A combination of deep learning and genetic algorithm for predicting the compressive strength of <scp>high‐performance</scp> concrete
Iman Ranjbar, Vahab Toufigh, Mehrdad Boroushaki
Abstract
Abstract This article presented an efficient deep learning technique to predict the compressive strength of high‐performance concrete (HPC). This technique combined the convolutional neural network (CNN) and genetic algorithm (GA). Six CNN architectures were considered with different hyper‐parameters. GA was employed to determine the optimum number of filters in each convolutional layer of the CNN architectures. The resulted CNN architectures were then compared to each other to find the best architecture in terms of accuracy and capability of generalization. It was shown that all of the proposed CNN models are capable of predicting the HPC compressive strength with high accuracy. Finally, the best of the six considered models was validated through the 10‐fold cross‐validation method and compared to the previous studies on the same data set. Models were developed through a comprehensive data set consisting of 1030 HPC compressive strength test data. Comparing the proposed technique with previous studies showed that the proposed technique has a considerable advantage over previous methods and can be employed for reliable estimation of the mechanical properties of different engineering materials.