Artificial intelligence in camera detection

Using image-based methods, a deep learning pipeline for damage detection was developed between April 2020 and December 2020.


  • It should be possible to carry out the steps of the inspection at the ICE4 using AI procedures in order to enable the automation of the inspection.

  • The approaches developed should be so universal that they can be transferred to other series with less effort


  • Object recognition for all components to be found

  • Anomaly detection for all operations performed

  • Graphical tools for testing and evaluating the generated models

  • Tensorflow-based machine learning environment for computer vision use cases (other technologies: MLFlow, CVAT, S3, MSSQL)


  • In-depth understanding of the structure of the ICE4 and the work steps to be carried out

  • Modular and flexibly adaptable to other/new work steps toolbox for camera exploration allows porting our solution to other train types and other use cases

  • Even the intermediate results of the anomaly detection can support factory employees in their findings

Are you interested in your own use cases?