-learning methodologies resulted in effective noise reduction.