Nanomodification analysis of pore structure in GO-enhanced CWRB based on metal intrusion and BSE imaging with deep learning

Understanding the microstructural reinforcing mechanism benefits for graphene oxide (GO) on cemented waste rock backfill (CWRB) strengthening. However, quantitatively characterizing the reinforcing effects of GO and locating the modified nano/microscale features remain critical challenges due to the disorderliness of the composites. This work proposes an innovative approach based on metal intrusion technology, backscattered electron (BSE) images, and deep learning to analyze the micro/nanoscale GO-modified characteristics of the microstructure of CWRB. The results imply that by nucleation and pore-infilling effects, GO can promote the hydrate reaction and cooperate with the generated hydration products to split the large pores into independent units, thus optimizing the microstructure of CWRB. The reinforcing effects of GO tend to be more efficient under a low Talbot grading index. The proposed BSE characterization combined with the deep learning-based approach can achieve up to 91 % recognition accuracy to identify the GO-reinforced specimens. The deep Taylor decomposition (DTD) algorithm successfully locates the reinforced characteristics of the GO modification in CWRB specimens under the resolution of 340 nm. The extracting feature analysis proves the GO reinforcement is inclined in the matrix's ITZ, with about 10.3 % optimizing efficiency improvement compared with the other regions. This study not only boards the understanding of the GO reinforcing mechanisms in cement composites but could also provide insights into GO modification for structural application.