Pt1000 thermistor with the ADC Pi V2
This post was submitted by Craig Edwards Dagda, who has been using a Pt1000 thermistor with the ADC Pi board and was getting unexpected results. He explains how he resolved the problems:
Here is what I’m doing.
I wanted to use a Pt1000 thermistor to take temperature readings, it sounds simple enough.
I thought using an ADC Pi would be ideal. Just put the thermistor as half a potential divider and read off the voltage. Bit of maths to work out the resistance and convert it into temperature.
Got it all set up, read the voltages, did the math and came out that I’m working in a room at 5'C!
So checked all again and found my problem (the first problem, but didn't know this at the time) seemed to be the voltage checked with a multimeter across the ADC Pi pins was not quite the same as reported from the Pi. So read the data sheet for the MCP3424 several times, and was getting the right number out, but when converting it to a voltage, the error crept in. So it turned out to be the full scale being defined as 5v, with the tolerance and standard values of the resistors on the front end of the chip, it actually comes to between 5.06v and 5.1v depending on the board (I've got 3 of them). So, the problem is solved, and away we go.
Well, not quite. It got me a bit closer.
I tested the little board with a thermistor on it with a power supply and multimeter, and all was fine, readings and calculations all fine.
Plug it into the Pi, and the voltage across the thermistor was low compared to stand-alone.
Played around quite a while with my wiring but couldn't find anything wrong, even went to using 4 wires (like a Wheatstone bridge) to make sure the wires' resistance wasn't giving me grief.
Turns out that my original maths was wrong.
V2 * R1 = R2
V1-V2
V1 = supply voltage
V2 = voltage across the thermistor
R1 = fixed resistor
R2 = thermistor
Standard PD. Nothing is immediately obviously wrong. Worked okay standalone, too.
However, it's all in the implementation.
The Pt1000 thermistor has a resistance of 1000ohm at 0'C (about 1080ohm at room temp).
So I put
R1 = 1KOhm, giving me a nice middle of the road output at 0'C (it actually was 1000ohm too! on the dot)
V1 = 4.93V, measured on a meter.
V2 = 2.49V, measured on a meter and the Pi.
Giving me
R2=1020ohm
The thermistor has 385ohm per 100'C, so I'm at 5.7'C!!
Where I went wrong was that R2 is not the resistance of the thermistor, it's the resistance of the thermistor and the input resistors on the ADC Pi.
So
1/R2 = 1/Rt + 1/Ri
Rt = thermistor resistance
Ri = input resistance (10k + 6.8k)
1/1020 = 1/Rt + 1/16800
It makes a huge difference
Rt = 1086 ohm, which is a more believable 24'C
So, the moral of my story is
Buffering is great (luckily, the op-amp won't need to swing to the supply rails).
Either that or I may swap the input resistors for 1Mohm and 680kohm, giving the same operation, but 1/1680000 becomes manageable (65>5>1ohm difference in calculated values by increasing resistors by x1, x10 and x100).
Further changes and information:
Trying to calibrate at 100'C does put steam on the circuit board, reducing the resistance, so beware!!
Thermistor calibration is close to the specification sheet (occasional ohm here and there) so you don't need to change any of the parameters to get a good reading.
Upped the fixed resistor to 2K2 (2170 exact) to give a voltage range that can be sampled directly into the MCP3424, keeping the supply voltage at 5V.
However, this will have a problem if the temp goes above 110'C as the voltage will be over the 2.048v reference maximum. I can't see that happening today unless there is a fire!
Oh, and instead of converting the integer reply from the ADC to a voltage, I have used the integers directly instead of the V; it's only used as a ratio anyway. But for direct reading with the 2K2, the V1 of 5v becomes (5/2.048)*maxcount, with maxcount being dependent on the bit resolution. So for 12-bit, it is 5000.
The other alternative is to get the DELTAV1 and put my own resistors and voltage regulator to have a direct reading into the MCP3424.
Just thought you'd like to know any pitfalls in case anyone else gets in touch. It would be a beach to diagnose remotely!
We would like to thank Craig for his detailed explanation of the solution to the problem and for allowing us to publish this on our blog.