Proportioning and curing optimization of carbonated calcium silicate-based materials via machine learning and genetic algorithm

Calcium silicate can react with CO2 in the presence of water to gain strength and thus can be both taken as an alternative binder and a carbon sink. A series of calcium silicates exists in industrial products and natural rocks with diverse material properties, implying its promising sustainable potential and yet foreseeable obstacles in quality control. To tackle that, a data-driven prediction-optimization computational framework was proposed based on a high-fidelity dataset of carbonated calcium silicate-based materials. Ensemble machine learning models were built based on the curated data for CO2 uptake and strength predictions. Pareto front searching was further conducted using a genetic algorithm to pursue better sustainability, strength, and curing efficiency of the materials. Reasonable decisions can be made once the trade-offs are considered among the optimal solutions.