Optimizing mix design of concrete with manufactured sand for low embodied carbon and desired strength using machine learning

Abstract

The aim of this paper is to propose a mix design methodology for concrete with manufactured sand (CMS), considering both the desired compressive strength and minimal embodied carbon, utilizing a machine learning approach. Initially, a compressive strength prediction model for CMS is developed using artificial neural network (ANN) based on a dataset containing 1810 measured strengths from real projects. The hold-out validation and 10-fold cross-validation are performed in addition to the validation from literature, indicating the high accuracy, robustness and generalization of the proposed model. Notably, the stone powder content of MS, a variable unique to CMS compared to river sand concrete, is found to have less importance in relation to compressive strength but should still be considered during mix design. Furthermore, the study reveals that the embodied carbon of CMS exhibits an increasing trend with the increase of compressive strength due to the enhanced cement content, while it varies significantly for a fixed compressive strength. However, cement may not be a carbon intensive constituent from the viewpoint of carbon intensity. Finally, mix optimization is achieved using ANN by setting the compressive strength requirement as the constraint condition and minimizing embodied carbon as the optimization objective. This paper not only provides an efficient solution for estimating the compressive strength of CMS without the need for experiments but also highlights the promising optimization capabilities of ANN for low-carbon CMS design.