As reported in railway gazette, Rumo is running a more advanced infrastructure diagnosis system to
prevent broken rails.
Why is it relevant
Broken rails are an issue all across the world. For instance, Network Rail in the UK faces some 100
broken rails per year.
A broken rail can easily lead to a derailment which is, especially if the speed is high, the loaded goods are heavy or
dangerous, or the place of derailment is not easily accessible or all together, a high cost event and
might even result in loss of life.
Broken rails can happen after time from smaller defects such as head checks, problems in the civil
works or excessive loads. Challenging environmental conditions, such as high temperatures or very
low temperatures can increase the speed of such defects forming by creating additional tension on
the rail and supporting metal fatigue.
Furthermore, the number trainsets running over a already weak point increase the likelihood of one
breaking. A typical high traffic railway experiences some 3 million axles per year, each with some 21
tons pressing on the wheel.
As reactions to broken rails, usually both parts are clamped together and operating speed reduced
temporarily in the area until a permanent solution, i.e. welding it together, can be fitted. However,
this means usually a temporary closure of the line and heavy civil works.
If a weak rail can be identified prematurely, either work can be prepared to fix it, or even the reason
for breaking can be prevented.
How such a condition monitoring and fixing system works
There are usually 3 components to such a system: Some sort of sensing; some sort of analytics; some
sort of maintenance or fixing process. The first component can for instance consist of a diagnostic
vehicle or fibre optical sensing along the line. The second component would be data science
activities, industrialized as far as possible. The third component would be mostly standard operating
procedures, proactive and reactive approaches on how to utilize the information gathered from the
first component and analyzed in the second component. Such a system only creates value in the third
one, i.e. by acting decisively based on the insights.
How it was implemented at Rumo
The objective of the system in place at Rumo seems to be focused on preventing derailments. Hence
the monitor and give immediate information to the train drivers to react appropriately, i.e. reduce
travel speed or even stop until the condition of the track could have been assessed.
Unfortunately, I could not find out which type of sensor was used, but from the articles available, I
understand that it could be for instance a closed cirquit sensor as part of their signaling installation,
placed every few kilometers. This sensor is submitting notifications about disruptions to a central
server or a database in the cloud.
This is where, apparently with some machine learning, a data scientist is reducing the number of
false positives in a data clensing task to a level that can be worked with from a false positive and false
The third block consists with Rumo apparently of an app for the locomotive drivers that gives them
insights in their track ahead in almost real time and they send out maintenance teams to fix any
If it pays out at the end is difficult to say, but certainly something that should be observed by looking
at some KPIs: Number of incidences due to broken rails; Avg. cost per incident; Number of false
alarms; Avg. cost per false alarm; Avg. maintenance cost per km; Avg. Speed per km; Number of false
negatives; Avg. cost per false negative.
I hope you are as excited as I am to learn and see what will be happening on their track in the future.