Exploring the rheological and mechanical properties of alkali activated mortar incorporating waste foundry sand: A comprehensive experimental and machine learning investigation

Alkali-activated materials, known for their economical and eco-friendly nature, are becoming popular as a viable substitute for traditional cement-based materials in the field of construction. This research thoroughly examines the intricate rheological characteristics and compressive strength (CS) of alkali-activated mortar (AAM). It specifically explores the use of waste foundry sand (WFS) as an environmentally conscious and sustainable alternative to river sand. The addition of WFS significantly improves the CS of AAM, demonstrating 35 % enhancements at a 30 % sand replacement level. The rheological properties of customized AAM mixtures were targeted to achieve optimal use in 3D printing. The increasing replacement level of WFS in the matrix induced a non-linear reduction in yield stress (37 % decrease at 30 % replacement) accompanied by a concurrent rise in plastic viscosity (38 % increase at 30 % replacement). Considering the intricate interplay between rheological parameters and CS, AAM containing 15 % sand replacement exhibited optimal properties. Moreover, through rigorous experimentation, a comprehensive dataset of 64 samples was established by varying the amount of river sand, WFS, and water content in the mix. Machine learning models-including AdaBoost, decision tree, K-nearest neighbors, stochastic gradient descent, gradient boosting, random forest, neural network, gene expression programming, and support vector machine-were developed to predict CS, yield stress, and plastic viscosity of the AAM. Modeling errors, including MSE, RMSE, MAE, and MAPE, were falling in the acceptable limits. Specifically, most models demonstrated a high level of prediction having R-2 > 0.90 across training and validation datasets. In conclusion, this study not only highlights the optimized properties of AAM but also validates the efficacy of various machine learning models in predicting crucial parameters for AAM formulation.