ABSTRACT
The Scope of the presented study is a Bridge-Weigh-in-Motion (B-WIM) method for the identification and quantification of train axles (loads and distances) using a regression-based machine learning model trained with synthetic generated train responses from trains passing over the bridge structure and data from structural health monitoring (SHM).
The B-WIM method uses synthetic train responses, generated with a calibrated numerical model of the bridge structure. Using a database with a large set of different generic train configurations (axle loads and distances) synthetic structural responses at different sensor positions are generated. The synthetic responses are downsampled and Peak-based features are derived and split into a training and a validation dataset for a regression-based machine learning model.
The trained machine learning model is used to calculate predicted axle loads and distances. Validation for synthetic data is then done by comparison with the known axle configurations for the generic train database. In upcoming steps of the ongoing research project, the trained machine learning model will then be used for identification of the axle loads and distances of the real trains passing over the bridge structure using features derived from the recorded monitoring data.
