Predicting Stress in Structures Using Convolutional Neural Networks: the Case for 2D Plates

Abstract

Abstract Aeronautical structures are subjected to various loadings and damage initiators, such as bird strikes, which can reduce their lifespan. High internal stresses can contribute to potential failures. Therefore, high-fidelity modeling methods, such as finite element analysis (FEA), are used to study these structures under various loading conditions to design more resilient systems. However, these methods are computationally expensive and time-consuming, especially for large projects. Reliable, efficient techniques that can analyze the performance of these structures can improve the design process and reduce the risk of incidents. As an attempt to reduce analysis time, we propose using deep neural network models, specifically an encoder-decoder based convolutional neural network (CNN), to predict stress distribution in a structure based on geometry, external loadings, and boundary conditions. To implement this technique, we study stress distribution in 2D plates with different geometries and loading conditions using a Python script and FEA software. The script creates plates with unique conditions and runs FEA to obtain stress data, which is used to train and test the CNN model. The results of our analysis show good accuracy for stress field predictions and offer a reliable, fast technique for analyzing engineering structures. This method can be modified to study stress fields under various loading conditions for more complex structures in a fraction of the time it would take the FEA software to run the required simulations.