Fourier neural operator-based data-driven method for predicting structural dynamic response

Deep learning methods have made significant advancements in solving partial differential equations (PDEs), where physics-informed neural networks (PINN) are the most widely applied methods across various fields. However, for structural dynamic response prediction, employing PINN to solve the PDEs of structural dynamic models may result in poor accuracy and low training efficiency. To address these challenges, a novel method based on Fourier Neural Operator (FNO) for predicting structural dynamic responses is presented. Specifically, the FNO is harnessed to capture the main characteristics of the structural dynamic response in the frequency domain, achieving an accurate and efficient solution. Numerical experiments with a cantilever beam structure demonstrate that the proposed method can predict the structural response with high accuracy and low computational cost, effectively capturing the dynamic characteristics of the structure. The proposed method achieves a prediction error for both low-frequency and high-frequency responses that is one to two orders of magnitude lower than the PINN method. Additionally, its prediction time is six orders of magnitude faster than the Finite Element Method and four orders of magnitude faster than PINN.