New on Railfleet: refuel alerts

Théo Delecour
Railnova Blog
Published in
5 min readJan 15, 2018

Fuel consumption and fuel supply are central aspects of diesel fleet management. Railnova monitors fuel gauge signals to automatically inform operators of relevant events, such as when a locomotive is running low on fuel and when it has been refuelled.

Recently, we developed a complete refuel alert to give operators accurate information about when and where a refuel happened, and how much fuel was supplied. This data can be helpful to check fuel billings and to keep track of the fuel consumption of the fleet.

Let’s take a look at how Railnova’s refuel alert works!

Filtering the fuel signal

Fuel gauges give us the measured fuel quantity available on the assets, but their signal can be pretty noisy. As the fuel in the tank of a moving locomotive moves randomly, reading values can show changes (drops and spikes in the measured fuel levels) of up to 20%. Fuel gauges can also be buggy and send wrong values.

To get a better estimation of the current fuel level and to avoid sending wrong values, Railnova performs a real-time filtering of the incoming data. This filter was designed with knowledge of typical fuel level evolutions:

  • a slow decrease when the locomotive is moving, and
  • a very fast growth during a refuel

By filtering the real-time fuel data from the train we’re able to prevent refuel alerts being triggered by noise (e.g. caused by fuel moving when the train is accelerating or braking, or by random errors). The images below shows you an example of noise being filtered out so no false alerts are sent, and an example of a refuel being detected despite a noisy fuel gauge.

Filtering the fuel signal: raw data is in blue and filtered signal in orange, sudden drops and noise are filtered out.
Case of a very noisy fuel gauge which disturbs the filter. The refuel was detected, other sudden changes did not trigger an alert.

Accurately detecting refuels

A refuel is detected by watching sudden changes in the filtered fuel signal. It is then confirmed using several criteria.

1. Is the locomotive moving?

If the locomotive was moving when the detected refuel occurred, it means that the alert was actually triggered by noise that passed through the filter, and thus the alert is discarded.

2. Is the locomotive close to a fuel station?

Another very relevant factor is the geolocation of the asset: an asset must be near a fuel station to be refueled. We can measure the position of a refueling asset in real-time using the GPS signal sent by the Railster device (Railnova’s remote monitoring equipment) installed on the asset. We then check that this position is close enough to a station that we previously registered in our database.

Knowing the locations of fuel stations is another challenge that Railnova completed with the power of data: to reduce the number of configurations for our users, we automatically cluster the frequent refuel locations using machine learning. The locations of the fuel stations are automatically found by feeding the geolocations of thousands of verified refuels performed by Railnova clients in the past to complex algorithms. This automated process is regularly conducted to update the location of fuel stations.

Clustering refuel locations lets us compute station locations. Computed station locations can easily be checked by consulting satellite imagery.

However, in some cases a locomotive might be refuelled in an unusual place — after an exceptional intervention for example. These cases are also considered by the Railnova system, and special confirmation criteria have been set for this case.

Our strategy lets us detect refuels as low as 5% of the tank capacity with close to zero false positives. The image below shows a particular case of what could be interpreted as a refuel, but was righteously recognised as a false alert thanks to GPS confirmation.

The locomotive was not moving when the sudden change occurred, however it later turned out to be an error and not a refuel. The alert was righteously discarded using GPS confirmation.

Giving relevant information

The complexity of our refuel detection algorithm allows us to give our clients very relevant data to enrich the alert content. Operators are informed of when the refuel took place, and of how much fuel was supplied to the asset in question (both in the percentage of the tank capacity and in liters).

An email alert informs operators of refuels.

When compared to actual refuel data from fleet managers, our alert gives very accurate and precise information: the error on supplied fuel quantity is below 5% for 90% of cases, and less than 2% of detected refuels were false positives.

Clients can access the alerts at any time on the real-time Operational Dashboard, and/or they receive the alerts through email. Both then let them easily access the raw fuel data to get more information and context about the refuel and verify the alert.

The alert invites users to use the Data Inspector to see the extended context of the refuel.

Besides real-time data monitoring, clients running diesel locomotives can also keep track of their monthly fuel consumption, get a list of all refuels and download these reports in Excel for further analysis or comparisons on the Railfleet reporting page:

Overview of monthly fuel consumption
Overview of refuels per asset
Excel export of all refuels from the last 3 months

These reports enable our clients to keep track of their total fuel consumption and to identify fuel saving opportunities.

Take a look at our the following help guide article to find out how you can set up a refuel alert yourself on your Railfleet account: Being alerted of refuels

A diverse range of alerts can already be set up on Railfleet, be sure to contact us if you’d have any questions about fuel alerts, or any other type of alert, we’re happy to help!

--

--