Fatigue Reliability Assessment of Welded Steel Bridge Decks under Stochastic Truck Loads via Machine Learning

Welded joints in steel bridge decks are vulnerable to the fatigue damage caused by heavy-loaded trucks. A realistic probabilistic model of truck loads provides a basis for simulating the fatigue stress spectrum of these welded joints, where the fatigue reliability assessment can subsequently be carried out. In this paper, a stochastic fatigue truck load model was developed for probabilistic modeling of fatigue stress ranges to investigate the fatigue reliability of welded steel girder bridges. To deal with the uncertainty-induced computational complexity, a framework including deterministic finite-element-based hot-spot analysis and probabilistic modeling approaches is presented. In addition, a learning machine integrating uniform design and support vector regression is used to substitute the time-consuming finite-element model. The development of both the framework and the learning machine provides a reasonable, efficient, and accurate probabilistic fatigue damage model. Finally, a limit-state function of fatigue damage is established with the consideration of traffic parameters, including the growth factors of traffic volume and the vehicle weight. A prototype steel box-girder bridge is presented as a demonstration to illustrate the feasibility of the proposed stochastic fatigue truck load model and the corresponding framework. Parametric studies indicate the impact of traffic parameters on fatigue reliability indices of the welded joint in the lifecycle. The stochastic fatigue truck load model provides a new approach for probabilistic modeling of fatigue damage and reliability assessment of welded steel bridges.