Fuel level sensors: types, calibration tables, and noise smoothing

Why the dashboard float is not enough

Every vehicle leaves the factory with a "fuel sensor": the float that drives the dashboard gauge. But that gauge was built for one purpose — letting the driver know when to head for a station — not to be a measuring instrument. Its response is non-linear, it saturates near full and near empty, it shifts with vehicle tilt, and it can sit on "full" for the first quarter of a trip before dropping quickly afterwards. Building fuel control on that signal is building on sand.

Serious fuel management — actual consumption figures, refill and drain detection, invoice reconciliation — needs measurement in liters with a known, stable error margin. That leads to one of two paths: a dedicated level sensor installed in the tank, or reading fuel data from the CAN bus on vehicles that expose it with sufficient quality. In both cases, the reading then passes through two indispensable steps, which are the subject of this article: calibration, then smoothing.

Sensor types: a practical comparison

Four main approaches to fuel-level measurement compete in the market, and each has its rightful place:

The practical rule: start with what the vehicle already provides over CAN and evaluate its quality across a full week of operation. If it is intermittent or coarse-grained, a capacitive probe is the right investment for the vehicles where you want dependable theft detection. Many fleets end up with a mix: CAN for modern vehicles, dedicated sensors for heavy trucks and equipment. See the devices page for supported protocol families — 18 tested families today, including fuel-sensor specialist Omnicomm alongside Teltonika, Ruptela, and others.

Calibration tables: how to build one, step by step

Whatever its technology, a sensor outputs a raw number: a voltage, a frequency, a percentage. Turning that number into liters requires a calibration table that says: reading 1200 means 40 liters, reading 2400 means 95 liters, and so on. This table cannot be derived mathematically from tank dimensions except for perfectly regular tanks — which are rare. Most tanks have curves, recesses, and internal baffles that make the height-to-volume relationship non-linear.

Building the table properly is a disciplined field procedure:

• Start from a near-empty tank and record the sensor reading there — that is your practical zero point.
• Fill in known increments (20 to 40 liters from the pump meter), and wait after each increment until the reading settles before recording it.
• Densify the points near the bottom and the top, where the tank's shape curves and the relationship accelerates or flattens; the middle section tolerates wider spacing.
• Record every pair (raw reading → liters) in the table; the platform interpolates linearly between points at runtime.
• Recalibrate after any intervention: a sensor replacement, a tank repair, or a systematic drift you notice between invoices and measurements.

And remember: the table belongs to the individual tank, not the model. Two trucks of the same model can carry tanks of different shape or capacity depending on trim. Copying a table from one vehicle to another "to save time" is the fastest route to a measurement you believe is accurate while it quietly misleads you.

Noise smoothing: why, when, and how

Assume a perfect sensor and a meticulous table; a purely physical problem remains: fuel sloshes. Every turn, braking event, and pothole sends a wave through the tank, and the reading jumps by whole liters within seconds. Add fuel expansion and density change with temperature, plus electrical noise on the supply wiring, and the raw curve looks more like a cardiogram than a consumption line.

Smoothing is the answer, and its essence is a careful balance between two opposing sensitivities:

• Too little smoothing: noise passes through, the log fills with false "drain" and "refill" events, and alerts lose credibility within days.
• Too much smoothing: the curve goes numb, real drains are detected late, or a small event dissolves into the average and is never seen at all.

Sound practice combines a moving time window that absorbs slosh with awareness of vehicle state: a reading taken while parked with a stable engine state is more trustworthy than one taken mid-turn, and a "sharp drop during a long stop" deserves different sensitivity than a gradual slope while driving. On the Pixa platform, anti-fluctuation smoothing is applied to the series after calibration and before event detection, so the fuel screen receives a clean curve and refill/drain events classified by volume, time, and location — with an instant alert on drains.

From raw parameter to the vehicle card

None of the above logic is exclusive to fuel. Pixa builds it on a general sensor-abstraction layer: any raw parameter arriving from a device is defined by a name, a formula, and a calibration table — fuel, weight, multi-zone temperature, doors, PTO, passenger count, RPM, and more. For a fleet operator, the practical benefit is threefold:

• Freedom in hardware choice: a new sensor or an unusual parameter does not wait for "official support"; it is defined with its formula and table, and it works.
• A raw-data viewer: during installation and calibration you see the raw reading exactly as the device sent it, side by side with the converted value — so a wiring or calibration mistake is caught in minutes, not weeks.
• One vehicle card: all calibrated sensors appear on the vehicle card on the live tracking screen, so a supervisor reads fuel, temperature, and doors at a single glance.

A checklist before final sign-off

Before declaring a vehicle's sensor "ready" and enabling alerts on it, verify these items:

• Solid mechanical mounting and a sound seal, with the capacitive rod cut to the tank's actual depth — not the assumed one.
• A documented calibration table with enough points, densified near the bottom and top, stored under the vehicle's name rather than the model's.
• A test refill: add a known quantity from the pump meter and compare it with what the system recorded — a constant offset is acceptable and correctable; an erratic one means an installation or calibration problem.
• A full week of real operation under observation before enabling drain alerts, to tune detection thresholds to the vehicle's actual behavior.
• A recheck scheduled within maintenance plans — by kilometers, engine hours, or date — so measurement accuracy does not erode silently over time.

Planning to fit fuel sensors across your fleet, or doubting the accuracy of the ones you have? Contact the Pixa team via the contact page to review the right measurement options for your vehicles and calibrate them correctly from day one.