Calibration of TMP36 sensors on Arduino?

Question

So I had a chance to try this out:

I put 3x TMP36 sensors side by side on a breadboard, hooked it up to use the 3.3v reference for better accuracy, set up the code to discard the first reading (to settle the ADC) and then take the average of 8 analogRead's with 20 ms delays between, and this is the result:

Temperatures = 25.04, 24.72, 24.40
Temperatures = 25.37, 24.40, 24.08
Temperatures = 25.04, 24.40, 24.08

The list is much more steady than that, but I copied the part that shows the bit of fluctuation that happens every so often - it steps up/down in increments of (reference voltage / 1024), which for my 3.3v reference is about 0.32 degrees. In other words, one point difference in the analogRead result equals 0.3 degrees. Using a 5 volt reference therefore would make that about a 0.5 degree increment.

Now, my handheld laser-guided temperature reading doohickey says it's about 24 degrees C in here - I know! it's a warm day today!! (I live on the very temperate West Coast of Canada)

So obviously there is a difference between sensors, as the first one is nearly a degree higher than the last one.

How would you normally adjust for specific sensors? Or would you adjust?

For example, if I trust my handheld reader, I would say that (based on the values of the last line) sensor 3 is perfect, and for sensor 1 I would (in the code) always subtract a degree, and for sensor 2 I'd subtract 0.3 of a degree.

But I'm a software guy and I'm betting there's a very clever hardware solution people would use for this.

This is for a one-off project, not sure if that would change the approach.

Also, the plan is to use 2 of these sensors to track temperatures inside and outside of my greenhouse. In my area I would not expect temperatures to exceed the -20 to +50 C range. Doing the math, +50 comes to 1 volt. Is there a way (an easy way?) to set the reference voltage to 1 volt so I can get accuracy of 0.1?

Thanks!

-Nico

neilk · Accepted Answer

Hi Nico Yes I have been quiet for a long time and no, sadly, I haven't completed my Greenhouse temperature control project yet. Most aspects of it are designed and in most case tested and proven. I haven't yet integrated the whole system! I too gave up on the idea of Gobetwino for this particular project and am attracted to the idea of of using a WiFi shield with the Arduino - maybe! You asked a number of questions about error, accuracy and calibration. Measurement error is normally divided into Systematic and Random : Systematic error is like a constant offset, eg a 50g weight that has had achunk filed off it so it only weighs 49.5g Random error is the fluctuation you get between readings that you hoped would be the same value. Accuracy is around making measurements which are appropriate to the measuring device AND to the reason for taking the measurement in the first place. So, if, at the ADC level, 1 point in 1024 represents 0.3 degrees, that suggests to me that quoting the temperature more accurately than the nearest degree is pretty optimistic. Then we come to why are we measuring in the first place? To control my Greenhouse, I want to switch on a heater when the temperature falls below about 5 degrees and back off again when it rises to about 10 degrees. I also want to switch on a cooling fan when the temperature rises above about 25 degrees and back off again when the temperature falls below about 20 degrees. So an accuracy of + or - 1 degree is more than adequate (for my application). Applying my crude approach to your readings, above, would give: Temperatures = 25, 25, 24 Temperatures = 25, 24. 24 Temperatures = 25, 24, 24 In otherwords, within + or - 1 degree for a device and even between devices. As for calibration, my memory of School Physics (which was a very long time ago!) tells me that: 0 degrees C is pure water containing ice made from pure water, well stirred, of course. 100 degrees C is steam over boiling water. Both are assumed to be at 1 standard atomsphere ie 760mm Hg. I guess anywhere close to sea level is near enough! See here to become a temperature calibration expert: http://en.wikipedia.org/wiki/Thermometer#Calibration And finally, if I was happy that my old fashioned thermometer (double checked, as above) was accurate, or, if it had a systematic error then I knew what that error was and could correct for it ( I have one that consistently over reads by 2 degrees C), then I would compare my sensors with the thermometer at a series of temperatures in the range I was expecting them to work. If I found that a sensor regularly disagreed with the thermometer by a significant amount, then I could correct for that in my software.........however, that would complicate the software and require different segments of code, which would need to line up with the correct Arduino pins and sensors.....could get a bit messy; might be easier to get hold of some more sensors and select a group which all have the same systematic error, which could be zero if you are lucky. Oh dear, I have just read this through again and it sounds really pedantic and maybe a little patronisisng - it's not meant to! Again, many years ago, measurement error and appropriate accuracy was part of my life! Nico, I hope you find some use in this Best wishes Neil

mcb1 · Answer

Nico

My calculation is that you are somewhere in the (750mV) A0=232 at 25degC, which is less than quarter of the fullscale (assuming 3v3 supply)

Using a 3v3 supply I calculate each step is 0.0032258mV

Another method is to change the analogue reference to 1.1v

http://arduino.cc/en/Reference/AnalogReference

My calculations are that you could read up to 60deg C using the 1.1v internal reference available on a 328 and the adafruit formula.

Temp in °C = [(Vout in mV) - 500] / 10

Mark

I also answered your other post.

mcb1 · Answer

Neil/Nico I did something you describe for 'The Shed' magazine as part of our 101 series. The article and software is here (part 4 included the LCD in both normal and I2C version) http://www.theshedmag.co.nz/online/arduino I used Dallas temp chips to ovecome most of the calibration issues There are some around that are already encapsulated and have a lead on them, to make it easier. The other option is to solder leads and insulate the wires, then apply heat transfer compound that is used for transistors, and fit them into a piece of metal tubing. The tube is sealed at one end and the other end fits into a compression gland commonly used in the electrical industry to hold a cable as it passes through an enclosure. The tubing should be close to the size of the device, but as long as its making contact it will take on the temperature of the metal tube. Mark

Calibration of TMP36 sensors on Arduino?

Top Replies