A Generative adversarial network model for estimating temporal frequency variation of vehicle-bridge interaction using modified Stockwell transform

The presence of massive and directional vehicles moving on a railroad bridge causes temporal fluctuations in the fundamental frequencies of both bridge and vehicle systems, making traditional structural inspections difficult to employ effectively. Thus, this paper proposes an image-to-image Generative Adversarial Network (GAN) that estimates temporal frequency variation for the vehicle-bridge interaction system. To address the challenges associated with conventional approaches, eigenvalue analysis, numerical substitution, and Fourier series approximations are examined using a simple degree of freedom and a more complex model. Thus, the GAN model is proposed to overcome those limitations. In the framework, based on the properties of a real bridge and vehicle model, the modified Stockwell transform of bridge acceleration from the dynamic simulation is used as input data with the paired data from the numerical substitution approach. Then, the model is validated through quantitative measures and laboratory scale experiments. Results showed that the model has strong performances across the various scenarios, demonstrating the potential for bridge condition assessment under operational conditions.