This study presents a novel approach to assessing electrical resistivity (ER) in nano-graphite-modified cementitious composites, a key factor for enhancing their durability and integrity. This innovative framework establishes a new paradigm in the application of ML to material science, highlighting the ability to bridge experimental and computational methods for more reliable and efficient outcomes. Leveraging a dataset of 539 cement-based materials, advanced machine learning (ML) techniques were applied, including the extremely randomized trees (XRF) algorithm, which outperformed traditional models with minimal error metrics (MAE: 37.093, MSE: 4992.581, RMSE: 8.406). The integration of Shapley’s additive explanations (SHAP) analysis further highlights the novelty of this work, revealing testing protocol (MM), ultrasonication time (SN), and graphene dosage (GN) as the most significant factors influencing ER. Longer SN times and higher GN doses reduced ER, while smaller graphene particles (2µm) increased it. Additionally, the water-to-binder (W-B) ratio and sand-to-cement (S-C) ratio showed peak ER values at 0.47 and 2.8, respectively. The four-probe method and lime-water curing produced more reliable and lower ER values compared to other methods. This data-driven methodology offers unprecedented insights into the complex interactions between material properties and ER, providing a powerful tool for optimizing composite formulations. Future research may focus on scaling this approach to larger datasets, investigating the long-term durability of composites, and refining ultrasonication techniques for better graphene dispersion.
Read full abstract