A Machine Learning-Based Computational Architecture for Unlocking Water Dynamics in Saturated Calcium Silicate...

Area of Science:

• Materials Science
• Computational Chemistry
• Civil Engineering

Background:

• Concrete durability depends on ion and water transport in calcium silicate hydrate nanopores.
• Molecular dynamics (MD) simulations offer atomistic insights but are computationally expensive for large-scale problems.

Purpose of the Study:

• To develop a computationally efficient framework for predicting transport properties in concrete nanoporous networks.
• To overcome the scalability limitations of traditional MD simulations.

Main Methods:

• Synergistic integration of high-throughput molecular dynamics with graph neural networks (GNNs).
• Training GNNs to learn the relationship between atomic environments and kinetic transport behaviors.
• Validation of the framework on saturated calcium silicate hydrate and multiparticle pore structures.

Main Results:

• Achieved high-fidelity predictions of pore water diffusion coefficients in calcium silicate hydrate.
• Improved computational efficiency by three orders of magnitude compared to conventional force field methods.
• Demonstrated strong transferability and accuracy in capturing nonlinear diffusion in complex pore structures.

Conclusions:

• The proposed framework offers an ultrafast and accurate tool for screening transport properties in diverse concrete structures.
• Lays the foundation for multiscale modeling of aggressive ion ingress, crucial for resilient concrete design.
• Advances the development of sustainable and durable reinforced concrete materials.