Exploring the Correlation Between Microstructure and Compressive Strength in Portland Cement-Based Materials: A Deep Learning-Based Comparison of μCT and SEM-BSE Microstructure Images

The compressive strength of portland cement-based material is closely linked with its microstructure. Scanning electron microscopy (SEM) with backscattered electrons (BSE) and x-ray microcomputed tomography (μCT) are capable of visualizing the microstructure. Processing microstructural images using techniques such as threshold segmentation relies heavily on expert knowledge and is prone to bias. This study uniquely employs convolutional neural networks to comparatively analyze the correlation between compressive strength and microstructural images obtained from μCT and BSE. The proposed method can automatically extract features from the images. The findings indicate that BSE images exhibit a stronger correlation with compressive strength than 2D μCT images. Under water-to-cement ratios of 0.25 or 0.7, the correlation between 2D μCT images and compressive strength decreases with increasing curing age. Additionally, the 3D spatial information in 3D μCT images is also correlated with compressive strength, while its correlation is lower than that of SEM-BSE images. This comparative study highlights the advantages and limitations of SEM-BSE and μCT in microstructural analysis of portland cement-based materials, offering new insights into understanding compressive strength through deep learning techniques.