Deep learning–assisted quantitative analysis of microstructural evolution in nano-reinforced cement composites under elevated temperatures

Effective modification of microstructure through nanomaterials is a key strategy for enhancing the performance of cementitious materials in high-temperature environments. However, the enhancement effects and mechanisms at different temperatures remain unclear, highlighting the urgent need for improvement of analysis in quantification, objectivity, and data throughput. This study trained a deep learning model to distinguish microstructural images of cement samples subjected to five levels of thermal treatment, with or without nanomaterial modification, achieving classification accuracies of 82.5% for temperature effects and 94.4% for carbon nanotubes (CNT) modification. Microstructure complexity initially increases and then decreases with temperature rise, but shows no explicit correlation with accuracy. Input microstructural images with scales exceeding 11.6 μm adequately capture feature information. Visualization algorithms and statistical analysis highlighted and isolated temperature and nanomaterial-related key features, revealing that thermal treatment below 400°C primarily induces localized pore expansion, while the damage accumulates rapidly and forms interconnected large-scale defects above 600°C. CNT enhancement features manifest as reduced defect frequency across all temperatures, maintaining localized microstructural reinforcement exceeding 6 μm when temperatures are above 600°C, despite limited overall mechanical performance improvement. This study establishes a connection between extreme environmental conditions and the microstructure of cementitious materials, advancing the analysis and application of nanomaterial enhancement technologies.