Deep mining the mechanical behavior of graphene oxide–cement composites using data-driven artificial intelligence techniques

Abstract

This study presents a data-mining workflow for mining potential information in graphene oxide-cement composite (GOCC) mechanics datasets using data-driven artificial intelligence techniques for enhancing data integrity and for data augmentation using semi-supervised learning. The obtained prediction models for the compressive, flexural, and tensile strengths of the GOCC had high R2 (0.98, 0.94, and 0.95) and low root mean square error (2.84, 0.80, and 0.34) in the real test set. The interpretability analysis of the black-box model showed that an appropriate GO admixture is critical; the GO thickness and linear size mainly affected its compressive strength and flexural strength, respectively. The 50 elites resulting from the search for the optimal mechanical properties of the GOCC in the developed model constituted the Pareto front, providing a suitable parameter design range for the GOCC. This study is essential for understanding the mechanical laws of GOCC and developing efficient design strategies.