An engineered ML model for prediction of the compressive strength of Eco-SCC based on type and proportions of materials

Abstract

Recently, various waste materials and industrial by-products such as supplementary cementitious materials (SCMs) have been proposed to improve the properties of self-compacting concrete (SCC). This profitable waste management strategy results in lowering the costs and carbon emission, and a more sustainable, cleaner and eco-friendly production of SCC (Eco-SCC). The properties of such a complex material are commonly measured through costly experiments. Researchers also proposed experimental data analysis and predictive modeling methods such as machine learning (ML) algorithms for prediction of the properties of concrete. However, proposed models commonly relate the properties to the proportion of constituents only and ignore the effect of their type and properties, and other influential factors. This paper aims to engineer the concept and develop a more efficient ML model for prediction of the 28-day uniaxial compressive strength (UCS28d) of SCC containing SCMs. A comprehensive dataset is collected through a precise literature survey. Some dimensionless ratios are proposed to reduce the dimensionality of variables and reflect the effects of considered influential factors in different ML models. Two separate datasets are considered to test the predictability of models where one has new proportions of materials only and the other contains new type of material with new properties. After validation and comparison between various ML models, Gaussian process regression (GPR) model proved to perform well on both considered Test datasets with R2, RMSE and MAE of around 0.96, 3.66 and 2.49 respectively. Sensitivity analysis results confirm the contribution and importance of considering type and properties of materials as model variables. This paper demonstrates and highlights that all influential factors must be considered to develop engineered ML models to use as universal tools for indirect estimation of properties of composite materials such as Eco-SCC.