Local stress fields prediction using global displacement through Fourier neural operators

Abstract Accurately capturing the stress distribution within complex structures is crucial for reliability and performance, particularly in high-stress environments. Traditional finite element analysis methods require a fine mesh to achieve highly accurate stress fields, which is computationally expensive and time-consuming. This paper presents a novel neural network-based approach that learns the mapping between global displacement values on coarse global elements and local von Mises stress on fine local elements. By training the neural network on a dataset of paired global displacement and local stress fields, our method effectively predicts high-resolution stress distributions from coarse-grained inputs. This approach significantly reduces the computational burden associated with fine mesh finite element analysis, enabling real-time simulations and optimizations. The proposed method has been validated by the test results that demonstrate the ability to accurately predict local stress fields while maintaining computational efficiency.