| About this Abstract |
| Meeting |
MS&T22: Materials Science & Technology
|
| Symposium
|
Ceramics and Glasses Modeling by Simulations and Machine Learning
|
| Presentation Title |
Machine Learning to Design and Discover Sustainable Cementitious Binders: Learning from Small Databases and Developing Closed-form Analytical Models |
| Author(s) |
Aditya Kumar, Taihao Han |
| On-Site Speaker (Planned) |
Taihao Han |
| Abstract Scope |
To reduce the carbon footprint of Portland cement, the prevailing practice embraced by concrete technologists is to partially replace the cement in concrete with supplementary cementitious materials (SCMs). Chemistry of the SCM profoundly affects all subprocesses leading up to property development in concrete. Owing to the substantial diversity in SCMs’ compositions, computational models are unable to produce a priori predictions of properties of [cement + SCM] mixtures. This study presents a deep learning (DL) model capable of producing a priori, high-fidelity predictions of cement hydration kinetics and phase assemblage development in [cement + SCM] mixtures. The DL is coupled with: (1) A fast Fourier transformation algorithm that reduces the dimensionality of training datasets; and (2) A thermodynamic model that constrains the DL. The training of the DL is leveraged to develop a closed-form analytical model capable of predicting cement hydration kinetics in [PC + SCM] mixtures. |