I’ve got a bunch of temperature/humidity sensors - and when I say a bunch I mean a lot, everyone seems to be integrating them into their devices because they’re so cheap - of unknown accuracy beyond the manufacturers claimed specs. You can get precision sensors like the Seeed S-THP-01A which for temp/ humidity uses an SHT45 rather than the usual DHT11 or similar, but they’re not cheap (their I2C models are cheaper but I like the SenseCAP’s Modbus interface since I have Modbus run everywhere so can easily compare to existing sensors, and I know about salt calibration but just want to do an easy A/B comparison next to existing sensors).
So in theory I could check the accuracy of everything I’ve got, but does it matter? For example this thread covers it peripherally but it’s more about precision, getting the same (or close enough) readings from a group of sensors. Has anyone else bothered to check the accuracy of their cheap sensors against a more accurate one, and what benefits did it bring apart from scratching a geek itch?
The reason you check accuracy is that in some instances the small differences and reporting times of the sensors vs others can be a critical aspect to certain setups.
This testing is also done with other types of sensors and devices to ensure reliability and quality of them rather than just saying here buy the cheapest one you can get your hands on as you don’t need to spend too much and you don’t want to cheap out since in both ends of the spectrum it can cause headaches in the long and short term.
I’m using them more for curiosity than anything else, I’m not (for example) keeping a wine cellar at exactly the right temperature but more checking up on things. So ATM the best justification I can find for getting a precision sensor is not being annoyed when two existing sensors show slightly different readings, which isn’t a strong motivator for getting one.
I happen to have a high-precision old-style “analogue” mercury thermometer with 0.2 °C resolution (but you can literally read between the lines for 0.1). Obviously I can’t know if the absolute value is precise as it isn’t calibrated, but its quality and linearity itself is most likely much better than a Chinese 5$ sensor’s readings.
I’ve compared all my 10 Tuya T/H sensors against it and found that except one they all give 0.3-0.5 °C higher readings. Three of them have a digital display (those are ~10$), and I simply used a marker to put a “+0.3 °C” comment on it.
I could create helpers in HA to apply the temp correction and get an excellent result, however, there is no way of adjusting what a users sees on the digital displays.
Sounds like a fun project.
If you can find 1 sensor that you trust, you could ‘calibrate’ others to match that sensor.
Take a number of measurements and record the difference. The more measurements the better.
Then take the compensation integration and apply the table, and use the output of that instead of the actual value from the sensor. Now this assumes the sensor is stable and will provide you a repeatable error value. Some might not be repeatable enough but you get the idea.
Usefulness, debatable. Nerdy and OCD, fun, absolutely.
That delta could be due to the heat of the electronics in most cases whilst one of them could of been calibrated to account for it whilst the others were not.
Like you (OP), I have many temp sensors - 21 in all.
I use 12 of them which I calibrate annually by placing them beside my thermostat (which I take as being my ‘standard’) for 48 hours and comparing the difference. These are either Hue motion or Aqara temp/humidity sensors and I have 1 per room between them. The Aqara ones are the most accurate and have never needed to be adjusted. The Hue are typically adjusted +/- 1°C.
I ignore all the others as, like you, they were available on devices that I use for other things (like contact sensors). I do have a dashboard where I could see them in case something weird happens (but never needed to so far).
Not quite right. Digital thermometers (also hygrometers) with displays can be owned if one actually cares. Therfor it is possible to calibrate them directly on the device.
But obviously that is not the case when just buying random stuff. We have plenty of this xiaomi thingies which are running pvvx firmware and allow local calibration beside many other features
It goes even that far that states from HA can be shown on the display which is quite a usefull addition we make use of
I use a calibrated kitchen thermometer to have a quick look if new temp sensors are reporting correctly. If anything is off by 1 degree or higher, I calibrate. Lower than that, I don’t bother - it’s not like I can tell the difference anyway.
The only benefit to calibration is that you can use the same temp values in automations for all sensors without having to remember that some need higher/lower values.
Other than that, there’s no real benefit, especially if you’re mostly interested in the temperature change rather than an absolute value.
Thanks. The first link in particular was informative, I hadn’t seen that one when I searched. So it was just my geek OCD telling me I had to do it but no particularly strong argument for it.
Was going to post the same thing, a lot of the cheap sensors read consistently high for temps and low for humidity because the DC/DC converters or other electronics are heating the sensors. Better-quality ones, or even cheap ones but with a bit of design thought put into them, isolate the sensor from the other electronics to avoid this problem. For example this style, which are as cheap as you can make them, nevertheless isolate the sensor package, electronics inside the case, sensor outside.
The Modbus sensors I used all have calibration offsets that you can set to compensate for readings that are off by a bit, but you’d need to verify that the offset is consistent over the expected temperature range rather than just at the one point where you did the comparison to a better-quality sensor, at which point you’d need something like a Peltier/TEC to get the readings at various elevated/lowered temps, and at this point it’s starting to look too much like actual work. Possibly if you ran some electronics blog or YT channel you could justify the effort by doing a writeup on different products, but certainly not worth it for “is my basement too damp” measurements.
This is really interesting. I was not aware of what is possible before following up on your comment and discovering the ESPHome project. Wished I had known this earlier. Well, I only bought 11 devices in total for maybe 60 EUR all together.
I may give your XIaomi recommendation a try when time allows. That said, my focus is to build on Matter/Thread devices with a ZBT1 as the radio unit. That’s why I chose WiFi for the temp sensors rather than Zigbee.
But getting a little bit off-topic in this thread now.
