Short answer
A generator tank loses fuel legitimately whenever the engine runs, so a falling level is not by itself evidence of theft. Good alerting works by comparing the observed drop against the running hours and expected burn rate for that machine, and raising an alert only when fuel leaves faster than the engine could possibly consume it.
Key takeaways
- Level alone cannot separate theft from burn on a generator, run-hours is the missing variable.
- Expected burn is a function of load: a 100 kVA set uses roughly twice as much fuel at full load as at half.
- The strongest theft signal is fuel leaving while the run-hours counter is static.
- A useful alert window is minutes, not hours, after that the vehicle has left the site.
- Non-contact ultrasonic sensing suits tanks where cutting an opening for a probe is not acceptable.
01Why is a generator tank different from a vehicle tank?
A parked vehicle has a simple property that makes theft detection easy: its tank should not change. Any drop while the ignition is off is worth investigating, and the rule needs almost no tuning to be useful.
A generator has no such quiet state. It is installed precisely so that it can run for long unattended periods, and while it runs it empties its tank continuously and legitimately. The level falling overnight is the expected behaviour, not the anomaly, so a rule written for vehicles produces an alert every single night on every generator on the estate.
That difference is why generator tanks are so often the weakest point on a monitored site. Teams either apply the vehicle rule and drown, or exclude generators from monitoring entirely and leave the largest, most accessible fuel store on the site unwatched. Neither is necessary once run-hours are part of the picture.
02What does normal generator fuel consumption look like?
Consumption is driven by load, not by time. Published planning figures for a 100 kVA set, roughly 80 kW at the 0.8 power factor gensets are normally rated at, put full-load burn at about 20 litres per hour, around 10 to 11 litres per hour at half load, and near 6 litres per hour at quarter load.
Expressed per unit of energy rather than per hour, modern diesel engines sit near 0.21 to 0.26 litres per kWh, and the efficiency curve is close to flat between 75 and 100 percent load. That flatness is useful: within the normal operating band, expected burn scales predictably with output.
The practical consequence is that a generator does not have one correct consumption number. It has a range that depends on what it was doing, and any rule that assumes a single figure will misjudge the machine at both ends of its load range.
03Why do run-hours change the whole problem?
Run-hours convert an ambiguous measurement into a testable one. Without them, a forty-litre drop is just a number. With them, the same drop is either consistent with two hours of running at load or wildly inconsistent with an engine that never started.
The test the system performs is a comparison, not a threshold: take the fuel that left the tank over an interval, take the running time and load over the same interval, and ask whether the first is explainable by the second. When it is, the event is consumption. When fuel left and the hour meter did not move, no amount of legitimate operation accounts for it.
This is why the run-hours signal is worth wiring even when it is inconvenient to obtain. A generator monitored on level alone can only ever produce suspicion; a generator monitored on level and running time produces a conclusion.
04What does a theft alert actually need to contain?
The recipient of a theft alert is usually a site supervisor holding a phone, not an analyst at a desk. The alert has to be self-contained: which generator, which site, how many litres, over what interval, and what the run-hours counter did during that same interval.
That last field is what turns a notification into a decision. An alert that says forty litres left in twelve minutes is interesting. An alert that says forty litres left in twelve minutes while the engine was not running is actionable, because it has already ruled out the innocent explanation before anyone opens the app.
Everything else is secondary. Charts, history and trend views matter for the investigation that follows, but the first message has to carry enough to justify walking to the yard. If the supervisor has to open a dashboard to decide whether the alert is real, the alert has failed at the only moment that mattered.
05How fast does the alert have to be?
Fuel theft has a short window in which intervention is still possible. Someone drawing fuel into a container or a vehicle is on site for a bounded period, and once they leave, the alert stops being an operational event and becomes a report.
That sets the design target in minutes. It also sets a requirement on the reporting interval: a sensor that reports once an hour cannot produce a minutes-old alert, no matter how good the analysis behind it is. Reporting cadence, not analytical sophistication, is usually the binding constraint on how quickly a site can respond.
There is a cost to that cadence in power and data, which is why it is worth setting deliberately rather than by default. A tank that is genuinely at risk warrants frequent reporting; a remote standby tank with no history of loss may not, and treating them identically wastes the budget on the wrong asset.
06What sensor suits a generator day tank?
Two approaches dominate. A capacitive probe is inserted through the top of the tank and measures level continuously along its length. The Eurosens Dominator RS is a representative unit: a published maximum error of ±1%, an operating range of −40 to +85 °C, a 10–60 V supply, IP67 ingress protection under IEC 60529, and RS-232, RS-485 and ISO 9141 K-Line interfaces.
The alternative is non-contact ultrasonic sensing, which measures the distance to the fuel surface from above without any part of the sensor entering the liquid. That matters when the tank is under warranty, when cutting an opening is not permitted, or when the fuel carries contamination that would foul an immersed probe.
Neither is universally better. The probe generally gives a steadier reading in a tank with surface movement; the ultrasonic unit avoids the installation and maintenance burden of an immersed element. The choice is driven by the tank, not by the sensor datasheet.
07How do you handle the refuelling cycle?
Refuelling is the event most likely to be misclassified, because it is large, fast and looks nothing like normal operation. A system that flags every sharp change will flag every delivery, and a supervisor who dismisses ten deliveries will dismiss the eleventh event without reading it.
Direction resolves most of it, a delivery raises the level, but the more valuable check is against the delivery note. A genuine delivery is a sustained rise ending at a level consistent with the litres invoiced. A rise that stops short of that is a short delivery, which is a loss the site is paying for and which no falling-level rule will ever catch.
Immediately after a fill, the reading needs a settling period before it is trusted. Fuel entering a tank moves, and a sensor sampled during that movement reports a level that does not yet correspond to a stable volume. Alerting during the settle window generates confident nonsense.
08What about generators at unmanned remote sites?
Remote generators are where fuel loss is both most likely and hardest to observe, because nobody is present to notice and connectivity is often intermittent. The design has to assume the link will drop rather than treat an outage as an exception.
That means the device must buffer readings locally and resend them in order once the connection returns. A system that silently discards what it could not transmit will show a flat line across the outage, which is precisely the window an investigator will want to examine. Worse, a gap and a genuinely quiet tank look identical unless the platform distinguishes them.
Monitoring the reporting itself is therefore part of the design, not an afterthought. A sensor that has stopped reporting is a fault that needs its own alert, because from the operations screen an unreporting tank and a tank that nothing is happening to are the same picture.
09How do you avoid alert fatigue across a generator fleet?
Alert fatigue is not a user-discipline problem; it is a design outcome. A stream in which most entries turn out to be nothing trains the recipient to dismiss without reading, and once that habit forms the system stops working regardless of how accurate it later becomes.
The most effective control is ranking by context rather than by size. A modest loss on a machine that was idle deserves more attention than a large one on a machine that ran all night, and a system that sorts by litres inverts that order every time.
The second control is measuring the alerts themselves. Track how many are dismissed without action; a persistently high dismissal rate is direct evidence that the baselines are wrong, and it is the number that should drive tuning. Treating the alert stream as something to be measured, rather than only produced, is what keeps it credible.
10What should a Saudi operator plan for specifically?
Heat acts on the measurement before it acts on the hardware. Diesel changes volume with temperature, which is why the petroleum industry corrects observed volumes to a standard reference temperature under ASTM D1250 and API MPMS Chapter 11.1. A generator tank in direct sun and the same tank before dawn do not report the same volume even when the engine has not started.
Hardware has to be specified for the environment rather than assumed to survive it. A sensor rated to +85 °C with IP67 protection under IEC 60529 is specified for dust and temporary immersion, which is the combination an exposed generator enclosure actually presents.
Site access is the third factor. Generators are frequently installed where a fuel road tanker can reach them easily, which is convenient for delivery and equally convenient for removal. Physical access should inform where monitoring is deployed first, because the easiest tank to fill is usually the easiest tank to empty.
11What does the investigation look like after an alert?
The first step is almost always physical: someone goes and looks at the tank and the machine. The alert has narrowed the question, but it has not answered it, and a level reading cannot distinguish a siphoning hose from a burst return line or a drain valve left open.
What the platform should supply at that point is the surrounding record rather than a verdict. The sequence of readings either side of the event, the run-hours trace, the last confirmed delivery, and whether the same asset has produced similar events before are together enough to tell a supervisor whether they are looking at a pattern or an incident.
Closing the loop matters as much as opening it. Recording what the investigation found, theft, leak, meter fault or nothing, is what allows the thresholds to be tuned against outcomes instead of guesses. Without that feedback the system never improves, and the same false positive recurs indefinitely.
12How ENTEK.AI handles generator tank alerts
On ENTEK.AI, a generator tank is watched through the Fuel Tank solution. It shows the live fill level of each tank, detects refuel and drop events automatically, and raises a Critical alert below 20% and a Low alert below 35%. It does not read the generator's running state, so the run-hours comparison described above is a check the site team makes against the engine's own hour meter or log.
Because tanks, vehicles and locks sit on the same map, a drop event can also be read against what else was on site. A drop that coincides with a vehicle arriving is a different investigation from one that does not, and having both facts in one system is what makes that distinction cheap.
The intended outcome is deliberately modest: fewer alerts, each carrying enough context that a site supervisor can act without a desk investigation first. Alert rules are checked every 30 seconds, and notifications go out within about a minute by email, SMS or WhatsApp.
| What the system knows | Engine ran 2 h at load | Engine ran 12 min | Engine did not run |
|---|---|---|---|
| Level only | Alert | Alert | Alert |
| Level + run-hours | No alert, consistent with burn | Alert, far above expected burn | Alert, no legitimate explanation |
| Supervisor action | None needed | Check the yard | Check the yard immediately |
Figures and sources
- Approximate full-load fuel consumption, 100 kVA diesel generator (about 80 kW; estimated between the chart’s 75 kW and 100 kW rows, as the chart has no 80 kW row): ~20 L/hour (Source: Generator Source, diesel generator fuel consumption charts, as of 2026-08-30)
- Approximate half-load fuel consumption, 100 kVA diesel generator: ~10–11 L/hour (Source: Generator Source, diesel generator fuel consumption charts, as of 2026-08-30)
- Approximate quarter-load fuel consumption, 100 kVA diesel generator: ~6 L/hour (Source: Generator Source, diesel generator fuel consumption charts, as of 2026-08-30)
- Typical genset power factor used to convert kVA to kW: 0.8 (100 kVA ≈ 80 kW) (Source: Walt Power, diesel generator fuel consumption, as of 2026-08-30)
- Specific fuel consumption of a modern diesel engine: ~0.21–0.26 L/kWh (Source: Walt Power, diesel generator fuel consumption, as of 2026-08-30)
- Load band across which the genset efficiency curve is near flat: 75–100% load (Source: Walt Power, diesel generator fuel consumption, as of 2026-08-30)
- Eurosens Dominator RS published maximum measurement error: ±1% (Source: Wialon sensor catalogue, Eurosens Dominator RS, as of 2026-08-30)
- Eurosens Dominator RS operating temperature range: −40 to +85 °C (Source: Wialon sensor catalogue, Eurosens Dominator RS, as of 2026-08-30)
- Eurosens Dominator RS supply voltage range: 10–60 V (Source: Wialon sensor catalogue, Eurosens Dominator RS, as of 2026-08-30)
- Eurosens Dominator RS ingress protection rating: IP67 (Source: Wialon sensor catalogue, Eurosens Dominator RS, as of 2026-08-30)
- IP67, second digit: protection against temporary immersion: immersion up to 1 m for 30 minutes (Source: IEC 60529, Degrees of protection provided by enclosures (IP Code), as of 2026-08-30)
- Petroleum volumes are corrected to a standard reference temperature under: ASTM D1250 / API MPMS Chapter 11.1 (Source: ASTM D1250, Temperature and Pressure Volume Correction Factors, as of 2026-08-30)
- Standard governing the Dominator K-Line service interface: ISO 9141 (Source: Wialon sensor catalogue, Eurosens Dominator RS, as of 2026-08-30)
Questions
Diesel keeps disappearing from the generator tank and I have no run-hours signal. Can I still catch it?
Partly. Without run-hours you can still catch losses that exceed any plausible burn rate, fuel leaving faster than the engine could consume it at full load. What you cannot do is distinguish a moderate theft from normal running, which is where most losses sit.
How often should a generator tank sensor report?
Frequently enough that an alert is minutes old rather than hours. The reporting interval, not the analysis, sets the floor on response time, an hourly reporter cannot produce a minutes-old alert. Balance it against power and data cost per tank rather than applying one interval everywhere.
Does an ultrasonic sensor work as well as a probe?
For different situations. A non-contact ultrasonic unit avoids opening the tank, which matters under warranty or where contamination would foul an immersed element. A capacitive probe generally gives a steadier reading where the fuel surface moves. Choose from the tank and its constraints, not the datasheet.
The site lost its signal. Do I miss the alert if diesel goes missing then?
The device should buffer readings and resend them in order when the link returns, so the outage becomes a delay rather than a hole in the record. The platform should also alert on the silence itself, an unreporting sensor and a tank where nothing is happening look identical otherwise.