Quantification of Fatigue Damage for Structural Details in Slender Coastal Bridges Using Machine Learning-Based Methods

Abstract

Exposed to the challenging coastal environment, slender bridges could experience significant dynamic responses and complex stress states resulting from the coupled dynamic impacts of wind, wave, and vehicle loads. Cracks could gradually initiate and propagate at structural details that might trigger failures of the structural members or the entire structural system. To predict the remaining fatigue life of slender coastal bridges, stochastic fatigue damage for structural details is quantified using machine learning (ML)-based methods, such as support vector machines (SVM), Gaussian process (GP), neural network (NN), and random forest (RF). Parametric probabilistic models for vehicles, defined based on long-term field measurements, and stochastic loadings from wind and waves, parameterized for various loading scenarios, serve as the input parameters. As for the output of ML models, equivalent fatigue damage accumulation is obtained based on the coupled vehicle-bridge-wind-wave (VBWW) system and stress analysis for complex structural details using multiscale finite-element analysis (FEA). With different training strategies, fatigue life for critical local details is obtained considering the ever-changing coastal environmental conditions. Training and testing results show that the GP algorithm outperforms other algorithms even though all algorithms exhibit the reasonable capability of predicting the fatigue damage accumulation.