Deep learning-based realtime multiload response prediction and inverse analysis of offshore bridges

Offshore bridges operate in complex ocean environments, making structural analysis, design, and monitoring more challenging. Existing typical time history analysis and nonlinear model updating based on finite-element methods are computationally intensive and time consuming, limiting the usage in many scenarios. To create a more efficient analytical tool, a Deep Learning-based Offshore Bridge Predictor (DeepOBP) is proposed. The model integrates structural characteristics and coupled dynamic loads in the ocean environments, enabling millisecond level and high precision nonlinear dynamic offshore bridge response predictions. A differentiable structural inverse framework (Inverse DeepOBP) couples the surrogate model with gradient-based optimization is further developed to enable rapid damage identification and model calibration for structural health monitoring. The experimental results show that DeepOBP demonstrates high accuracy under both normal operating conditions and multihazard coupled conditions, with R² = 0.93 and 0.92. Inverse DeepOBP delivers more than a 10-fold and more than a 10⁴-fold speed-up over surrogate-based model updating with the heuristic algorithm and nonlinear finite element model updating, respectively, while maintaining relative errors below 7 % for each identified parameter, enabling efficient structural analyses and real-time monitoring.