Thermocouple CJC puzzle

with-the-ADS1118.pdf

But, most of these schemes assume you convert from thermocouple extension wire to copper-copper before going to the voltage mesurement function (amp and ADC). That throws in two more offsets at the "cold junction".

Jon

Jon Elson wrote

Not AIUI. The cold junction has to be *somewhere*. You can't make the A-D converter chip out of thermocouple materials :)

Hence the CJC process involves

• reading the thermocouple voltage
• reading the cold junction temp (the terminal block)
• converting the cj temp into the equivalent thermocouple voltage (reverse linearisation)
• adding the two voltages together
• linearising the resulting voltage

It is rather counter-intuitive how exactly this works.

The simpler version is to read the thermocouple voltage, linearise it, and simply add the cj temperature to that, but that is wrong, unless the cj temp range is very narrow.

Right. That works. I think you essentially did the cold junction compensation twice.

If the terminal block is, say, 25c, the electronics sees *more* voltage than if it were 0c, so the correction is actually, numerically, a subtraction.

A lot of instruments uses a diode or something as the cj sensor, and assume the correction is linear, which it's not. The biggest error in most cheap t/c meters is the cold junction compensation, at least for moderate temperatures.

And a lot of people assume the voltage is a function of the temperature difference, which it's not.

snipped-for-privacy@highlandsniptechnology.com wrote

Obviously the tc voltage is based on the difference between the two junctions.

What is less obvious to me, but apparently true, is that these two cases will produce the same voltage at the ADC

a) +200 +20 b) +240 +60

They do however result in a different voltage being fed into the linearisation function, and thus a different resulting tip temperature, because the cj voltage is linearised on a different curve to the adc output voltage.

I know of a lot of industrial instruments which have for decades just used a 1N4148, and they probably aren't doing the reverse v -> cjv conversion either. It works because if say you are controlling +300C, in say injection moulding, a few degrees off is not noticed, precise temp readings are very hard to arrange (sensor location etc) and somebody wanting precision will be smart enough to use a PT100, which is amazingly accurate *and* since the measurement is (if 4w is used) totally ratiometric, you can get amazing accuracy. I have been testing with a 100R 0.005% reference resistor (20 quid) and one easily gets 16 bits out of it. A Class A PT100 is much worse than 0.005%...

We use inexpensive 1206 surface-mount thinfilm platinum RTDs as cj sensors. They are amazingly accurate.

But for a bench instrument, or kitchen thermometer, you have a climate-controlled ambient, and for any small range of temperatures, a linear approximation DOES accurately cover a cold junction , i.e. 'the cj temp range is very narrow'.

Absolutely, it is not. Approximately, it is, for my kitchen. Hey, getting the oven within

How is this normally done, say in a DIN style temp controller or even DMM with a temperature range thrown in?

If the engineer is lazy, he goes with an off-the-shelf module,

where a hidden magic is applied. Otherwise, he digitizes everything, and applies his own software magic, which is rarely disclosed, and tries to impress the customer with a long list of thermocouple types supported.

All sorts of ways, the cheapest being a diode that directly adds to the t/c voltage.

We use a selectable local/remote 4-wire platinum RTD, separately measure it, and do the math for RTD linearization and thermocouple linearization. Or allow the user to specify the cold junction temp, if he can acquire that some other way.

We generally use isolated delta-sigma ADCs for all the analog acquisition, including the RTD current.

I've never done the compensation. (except for a three junction thing, with a thermos of ice water as ref, and microvoltmeter.) But on cheapish controllers I've used, the cold junction is obviously on the back end terminal block, and the T sensor (diode or other) elsewhere... often upfront with user interface. This can lead to little T wiggles when something makes a thermal gradient across it... fan, heater, fridge...

Thermocouples are a bit weird. I bought a book about 'em (MacDonald?) but I still don't understand copper. Which has a 'backwards' sign over part of the T range. IIRC. George H. whose always favored copper /constantin for least magnetic effects.

Some of the more exotic thermocouples, like B, change slope over their temperature range. That is mathematically tricky.

nsion

n (amp

the

...

Yes, ISA type B slope changes *polarity* near room temperature, which means that you cannot determine the temperature unambiguously near room temperature (there's two possible answe rs). The output is so low near room temperature that there's not much point in trying until you get a couple hundred degrees C higher so it's not really a practical problem, but there's an iss ue deciding what to report or display.

Anyway, there are two readings required to properly determine the temperatu re in most cases.

#1 the millivolts from the thermocouple. The extension wire and those junct ions are chosen to minimize the error over typical temperatures where those junction s take

need platinum alloy wires coming off the terminals). Usually base metal T/Cs use the same alloys for extension wire. Somewhere (usually at a terminal block on the instrumen t) there are two junctions between thermocouple wire (one junction if one of the wir es is copper) and the copper conductors in the controller or indicator. The junct ions are designed be isothermal with each other and with a temperature sensor. T hat voltage reading needs to be accurate, in millivolts.

#2 the temperature at that terminal block, say in degrees C. It doesn't mat ter how nonlinear or linear that sensor is, only that you can convert the reading into an accurate reading in degrees. Any error will typically result in almost t he same error in the measured temperature at the 'hot' junction (which might be cryogenic, of course). So there's little point in using a thermocouple if y ou can place that cold junction sensor where you are actually trying to measure th e temperature.

You take that temperature reading and apply the reverse linearization equat ion(s) to it and get the theoretical millivolts for a thermocouple with a known CJC r eference

You take those millivolts and add them to the millivolts from the sensor, t hen

erence temperature as the reverse linearization. That gives you the temperature at the 'hot' junction.

As far as 1/4-DIN temperature controllers, that's exactly how I did it with digital controllers and selectable sensor types.

In the old days of analog controllers we would add a linear temperature sen sitive voltage (we used a linear temperature sensitive resistance- an RTD or preci sion temperature sensitive resistor- and add that, adjusted for the average uV/

thermocouple near room temperature. And that was good enough for most purposes, and only zero and span calibration was required (as opposed to tw o measurements).

Best regards, Spehro Pefhany

hold on a second. Platinum RTDs are not thermocouples, so different junctions don't matter (much) and cold junction isn't needed. I deal with the three wire versions and there's never really any surprsises.

It's true thermocouples that get all wonky and seem to need a reference temperature, somewhere. I've never been clear on how that part works in an say a spiffy Omrom controller. They don't explain the internals, just the convoluted programming steps.

We use thinfilm RTDs as the thermocouple reference junction temp sensors.

A t/c would be awful without a reference junction sensor and compensation. There would be about a 1:1 error induced by ambient temperature changes (except for some exotic cases.)

The ref junction is the place where the two t/c wires transition to copper+copper. The sensor should be isothermal to that zone. That is often done badly.

AFAIK no. Logically one can think that there are two termocoples, on going from measured temperature to say 20 C, second from 20 C to actual junction in the instrument. So in the first case you get

V(200, 20) + V(20, 20) = V(200, 20)

where V(t_1, t_2) is termocuple voltage between t_1 and t_2. In the second case you get

V(240, 20) + V(20, 60) = V(240, 20) - V(60, 20)

V is nonlinear so the two values are different. Theoretically correct procedure first uses known temperature of cold junction to determine correction to voltage (the V(60, 20) term above) using known termocouple tranfer function, then one gets voltage normalized to say 20 C (by adding correction to measured voltage) then one inverts V(t, 20) to get t. In practice nonlinarity may be small enough that simpler procedure gives acceptable result. But even if error due to incorrect procedure is small is is quite cheap to do correct thing in software.

OK, that makes sense now.

There's definitely no platinum RTD in the terminal block or octal pins of a DIN temp controller controller. Is the PCB assumed to be good enough? I've never really liked TC probes for anything where an accurate reading is really needed. I keep some fancy mercury thermometers around as the tie breaker and reference. Even those are getting real hard to find.

If the octal pins of the controller are stuffed with extension wire of the thermocouple type you're using, that terminal block is NOT temperature-critical (as long as its various parts are not under extreme thermal stress). The cold junction can be buried inside the controller.

It's common to use a semiconductor sensor, an LM45 type part, on the PCB near the thermocouple terminals. Sometimes just a diode.

That's OK as long as there are no temperature gradients anywhere.

A good RTD and a good DVM are probably better than any mercury thermometer.

I think that's right.

The standard temperature scales consist of a number of fixed-point temperatures, with interpolation between these standard temps. The fixed points are melting/freezing, boiling/condensing, and triple points of various pure substances.

Interpolation is performed using either thermocouples (with about one degree error) or RTDs (much better). Mercury thermometers were originally used, but this is pretty fussy compared to an electronic standard.

With a proper DVM, one compensates for the actual RTD resistance, so that drops out. Thermocouples go to much higher temps than RTDs.

Anyway, the whole story is in the following:

This is likely far more information than wanted. There may be a later version, but only details will have changed.

Joe Gwinn