Skip to content
Sign inGet Started

Why a Fuel Reading Is Less Accurate While Moving, and What to Do About It

By Saeed Baatiah · Published:

Short answer

A level sensor measures the liquid surface, and motion tilts and disturbs that surface. Technoton publishes an inaccuracy of not more than 0.2% for stationary tanks and standing vehicles, rising to 1% for moving vehicles. That is the same hardware, a fivefold change, caused only by movement, so judge level changes at rest.

Key takeaways

  • Technoton publishes two accuracy figures for one sensor: not more than 0.2% stationary, and 1% moving, a fivefold degradation caused only by motion.
  • The figure on a datasheet is usually the stationary one, so a theft threshold sized against it is five times tighter than the moving signal can support.
  • OIML R 117-1 treats liquid motion as a measurement corruption mechanism, requiring an anti-swirl device in each road tanker compartment.
  • Teltonika checks movement criteria every second, so the platform already knows whether a reading was taken at rest, the state is available, it just has to be used.
  • Smoothing is the manufacturer-prescribed cure, but Teltonika warns in its own documentation that setting it too high delays the response to real changes.

01One sensor, two accuracy figures

The clearest evidence that motion is the problem comes from a manufacturer publishing two numbers for the same device. Technoton states an inaccuracy of not more than 0.2% for stationary tanks and standing vehicles, and 1% for moving vehicles. Nothing about the probe changes between those two rows. The electronics, the calibration and the installation are identical; only the state of the vehicle differs, and the specification degrades fivefold. That is a remarkable admission to find in a datasheet, and it is worth pausing on, because most buyers read only one of those two rows. The habit is to note the best figure a vendor publishes and carry it into every calculation afterwards. But the two rows answer different questions: one tells you how well the sensor can measure a still surface, and the other tells you what it can do on a road.

02What motion actually does to the surface

A level sensor does not measure fuel. It measures the height of a surface and converts that height to a volume through a calibration table. The conversion is only valid while the surface is flat and level, and a moving tank guarantees it is neither. Acceleration tips the surface toward the rear, braking tips it forward, and a corner tips it sideways. A partially filled tank on an uneven road adds sloshing, a travelling wave that raises the level at one end while lowering it at the other. A probe fixed at one point in the tank sees only its local column, so the reading rises and falls with a wave that has nothing to do with how much diesel is present. On a slope the effect persists rather than oscillating, which is worse: a parked truck on a ramp reports a steady, confident and wrong number.

03The standards body agrees, in its own way

This is not a telematics observation. It is recognised in legal metrology, where the money at stake makes the requirements explicit. OIML R 117-1 requires an anti-swirl device in each compartment of a road tanker unless a compliant gas separator is fitted. That clause exists because a rotating or swirling liquid entrains gas and corrupts a volumetric measurement, and the standard treats it as serious enough to mandate physical hardware inside the vessel rather than a software correction afterwards. The lesson transfers directly. When an organisation whose job is to defend the fairness of a fuel transaction responds to liquid motion by requiring a baffle, a fleet operator should not expect a level reading taken mid-journey to carry the same weight as one taken at a standstill. The physics is the same physics; only the consequence of getting it wrong differs.

04Smoothing helps, and it costs something

The standard cure for a jumping level is a moving average, and Teltonika prescribes it for exactly this problem, recommending an averaging constant of 2,500 to 5,000 for moderate instability and 6,000 or more for very unstable signals. It works. A wave that raises and lowers the surface averages toward zero over a long enough window. But the same document carries the manufacturer's own warning that setting it too high could delay the response to real changes in fuel consumption, and that warning is the whole difficulty. The filter cannot distinguish a slosh from a siphon. Whatever window suppresses the noise also suppresses a genuine drop of similar duration, so a heavily smoothed channel is a quiet channel in both senses. Smoothing buys stability by spending sensitivity, and the exchange rate is not disclosed anywhere.

05The state you need is already being recorded

The useful discovery is that the tracker already knows whether the vehicle was moving. Teltonika checks its movement criteria every second, a far finer granularity than the smoothed fuel channel offers, and that state travels with the record. So the discriminator a fuel rule needs is not something to add; it is something already present and usually ignored. The rule to build is therefore conditional rather than absolute: compare a level to a level captured in the same state. A drop measured between two stationary readings inherits the 0.2% figure. A drop measured between two moving readings inherits the 1% figure. A drop measured between one of each inherits neither, because the difference includes an unknown amount of tilt, and no threshold can separate that from a loss.

06Whether a reading exists at all

There is a second layer beneath accuracy, which is availability. Teltonika takes records through four simultaneous methods, a time period, a distance travelled, a heading angle and a speed delta, and a zero value disables each of them. A level change is visible only if a record happened to be taken while it was underway. So an operator asking why a stop was not evaluated at rest may find the answer is that no record was written during the stop, because the distance and angle triggers had nothing to fire on and the time period was long. Configuring for a stationary comparison therefore means configuring the time trigger to keep producing records when the vehicle is parked, which is precisely when a naive setup goes quiet to save data.

07What this changes in practice

The practical programme is short. Read both accuracy rows on the datasheet and size every threshold against the moving figure, not the stationary one, unless the rule genuinely only fires at rest. Record the movement state alongside every level sample and store it, so a comparison can be qualified later rather than reconstructed. Evaluate a suspected loss between two stationary samples wherever the operating pattern allows it, because that is the only comparison the better specification applies to. Keep the smoothing constant as low as the signal tolerates, and write down what it is, because it is a sensitivity setting disguised as a noise setting. And treat a reading taken on a slope as unqualified rather than merely noisy, since a persistent tilt does not average away.

What a level comparison is worth, depending on the state of the vehicle in each sample
ComparisonApplicable published inaccuracySafe to use for a theft rule?
Stationary sample vs stationary samplenot more than 0.2%Yes, the tightest comparison available
Moving sample vs moving sample1%Only with a threshold sized for 1%
Stationary vs moving (mixed)Neither figure appliesNo, the difference contains unknown tilt
Either sample taken on a slopeNot covered by either figureNo, a persistent tilt does not average away
Heavily smoothed channel, short eventAccuracy is not the limit hereNo, the filter may have removed the event
No record written during the eventNothing to compareNo, check the four acquisition triggers

Published fuel-level sensor accuracy

Technoton DUT-E, stationary0.2 %
Technoton DUT-E, moving1 %
OMNICOMM0.5 %
Teltonika UL2020.5 %

Statistics in this article: Technoton DUT-E specifications, ENTEK.AI product catalogue, Teltonika Telematics Wiki

Figures and sources

Questions

Why is the same sensor five times less accurate when the vehicle moves?

Because it measures the height of a surface, not a volume. Acceleration, braking, cornering and sloshing all tilt or disturb that surface, so the local column the probe sees no longer represents the tank. Technoton publishes not more than 0.2% for stationary tanks and standing vehicles, and 1% for moving vehicles, for exactly this reason.

Should I just apply more smoothing?

Only up to a point. Teltonika prescribes averaging constants of 2,500 to 5,000 for moderate instability, but its own documentation warns that setting the value too high could delay the response to real changes in fuel consumption. The filter cannot tell a slosh from a genuine drop of similar duration, so extra smoothing quietly costs you sensitivity.

How do I know whether a reading was taken at rest?

The tracker already knows. Teltonika checks its movement criteria every second, and that state travels with the record, so the discriminator is available rather than something you have to add. The work is to store it beside each level sample and to require matching states before comparing two levels.

Does parking on a slope count as being at rest?

For the movement flag, yes, but not for accuracy. A slope produces a persistent tilt rather than an oscillation, so it does not average away over time the way sloshing does. A parked truck on a ramp will report a steady, confident and wrong level, which is harder to spot than a noisy one.

Sources

Related insights