Location and positioning in the railway (part 4)
Other sources of position
So, what can be done to correct/supplement the position to ensure correctness? There are a number of existing methods out there. The first one we have discussed is a correction service to try and remove GNSS errors (RTK, PPK, Terrastar) - there are constraints with these, availability of correction service, time constraints and satellite availability. There is the option to use physical infrastructure such as balises, RFID tags - these rely on physical installation (and survey), maintenance and are not available throughout the network. Care must also be given to ensuring that the events generated by these are consumed at the correct level by dependent systems (so not to introduce a time-based error). These also must be “mapped” to the network model being used (this not only involves traditional survey, but a potential shift and attribution to the network model so that these “speak” the same language as the positioning and measurement systems).
The signalling system itself could also be used, as a train travels through the network it passes through signal berths and this information is made available via a service. There are issues with this:
The mapping between berths and the network model may not be accurate
The time and location resolution of this is not precise (i.e., we get an approximate time a train passed a location and the location could also be approximate)
It is not trivial to work out which train is which and we often need a lag of time to calculate this
The coverage is not necessarily granular enough to be used everywhere
Other on train sensors can be used, Lidar has been used to work out which direction a train travels through a switch for instance. This can work well but does not always work through complicated areas of infrastructure and requires the network model to always match reality. It also requires the position (and track attribution) before the switch to be correct.
Lineside fibre optic cable has also been used to position trains; this obviously requires the installation of the fibre but also requires calibration. For IM, the drawbacks around timing and synchronisation would prove an issue.
It is worth revisiting the problem, positioning of IM data is more than a pair of Long/Lat coordinates, it is also a problem of time. If we imagine a system that is capable of synchronisation to another with an error of 10 milliseconds, this at face value sounds great. However, if we consider a train running at 125mph - this equates to an error of nearly 56cm. Add this to our positional error and it’s clear this is not ideal. With this in mind, how systems are synchronised is paramount as we can quickly introduce huge positional errors if it is not done correctly. It’s very easy to get this wrong and consideration of, not only computer clocks, but also stamping of data correctly from input sensors, must be paramount.
Ideally, we would like to keep boots off ballast and avoid installation of physical assets, we would also like a solution that works everywhere with maintenance of, at most, one system (keeping multiple systems in line is difficult - especially when dealing with data for IM which requires up to date information at all times).
A note on relative and absolute accuracy
What does having accurate position actually mean – there are two metrics to quantify? Absolute accuracy - i.e., where the position is in absolute space. A good question woudl be - is this important? Absolute accuracy is only ever really important if we want to measure the same thing (or locate the thing) with two different methods - i.e., absolute accuracy gives us the mapping between the two, if we go to the same place with different systems/devices or at different times, do we get the same result.
Relative accuracy is often interchanged with repeatability, if we measure something with a device, can we measure it again with the same device and get the same answer. Consider how Track Geometry (TG) is currently validated (and therefore often positioned). A reference run is recorded, subsequent runs are then recorded and “aligned” to the initial run. This works well if the track geometry doesn’t change much between runs and the measurement system has no (or few) erroneous measurements. However, there are assumptions made here, namely that the base run is correct - both in terms of measurement AND position, so that the position of subsequent runs will be the same as the base run. If the geometry changes (either through physical track change or erroneous measurement) then this will introduce an error into the alignment/position of the runs. The biggest drawback of this approach is its lack of relatability back to an absolute truth, we potentially have no (easy) way of directing a maintenance worker to the exact location of a defect and/or correlating other measurements (e.g., an OLE measurement from a contact-based system) to the same data as there is no absolute positional reference - that we can quantify.
