With the exception of the post-norm validation, the process is fairly generic, and can be applied to any kind of sensor that you are deploying in groups of 10 or more.
I always do my calibration to a traceable known standard which is at least one decade better then what my unit will measure. Simple example: a 3.5 digit is checked and possibly calibrated with a 4.5 digit that is in calibration. From what I could get from your post it appears it does not use the absolute value but the mean value of the sensors.
This was not a calibration - just normalization. When you deploy a set of sensors into, for example, a cave system what you are concerned about is the differential between the passages which gives you the direction of flow. This can be small enough to be below the offset spread of the set. So we normalize those sensors to each other because -in this use case- its only difference between the sensors that matter.
I simply showed the mean vs the airport sensors to demonstrate that, with a random set of sensor offsets, normalization can give you an accuracy improvement, though that's rarely the case if you buy all your sensors from the same production run. Also we are aiming at typical classroom settings where they don't have the budget for a nice Vaisala climate station.
What you would suggest as an affordable calibration reference sensor for barometric sensors?
@EKMallon Very helpful post, thanks! I hope it gets the attention that it deserves, as there is a lot of confusion over what sensor data sheet specs actually mean, and what range of sensor-to-sensor variations to expect.
I recently saw a very angry thread on the Adafruit forum, from a user who demanded a refund for a properly functioning barometric pressure sensor, because it showed a (correctable) offset to the values produced by another of the same type.
I remember many years ago when Motorola came out with pressure transducers. I got a few and yes they are still right on for over 20 years 24/7/365. I beleive that process is now owned by NXP Integrated Pressure Sensor (0 to 10 kPa) | NXP Semiconductors I got several of the NXP a few years back and they are still working flawlessly. They make both Absolute and Gauge types available in single ported and duel ported (differential). They are very reasonable and of great quality. To make life easier pick one that has the amplifier on board, then no fancy external analog circuits and they will work directly with an Arduino.
I will need to think for a while about how to build a calibration setup with those NXP's. I want to design a box that the students can just put all their builds into for a few days between classes, so we can do the correction exercise with that data (ideally thermal as well)
It appears the NXP's have 5-7% accuracy error, or about ±5 millibar. That's on par with the datasheet spec of most barometric sensors so doesn't seem to provide that decade you were mentioning? Or did I misread something there?
I'm hoping to get more posts out this year on "What real-world data looks like.". That material isn't very flashy, so it probably won't get much traffic. But one of the biggest challenges we face is students think their loggers are working 'perfectly' once they are running & gathering data. People are so used to technological things that 'just work', because they rarely get exposed to the staggering amount of error correction that is going on in the background around them.
That should be relative easy, following boils law. Use whatever the most accurate device to calibrate against you have. I have used water columns many times. Since are mainly interested in the difference which can be adjusted in software. I think you have a good solution. Hint connect all of the sensors to the cal setup at the same time, eliminates lots of errors.