Fatigue crack detection in welded structural components of steel bridges using artificial neural network

Under the cyclic traffic loads, welded structural components of steel bridges may encounter fatigue, which can cause a shorter service life and lead to fracture. A precise fatigue life prediction of structural components requires an accurate collection of stress cycles of the respective component. The density of sensors installed for monitoring the component and the distance to the concentrated stress areas are the features, which impact the efficacy of the estimated fatigue life. In this study, a platform is developed for the data-driven fatigue assessment of welded structural components of steel bridges, using artificial neural networks (ANN). The proposed algorithm is implemented for a case study, vertical lift truss bridge, the Memorial Bridge, Portsmouth, NH. A 12-month data collection period is utilized for the algorithm, from the long-term SHM program of the bridge. The stress cycles are used to estimate fatigue responses of an instrumented structural component of the bridge and determine the correlation between the estimated fatigue responses at the instrumentation plan. Additionally, a validated finite element model of the bridge is utilized to investigate fatigue responses in the unhealthy condition of the objective component. Therefore, multiple physical damage cases are simulated to compute the damage-induced stresses and the resulting fatigue life. The healthy and damaged fatigue responses are the ANN inputs, to detect crack-induced variation in the estimated fatigue responses at the instrumented locations. It is demonstrated that the proposed damage detection method can effectively detect possible fatigue cracks using a detailed database of damaged and healthy fatigue damage indices for training ANNs.