The 7th National Conference on Computational Mechanics of Granular Materials in conjunction with The 4th International Symposium on Computational Particle Technology (CMGM-2024 & CPT-2024)

Computational microstructure characterisation and reconstruction of materials with a cementitious nature are essential for understanding their behaviour and predicting properties in the macro scale. Modelling cementitious materials with a representative spatial scale with precise characterisation has troubled researchers for decades. Although numerous physical descriptors have been applied for the characterisation of cementitious materials, only a few of them describe high-order information within the microstructure such as spatial arrangements. In this work, we demonstrated the capturing of the spatial correlation in cementitious material using deep convolutional neural networks at multiple scales based on imaging data with nanoscale resolution. Our results revealed the presence of a spatial correlation in the cementitious system and give the first indication of its distribution among the diverse features of the microstructure. We also propose functions of the discovered correlation for representative scale determination of cement materials and suggest the implications for the reconstruction of the cement microstructure based on its spatial correlation.
 

deep learning,cementitious composites