AI-guided design of low-carbon high-packing-density self-compacting concrete

Self-compacting concrete (SCC) has gained substantial traction in modern engineering due to its exceptional fresh and hardened properties. However, traditional SCC design methods encounter significant challenges. Conventional experimental design approaches often necessitate a considerable number of trial mixes to fulfill diverse performance objectives, incurring escalated material, time, and labor costs. Additionally, conventional SCC designs tend to use high cementitious material content, leading to elevated carbon emissions and energy consumption compared to ordinary concrete. To address these issues, this study proposes a novel approach that combines the compressible packing model (CPM) with machine learning (ML) techniques. This approach innovatively utilizes particle packing theory to guide ML in optimizing SCC aggregate grading and mix proportions. By integrating physical principles into artificial intelligence, this approach facilitates the intelligent design of low-carbon, high-packing-density SCC. Compared to the conventional method, SCC designed using the innovative AI approach demonstrates a 57.2% reduction in embodied carbon emissions. [Display omitted] •Integration of CPM and DE algorithm leverages physical knowledge in machine learning guidance.•Aggregate gradation for low-carbon SCC is optimized using DE and CPM.•Automatic optimization of low-carbon SCC mix proportion is achieved.•Random Forest accurately predicts slump flow and 28-day compressive strength of SCC.•The proposed methodology achieves up to a 57.2% reduction in embodied carbon for SCC.