Abstract
Many persistent challenges in civil and environmental engineering require innovative materials—such as creep-resistant cement for structural durability, materials that incorporate CO₂ into building components, and low-cost catalysts for water treatment. Traditional materials discovery is slow and costly, relying on trial-and-error. While physics-based computational methods improve understanding, they are resource-intensive and unsuited for high-throughput screening. Machine learning (ML) offers a powerful alternative, enabling efficient exploration of material properties and recognition of patterns beyond human capability.
This thesis applies ML techniques to accelerate material innovation across three key areas:
I. Creep Prediction in Disordered Solids: Using molecular dynamics and ML classification, we identify "looseness", a structural descriptor strongly correlated with early-stage creep in calcium-silicate-hydrate (C-S-H), guiding microstructural design.
II. High-Entropy Alloy (HEA) Electrocatalysts: ML regression models trained on DFT data predict O* and HO* adsorption on CoFeNi-X (X = Mo, Mn, Cr) HEAs, enabling rapid screening and design for water treatment catalysts.
III. Earth-Abundant Oxides for CO₂ Mineralization: Pre-trained ML models from the Open Catalyst Project are used to screen CO₂ adsorption on 21 metal oxides. MnO (2, -1, 2) and Ti₂O (1, 1, 0) with oxygen vacancies emerge as promising catalysts.
By integrating ML with simulations and open databases, this thesis establishes predictive frameworks for rational material design, supporting advances in civil engineering and sustainable technologies.