A Deep Learning-Based Study of the Role of Graphene Oxide Nanosheets on the Microstructure of Cement Paste
Mingrui Du, Yunke Yang, Jianxiong Miao, Rongbin Hou, Kejie Zhai, Xupei Yao
Abstract
Despite graphene oxide (GO)-based nanosheets exhibiting a great potential for reinforcing cement composites, the difficulty of observing nanomaterials in backscattered electron (BSE) images hampers our understanding of their microstructural characteristics. In this study, we employed a convolutional neural network to extract the key features of the GO-modified microstructures and utilized three deep-learning-based approaches to identify its characteristic patterns. These three methods cross-validated each other, demonstrating high accuracy in identifying GO characteristic patterns. Moreover, we used microstructural descriptors, including pore size distribution, solidity, and circularity, to analyze the characteristic patterns. Results show that GO modification of the microstructure of cement paste mainly affects the pore phase, reducing the total porosity by about 12% at 28 days of hydration. GO leads to a decrease in the number of micropores, with more pores having diameters concentrated between 0.2 and 0.6 μm and an increase of 10–20% in the proportion of small pores. This microstructural optimization increases the compressive strength by 18–35% and the tensile strength by 24–35%. Additionally, the analysis of results extracted using deep learning provides more precise insights, which further elucidates the mechanism of GO microstructural modification. These findings provide a deep understanding of the role of GO in cement, which paved the way for the design of high-performance cementitious nanocomposites.