Effect of inclusion of natural pozzolan and silica fume in cement - based mortars on the compressive strength utilizing artificial neural networks and support vector machine

Abstract

This study illustrates the effect of incorporating natural pozzolan (NP) and silica fume (SF) in cement-based mortars on the compressive strength. Up to 40% of the weight of cement in cement mortars can be replaced by NP powder. NP in its natural state can be used to replace up to 40% of the volume of sand in cement mortars. The replacement levels of cement by weight with SF are 5% and 10% to improve the mechanical properties of the NP-based cement mortars. Experiments were conducted to assess the compressive strength of cement mortars containing NP at 28 and 180 days of curing. The water/cement ratio was 0.49 for all the specimens. Models for the compressive strength of NP-based cement mortars and NP-SF-based cement mortars were formulated based on nonlinear regression analysis, artificial neural network (ANN) , and support vector machine (SVM) approach to explore the impact and exchanges of cement, NP, and SF mixtures. The RMSE of the nonlinear, ANN, and SVM models revealed that the RMSE of the ANN is significantly less than that of the nonlinear and SVM models. A sensitivity analysis for the predicted compressive strength of NP-based cement mortars and NP-SF-based cement mortars was conducted. The application of the SVM technique was evaluated through four different kernels, including sigmoid, linear, polynomial, and radial basis function (RBF). The RBF kernel shows better performance and offers more accurate results than the sigmoid, linear, and polynomial kernels. The results showed that the NP replacing cement and the SF replacing cement are the most dominant parameters for the prediction of compressive strength. Finally, the developed models from this study can be used to predict the compressive strength of cement mortars containing NP and SF.