Sensor calibration
When to do it, how, what to record
OpenSense does not require calibrated sensors. The default Shelly / Aqara / Efento ship calibrated by the chip vendor (Sensirion, typically ±0.4 °C ±2 %RH out of the box) and that is sufficient for most HACCP and climate use cases.
Calibration becomes relevant when:
- A pharmacy or hospital pharmacy auditor asks for traceability.
- A sensor has been in service long enough to drift (typically
2 years).
- A discrepancy between two adjacent sensors makes you suspect one is wrong.
- You replaced a sensor and the new one reads systematically differently than the chart you have on file.
What "calibration" actually means
Three different things use the word:
- Factory calibration — done in production, baked into the sensor's internal offset registers. Sensirion publishes the datasheet accuracy.
- Field check — comparing the installed sensor against a portable reference for 5–10 min, recording the delta. No adjustment is made.
- Field calibration with offset — same as field check, plus applying the measured delta as a configured offset in OpenSense.
In OpenSense we support all three; the first is the chip's responsibility, the second is operator hygiene, the third is the "channel offset" field.
What you need for a field check
- A reference probe: a calibrated thermistor or PT100 with a
recent (< 12 months) calibration certificate. Examples:
- Fluke 1551A "Stik" — €700, ISO 17025 cert from Fluke.
- OMEGA HH377 + Type-T probe — €350.
- For non-pharma use: a Sous Vide-grade ThermoWorks ThermaPen — €100; not certified but reliable to ±0.2 °C against a Fluke.
- A stable environment: not in a fridge being opened, not in direct sunlight, not next to a heat source. A still-air room at 20–22 °C is ideal; an ice bath at 0 °C is the second reference point (single-point calibration on ice + a second at room temperature gives you a two-point line).
The procedure
- Note the installed sensor's reading right now.
- Place the reference probe next to it (within 5 cm). Insulate both from local air movement if you can.
- Wait 10 minutes. Sensors have thermal mass; the reading you see in minute 1 is not the steady-state reading.
- Record both readings simultaneously. Compute delta.
- Repeat at a different temperature if practical (ice bath; a freezer at −18 °C; a fridge at +4 °C).
- Compute the offset and the slope (if you took two points).
Apply the offset
In OpenSense:
PATCH /v1/sensors/snr_a1b2
{
"offset": -0.3,
"offset_recorded_at": "2026-05-17T10:00:00+02:00",
"offset_recorded_by": "ops@cafe-bratislava.sk",
"offset_recorded_against": "Fluke 1551A s/n 12345 cert 2026-02"
}
The offset is applied post-ingest — the raw reading is stored
unchanged; the displayed and rule-evaluated value is raw + offset.
This is intentional: if you ever doubt the offset is correct, you
can recompute history with a different one without losing data.
The audit trail records the offset change as a configuration event.
How often
| Use case | Recommended cadence |
|---|---|
| HACCP, food retail | Annually |
| Legionella, hot water | Annually |
| Server room | Every 2 years |
| Indoor climate (museum) | Annually |
| Indoor climate (school) | Every 2–3 years |
| Pharma cold chain | Bi-annually + after any service |
If you cannot afford a calibrated reference probe but want a zero-cost sanity check, every 6 months put all your sensors in the same room for 60 minutes and compare their readings. If one drifts
1 °C from the median, it is the suspect; field-check it against a borrowed thermometer or buy a replacement.
What the auditor wants to see
For HACCP / Slovak ŠVPS in our experience:
- A note that "calibration was performed on YYYY-MM-DD with reference X, against probe Y, delta was Z °C". The dashboard's calibration field accepts free-text; auditors read it.
- A copy of the reference probe's certificate (or a serial number pointing at one). The certificate need not be in OpenSense; we have a per-account "documents" attachment area you can put it in.
For pharma / GxP — which we do not service today — the auditor will want a full traceability chain back to a national metrology institute. Buy Eupry or Vaisala for that case.
When in doubt
If a sensor reading does not match reality and you cannot calibrate it, replace it. A Shelly H&T is €30. It is cheaper than an hour of your time arguing with chip drift. Keep the old sensor in a drawer labeled "for the office plant" so you do not throw away a working device — but do not have it on the food.