Prediction of rheological behavior of self-compacting concrete by multi-variable regression and artificial neural networks

To control the rheological behavior of self compacting concrete, this study attempts to provide predictive models using a multi-variable regression model and artificial neural networks such as machine learning techniques on a database of 59 samples (average of 354 specimens), collated from literature. The statistical significance of the model coefficients and the coefficient of determination are adopted as criteria to check the performance of the model on test data. Interaction terms such as slump flow diameter, V-Funnel flow time, and L-Box ratio were inferred from literature to improve the efficacy of the predictive modeling. Two approaches for the prediction of yield stress and plastic viscosity are proposed and validated. With an increase in slump flow diameter and V-Funnel flow time, there was a significant increase in yield stress and plastic viscosity from 2.1 Pa and 18.2 Pa.s to 98.6 Pa and 198.8 Pa. s, respectively. A significant decrease in these rheological parameters was noted with an increase in L-Box ratio from 0.5 to 1. Thus, the comparison between the two numerical methods proposed shows that, for datasets with only one output, the multi-variable regression presents an important advantage. Thus, the regression model for the plastic viscosity represents values of R of the order of 0.955, 0.968, and 0.916 in correlation with slump flow diameter, V-Funnel flow time, and L-Box ratio, respectively. [Display omitted] •Control the rheological behavior of SCC by ANN.•Prediction of rheological behavior by empirical tests.•Predictive models based on multivariable linear regression analyses.•Use of SGD Optimizer with 405 epochs for the ANN system.