Uneven spatial distribution of fatigue cracks on steel box-girder bridges: a data-driven approach based on Bayesian networks

Orthotropic steel decks (OSD) of bridges are subjected to cyclic vehicle loads, leading to the growth of fatigue cracks. Fatigue strengths are dependent on fatigue details and the load effects vary from position to position; consequently, the spatial distribution of fatigue cracks on the bridge deck is not uniform. Previous studies on fatigue effects of OSD are mainly limited to a single fatigue detail. This article presents a Bayesian network (BN)-based method to investigate the uneven distribution of fatigue cracks on a cable-stayed bridge deck. The bridge deck is divided into girder-elements, and types of fatigue cracks are considered. Critical variables controlling the uneven distribution of fatigue cracks are represented by root nodes, while the fatigue states of girder-elements are represented by leaf nodes; the causal relationships are described by directed edges between nodes. The proposed approach is validated using field inspection results. A sensitivity analysis is conducted to reveal the importance of each node in the BN. Compared to previous researches involving fatigue cracks, the proposed approach is superior in calculating the occurrence probabilities of types of fatigue cracks given critical variables as well as quantitatively analysing the main causes for types of fatigue cracks.