Convolutional Neural Network Approach for Surrogate Modelling of the Torsion Problem

This study explores the use of Convolutional Neural Networks (CNNs) in surrogate modelling for solving the torsion problem of elastic bars with arbitrary cross-section profiles. The conventional approach to solving this problem, the finite elements method (FEM), is computationally expensive and time-consuming, limiting its use in real-time applications. To address this, Convolutional Neural Networks (CNNs), a deep learning technique that has shown success in various domains is used. By training CNN models on datasets generated using FEM, this study bypass the FEM process and efficiently obtain the desired stress information. Further, a comparative analysis of different CNN architectures are done to obtain the optimal trade-off between accuracy and efficiency. This study adds to the expanding collection of literature exploring the application of deep learning methods in the field of computational mechanics.