Connect with us

Thermocouple simulator

Discussion in 'Electronic Design' started by Richard, Oct 15, 2005.

Scroll to continue with content
  1. Richard

    Richard Guest

    I need to build a thermocouple simulator for the test bench. It will be for
    type K and will simulate 32 F to 1600 Deg F or there abouts.
    It is NOT a calibrator, just to be able to simulate some adjustable inputs.
    I have to build a box with 24 pots in it for 24 channels.

    No whiz at all on this so give me hand if I'm getting in trouble. I'm
    probably doing the hard way. My supply is from a 7805 so I was going to
    take 5 volts to each pot like this..

    5v ---10K----100---- Ground

    Between the 10 k resistor and the 100 ohm resistor, I would put a 10 K pot.
    so that one end of the 10 pot went to ground and my center tap would be an
    adjustable 0 to 50 millivolt output.

    Is there a better arrangement of resistors to use?

  2. How do you plan to simulate 32°F if your minimum voltage output is
    equalt to or greater than 0V? You'll need to go negative by a mV or so
    depending on your ambient temperature.
    That will give a pretty high source impedance (from almost 0 to about
    2.5K). Are you sure that will be okay with whatever it is you're going
    into? Typically thermocouple instruments detect sensor breaks by
    passing a small current thorough the sensor.
    With wirewound pots, there's typically a trade-off between resolution
    and resistance value.

    Best regards,
    Spehro Pefhany
  3. John Larkin

    John Larkin Guest

    Hey, I'm just now finishing up the firmware for a 16-channel t/c
    simulator module, J K E T R S B N, rtd reference junction sensors,
    floating outputs. Pretty serious pita.

    Don't forget reference junction compensation. If you were to do...

    | |
    | |
    | /
    | /
    10k pot /<------1k-------+-----------A=================== tc+
    | / |
    | / |
    | | 10 r
    | | |
    | | |
    +--------- | -------------- | ----------B=================== tc-
    | | |
    | | |
    10r | |
    | | |
    | | |

    you'd get -5..+45 mV out. If everything to the left of A:B is copper,
    and the leads ====== to the right are t/c wire, and you measure the
    voltage at A:B with a DVM or whatever and set the voltage per the NIST
    thermocouple tables, you then have to compensate for the temperature
    at A:B, or else keep A:B in a good ice bath.

    You can buy a decent t/c simulator box for a couple hundred bucks or

  4. DaveM

    DaveM Guest

    I used to calibrate thermocouple potentiometers in a cal lab (late 60's)
    with a voltage standard such as the Fluke 332. If you have access to a lab
    voltage standard, all you need is a temperature-voltage table for the
    thermocouple types you need to simulate. Set the voltage standard to the
    voltage corresponding to the desired temperature and there you are.

    Using your method, I can see a problem immediately. The circuit that you
    are going to drive with the output of your resistor string will need to have
    a very high impedance in order to avoid loading problems. A better way, if
    you insist on using this method, is to follow the resistor string output
    with a quiet, low-drift op amp that can easily drive the circuit that will
    measure the voltage. You can find a good assortment of those at Linear,
    Analog Devices, Maxim, etc. Make sure that you select a model that has
    offset voltage adjust facilities, or provide another method to adjust the
    output offset to zero volts for zero volts input. You also would need to
    use a bipolar power source for the op amp.

    A third method would be to use a stable power source into a 15- or 25-turn
    wirewound pot. Follow the output of the pot with a stable op amp with
    suitable gain to drive a 3 1/2- or 4 1/2-digit digital panel meter to give
    you a scaled reading. Use the voltage table to adjust to the voltage you

    Whichever method you decide to use, you need to pay close attention to the
    quality of the components. Use low tempco resistors and pots. Use
    low-drift op amps. Use very stable voltage sources. Otherwise, all the
    detail you put into your simulator will be wasted by temperature drift and

    All that said, if you're doing this for your company, I'd suggest that
    buying a simulator might be cheaper than building one.
    Omega has a cheap one at It
    doesn't have the accuracy of the higher-priced models, but you didn't state
    that requirement in your post. Googling for "thermocouple simulator"
    returned a lot of possibilities.

    Dave M
    MasonDG44 at comcast dot net (Just substitute the appropriate characters in
    the address)

    Never take a laxative and a sleeping pill at the same time!!
  5. Richard

    Richard Guest

    Thanks for the input, we have several calibration and simulator tools, this
    is not about precision or dift or anything else.

    It's just a rough way for us to hook up a panel and dial some temps in above
    our shutdowns, and be able to slightly adjust them in order for our panel to
    stay in run mode without flagging an error.

    Pots are what we are using for all the analog inputs and although they are
    scrathy and noisy, they do fine for the intent.

  6. Fred Bloggs

    Fred Bloggs Guest

    What is your problem that you can't overcome this electronically? Too
    many parts?
  7. John Larkin

    John Larkin Guest

    Are you addressing me? I am overcoming it electronically, as noted.
    1050 parts, 2 fpga's, 1 32-bit uP, 12k lines of code, 16 isolated
    outputs, 4 rtd inputs, zero cuts/jumpers first try. See pic in

  8. Richard

    Richard Guest

    Sorry, about time for a new keyboard, its leaving too many letters out.
  9. xray

    xray Guest

    Didn't see such a post in abse. Usually get good coverage from my news
    providers. Is it me or is it missing?

    Not really following the thread. What in your app makes all this

  10. xray

    xray Guest

    Nevermind. Looking in more detail tonight, I see my coverage of the
    binary group is all but binaries. Sorry for the message.

    Looking to my provider for a change or solution.
  11. Fred Bloggs

    Fred Bloggs Guest

    1,050 parts?! that seems ridiculous...Is this for someone's production
    calibration/ checkout line?
  12. John Larkin

    John Larkin Guest

    Gosh, Fred, I'm starting to lose hope of ever pleasing you.

    It has 16 channels, each with an isolated dc/dc supply and isolated
    data paths. Each has a 16-bit dac with programmable ranges from +-12.5
    volts to +-25 mV full-scale. Outputs are protected to +-35 volts
    forced differentially, +-750 common-mode. There are four 24-bit RTD
    acquisition channels for external reference junction boxes and one
    internal temperature sensor. Each channel has a relay that can be
    programmed to switch it to the D9 cal connector for in-crate
    calibration check. There are 85 distinct power rails on the board.

    Net 65 parts per channel ain't too bad, and I think we can maybe
    delete about 100 parts from the next rev if we can simplify the dc/dc
    stuff; we always figured that would be the toughest part, and it was.

    Our initial customer will use it for testing and certifying jet engine
    control computers; two other vendors tried and screwed up one way or
    another, so they asked us to do this ASAP... nobody can do decent
    analog stuff any more! I think we have a helicopter developer sold,
    too, and maybe a couple other aerospace apps like JSF.

    Next project will be the complement, a 16-channel isolated
    analog/thermocouple input board, using the same module mechanics and
    uP/control stuff.

  13. Fred Bloggs

    Fred Bloggs Guest

    How did you achieve that +/-35V differential protection- if it's not
    proprietary- I like this clever scheme from LT, (they omit the pull-up
    emitter SD because they can diffuse for high Veb breakdown):
    View in a fixed-width font such as Courier.
  14. Fred Bloggs

    Fred Bloggs Guest

    whoops- should be:
    View in a fixed-width font such as Courier.
  15. Fred Bloggs

    Fred Bloggs Guest

    Highest speed development would be to modularize into say 4-channel
    self-contained boards depending on economics and available parts- then
    making multiple smaller boards is much easier than one humongous thing-
    the control part is easily scaled. I have seen schemes where a S/H is
    refreshed by a central DAC to drive multiple channels like that but
    16-bits might be pushing it, at least for you, best to stay out of
    unknown territory when you're in a rush:)
  16. John Larkin

    John Larkin Guest

    Sadly, the density of this board doesn't leave room for much "circuit
    design"... so we have to use ICs wherever we can. More and more, we
    just connect boxes. We did look into polyfuses and transzorbs, but the
    polys are very sloppy parts and we couldn't make the numbers work.

    We used these...,,768_833_ADG465,00.html

    which is, actually, an oldish idea using cascaded enhancement mosfets.

    nfet pfet
    in __________________ _____________ __________ out
    ----- -----
    --- ---
    | |
    | |
    v+ v-

    which works great until you blow the gates out.

    Also, the final pole of the output amp loop is a biggish cap directly
    across the output. This allow microfarads of external caps to not
    destabilize the loop, and serves as the gross ESD protector.

    Dang, time to take the girls to the opera, "Dr Atomic."

    ("take" as in "drive", not as in "join them"!)

  17. John Larkin

    John Larkin Guest

    At least for me? Bizarre.

    We did propose a couple of different architectures, with my favorites
    promptly shot down, and settled on this one in a meeting with our
    prime customer in Hartford. This is what they want to buy, so I can't
    argue. Their test cells run to high channel counts, so 16 dacs on a
    board isn't especially inefficient.

    The mezzanine board thing sounds good (PMC, IndustryPak, several
    others) but in real life it winds up with higher cost and lower
    density than dedicated single boards, and the mez things are usually a
    programming and packaging nightmare.

    The scanned s/h made more sense when 16-bit dacs were very expensive,
    but they're dirt cheap now, and isolating the s/h thing (flying caps?)
    is nasty, too. Done that, don't expect to do it again.

    I guess you could put the digital part of a delta-sigma dac in an
    fpga, optoisolate the 1-bit result, and get a floating dac with just a
    switched reference and a lowpass on the isolated side. If we'd had
    more time, we might have investigated that, but I think it would be
    inherently slow, and we want the option to have a fastish version.

Ask a Question
Want to reply to this thread or ask your own question?
You'll need to choose a username for the site, which only take a couple of moments (here). After that, you can post your question and our members will help you out.
Electronics Point Logo
Continue to site
Quote of the day