Molecular Dynamics and Machine Learning Approaches for Cementitious Material Design

The brittleness and cracking issues in concrete significantly compromise its structural safety and durability. This study combines molecular simulations and machine learning to address these challenges and enhance cementitious material performance. First-principles calculations, reactive force field, and classical molecular dynamics were employed to decode hydration mechanisms, mechanical properties, and interfacial behaviors, with the aid of a novel machine learning force field achieving near-DFT accuracy at reactive force field speeds. Concurrently, generative adversarial networks (GANs) were used to augment data for predicting concrete strength, enabling the development of high-accuracy machine learning models. These integrated approaches reveal fundamental mechanisms and establish predictive frameworks, offering a transformative pathway for the rational design of next-generation high-performance cementitious materials, thereby contributing to improved structural safety and sustainability in construction.