Application of Response Surface-Corrected Finite Element Model and Bayesian Neural Networks to Predict the Dynamic Response of Forth Road Bridges under Strong Winds

Bayesian Neural Networks (BNNs) outperform traditional Finite Element Models (FEMs) in predicting the dynamic response of large-span bridges under strong winds, offering higher accuracy and computational efficiency. A study on the Forth Road Bridge demonstrated that BNNs achieved R² values of 0.9073 for Y-displacement and 0.7969 for Z-displacement, significantly surpassing FEMs (0.6167 and 0.6283, respectively). While BNNs completed predictions in just 20 seconds, FEMs took five hours. However, FEMs provided superior interpretability and comprehensive structural response predictions. The research integrates machine learning with FEMs, leveraging response surface methodology (RSM) and genetic algorithms (GAs), to advance dual-driven approaches for large-scale bridge monitoring under extreme conditions.

Liu, Y., Meng, X., Hu, L., Bao, Y., & Hancock, C. (2024). Application of Response Surface-Corrected Finite Element Model and Bayesian Neural Networks to Predict the Dynamic Response of Forth Road Bridges under Strong Winds. Sensors, 24(7), 2091. https://doi.org/10.3390/s24072091