Intelligent recurrence and prediction of long-term mechanical properties of in-service cable-stayed bridges with ambient temperature and concrete time-dependent effects

The awareness of mechanical properties evolution of cable-stayed bridges in service stage is crucial for structural condition assessment. However, the ambient temperature and concrete time-dependent effects are indispensable for the accurate calculation of long-term mechanical properties, which greatly increase the computational cost. Thus, an intelligent method based on the artificial neural network is presented to recur and predict long-term mechanical properties of in-service cable-stayed bridges. Sequences of daily maximum and minimum ambient temperature of an in-service cable-stayed bridge in 1 year are analyzed. Periodic field measurements and step-by-step finite element method are used to collect sample data of long-term mechanical properties. Artificial neural networks are then constructed to map mechanical properties of an in-service cable-stayed bridge. The generalization capabilities of neural network models are examined with the linear regression analysis. Comparing reproduced results with on-site measurements, the maximum relative cable force difference is about ± 10%, while the maximum difference of girder deflections is 0.029 m. The results indicate that the presented method can reproduce mechanical properties of in-service cable-stayed bridges. The evolution pattern of long-term mechanical properties of an actual in-service cable-stayed bridge within the 10 years of service are then determined. The presented intelligent method can effectively recur and predict long-term mechanical properties evolution of in-service cable-stayed bridges with a limited amount of sample data under an acceptable computation cost.