Prediction of strength parameters of concrete containing different additives using optimized neural network algorithm.

In this research, a multilayer feed-forward backpropagation error neural network has been used to predict the strength parameters of a concrete sample containing different additives. To achieve the most optimal neural network structure, the strength parameters of the concrete have been evaluated for different neural network arrangements. Control criteria are the use ofnumerical values of performance, the correlation between training functions, validation and,testing in the neural network, gradient and results of regression diagram to determine the most optimal neural network structure. It was found that the function of the neural network largely depends on its geometric structure. Revealed by the research findings, the most optimal prediction of the neural network has occurred in the case of using three layers with 30 neurons in each layer in the neural network. In this case, the numerical value of the neural network performance and the regression were obtained as 58.5 × (10-9) and 0.9846 , respectively. By determining the optimal neural network, different percentages of concrete raw materials based on the pre-performed experimental study are introduced to the selected neural network and the considered resistance parameters are predicted through residual analysis. According to the results, the differences between the predicted values of the neural network and the numerical values of the experimental study concerning the parameters of compressive, flexural, and tensile strength were also found to be equal to 1.68%, 1.92%, and 0.21%, respectively. Such a slight difference reflects the optimal accuracy of the chosen neural network in predicting the strength parameters.