Effects of Temperature and NaCl Concentration on the Adsorption of C-S-H Gel in Cement Paste: A Multi-fidelity Molecular Dynamics Simulation

Abstract

The durability and compressive strength of concrete will vary with the material components, ambient temperature, external intrusion. Using molecular dynamics (MD) methods to study the dynamic behavior of particles in cement-based materials can help us understand the underlying mechanism of property changes in concrete caused by above factors at the atomic level. So far, MD methods have been widely used to analyze the physical and chemical properties of concrete materials and the interaction mechanism between different interfaces at the nanoscale. However, too much complexity in the models will reduce the result accuracy and increase the computational cost. A suitable neural network structure can not only ensure the accuracy of analysis results, but also reduce the computational cost. In this work, MD methods are applied to build the models to explore the diffusivity of Na+ and Cl− in the calcium silicate hydrate (C-S-H) gel pores at different concentration and temperatures. In the process of running models, part of the MD models’ fidelity is reduced to save the computational cost, then the trained multi-fidelity physics informed neural network framework was used to obtain more accurate analysis results. The combination of MD simulations and deep learning methods expands the application range of MD in the field of concrete structure, has good development prospect and application value.