Abstract Summary
The presence of unbalanced loads on freight trains can potentially cause higher levels of deterioration or even the failure of railway track components, as well as situations of risk of derailment. This paper aims to develop an unsupervised methodology for the early detection of unbalanced load scenarios, both in lateral and longitudinal directions. The study relies on a 3D numerical simulation of the train-track dynamic response to the presence of unbalanced vertical loads. Different measurements layouts were tested, in terms of type (strain gauges and accelerometers) and position of the sensors. The proposed methodology involves: i) the feature extraction based on an Autoregressive Exogenous model (ARX); ii) the feature modelling based on a Principal Component Analysis (PCA) to remove the influence of operational effects; iii) feature fusion, and iv) feature classification based on a cluster analysis to distinguish a regular loading situation from a faulty one.
