A Surrogate Model Using Deep Learning for 2D Stress Distribution Prediction of Satellites

In engineering applications, the real-time calculation of stress distribution in satellite structures poses a significant computational burden when using the finite element method (FEM) as the calculation tool. To address this challenge, this paper presents a novel deep learning-based framework for efficiently calculating the stress distribution of satellites. The proposed framework utilizes a surrogate model constructed through neural networks, which performs an image-to-image regression task to learn the mapping between the component loading conditions and the corresponding stress distribution. By leveraging the surrogate model, the stress distribution of satellites can be quickly calculated and analyzed, assuming the motion state is known. Typical two-dimensional stress distribution of satellites is investigated to demonstrate the feasibility and effectiveness of the proposed deep learning-based framework. The experimental results demonstrate that the developed surrogate model not only achieves high-precision prediction but also exhibits strong generalization ability.