Abstract Summary
Nowadays, the society is highly dependent on railways, since it is the most effective mode of transportation that has the lowest greenhouse gas emissions in the transports sector. Because of that, there has been an increase in rail transport, with great dependence on transport infrastructure such as railway bridges. For that reason, and to reduce inspection and maintenance costs, some techniques using damage detection, that is part of the Structural Health Monitoring (SHM), are being developed and implemented in railway bridges, in order to these infrastructures can be used safely, thus extending their lifespan. This work is focused on the application of machine learning methodologies for damage identification in a filler-beam railway bridge, based on numerical simulations. The numerical simulations required the development of an advanced Finite Element (FE) numerical model of the train and the bridge, including the track. The dynamic analyses were performed using a train-bridge dynamic interaction method, including the track irregularities, and considering the baseline scenario of the bridge and the damage scenarios. The features extraction from the dynamic analyses will be performed based on Autoregressive models (AR) and Principal Component Analysis (PCA) applied to the acceleration records from the passing trains. Classification methodologies will be applied to select the most sensitive features for damage detection. The proposed methodology will be useful to detect early damage levels in the structure, without interfering with the normal service condition of the trains and bridge.
