Characterization of the impact of graphene oxide on UHPC pore structure and strength via deep learning-based image analysis

Understanding the dispersion characteristics of graphene oxide (GO) in ultra-high-performance concrete (UHPC) and elucidating its reinforcement mechanisms are prerequisites for further enhancing UHPC performance and promoting its widespread industrial application. However, due to the nanoscale dimensions, extremely low content, and opacity of cementitious materials, comprehensively analyzing GO's impact on pore structure poses significant challenges. In this study, We adopt the sol-gel method (RCGO), the three-step dip coating method (SCGO), and the physical agitation method (BCGO) coated GO on steel fibers to optimize UHPC. In addition, we present an integrated approach combining metal infiltration, backscattered electron (BSE) characterization, and deep learning analysis to investigate GO-enhanced UHPC. After GO enhancement, the porosity of the UHPC matrix and the interfacial transition zone (ITZ) decreased significantly, ranging from 12.6% to 32.6%. Classification recognition deep learning was used to extract the pore structure model of locally enhanced regions, achieving an average accuracy of 95%. Analysis showed that the three step dip coating method improved by up to 27.2%. More importantly, the linear fitting relationship between macroscopic mechanics and microscopic porosity is established. The high goodness fitting between compressive strength and porosity reached 0.85. GO significantly influences the pore structure of UHPC within the specific pore size range of 10-15μm. The fitting goodness of pore size distribution to porosity in this range reaches 0.915, indicating that pores in this size range play a crucial role in determining the mechanical strength of UHPC. This study provides new avenues for a deeper understanding of the GO reinforcement mechanisms and offers scientific insights and technical support for the preparation and performance optimization of UHPC. • The pore size distribution of GO-reinforced UHPC is accurately characterized. • The deep learning-based method recognized the GO reinforcing characteristics. • The GO enhancement and optimization of UHPC pore structure are confirmed. • A close fit between UHPC pore structure and mechanical strength is achieved.