Thermocouple mux

Use ADS1248 from TI. This single IC does everything you need and replaces the whole AD circuit.

Vladimir Vassilevsky DSP and Mixed Signal Consultant

On a sunny day (Sun, 25 Nov 2012 19:48:00 +0100) it happened Marco Trapanese wrote in :

I have 4 tcs like this:

They are obviously trying to sell chips. A decent delta-sigma ADC with PGA can handle a tc input directly, without those two silly in-amps. If you want to mux a bunch of thermocouples, use some HC4051s.

Use something like an LM45 or an LM71 to pick up the local ref-junction temperature, ideally isothermal with all your tc-to-copper transitions. Do the ref junction compensation in software.

If you only want to do two tcs, you could use a dual-channel ADC, like AD7793 or some such, with an internal reference.

I don't know how expensive this circuit is, but I seem to recall that you can get a decent TC chip for $5 that gives you a digital output with cold junction compensation. Turns out the hard part of a TC measurement is the cold junction compensation. How do you measure the temperature of the cold junction to ±0.5­°C? The last time I looked at this I found that ±2°C was about as good as you can get without spending more bucks.

You can mux the TC wires using an analog switch, but you need to decide up front how you are going to handle the grounding. Some probes expect one side of the TC to be hard grounded, others need to be floated. The most common compromise is to ground one side with a 1 Mohm resistor like show in the app schematic. So you may want to switch both legs so the two probes can be different.

Rick

Il 25/11/2012 23:29, rickman ha scritto:

Do you recall any part number?

Ok thanks! Marco

Il 25/11/2012 22:06, John Larkin ha scritto:

I was afraid about the change of materials, but I was wrong because all tc will change at the same point. Very easy, indeed.

Got it.

The ds proposes an hardware solution to compensate the cold-junction using a thermistor and a precision resistor. I guess this requires a precise selection of them (and perhaps I need to adjust values among production batches).

In fact, you suggested to do the compensation in software.

How to handle this? I mean: I read the absolute Vtc using ADC (e.g. in uV). Then I know the temperature of the cold-junction in °C.

In order to get the temperature at the reference junction do I need to "invert" the tc equation? So I get the voltage a tc should exhibit at the cold-junction temperature, I subtract it from the actual Vtc and finally convert this into a temperature again.

Am I right?

Thanks Marco

Perhaps the Cold-Junction-Compensated K-Thermocoupleto-Digital Converter by Maxim

Il 26/11/2012 10:14, Spuckle ha scritto:

Very interesting! The Larkin's solution is more accurate and flexible. This one is very straightforward. Anyway some parameters are provided by design only - not tested.

Have you ever used this chip?

Marco

I haven't used it, it looks nice & simple for low volume use. It does the CJ, all you do is read it.

I always go for easy where possible.......

Right.

Most cold-junction compensation just adds a linear slope of voltage-vs-temp to the tc voltage, often by putting it electrically in series with the thermocouple. That creates a linear approximation of the actual tc curve, maybe good enough in a small range of expected ref junction temps. It's more accurate to measure cj temperature, run the tc curve "in reverse" (use the actual tc equations or lookup table to convert ref junction temp to millivolts) and add that to the measured tc voltage. I usually have two lookup tables, the main table that converts millivolts to temperature over the full range, and a smaller reverse table to convert ref junction temp to offset voltage, over the expected ref junction range. I usually lookup/interpolate both tables, and use a thinfilm platinum RTD to pick up the cold junction temp.

Whether you want to do all that depends on your accuracy requirements.

It's impressive how bad the cjc is on commercial thermocouple instruments. It's common for people to use the linear approximation (with a diode or thermistor) and miss the correct cold junction slope by 10% or so. And put the sensor only vaguely near the actual cold junction.

It's usually the case that the cold junction is expected to be at room temperature (like, 20C +/- 5) but the hot junction can be at -200C to 800C... so one can get away with crude compensation. That's a good reason not to trust the all-in-one chip solutions, unless you really HAVE to make the output show up on a moving-needle meter (or other panel voltmeter).

I'm not sure this is the right question. It is easy to deal with the cold junction if you know its temperature. How do you plan to get the cold junction temperature? That is the hard part, at least if you want to get an accurate reading of temperature (better than ±1°C). Getting the local temperature accurately is not so easy to do inexpensively and accurately. That is why the TC specific devices seem a little pricey.

Rick

Careful, that's not quite correct.

To be a bit more explicit, the thermocouple transfer function or tables assume a certain reference (cold-junction) temperature, usually (but not always) 0°C (the temperature of an distilled ice-water slurry, generally close enough anyway).

The inverse transfer function gives you the voltage you would read if the "hot" junction was at the terminal block temperature and the cold junction was at the reference temperature (say 0°C). For terminal block temperature > 0°C that will be a positive voltage.

You then ADD that voltage to the measured voltage from the thermocouple, and turn the crank attached to the transfer function to get the temperature at the hot junction.

For example, suppose the terminal block is at 20°C we measure 4.360mV from a type R (Platinum-Rhodium) thermocouple.

For 20°C. the corresponding voltage is +111uV

Add 4.360mV + 0.111mV = 4.471mV

Turn the crank, and get +500°C for the hot junction temperature.

What do you pay (approx.) for a platimun RTD? And where do you get them.

What about a 'regular' ~10k thermistor? (Yeah I know, the 1 degree ones cost more.)

Or just a diode, the slope is (pretty much) known it's the 'offset' that is different piece to piece. So maybe a single point 'zero' where you stick the TC onto the cold junction would work.

George H.

The thin film ones are cheap and plentiful these days- even Digikey has them for a couple dollars. They don't have much hysteresis (their big failing) if you keep them around room temperature. Wirewound ones are relatively expensive (and slow).

The slope is more important than the absolute accuracy, because you want to avoid having to test each instance of the product in a thermal chamber- too slow and expensive. Platinum RTDs are pretty good. Cheapie "accurate" thermistors seem to be trimmed for the zero, but have rather poorly controlled slope.

Same thing applies to the RTD, in spades. Just zero it out in an ice bath and Robert's your uncle. You don't need really high precision parts.

If you really want to use a junction, a transistor-connected diode is better than a diode, and transistor-connected diode measured at two or three currents is even better (but the signal level goes down, so the measurements have to be better).

For most food and paint baking and most plastics machinery, few care about a few degrees C. They're good enough, and the whole BOM including power supply, case and output stage has to come in at less than those boutique chips cost. It's hard to make the device that isothermal without adding cost and subtracting ease of replacement and multi-sensor type ability, so the accuracy of measurement is a moot point in many cases.

We used a small coil of copper wire as well as special resistors as RTDs. Worked pretty well. Platinum for the rich customers. These days, they just have to be well-heeled.

Enercorp 1206 100r $2.22

Minco ceramic slab 1K $4 in small quantities.

OK dat's nice. I'm use to the sire wound variety ~$10.

OK, I've

Grin, Yeah I'm using pnp's in to-220 pac's as T sensor's (positive current source) Using the decade current difference measurement gives the absolute T to something better than 1% (a degree or two).

But mostly I'd like to put the TC and cjc inside a temperature control loop.

An all analog circuit would make things simple for me,

For my control loop application I need stability ... but no accuracy.

George H.

- Hide quoted text -

ith

ll

ike

Much thanks,

(It's seems weird paying more for some special R's and C's than for opamps floating around 'em.)

George H.

.highlandtechnology.com  jlarkin at highlandtechnology dot com

Il 26/11/2012 21:02, rickman ha scritto:

Good point. In my application the "cold" junction (by the way about 50 °C) and the board are enclosed in a small thermally-isolated box. I need to know the internal temperature for other purposes and I'm going to use a quite good temperature sensor to get a measure better than ±1 °C.

I will put the sensor very close to the tc connector. Due to the small box I may assume the temperature will not change a lot in few centimeters.

Marco