Good used (or new) temperature sensing equipment?

[madshaman]

Hi, I'm looking to add some high accuracy and precision temperature sensing ability to my lab without breaking the bank.

Some basic requirements (a little fuzzy as I don't know much about what's available):

- at least four temperature channels (don't mind having more than one device) at least K-type thermocouple supported.
- very good accuracy and good resolution (1/100 degrees Kelvin or better?)
- GPIB/RS232/USB, basically any way to get the raw reading into a computer program
- something that will just work, I don't want to roll my own for this

I've looked around for what's new but was wondering if there was an equivalent to a HP3457/HP3456/Keithley 196 in the way of old gear that will measure temperature, something I could nab from eBay.

Any suggestions or recommendations are welcome.  I want to be able to monitor temperature during long-term circuit testing.

Thanks!

[madshaman]

Been looking around, haven't found much in the way of older equipment.  Half thinking of opening the wallet and settling for a couple Fluke 54s. :-(, it's not small change and I *hate* opening the wallet, but I don't want a poor half assed tool either.

[mzzj]

How much accuracy you need or expect to get?
Fluke 54 + thermocouple is accurate to 0,3 K or so in best case, error is easily 0,3 degrees +1% of the measured value.
K-type is not any good for precision, T-type would be better if you want to measure with thermocouples.

For better accuracy Agilent 34970A  +4pcs Pt100 platinum resistance probes would be next option but it is NOT going to be cheap.

[madshaman]

How much accuracy you need or expect to get?
Fluke 54 + thermocouple is accurate to 0,3 K or so in best case, error is easily 0,3 degrees +1% of the measured value.
K-type is not any good for precision, T-type would be better if you want to measure with thermocouples.

For better accuracy Agilent 34970A  +4pcs Pt100 platinum resistance probes would be next option but it is NOT going to be cheap.

Good suggestion.  This has caused me a lot of thought.  I'm a hobbyist, but, with the projects I'm getting into and the things I want to do in the future, I find myself requiring more and more accurate and flexible equipment.  I've not considered a data acquisition multiplexor before, and I'm sure they're used in industry all the time.  I've taken a look around eBay for this 34970A controller and a 34901A plugin to go with it.

To answer the question about precision, the answer is really "as good as possible".  I'll be using this to do long term testing of my own simple circuits as well as some experiments I'm planning to do with homebrewed NiFe cells, what I will do in the future is unknown to me; the accuracy and precision are important for being able to correlate events I see in voltage/current etc. graphs with events I see in temperature graphs, but, I have no idea what kind of events I'm going to see (and will never know what kind of events I'm going to miss due to lack of precision in both metrics and time), a lot of this will be learn as I go research.

One thing that bothers me about a multiplexor is that it reduces my sample-rate as the single built-in meter needs to do every measurement.  Sure, I might be logging data for hours, maybe even days, so maybe I can live with this as it would also free up my other equipment during that time (as I could also use it for other measurements, again, losing time resolution), but then I run the risk of missing transient events.

I'm half on the fence about going for this anyway, although it seems a lot of money to pay for what's essentially a glorified computer controlled switch with a DMM thrown in.  Given infinite time, I'd probably roll my own solution that meets my eccentric tastes and requirements better.

Ballparking around $1500 buying used, I'm thinking it's definitely worth the money versus my time, the only thing that troubles me is I won't know how useful adding this equipment to my lab will be until I start using it.  Actually, I'm more than half convinced to pull the trigger and just go for it.

Thoughts?

P.S. I assume the lack of accuracy of thermocouples must be related to the signal to noise ratio of their output, is it really that bad?  What's the source of the noise?  Is there any site or paper someone could point me at which gives a good survey of temperature measuring technology which goes into some details about the precision available and what limits it?

[madshaman]

Read through Agilent's tech note on temperature measurement; pretty good intro.  I think I'm going to pull the trigger on this option (34970A + 34901A + probes) via eBay (maybe not for the probes).  Pray I don't get e-screwed, going to be very careful to buy one with the DMM built in.

[mzzj]

P.S. I assume the lack of accuracy of thermocouples must be related to the signal to noise ratio of their output, is it really that bad?  What's the source of the noise?  Is there any site or paper someone could point me at which gives a good survey of temperature measuring technology which goes into some details about the precision available and what limits it?

Paying extra for unrealistic accuracy ends up really expensive soon.  No problem getting system to measure accurately down to 0,005K but the measurement system is going to cost you tens of thousands $.

Thermocouple accuracy is limited by other factors than noise and noise limits mostly resolution. You can get something like 0.01K resolution with +-0,02K short term stability with good equipment and proper techniques.

Biggest problem is the thermocouple sensor wire, manufacturing tolerances are HUGE. If you stay something like below 200 °C you can theoretically calibrate and correct the error to better than 0,05K
T-type thermocouple error is usually more linear than K-type and you can interpolate your results more easily. K-type might require calibration every 20K to be accurate to 0,1K and with T-type you can get away with calibration every 100K or so.

Above 200 °C thermocouple stability is  a bit problematic and you have more or less problems as thermocouple error will depend on the previous usage history, temperature gradient along thermocouple, immersion depth in use vs. calibration etc etc.

Second biggest problem is the measurement equipment itself, cold junction compensation is usually accurate only to something like 0,3..0,5K
This is a problem especially in general-purpose dataloggers like 34970A as cold junction temperature is measured only at 1 point and connector terminals are spread along large area inside relay/connector card/plugin.

You can circumvent this by using thermocouple-to-copper wire cold junction in accurately known temperature (cold junction immersed in  thermos bottle full of crushed ice and water) and manually set your datalogger to measure with cold junction temperature set to 0,0°C or measure in microvolts and convert microvolts to temperature based on NIST-published equations.

Here is some good material on thermocouples http://www.bipm.org/utils/common/pdf/its-90/TECChapter18.pdf

[madshaman]

Paying extra for unrealistic accuracy ends up really expensive soon.  No problem getting system to measure accurately down to 0,005K but the measurement system is going to cost you tens of thousands $.

Yep, I saw the same trouble coming when wanting a high bandwidth oscilloscope, frequency counter, etc.  Solving that by taking matters into my own hands, being a hobbyist I can afford that luxury, but, can't afford to do that with absolutely everything or I'll have too much on my plate.

One thing I've always had confidence in, is that with knowledge, time and care, helpful guidance from peers and proper application of science, one can usually come close to duplicating results that could otherwise be obtained by spending megabucks; I actually feel lucky I don't do this stuff for my job, it's going to force me to learn more things more in depth over time.

Thermocouple accuracy is limited by other factors than noise and noise limits mostly resolution. You can get something like 0.01K resolution with +-0,02K short term stability with good equipment and proper techniques.

Yeah, I need to be more careful, I meant precision, not accuracy, the resolution is obviously limited by the resolution of the voltage measuring device and noise will limit the precision to less than the resolution (oversampling w/ averaging aside).

I think I'd be happy with 0.01K resolution and +-0.02K stability to begin with, and maybe I'll find that's all I'll ever need.

Biggest problem is the thermocouple sensor wire, manufacturing tolerances are HUGE. If you stay something like below 200 °C you can theoretically calibrate and correct the error to better than 0,05K
T-type thermocouple error is usually more linear than K-type and you can interpolate your results more easily. K-type might require calibration every 20K to be accurate to 0,1K and with T-type you can get away with calibration every 100K or so.

Yeah, and it's not like I have a vast array of temperature references, pretty much limited to latent heat capacity buffering to get a reference (is there a technical name for this?), i.e. ice water or other two-phase mixes of pure substances, even the water is technically a problem, it will have to be very pure (I know that impurities will change the melting point).  A water purification system has been on my list for quite a while (how the hell did people do science before science supply companies, always need more things..); not that it's overly complicated, but I'm not made of time :-(.

Above 200 °C thermocouple stability is  a bit problematic and you have more or less problems as thermocouple error will depend on the previous usage history, temperature gradient along thermocouple, immersion depth in use vs. calibration etc etc.

I probably won't be doing much measurement over 200 °C, but I'm pretty sure an exception will come up in the next couple years if not sooner.

Second biggest problem is the measurement equipment itself, cold junction compensation is usually accurate only to something like 0,3..0,5K
This is a problem especially in general-purpose dataloggers like 34970A as cold junction temperature is measured only at 1 point and connector terminals are spread along large area inside relay/connector card/plugin.

You can circumvent this by using thermocouple-to-copper wire cold junction in accurately known temperature (cold junction immersed in  thermos bottle full of crushed ice and water) and manually set your datalogger to measure with cold junction temperature set to 0,0°C or measure in microvolts and convert microvolts to temperature based on NIST-published equations.

Thanks, this is looking to be my best option at the moment, at least the datalogger will take care of a lot of the work I'd have to do if I built the system from scratch, and, I'm sure I'll find many more uses for it as time goes on.

Here is some good material on thermocouples http://www.bipm.org/utils/common/pdf/its-90/TECChapter18.pdf

Thank-you for the link!  And thanks for the input!

[mzzj]

0,00 Celsius is really easy to calibrate as reasonable quality tap water melting point is usually better than 0,02K accuracy. Spend 1 dollar for liter of distilled water  and you can do better than 0,003K with proper technique.
(distilled or reverse osmosis treated water is commonly available in car part shop for  batteries)

Any other point for calibration is way more difficult or  inaccurate or expensive. Boiling point of water is reasonably accurate if you compensate for the local atmospheric pressure. I cant  remember actual numbers for attainable accuracy because it is not used  in calibration laboratories any more because other methods are more accurate.

[madshaman]

Okay, so started hunting around for a decent multiplexer, I had a couple of questions:

1) I've never used a switch mainframe before, apart from figuring it out directly from the specs myself, which is better (well, what are the pros/cons) of a solid state switching card vs a relay based one?  (I immediately leaned towards solid state because the cycle life of most relay-based cards I looked at seemed almost useless in that I'd need a new card every year or so)

2) When measuring using thermocouples, is it feasible to connect the cold junction in parallel with the hot junction connected to each DUT?  like (please excuse my ASCII art):

  ____(alloy "x")______ -> Multiplexer: Ch 1, post 1
 /
<  DUT 1
 \____(alloy "y")______ -> A


  ____(alloy "x")______ -> Multiplexer: Ch 2, post 1
 /
<  DUT 2
 \____(alloy "y")______ -> A


  ____(alloy "x")______ -> Multiplexer: Ch 3, post 1
 /
<  DUT 3
 \____(alloy "y")______ -> A

…

Multiplexor: post 1 out --.
                          |
                          V (DMM: LO)


----------------------------------------------------------------------


                               .----/\/\/\---.
                               |             |
                               |     |\      |
                               |     | \     |
  ____(alloy "x")______________|_____|- \    |
 /                                   |   \___|__________ -> DMM:HI
<  Cold Junction in ice bath         |   /
 \____(alloy "y")_____ <- A       .--|+ /
                                  |  | /
                                  |  |/
                                  |
                                  V (DMM: LO)

Here was my thinking (although after some reading I might want to go the RTD route):

Most used temperature/thermocouple mux cards seem to use relays, even at a lifespan of 10^8 cycles, they'd wear out pretty fast with the sample rate I'd want.  So instead I'd use a solid state mux card, my own amplifier (which I'd need to calibrate; calibration pot, decoupling caps etc. not shown in the schematic), a shared ice bath and do the voltage to temperature conversion in software according to NIST data.

The idea would be to connect all the alloy "y" terminals together at the shared ice bath by physically crimping them together with some kind of non-reactive, non-conductive jig.  Seems like a lot of work to be honest, which makes me wonder why thermocouples are used at all when it's not absolutely necessary.

[charlespax]

I know this is an old thread, but it seems like a good place to mentions this. In September I'm launching a Kickstarter for an open-source four-channel thermocouple datalogger that saves to SD, outputs to USB, and displays the last 100 readings. I'm still building the site right now, but there is some information at http://paxinstruments.com/.

[Carrington]

I know this is an old thread, but it seems like a good place to mentions this. In September I'm launching a Kickstarter for an open-source four-channel thermocouple datalogger that saves to SD, outputs to USB, and displays the last 100 readings. I'm still building the site right now, but there is some inforation at http://paxinstruments.com/.

Is it based on two ADS1118 a MCP3424 or 4 AO?

[retrolefty]

At our refinery we considered any TC sensor measurement good if within +/- 5 degrees F, especially at elevated temps, just too much variations is TC material (Usually K types, some J) to obtain the accuracy you are looking for. On important temperature control loops we always used PT100 sensors, usually good to +/- 1 degree F.

 

[David Hess]

Those multimeters like the Keithley 196 which support RTDs directly are slick.  I am disappointed that all 4 wire multimeters do not support this.

For accuracy, I would avoid multiplexing thermocouples (or avoid thermocouples all together) because the extra junctions will just make things worse.

If I wanted to do multichannel accurate temperature measurement on a budget, I would replicate one of the simple analog signal conditioning circuits for RTDs to generate a high level signal which can be multiplexed.

[Carrington]

8 Input Portable Thermometer/Data Logger with SD card:
http://www.omega.co.uk/vhle/index.html
Is just an example, there are many other...

I love these thermocouples:
http://www.omega.co.uk/Temperature/pdf/IRCO_CHAL_P13R_P10R.pdf

[David Hess]

Omega is the last place I consider when thinking budget.

I have considered one of those cheap Chinese RTD meters but am leery of their quality.

[charlespax]

Is it based on two ADS1118 a MCP3424 or 4 AO?

It's based on the MCP3424 ADC and MCP9800 temperature sensor for junction temperature.