This sub project has investigated the possibilities of using mobile data collection from railway vehicles to utilise collected data for safety inspection in a railway facility according to TDOK 2014: 0240, without physically being in the facility (off-site).
The project has used existing data that was collected in connection with an ERTMS design and which followed the specification used by Trafikverket for mobile data capture by railway. In parallel with the data capture, more high-resolution data was also collected to evaluate how usability can increase with changing requirements.
Every step in the inspection of TDOK 2014: 0240 has been evaluated and the opportunities for off-site inspection have been calculated statistically for each type of technology. There greatest opportunities was for inspection of tracks where approximately 85% of the inspection was estimated to be possible. There were also relevant opportunities for other technology areas where an overall measure of the condition could be achieved. Other accompanying benefits are highlighted for this type of machine measurement such as increased objectivity and an opportunity to inspect a railway asset at such a level that it is possible to monitor it from a life cycle perspective.
The possibilities for automation of each inspection step were evaluated, mainly with regard to methods based on automatic image recognition, but also for other algorithms for geometric analysis of laser data. Automation was most developed in some cases where you can see changes between images from different data collections, for example for the detection of missing or damaged fasteners and slipers, damaged signals and signs, cracking on various asset objects.
Other areas where automation of inspection proved to be possible based on analysis of laser data were, for example, wear on rail heads, deviating track geometry, control of ballast levels, erosion and damage to the embankment and bridges, free space, and control of catenary.
Sharp testing of automation was performed for the following inspections to be able to evaluate the accuracy of image recognition and geometric analysis of laser data:
- Automatic image recognition of rail damages, similar to squats. The methodology worked well with an accuracy of 88% for small injuries and significantly better for larger damages with the possibility of being improved with a larger amount of data.
- Automatic image recognition for missing fastenings. The methodology worked with very high accuracy and also had the potential to see changes in fastenings before they loosened.
- Detection of tilted sleepers worked well.
- Automatic sight length analysis. The methodology worked well and could be used both to detect deviations both for the visibility of signals and boards, as well as the visibility for road users at level crossing.
- Detection of vegetation that may fall on lines or tracks. The methodology worked, but requires a number of preparatory steps before it is applicable.
Stakeholders: Luleå tekniska universitet, Trafikverket, WSP och NRC.
Project leader: Peter Östrand, WSP