Elucidating Rheological Properties of Cementitious Materials Containing Fly Ash and Nanosilica by Machine Learning

Researching the rheology contributes to enhancing the physical and mechanical properties of concrete and promoting material sustainability. Despite the challenges posed by numerous factors influencing viscosity, leveraging machine learning in the era of big data emerges as a viable solution for predicting the general properties of construction materials. This study aims to create models to forecast the rheological properties of cementitious materials containing fly ash and nanosilica. Four models-Random Forest, XGBoost, ANN, and RNN (Stacked LSTM)-are employed to predict and assess shear rate versus shear stress and shear rate versus apparent viscosity curves. Through hyperparameter adjustments, RNN (Stacked LSTM) exhibits excellent performance, achieving a coefficient of determination (R²) of 0.9582 and 0.9257 for the two curves, demonstrating superior statistical parameters and fitting effects. The RNN (Stacked LSTM) exhibited a better generalization ability, suggesting it will be more reliable for future prediction in cementitious material viscosity.