Shear Design Optimization of Short Rectangular Reinforced Concrete Columns Using Deep Learning

This paper aims to show the effectiveness of artificial intelligence (AI) for structural design optimization. Design optimization of rectangular reinforced concrete (RC) columns using a deep neural network (DNN) results in a reduction of both time and monetary resources. The utilization of the DNN model prevents the iterative design process, which is common in a conventional approach. To create the necessary dataset of designs, parametric RC column designs are generated and analyzed automatically using a finite element model (FEM) of the OpenSeesPy Python library. The dataset spans five heights and six concrete classes. The data is pre-processed using equal-sized filtration to preserve the most economical column designs for specified ranges of loading conditions. Based on the given constraints of axial load, bending moments, and shear loads, the NN model predicts cross section geometry and longitudinal and transverse reinforcement. To evaluate the accuracy of the NN model predictions, thirty cases are run through the model and checked for compliance with the Eurocode building standard. A comparative analysis of the NN performance with manual designs demonstrates the overall effectiveness of the NN by 11.3% in terms of monetary price. As for the time aspect, the NN is faster by 8.57 min and 96% more efficient than manual design.