Prediction of self-consolidating concrete properties using ML Models: Rheological properties

Self-consolidating concrete, SCC, is a high flowability and non-segregating material that provides great benefits in complex construction applications. The prediction of yield stress and plastic viscosity of SCC is critical for ensuring its performance and quality during mixing, transportation, and placement. Conventional methods to evaluate rheological properties are long-time-consuming, expensive, and prone to human errors. The current work aims at exploring the applicability of m The ML models developed in this study, such as Decision Trees, Support Vector Machines, Random Forests, Gene Expression Program and Deep Neural Networks, were trained and validated by using a dataset containing mix design parameters and experimental measurements of rheological properties. In addition, feature selection techniques were used to determine critical influencing factors like water-to-cement ratio, aggregate composition, and admixture dosage. Performance of such models was checked with respect to Mean Absolute Error, Root Mean Square Error, and R² scores chine learning (ML) models for the accurate and efficient prediction of the rheological properties of Self-consolidating concrete (SCC) is a non-segregating, highly flowable concrete that improves construction efficiency, especially in complex shapes and high reinforcement density areas. Precise prediction of its rheological properties, i.e., yield stress and plastic viscosity, is critical to ensure quality during mixing, transportation, and placing operations. Traditional testing procedures are labor intensive, expensive, and prone to human errors. This research explores the use of machine learning (ML) models, i.e., Gene Expression Programming (GEP), Deep Neural Networks (DNN), Decision Trees (DT), Support Vector Machines (SVM), and Random Forests (RF), for accurate SCC rheological property prediction. Model performance was evaluated in terms of Mean Absolute Error (MAE), Root Mean Square Error (RMSE), and R². The GEP and DNN models were determined to be better, with R² values of 0.93 and 0.89 for V-funnel time, and 0.81 for slump flow prediction. To provide insights into model predictions and explore the contribution of influential mix design factors, SHAP and PDP analyses were performed. The results validate that ML models, i.e., GEP and DNN, can accurately predict SCC rheological properties, thus eliminating extensive experimental testing and providing useful insights for optimal mix design.