Announcement
Collapse
No announcement yet.
Measuring Roast Temperature with thermocouples
Collapse
X
-
Part 3. Testing.
TLDR: A couple of simple tests will tell you whether a thermocouple device with a non-standard termination is accurate.
The voltage produced by a thermocouple is tiny, with the Type K it is about 41 microvolts per oC so even with a 200 degree temperature difference end to end we are looking at reading 8 mV. It is not a trivial task to read a voltage that small with any accuracy. One reason thermocouples are used despite the difficulties this causes is that they are predictable: since the characteristic is well known, they don’t require fancy multipoint calibration: if it reads correctly at one point, it should read correctly over the range.
The opposite of this is also true: if it reads incorrectly at any test point you can safely assume it is inaccurate across the range so a single point accuracy test is very informative.
If the device uses a non-standard termination the most likely cause of inaccuracy will be due to variation in temperature between the actual cold junction temperature and the reference temperature used by the receiver chip. Fortunately it is not difficult to test for accuracy and to vary test conditions so that the source of the inaccuracy becomes evident.
Easily available standards for single point measurement are 100oC (the boiling point of water at sea level but not, for instance, in Canberra) and 0oC (the temperature of a 50 / 50 mixture of water and ice). Although the first is closer to the temperature range in which we are interested, to get accurate results you need to take account of local barometric pressure and apply a correction so the second is more practical. If we combine this with deliberately biasing the cold junction temperature away from the device temperature then letting them equilibrate we can get a lot of information about the performance of the device.
To achieve this I placed the thermocouple with its lead and plugs in the fridge whilst keeping the device box and USB lead at room temperature. Meanwhile, I prepared the 0oC temperature standard by mixing an equal weight of ice and cold water in an insulated container. I checked the temperature of this with a NATA traceable calibrated thermometer but this isn’t necessary, it really only confirmed that the thermometer was in calibration. When everything was nicely equilibrated I placed the thermocouple into the ice water, plugged it into the device, connected everything up and watched the reading.
I got an initial error of +24 oC which gradually drifted back until it settled at +6 oC. As a reality check I asked CSer Dimal, who uses the same device, to do the same experiment with his device. His result was a similar pattern with an even larger discrepancy.
Another test that demonstrates the problem is to simply track the drift caused by self-heating. All electronic devices self-heat in use because the energy they use is turned into waste heat which increases the temperature of the device until it loses as much heat to the environment as is being created by the energy consumption. If the cold junction is not thermally coupled to the receiver, the receiver will heat more than the cold junction and this should be detectable by tracking from a cold start. This has the advantage of not needing a reference temperature.
To check this, I simply placed the thermocouple and the device in a reasonably constant temperature air conditioned room while they were not connected to any power source and let them equilibrate to temperature for a few minutes. I then connected them up and started a dummy roast run, watching the temperature indication. It drifted upwards by about 5 oC over a period of a few minutes then settled at that level. This was quite repeatable, disconnecting the device and allowing it to cool then reconnecting it displayed the same behaviour.
If you have a thermocouple device with non-standard connectors I encourage you to test your device and post the result.

After I modified my device as outlined in part 2 ( Pic above) I repeated the tests.
On the iced water / fridge test I got an initial error of 6oC which gradually drifted back to about 1 oC plus noise: not perfect but much better.
The self-heating drift is now buried in the noise floor of the device which is around +/- 0.2 oC.
Does it make a practical difference? Yes.
Since I fixed my device I have had to move all my temperature profiles down by 8 - 10oC from the values I was using. The results above confirm that this is because the device was reading too high due to a temperature difference between the cold junction and the receiver chip.
I think the reason the difference is greater than the room tempreature drift error is tied to why I began questioning the accuracy in the first place. It is very cold in winter where I roast : the very thick granite walls mean the temperature inside the building stays around 4 oC all winter, the same temperature as your fridge.
If you want to fix it yourself the ANSI connector costs about $4 from RS, the part number is 769-1173. If you order such a small quantity from RS they will slug you for shipping, I just combined this with a larger order that I was making for work so shipping was free.
As mentioned I already had a suitable thermocouple but if you need to buy a new one they are pretty commonly available for $15 – 20 with the ANSI Type K plug: it’s the bright yellow connector used with common meters. Make sure the junction is insulated from the housing. If you have a functioning thermocouple but no plug the equivalent to the socket above is RS part no 769-1101 but this is $10, almost as much as a new thermocouple with plug.
If you plan on doing this and you need the two short thermocouple wires PM me and I will send you a couple of short pieces for free and for nix as a pay it forward for this forum. They’ll fit in a standard envelope so no charge for shipping. Australia only, offer expires when I run out of wire.Last edited by Lyrebird; 23 November 2021, 10:13 PM.
- Flag
- Likes 2
-
Part 2: How to mount them.
TLDR: Thermocouples should always be connected with specific connectors designed for the job.
These are available to several standards, most commonly ANSI, JIS or IEC (don’t you love non-standardisation of standards?) The differences lie in the colour codes for different thermocouple types and the plug shape, the common feature is an internal architecture using the same alloys as the thermocouple wires, thus eliminating unnecessary extra junctions and contact potential differences. FWIW this is part of the reason the plugs and sockets are colour coded, using the wrong part and thus the wrong alloys would throw the results off. An added bonus is that since these things are designed for use with thin, hard thermocouple wires they provide a much more secure termination than screw terminals or banana plugs which were designed for use with thicker, softer copper wires that deform under the screw terminal.
On a new design the best solution is one of the specifically designed on board terminals which are designed to be soldered next to the receiver chip on a PCB with internal copper planes as heat spreaders. This is the industry standard but this solution is obviously not available if you are retrofitting an existing design, so you need to choose one of the external mount connectors from the range described above.
I did this recently and I used an ANSI flat pin socket since I already had a suitable type K thermocouple with an ANSI plug. I went with a screw terminal connector as this makes it a simple matter to ensure the wires from the connector to the board are again the correct alloys, thus moving the cold junction onto the PCB as per MAXIM’s procedure. Since I had a dead thermocouple with bare wire termination to hand I simply cut a few cm of each of the wires and used that.
Attaching them to the board isn’t as simple as it sounds because neither chromel nor alumel are wettable with standard electrical solder / flux combinations. One option is to add another pair of screw terminals but this is also not easy on a pre-existing board. Another is to use direct PCB mount thermocouple connectors that are designed to thermally couple with the receiver but there’s no space on the existing board for this.
I tried pre-plating the wire terminations using Harris 56% silver alloy and the appropriate flux since Harris 56 will wet anything this side of titanium. This allowed me to solder these terminations directly to the board next to the input pins with 3% silver (lead free) electrical solder. This worked but only just: the alumel wire was very difficult to coat and solder and the end result looked like a pig’s breakfast so I removed it and tried a different tack.
I found that the thermocouple wires I have are a good fit into turned pin IC sockets so I cut a pair of pins off an IC socket and attached this to the board: again I soldered the pins as close as I could to the input pins on the MAX31855 and glued the body to the board with superglue.
In a spirit of total overkill I then fabricated a thermal bridge between the socket and the MAX31855 using an aluminium oxide based thermal paste, some silicone adhesive thermal tape and a small sheet of four nines pure silver (silver has the best thermal conductivity this side of diamond). I did say it was overkill, in my defence I used scrap silver sheet from making my bikes’ head badges. I secured this with a couple of over-wraps of Kapton tape.
I must apologise I didn’t take any pics until I’d finished this and now it just looks like the PCB with a chunk of silver on it.
There should, however, now be minimal temperature gradient between the cold junction and the chip which will improve the accuracy of compensation.
TBC with some experimentation measuring the results obtained.
- Flag
- Likes 2
Leave a comment:
-
3-Wire RTDs are so much simpler to employ by comparison.
Nice explanation Lb...
Mal.
- Flag
- Likes 1
Leave a comment:
-
Measuring Roast Temperature with thermocouples
Part 1:
How thermocouples work.
TLDR: they measure the temperature difference between the hot end and the cold end. For this to be accurate attention must be paid to both ends.
Long Version:
Thermocouples exploit the way the potential difference generated at a junction between two metal wires changes with temperature. For type K thermocouples the two metals are two nickel alloys, one doped with chromium (chromel) and the other with aluminium (alumel).
All materials have a contact potential which is related to the work function, the energy required to remove a charged particle from the surface. If you join two dissimilar metals the difference in contact potentials becomes the junction potential. Since the work functions depend on temperature, the junction potential also changes with temperature and we can measure the temperature by measuring the junction potential and doing the appropriate calculation.
Alas it’s not quite so simple, the other ends of the two pieces of metal (typically wires) also have contact potentials so this needs to be included in our calculation.
The actual welded junction between the two wires is the “hot” junction, the other ends of the two wires are collectively the “cold” junction. The output is the potential difference between the junctions which is a function of the temperature difference between the two junctions. If you know the potential difference and one temperature you can calculate the other temperature.
This is an important point: to get accurate results we must know the cold junction temperature accurately.
The classical way to perform this measurement is to have the cold junction at a constant 0oC so the potential difference between the two wires represents the temperature difference between the hot junction and 0oC. This is both arithmetically convenient and corresponds to the most commonly used frigorific mixture, 50: 50 ice in water.

Note that in the pic the cold junction is where the thermocouple alloys are connected to the wires to the receiver (shown as copper here). This is an important point and is an inevitable consequence of how contact potential works. For brevity (Huh!) I’ve left out the explanation of why this is; if you want it just ask.
In practice nobody uses a constant temperature cold junction outside a lab environment, the usual way around it is to use a thermocouple receiver which measures the cold junction temperature and adds a bias voltage that compensates for the difference between this and 0oC. A typical example is the Maxim MAX31855 which simply uses the temperature of the chip itself,* so the datasheet
https://datasheets.maximintegrated.c...s/MAX31855.pdf
calls out maintaining the cold junction at the same temperature as the chip, preferably by ensuring they are thermally coupled.
A typical circuit board is constructed of a fibreglass material known as FR4 with 2 layers of 1 oz copper. PCB mount screw terminals and several other connectors such as 4mm banana plugs are typically made from steel. FR4 itself is almost as good a thermal insulator as it is an electrical insulator, it has a U value of 0.25 W m-1K-1. Copper is about 380 and ordinary steel is about 40, so steel is about 1/10th as good at conducting heat as copper and FR4 is about 1 / 1500th.
The upshot of this is that relying on conduction through these materials will not allow the chip to read the cold junction correctly: in a typical scenario the thermocouple amplifier will read high by a few degrees in the mildest of conditions and can be off by over 20 degrees in more severe conditions.
*I’m guessing it uses a simple temperature dependent bandgap reference in the chip architecture, this constrains the measurement to being that of the actual die temperature since the bandgap reference will be etched in the die.
TBC, I will explain how to design the cold junction connector so it works optimally.

Leave a comment: