How far to trust it
The ±1 % parts alone allow about half a degree either way near 25 °C, under a degree up to 50 °C, and more as it gets hotter. The B model, the ADC and the board it is plugged into come on top. Around a degree near room temperature is the honest claim.
What the parts allow
Three parts set the answer, and each is only ±1 %: the thermistor's 10 kΩ at 25 °C, its B value, and the fixed 10 kΩ above it. The sketch assumes all three are exact. The figure works out how far the answer lands from the truth if each one is off by its full tolerance, alone and all together at their worst.
Near 25 °C the B value hardly matters. The B equation is pinned to the thermistor's 25 °C resistance, so an error in B only shows as the temperature moves away from 25. What is left is the two 10 kΩ values, and together they allow about 0.45 °C either way.
Away from 25 °C the B tolerance grows. By 50 °C the three together allow about 0.8 °C, and by 100 °C about 1.7 °C.
These are derived from the equations, not measured. The ±1 % on B is itself read from the part number, not from a datasheet.
What it leaves out
Three things are not in the figure, and each adds to it.
- The B model itself. One number cannot follow a real thermistor's curve exactly. It is closest between 25 and 50 °C, where B is usually measured, and drifts further out. How far needs the part's resistance table, which was not available.
- The ADC. On an Uno one count is about 0.09 °C near room temperature, so counting is not the limit. On an ESP32 the reference and the converter's curve add their own error, and the assumed 3.3 V adds about half a degree per per cent.
- Where the block is. It measures its own temperature, which is the air's only if nothing nearby is warmer. How fast it follows comes back to this.
Put together, around a degree near room temperature is a fair expectation,
and worse far from it. For better, calibrate: compare it with a thermometer
you trust at the temperature you care about, and adjust R25 in the sketch
until they agree.
When it does not work
Calibrate it against a thermometer you trust. Put both in the same still air for a quarter of an hour, note the difference, and correct for it: nudge R25 in the sketch until the two agree. One point fixes the error near that temperature; it does less far from it.
No. Averaging made the number steady to about a tenth of a degree, which is why the second decimal stops jumping about. The parts are only good to a few tenths. Steady and correct are different things.
Not out of the box. A digital temperature sensor is calibrated at the factory and reports degrees directly. The thermistor is cheap, small and quick, and with a one-point calibration it can be good near the temperature you calibrated at.
A three-constant version of the curve that follows a real thermistor more closely than the one-number B model, especially far from 25 °C. It needs the part's resistance at three temperatures, from its datasheet table or your own measurements. With only the part number to go on, B is what this book uses.
Seconds for heat, half a millisecond for the capacitor, and why the current does not warm it.
How fast it follows →Edit this page — content/books/ntc-thermistor/how-far-to-trust-it.mdx
Questions about this product
See what other owners have asked, and read their solutions.
NTC Thermistor
Loading discussions…
Discuss this article
Ask about this page. The answer stays here, on the page it belongs to, for whoever hits the same wall next.