Corrosion fatigue analysis on in-service steel bridges subject to random traffic loads via long short-term memory-based surrogate model

The coupled effect of repetitive traffic loads and environmental corrosions will significantly accelerate the fatigue damage accumulation, hence reducing the fatigue life of steel bridges. In the present study, a surrogate model-based methodology that integrates field tests, finite element (FE) analysis, and long short-term memory (LSTM) neural networks, is proposed to evaluate the corrosion fatigue effects on the failure probability and fatigue life of in-service steel bridges subject to random traffic loads. An existing steel girder bridge and measured weigh-in-motion (WIM) data were employed for illustration. Specifically, based on the measured strain histories of the steel girders and the truck load spectrum derived from WIM data, a modified FE model of the bridge was established to compute the stress influence lines at the fatigue-prone detail. Subsequently, calibrated stress influence lines considering different corrosion conditions were derived, and random traffic loads were acted on the stress influence lines to obtain the stress time histories. The fatigue failure probability was then computed and trained as samples for the LSTM neural network, and the corrosion fatigue effects on life estimation of the steel bridge were further investigated via the proposed LSTM-based surrogate model. The parameter analysis shows that the effect of either overloading or corrosion applied separately will underestimate the fatigue damage accumulation, leading to an overestimated truck axle weight limit (TAWL) for steel bridges. Additionally, a rise in the traffic flow passing the fast lane will decrease the bridge fatigue damage. Furthermore, a detailed procedure for determining the TAWL that can ensure the target fatigue failure reliability of the steel bridge after its expected service life is proposed based on the coupled corrosion fatigue analysis.