Soft computing based formulations for slump, compressive strength, and elastic modulus of bentonite plastic concrete.

Utilizing bentonite in composites such as concrete mixture is one of the practical approaches for adsorption of heavy metals. The mixture of bentonite and normal concrete is known as bentonite plastic concrete (BPC). This type of concrete is commonly utilized as the material of cutoff walls under earth dams with the aim of minimizing or preventing water seepage. Plastic concrete should possess high workability and consistency since it is poured into trenches by tremie pipe; this fact highlights the importance of predicting the slump of BPC. Different strength parameters of BPC such as compressive strength and elastic modulus can be predicted by utilizing prediction models. This information is exceedingly useful for modifying mixing design of BPC which results in reducing the cost and time of constructing a project. Consequently, it is vital to propose models that can predict the parameters of BPC with high precision. In this research, 158, 169, and 119 data records respectively for slump, compressive strength of cubic samples (f' c,cube), and elastic modulus (E c) of BPC were collected from literature in order to propose prediction models. Three soft computing methods of artificial neural network (ANN), Multivariate adaptive regression splines (MARS), and M5 model tree (M5Tree) were utilized and compared in this paper. Then, parametric studies were conducted to study the effect of different variables such as silty clay addition, bentonite dosage, water content, and curing time on the outputs (i.e., slump, f' c,cube , and E c). Results indicate that ANN models are more accurate than the other models for prediction of all three parameters. The water variable produces the greatest effect on the slump of BPC while the sand variable has the least influence. In addition, both for f' c,cube and E c models, variables of cement and curing time have the maximum and minimum impact on the outputs, respectively. • Three datasets of Slump, f' c,cube , and E c of BPC were gathered from the literature. • Three methods of ANN, MARS, and M5Tree were compared. • ANN models were more precise than the other models. • The parametric studies were performed to investigate the behavior of ANN models. • The sensitivitsy analysis showed the influence of input parameters on models output. [ABSTRACT FROM AUTHOR]