Fuel Sensor Calibration Drift: The Correction Most Integrators Apply Twice

Short answer

A capacitive probe in diesel is almost temperature-immune: the dielectric coefficient of 0.032 percent per °F nearly cancels the density coefficient of 0.033, so it reads a mass-like quantity without compensation. Adding a thermal volume correction on top double-counts. Real drift comes from fuel blend and tank geometry instead.

Key takeaways

• Diesel's dielectric and density temperature coefficients are 0.032 and 0.033 percent per °F, so they nearly cancel in a capacitive probe.
• Adding an ASTM D1250 style thermal volume correction on top of a raw capacitive reading double-counts and makes accuracy worse.
• Biodiesel permittivity is 3.3 against diesel's 2.2, a 50 percent shift that dwarfs a ±1 percent sensor specification.
• A horizontal cylindrical tank holds 19.55 percent of capacity at quarter height, not 25, an error no sensor accuracy can rescue.
• No manufacturer we could read publishes a recalibration interval; every such figure online traces to secondary blogs.

01The correction that is applied twice

Ask an integrator what degrades fuel-level accuracy and temperature comes first. The physics of a capacitive probe in diesel says otherwise, and the reason is precise enough to be measured. A United States patent for non-intrusive fluid measurement records that the temperature coefficient of the dielectric constant for diesel fuel is 0.032 percent per degree Fahrenheit, whereas the temperature coefficient of density for diesel fuel is 0.033 percent per degree Fahrenheit. The same document explains why that matters: a decrease in fuel density also produces a decrease in the dielectric constant of the fuel, which directly offsets the density change. The two effects run in opposite directions and are nearly equal, so they cancel. The patent states the resulting property outright, noting substantial temperature independence for diesel fuel across temperature ranges between -10 °F and 190 °F.

02What the probe is actually measuring

The cancellation is not a coincidence, and understanding it explains what the reading means. A capacitive probe responds to the permittivity of the medium between its electrodes, and hydrocarbon fuels of this class generally have a dielectric constant in the range of 2 to 3. When diesel warms, it expands: the same mass occupies more volume, so the level in the tank rises, and its density falls. The dielectric constant falls with the density, and the probe sees a taller column of slightly less responsive fluid. Those two changes offset, which means the probe is reporting something closer to the mass of fuel present than to its volume. This is a desirable property, because mass is what a fleet is actually buying and burning, and it is the reason a well-installed capacitive probe holds its reading through a Gulf day without any temperature channel at all.

03Where the thermal correction does belong

None of this makes ASTM D1250 and API MPMS Chapter 11.1 wrong; it places them. Those standards exist precisely because volume changes with temperature, and they define how to correct a measured volume to a reference temperature. The metric procedures require the customary-unit procedure be used first to compute density at 60 °F, with that value then further corrected to give the metric output, and ASTM has added the base temperature of 20 °C alongside 15 °C. The correction is expressed as a factor driven by a thermal expansion coefficient at base temperature, and that coefficient is not a single constant: it is a density-dependent correlation whose K-coefficients change per commodity group, with diesel falling in the fuel-oils group defined between roughly -10 and 37 degrees API. All of that applies to a volumetric measurement. It does not apply a second time to a reading that already cancelled the effect.

04The first real drift: fuel composition

If temperature is not the culprit, what is? The largest documented mechanism is a change in the fuel itself, because permittivity is exactly what the probe measures. Peer-reviewed work in Sensors measured room-temperature dielectric constants of 2.2 for diesel and 3.3 for biodiesel, under controlled conditions at 21.0 ± 0.1 °C across a frequency range from 1 Hz to one megahertz. That is a 50 percent shift in the property the sensor depends on. A probe calibrated on pure diesel and then filled with a blend will read high across its span, and the error scales with blend fraction: roughly 2.5 percent of span on a B5 blend and about 10 percent on B20. Set against the ±1 percent measurement error the Eurosens Dominator publishes, or the ±0.5 percent SOJI states for the LIGO Pro, a blend change overwhelms the sensor specification by an order of magnitude.

05The second real drift: the shape of the tank

The second mechanism is not drift in the sensor at all; it is a conversion error that no sensor accuracy can rescue. A horizontal cylindrical tank does not hold volume in proportion to depth. From the exact circular-segment geometry, a horizontal cylinder filled to one quarter of its diameter holds 19.55 percent of its capacity, not 25 percent, and at three-quarter height it holds 80.45 percent rather than 75. Assuming level is linearly proportional to volume therefore errs by 5.45 percentage points of full capacity at those two points, and the deviation reaches its true maximum of about 5.77 percentage points near 19 percent of height. The error is zero only at the halfway mark, at the centreline. Against a sensor rated ±1 percent, a linear assumption contributes five times the sensor's own error budget, and it does so systematically rather than randomly.

06Which is why the calibration is a polynomial

The industry answer to tank geometry is visible in how calibration is actually performed, and it confirms the non-linearity rather than assuming it away. Teltonika's documented procedure fills the tank in increments of around 5 to 10 percent with the engine off, waits roughly thirty seconds after starting for the reading to settle, and records the pair. When the table is complete the process yields four coefficients, a0, a1, a2 and a3, which is a cubic polynomial, not a linear scale factor. The manufacturer states that more measurements provide more accurate readings later on, so the point count is a direct accuracy lever available at no hardware cost. The Eurosens manufacturer requires a calibration table for a related reason: a real vessel deviates from the ideal shape, can have dents and internal elements such as ladders and hatches, and may be installed at a tilt.

07Where a volumetric measurement does need the correction

The cancellation applies to a capacitive probe, not to every fuel measurement, and the distinction decides where the standard belongs. A flow meter, a dipstick reading converted through a strapping table, or a delivery invoice all measure or state volume directly, and volume genuinely does change with temperature. Those figures need the correction, which is why reconciliation between a delivery note and a tank reading has to be done in consistent terms: correct the volumetric side to a reference temperature before comparing it against a probe that is already reporting something mass-like. Getting this backwards is what produces a persistent, seasonal reconciliation gap that no amount of sensor recalibration closes, because the sensor was never the source of it. The rule is short: correct volumes, and leave the capacitive reading alone.

08The diagnostic that costs nothing

Teltonika documents one check during calibration that doubles as a fault detector for the life of the installation, and it is worth adopting as a standing rule. The voltages must increase or decrease consistently across the calibration points; if the values go up and down between different levels of fuel, the underlying issue must be found before continuing. A non-monotonic curve is physically impossible for a working level sensor in a sound tank, so its presence proves a defect: a loose connection, a failing probe, water in the tank, or a wiring fault. The value of this check is that it converts a vague suspicion of drift into a binary test. Any operator can rerun a few calibration points and see whether the curve is still monotonic, and if it is not, the sensor output is not evidence of anything until the fault is found.

09Fouling and coating, the slow mechanical paths

Two physical degradation paths are documented in the patent literature even though neither comes with a published rate. A patent for a capacitive filling-level sensor with a dielectric coating states plainly that insulating films of dirt falsify capacitance signals, and describes varying electrode coating thickness to eliminate that falsification, so contamination altering the reading is an acknowledged failure mode designed against, not a field rumour. The same patent treats long-term stability of the probe's own coating as an explicit requirement, ensuring the constancy of the dielectric constant of the coating even under exposure to water over many years, with the coating's dielectric constant between 2 and 7. Water in a diesel tank is common in humid coastal conditions, so both mechanisms are live in Gulf service. What no primary source provides is a drift rate, which means the honest posture is periodic verification rather than a predicted decay curve.

10The recalibration interval nobody publishes

A specific answer is expected here and we cannot give one honestly. We read the official Eurosens Dominator specification, its RS/CAN datasheet, and its full operation manual; none states a recalibration or verification interval. SOJI's LIGO datasheet likewise gives only an average service life, stated as ten years. The Eurosens warranty period is 60 months from the date of manufacture, which is a commercial term and not a calibration schedule. Figures circulating online, recalibrate annually, after the first 500 hours, or when drift exceeds 2 percent against a dipstick, appear exclusively on aggregator and SEO blogs, not in any manufacturer document we could reach. So a recalibration interval is an engineering decision the operator owns, and the defensible way to set it is from the mechanisms above rather than from a number with no provenance.

11A verification schedule built from the mechanisms

Since the mechanisms are known even where rates are not, the schedule can follow the events that actually change the reading rather than the calendar. Recalibrate on a fuel-supply or blend change, because that is the documented 50 percent permittivity shift and it is a step change, not a drift. Recalibrate after any tank work, mechanical damage, or a change in mounting angle, since the calibration table encodes the vessel's real shape and tilt. Rerun the monotonicity check periodically as a cheap fault detector, and treat a failure as a fault to find rather than a curve to refit. Verify against a manual dip at a level away from the halfway mark, because at the centreline the linear assumption and the true geometry agree and the comparison proves nothing. Do not add a thermal volume correction to a raw capacitive reading.

12What the sensor specification does and does not promise

Reading the published numbers precisely prevents a common misreading. The Eurosens Dominator states a measurement error of ±1 percent, an ambient operating range of -40 to +85 °C, IP67 protection, a 9 to 32 volt supply and RS-485 output. SOJI publishes ±0.5 percent measuring error for the LIGO Pro across its output variants, with an operating range of -40 to +85 °C, up to 6,000 millimetres of length, and an internal temperature sensor with error below 0.6 degrees. SOJI also markets accuracy of up to 99.5 percent, but qualifies it as achievable under particular conditions, that is the complement of the ±0.5 percent table figure, not an independent claim. None of those numbers is a long-term drift budget. They describe static conversion accuracy under stated conditions, which is why they cannot be used to predict how far a reading will have moved in a year.

13What ENTEK.AI does not claim here

Three absences are deliberate. We publish no recalibration interval, because no manufacturer document we could read states one and inventing a plausible number would be worse than the gap. We publish no numeric fouling rate, because the patent literature confirms the mechanism but no primary source quantifies it. And the tank-geometry figures here are our own computation from the exact circular-segment formula rather than a quotation from API Chapter 2.2, which is paywalled, they are stated as computed and are reproducible by anyone with the same formula, which is why we are comfortable publishing them where we would not publish an unverifiable vendor figure. We also publish no drift statistics from our own installed base, since we have not run that as a dated study with a stated method and sample size.

Figures and sources

• Temperature coefficients of diesel that nearly cancel in a capacitive probe: dielectric 0.032 %/°F versus density 0.033 %/°F (Source: US Patent 4,296,472, Non-intrusive fluid measuring system, as of 2026-08-31)
• The physical reason the two coefficients offset each other: a decrease in fuel density also decreases the dielectric constant, directly offsetting the density change (Source: US Patent 4,296,472, as of 2026-08-31)
• Temperature range over which diesel capacitance measurement is substantially temperature-independent: -10 °F to 190 °F (approx. -23 °C to 88 °C) (Source: US Patent 4,296,472, as of 2026-08-31)
• Dielectric constant range of the hydrocarbon fuel class a capacitive probe measures: 2 to 3 (Source: US Patent 4,296,472, as of 2026-08-31)
• Measured room-temperature dielectric constants of diesel versus biodiesel: diesel 2.2, biodiesel 3.3 (a 50% shift) (Source: Sensors (MDPI), PMC8588433, Electrical characterization of biodiesel–diesel blends, as of 2026-08-31)
• Controlled conditions under which those permittivity values were measured: 1 Hz to 1×10⁶ Hz, 0.2 V, at 21.0 ± 0.1 °C (Source: Sensors (MDPI), PMC8588433, as of 2026-08-31)
• ASTM D1250 / API MPMS 11.1 base temperatures and the required computation order: 60 °F computed first; metric output then corrected to 15 °C or 20 °C (Source: API MPMS Chapter 11.1 Addendum 1, as of 2026-08-31)
• ASTM D1250 metric base-temperature scope: 20 °C incorporated in addition to 15 °C (Source: ASTM D1250 official scope, as of 2026-08-31)
• The API MPMS 11.1 commodity group that governs diesel, by API gravity band: Fuel Oils: approximately -10 °API to 37 °API (Source: ASTM D1250 / API MPMS Ch. 11.1 scope, as of 2026-08-31)
• The thermal expansion coefficient in API MPMS 11.1 is a density-dependent correlation, not one constant: α60 = K0/ρ² + K1/ρ + K2, with K-coefficients per commodity group (Source: NFOGM / North Sea Flow Measurement Workshop, API/IP density referral constants, as of 2026-08-31)
• Volume held by a horizontal cylindrical tank at quarter height (exact circular-segment geometry): 19.55% of capacity at h/D = 0.25, and 80.45% at h/D = 0.75 (Source: Computed from V/Vtot = (θ − sin θ)/2π, θ = 2·arccos(1−2h/D); ENTEK.AI computation, independently reproducible, as of 2026-08-31)
• Maximum error of assuming level is linearly proportional to volume in a horizontal cylinder: 5.45 pp at quarter and three-quarter height; true maximum 5.77 pp near h/D = 0.19; zero only at half height (Source: Computed from the exact circular-segment formula; ENTEK.AI computation, independently reproducible, as of 2026-08-31)
• Teltonika documented calibration procedure and the polynomial it produces: 5–10% tank increments, ~30 s settling; yields cubic coefficients a0, a1, a2, a3 (Source: Teltonika Telematics Wiki, Analog Fuel Sensor, as of 2026-08-31)
• Teltonika monotonicity diagnostic during calibration: voltages must change consistently; up-and-down values mean a fault to find before continuing (Source: Teltonika Telematics Wiki, Analog Fuel Sensor, as of 2026-08-31)
• Eurosens Dominator published accuracy and operating envelope: ±1% measurement error; -40 to +85 °C; IP67; 9–32 V; RS-485 (Source: Mechatronics, eurosens Dominator Specification v.1.0, as of 2026-08-31)
• Eurosens Dominator warranty period, a commercial term, not a calibration schedule: 60 months from the date of manufacture (Source: Mechatronics, eurosens Dominator Specification v.1.0, as of 2026-08-31)
• SOJI LIGO Pro published measuring error across output variants: ±0.5% (AF, RS232 and RS485 variants) (Source: SOJI Electronics, LIGO Fuel Level Sensor datasheet, §IV, as of 2026-08-31)
• SOJI marketing accuracy claim and the condition attached to it: up to 99.5%, stated as achievable "under particular conditions" (Source: SOJI Electronics, LIGO Fuel Level Sensor datasheet, §I/II, as of 2026-08-31)
• SOJI LIGO Pro operating envelope, length limit and stated service life: -40 to +85 °C; 2,500 V isolation; up to 6,000 mm; internal temp sensor error <0.6°; 10 years average service life (Source: SOJI Electronics, LIGO Fuel Level Sensor datasheet, §II and §IV, as of 2026-08-31)
• Probe fouling as a documented capacitive-level failure mode: insulating films of dirt falsify capacitance signals (Source: US Patent 6,405,590 B1, Capacitive filling-level sensor with dielectric coating, as of 2026-08-31)
• Long-term coating stability as an explicit design requirement against multi-year water exposure: coating dielectric constant held constant over many years of water exposure; coating constant between 2 and 7 (Source: US Patent 6,405,590 B1, as of 2026-08-31)
• Eurosens manufacturer rationale for requiring a tank calibration table: a real vessel deviates from the ideal shape: dents, internal ladders and hatches, and installation tilt (Source: Mechatronics, Using tank calibration table utility, as of 2026-08-31)

Questions

Sources

• US Patent 4,296,472, the measured dielectric and density coefficients for diesel
• Sensors (MDPI), electrical characterization of commercial biodiesel–diesel blends
• API MPMS Chapter 11.1 Addendum 1, the volume correction procedure and its base temperatures
• US Patent 6,405,590 B1, dirt films falsifying capacitance, and coating stability

Related insights

• Fuel Sensor Showing the Wrong Litres? Datasheet Accuracy Versus the Field
• Refuelling or Theft? Why Flow Rate Alone Cannot Tell Them Apart in Tank Telemetry
• IoT Hardware at 50 °C: Why the Temperature Rating on the Datasheet Is Not the One That Applies