Predictive Analysis on Mortar Rheological Property Based on Machine Learning
Mortar rheological properties are not only influenced by the mixture component characteristics and mix design,but also varying with the duration of mixing with cementitious materials.To study the influence of these factors on mortar rheological performance,the rheometer was used to obtain the plastic viscosity and yield stress of mortar at 4 different time intervals,and with 30 different mix ratios.The generated 120 sets of rheological performance data were cleaned and analyzed.Subsequently,3 machine learning algorithms of support vector regression,K-nearest neighbor regression and random forest regression were used to predict the mortar flow properties.Parameters(e.g.,cement content,manufactured sand,fly ash,limestone powder,water-reducing agents,water,time after mortar mixing,water-cement ratio,aggregate-cement ratio,and water-cement ratio)were used as independent variables.The plastic viscosity and yield stress were as dependent variables.The result indicates that the support vector regression algorithm has the highest accuracy in predicting plastic viscosity and yield stress of mortar.The MAE,RMSE,MAPE and R2 show better generalization capability and applicability in the prediction field compared to other models.Through time-effect analysis,the significant coefficients for influence of different components on mortar rheological properties are identified.The water-reducing agents,cement,sand,and water have significant influence on mortar rheological properties.Finally,the single-feature analysis is conducted by using support vector regression on the variables with high feature importance(e.g.,cement,manufactured sand and water-reducing agents).The influencing curves of these variables on mortar rheological properties are obtained by varying each variable individually.
NETL
NSTL
万方数据
