A framework for low-carbon mix design of recycled aggregate concrete with supplementary cementitious materials using machine learning and optimization algorithms

This study presents a framework for designing low-carbon and cost-effective mixtures of recycled aggregate concrete (RAC) with supplementary cementitious materials by integrating machine learning and grey wolf optimizer algorithms. The concrete mix design process considers key performance parameters such as compressive strength, chloride ion penetration resistance, and carbonation resistance. A dataset comprising 5306 data samples from 154 scientific resources is collected from the literature to train the machine learning models. Four different techniques, namely random forest, extreme gradient boosting (XGBoost), light gradient boosting (LightBoost), and category boosting (CatBoost) are employed to model the compressive strength, chloride ion penetration, and carbonation resistances of RAC. The best-performing models are then utilized to optimize the RAC mix design by minimizing the cost and reducing the carbon footprint. The results indicate that the CatBoost model demonstrates better predictive performance for the compressive strength and carbonation resistance, while the XGBoost and LightBoost models perform better in estimating chloride ion penetration resistance. Furthermore, the adoption of low-carbon mix design principles leads to a reduction in carbon footprint by 5.0% to 31.5% compared to cost-effective mix design for different compressive strength targets, with varying permeability levels. The framework provides a promising approach for designing environmentally friendly RAC mixtures while considering economic and durability factors.