Decarbonizing concrete with artificial intelligence

Abstract

Concrete is a key enabler for modern infrastructures but also a top source of carbon emissions in societal development. Because of the high degree of freedom in concrete mixture design, the optimization of concrete formulas remains broadly empirical and inefficient. Further, the concrete carbon footprints is seldomly considered in the conventional mixture design protocols. Here, we approach the concrete optimization from a novel angle of artificial intelligence, where a machine learning model is trained based on a large dataset of 1,150 representative concrete formulas that are developed in a quality control lab for guiding real concrete production. The results demonstrate that our model achieved an unprecedented accuracy for predicting concrete strength at various ages. By further associating each model input with the corresponding carbon embodiment, the machine learning model is used for designing high-performance concrete mixtures that are optimized for both strength and sustainability.