Comparative study of evolutionary artificial intelligence approaches to predict the rheological properties of fresh concrete

Abstract

Rheology has been an essential tool to control the fresh state properties of concrete in case of self-compacting concrete, 3d printing of concrete, and ultra-high-performance concrete. Through proper control of rheology, it is possible to achieve desire green strength concrete and free from honeycombing, bleeding, and segregation for self-compacting concrete. The rheological properties of concrete were investigated in the study with the application of machine learning methods. The decision tree (DT) and bagging regressor (BR) were employed to predict the plastic viscosity (PV) and yield stress (YS) of the concrete with various mixes. Total 140 data points (mixes) for concrete were used to the run the selected models to obtain the forecasted result for both PV and YS. Six input variables were used for running the models for two outcomes (PV and YS). Results revealed that the BR was more effective in term of predicting both properties PV and YS of concrete by indicating the coefficient of determination values 0.90 and 0.95, respectively. However, the said results for PV (0.90) and YS (0.93) from DT model was also satisfactory. The lesser values of the errors, root mean square error, mean square error, mean absolute error and the indication of high performance of the BR towards the prediction. The sensitivity analysis reflected the importance of each parameter with water and gravels having more than 50 % impact on PV output values, while for YS, both medium and small size gravels were found having impact more than 65 %. The statistical checks and method of k-fold cross over validation also confirms the accuracy of models.