Search for the Best Architectural Models in Neural Network for Modeling Normalized Mechanical Parameters of Concrete

In this study, in order to evaluate the performance of neural network, two models of MLP neural network and RBF neural network were used to predict flexural, tensile and compressive strengths. The data used are taken from the results of models fitted to the results of tests performed on roller concrete samples containing different amounts of recycled crumb rubber, fly ash and nanosilica based on compressive, flexural and tensile strength tests. Different types of artificial neural networks have been used to predict the types of concrete strength. In each section, the structure of the neural network used is given along with the table of input information and output results of that network. In each type of neural network, the number of layers and the number of different neurons have been used for modeling. In the tables, the green rows represent the best structure that has been able to predict the strength of concrete well. Also, the best result (lowest error and highest correlation coefficient) has been selected by considering the network performance in simultaneous prediction of resistance types. The results of compressive and tensile strengths are in the same direction and generally in the same direction, but flexural strength usually shows different results.