Issue with temperature accuracy

[chafouin]

Hello,

I am a happy customer, although I am running into issues for the first time.

I have a breakout for the BMP280 (PID 2651) and also a Clue which has a BMP280 on it (PID 4500)

The temperature reading from the two boards are 4 to 5 Celsius apart: the clue has a much hotter higher. (similar position / orientation)
I have yet to do more testing for myself, but I did not see anywhere on your site or documentation that a the BMP280 REQUIRED a calibration for the Clue.
Granted that any components would benefit from a manual calibration (although some components' documentation like the TMP117's mention that they are calibrated by the manufacturer and the minimum electronic on the TMP117 breakout should allow us to believe the reading "at face value".), but in this case, it's a huge 4-5C difference for the same component on 2 different boards (and of course, this translates in crazy values in F )

So my question is the following:
Did I just get a Clue board that is a lemon?
OR
Are all the Clue boards like that and users are required to perform a calibration (hopefully just an offset) to get a meaningful temperature. (if so, you may want to consider a default and configurable offset in your Clue's API when reading the temperature)
AND/OR
Have the Clue boards some hardware design issues, distorting the temperature reading? (electronic noise, temperature emitted from nearby components, etc...)

Note: I have the most recent Stemma Qt sensor breakouts for temperature as well (BMP280, TMP117, Si7021, etc..), and they all are within a reasonable temperature range (given their spec). So I can confirm that it's the CLue that has an abnormally high reading for the temperature. I have not tested the other sensors embedded on the Clue boards.

Hopefully, your upcoming designs, which seem to include BMP280/BME280 a lot, do not suffer from this (I would prefer the TMP117 btw).

Thanks for your guidance !

[adafruit_support_bill]

The Clue board has a lot of other circuitry besides the temperature sensor. It is not unusual for the on-board temperature to read several degrees above ambient due to the other heat sources on the board.

In general, all sensors should be calibrated for critical applications: https://learn.adafruit.com/calibrating- ... -calibrate

[chafouin]

Looks like it's a known issue as reported here: https://learn.adafruit.com/clue-sensor- ... -3060365-2
What practical approach do you recommend to do an accurate calibration without buying an additional NIST certified thermometer?
(Dropping the board in boiling water or melting ice is not practical :) )

Would you recommend that the single BMP280 breakout to be used to calibrate the Clue (assuming the single BMP280 breakout's circuitry does not impact the measurements) ?
OR maybe using a TMP117 breakout ( also assuming its circuitry does not impact the measurements) ?
Would the variance in temperature be significantly different from the component's datasheet?

Thanks for your guidance.

[adafruit_support_bill]

How critical are your measurements? If you simply want it to agree with other temperature sensors in the room, calibrate it against those. A one-point calibration is probably sufficient. However, do note that the self-heating effect of circuitry on the Clue may vary with processor activity and air circulation.

If you are monitoring some temperature sensitive process, then a Clue is not your best choice. You should use a standalone temperature sensor isolated from other circuitry and in close contact to what you are measuring.

[chafouin]

I want to use it as a room temperature/humidity probe, relaying the metrics to a hub that would trigger the furnaces/ACs and smart vents accordingly.
This means that I would have a Clue, or similar boards, in each room of the house and I want ideally the temperature at a -+0.2C accuracy.
The Zigbee or Zwave temperature probes commercially available that I have tried are not satisfying in many ways. My project is to investigate doing it myself with components of my choice.

The BMP280 was a compromise for me since it does not have a +-0.2C/0.5F accuracy in the first place (I like better the tmp117 sensor for that), but it had the advantage of being embedded in designs such as the Clue that is giving the opportunity for the user to interact (could offer the opportunity for the user to bump the programmed temperature for example). Truth be told, my application does now require an "exact grand Truth" from a sensor, but a consistent and repeatable read overtime and across devices. There are different ways to mitigate the issue, I will test and experiment further with the Clue)

I would suggest however to make clearer that the temperature read out-of-the-box from the Clue is way off. I would think that people are checking the components' datasheets and expect the performance of the boards to offer similar characteristics, but a 4-5 C (10F) gap in temperature measurements is just too much.

I believe that a helpful and elegant way to address this could be to change your circuitpython API to offer a call for the component's raw read (which is what you have today) and a call for a calibrated read such as "self.clue.temperature( f_offset = default_offset) ". Since you are manufacturing these boards, I would believe that you can take a sample and offer a basic educated "default_offset" value that will go a long way in delivering a better-estimated value of the temperature for an out-of-the-box usage of the Clue boards.
This also makes a lot of sense from the perspective of "amateur tinkers" who would rather spend their focus and money on purchasing electronic components for their projects than spend on certified tools to calibrate them. Such certification tools (like a certified thermometer to test boards for example) are an insignificant cost at the scale of a manufacturer's usage, but it is not an insignificant cost for an amateur.

Thank you for your support, I don't think this post calls for an additional answer. I am looking forward to playing with the Clue and other Adafruit boards embedding sensors, and I will continue my testings.

[adafruit_support_bill]

I want to use it as a room temperature/humidity probe, . . . . and I want ideally the temperature at a -+0.2C accuracy.

Temperatures within a single room will vary by many times that much. Sunlight, human occupancy, electronic devices and convective currents create all kinds of microclimates within an enclosed space.

[kevinjwalters]

The Bosch data sheets actually mention that the temperature sensors are intended to be used to calibrate the other onboard sensors and not for ambient temperature measurement. I hadn't realised this til recently.

I've assembled a humble temperature sensor collection, TMP36, LM35 and Waterproof 1-Wire DS18B20, and plan to go through those when I get the time. This is to supplement an Enviro+ FeatherWing. I'd be interested to know how you get on.

There's an important detail for many analogue temperture sensors mentioned at the bottom of Adafruit Learn: TMP36 Temperature Sensor. Their high impedance requires care with the common multiplexing approach for ADC on microcontrollers.

I agree on the differences across a room. From an 8833 I can see my floor is a few degrees colder than the air and my (double-glazed) windows are colder still.