Rail Defect Classification Viewer

Disclaimer:

The assessment of rail artifacts in this overview is based on a neural network method from the research project AIFRI. Consequently, it only shows the current state of a research model, with room for improvements. Patterns detected as defects shouldn't necessarily be considered actual defects, but more as an example of how the model's output would look. The data has already been manually evaluated by experts from the field and was not used to train the model.


Further points should be considered when exploring the data and the model.

  • DCM data files are downloaded directly from the mobilithek . Please be nice to their server!
  • The data's origin is a regular inspection run from the field.
  • Before exploring any data, the corresponding file needs to be downloaded.
  • This overview compares the current Automatic Assessment (AA) with a neural network approach from AIFRI (AI).
  • The AA approach is bound to fixed moving windows and assesses gate data distribution hierarchically.
  • The AI approach assigns the most likely class to a pixel column that is detected as an artifact.
  • Assessments are only available for field inspection runs. Simulations are shown as they are.
  • Only UT data can be displayed for now.
  • Only data from the meter selected in the overview plot is visualized, and its immediate surroundings are visualised.
Select elements to be shown in the viewer?
Select an assessed meter.

The datapoints show the assessment of a designated meter. The category is a result of what assessments by AA and AI occur after the meter start. This leads to some misclassification in the visualisation, if the fixed AA bounding box starts in the previous meter and then overlaps with the AI assessment.