# MAX31855 and Type K sensor seem to top down at -60°C (should be measuring -80°C)

As a prelude, I'd like to offer that I have obviously read the famous question here: Sensor for ultralow (to -85°C) freezer monitor

I have set up a K-Type probe from Adafruit and their MAX 31855K breakout board with SPI. I can read the data all right, and get consistent results at room temperature.

What is problematic is that when I put this sensor in a -80°C freezer, the probe seems to bottom out at ~-65°C, whereas the actual temperature probe of the freezer shows -80°C.

I though my computations were false (the 14 bit temperature is tricky to bit-shift and 2's complement convert), but I have made the math myself by printing the bits and using paper and I find consistent results with my code.

So yeah.. do you guys know if this is a non-linearity of the sensor at these temperatures?
Is the K-Type thermocouple not adapted? (I thought it could go down to -200°C) Could this be a problem on the MAX 31855K side?
Any hint at what could be wrong?

I'm planning to check different places in the freezer (right now the probe is on the top compartment, I'll try the bottom tomorrow), and another freezer too to eliminate any "freezer-specific" influence, but if there is something I overlooked, can you help me?

• Is the interior wall thermally conductive or insulative? You have have a temperature gradient from top to bottom. Is there forced air? Dec 18 '18 at 18:48
• There is no forced air, these freezers (New Brunswick) are quite old and simple, and I think the interior wall is thermally insulative. Dec 19 '18 at 7:31
• Then I expect it will be colder at the bottom Dec 19 '18 at 7:48

The MAX31855 applies no linearization, the equation:

VOUT = (41.276µV/°C) x (TR - TAMB)

is used.

Thermocouples such as K are more-or-less linear so it's good enough for narrow ranges near where that average occurs.

But if we look at the millivolts for type K with a reference junction of 0°C we get an average of 40.03uV/K over -80 to +25°C so there is an error of about +3.2 degrees C resulting from the nonlinearity.

You may have additional errors due to heat flowing down the thermocouple wires, as @Tony suggested in a comment, but there are other errors-

There is also a span error, which can be as big as 6°C over this range, and another 2 or 3 degrees C from the cold junction error.

Make sure you have the thermocouple wired all the way to the terminal block with proper Type K thermocouple wire or thermocouple extension wire, connected with the proper polarity. If you use copper wire or have a double reversal, you'll have an error of 1-2x the difference in temperature between the two pairs of junctions. Use a small magnet to test the wires and join magnetic to magnetic and non-magnetic to non-magnetic, color codes vary from country to country. Usually, in North America, red is (-) and yellow is (+) for K. The (-) is magnetic.

Of course (+) and (-) refer to the voltage polarity when the tip of the thermocouple is hotter than the cold junction(s), in your case it will be the reverse.

• Hi Spehro, indeed from the sheet of Max 31855 you have a +/- 6°C span. Do you think that adding an offset (-13°C) will compromise the precision of the sensor? I want to monitor the temperature drop during week ends and periods where people are not in the lab, and if the freezer fails I want to be made aware. Dec 19 '18 at 7:41
• I think you should get to the bottom of the error. If it’s due to wiring, for example, you could have issues when the ambient changes. The easiest thing would be to swap the MAX for a known-good thermocouple indicator, though a voltmeter and thermometer could be used if you know how to use them to measure a thermocouple temperature. Dec 19 '18 at 12:26
• As a point of measurement, we dipped the probe in liquid nitrogen (-195°C) and the MAX31855 measured -136°C. I think this span is not linear. Dec 19 '18 at 13:41
• Yes, not linear for sure, however the sum of the errors does not quite add up to what you are seeing, so I am suspicious something else is going on. At -80°C the linearity contribution to the error is only a few °C. Dec 19 '18 at 13:46