Probe physics

A complement to wrong readings — where that page is about what the user sees, this one is about why. Each section is a physics phenomenon that shows up in the data looking like a software bug.

1 · Self-heating

A WiFi sensor in transmit mode draws ~150 mA at 3.3 V → ~0.5 W. That half watt is dissipated inside the same plastic shell as the temperature probe. In a sealed enclosure with no airflow, the sensor reads 1.0–1.5 °C warmer than the ambient air.

Symptom: a sealed-box installation reads consistently 1 °C higher than a separately-mounted reference probe.

Fix: ventilate the enclosure (drill two holes 5 mm in diameter, top and bottom — convection chimney) or accept the offset and configure it as the channel's offset. We see this most often when operators put a Shelly inside a small NEMA box for "weather protection"; the NEMA box is also a thermos.

2 · Thermal mass

A probe glued to a metal plate sees the plate temperature, which lags ambient air. In a fridge that holds 4 °C ± 1 °C, a plate-mounted probe will read 4 °C ± 0.3 °C — flatter, slower to respond, and delayed by 60–180 seconds against the air.

Symptom: a door-open event spike in the air is missed by the probe; the chart looks fine while a separate inspection finds the food was briefly warm.

Fix: mount probes in air, not in contact with metal. If the only mount point is metal, use a thermal standoff (3 mm of plastic between sensor and metal).

3 · Evaporator drip

Walk-in fridges defrost by warming the evaporator coil. Liquid water drips. A sensor mounted under the evaporator gets dripped on, periodically. The wet sensor reads colder (evaporative cooling) by 1–3 °C until it dries.

Symptom: regular sub-zero excursions on a fridge set to +4 °C, clustering at the same hour each day (the scheduled defrost).

Fix: move the probe away from the drip path. Or shroud the probe with a perforated cover that lets air through but blocks splashes.

4 · Door-opening transient

When the fridge door opens, a thin layer of room air pours over the sill and falls down (cold air weighs more, but the temperature gradient moves room-air down too once equilibrium starts). A sensor in this stream sees 12–15 °C briefly before the door closes and the compressor catches up.

Symptom: short (5–15 s) spikes during business hours, regular pattern (every 3–10 minutes during service).

This is not a bug; it is the real temperature next to the door. But it is not the temperature of the food in the back.

Fix: do not mount probes near the door if you are monitoring the food. Move to the back. Or accept the spikes and set the grace period to 60 seconds.

5 · Stratification

A walk-in fridge is not at one temperature. The air at the top can be 6 °C; the air at the floor can be 1 °C. Two probes at opposite vertical extremes can read 5 °C apart and both be correct.

Symptom: two sensors in "the same fridge" disagree by 4 °C persistently.

Fix: this is the design of the fridge, not a sensor problem. Either:

• Mount both probes at the same height so they are measuring the same air mass. Or:
• Mount one probe at the warmest stored item and use it as the compliance reference. The other probe is informational.

6 · Window draft (bonus)

For indoor climate sensors in offices and museums: a temperature sensor near a window reads ambient (cold in winter, hot in summer) not room average. A humidity sensor near a window reads condensation or evaporation depending on the season.

Fix: mount climate sensors away from windows, away from HVAC diffusers, away from radiators, on internal walls when possible.

Summary

If you can match your symptom to a row, you do not have a software bug. Move or shroud the probe, then look at the data again before opening a ticket.