Life-cycle seismic performance analysis of an offshore small-to-medium span bridge based on interpretable machine learning

In earthquake-prone offshore regions, small-to-medium-span bridges undergo significant deterioration in seismic performance due to chloride corrosion and aging impact over their service life. To address this concern, this paper proposes an interpretable machine-learning framework to assess the life-cycle seismic performance of offshore small-to-medium-span bridges. First, the time series on aging features, based on the chloride erosion model and the aged Laminated Rubber Bearing (LRB) mechanical model, are established by Latin Hypercube Sampling (LHS). Subsequently, a finite element model of an offshore bridge is updated to incorporate time-dependent material deterioration status. Pushover analysis and time-history analysis are employed to generate time series data, which is then input into the Long Short-Term Memory (LSTM) neural network to predict the seismic performance of the bridge at a given age. Finally, the interpretable machine learning Shapley value is combined to probe the aging evolution of the bridge and rank the importance based on their impact on the time-dependent seismic performance. The results indicate that the chloride corrosion primarily affects the strength and remaining diameter of the reinforcement, while aging effects influence the shear stiffness and frictional coefficient of the LRB. The designed LSTM framework achieves high-accuracy prediction of the life-cycle seismic performance of the offshore concrete bridge, with the Mean Square Error (MSE) of the training set less than 0.13 and the R 2 of the prediction set above 0.806. The analysis of interpretable machine learning shows that the performance of longitudinal steel plays a major role in the deterioration of the seismic capacity of piers, followed by the unconfined concrete and the hoop steel. Changes in the frictional coefficient and shear stiffness of the rubber bearing play a dominant role in the deterioration of bridge seismic demand, while the influence of aging factors related to piers is relatively limited. The LSTM-Shapley-based approach presented in this study enables bridge professionals to assess the impact of aging components on seismic performance, guiding maintenance and retrofitting decisions to ensure long-term safety and functionality of offshore bridges.