Abstract Summary
Vehicle-bridge systems are continuously subjected to environmental and operational variations that complicate the process of identifying structural damage. In general, damage is defined both as a substantial alteration of the physical characteristics of the material, as well as the modification of the geometric characteristics of any part of the structure, including changes in the boundary conditions and connectivity of the system itself. The loads that mostly affect the structural integrity of bridges are associated to vehicular traffic. This paper proposes a method for damage detection in railway bridges which combines two approaches of dynamic analysis: Model-Based and Data-Based. Two models of a bridge need to be created for this purpose. The first model is a finite element numerical model (FEM), which is calibrated and updated through the data collection campaign of a continuous or discontinuous monitoring system of the structure in terms of accelerations and deformations. The purpose of this model is to have a reference benchmark against which damage can be measured. The simulation is carried out in a supervised learning fashion, through which damage indexes can be determined through the comparison of the experimental data with the benchmark FEM. The second model is based on an autoregressive (AR) model, which uses both experimental and numerical data to conduct damage identification. The effectiveness of this approach arises as a result of the AR parameters being directly proportional to the stiffness of the structure. Therefore, the time series of continuously identified AR parameters can be used as damage-sensitive features. The effectiveness of the proposed approach is firstly validated through a simplified bridge model that takes into account the vertical and lateral dynamic interactions of the vehicle-structure system. Afterwards, the proposed methodology is applied to a real in-operation bridge, the Mascarat Viaduct in Alicante (Spain). The viaduct, built in the beginning of the 20th century, is part of the railway network of the Valencian community FGV (Ferrocarriles de la Generalitat Valenciana) belonging to line 9 of the TRAM of the province of Alicante. The presented results and discussion evidence that the fusion of the two approaches facilitates the damage identification problem in railway bridges. The usefulness of the proposed methodology lies in the fact that bridge data acquired in damaged conditions are generally scarce or even non-existent.
