RTDs suck!

[bigben2k]

RTDs suck!

I was set on using RTDs for my testbench, and I'm finding out that thermocouples would be more appropriate, for a number of reasons:

1-the response slope of an RTD is almost flat, making it virtually impossible to detect temperatures with a resolution of 0.01 or smaller. Thermocouples have a nice slope, but it's not quite linear. Omega does however carry "linear response" TCs...

2-Going over Omega.com 's PRTD elements, class A or class B, the response time is really poor, compared to thermocouples.

PRTDs have an advantage where their linearity varies very little, over a very wide temperature range. Useless to me, I'll be measuring temps within a very narrow range, and response time is actually important!!!

I might get one PRTD, but just for the sake of getting an accurate reference point, checking temps alongside thermocouples. BTW, a class A PRTD has an accuracy of +/- 0.15 deg C (element alone, not counting the temp reader).


Anyone want links to the PRTD info from Omega?

[pHaestus]

Why is response time important if you are truly at steady state?

//edit: do you plan to keep your setup in cal/operating within specs? platinum RTDs are much more stable over time.

[bigben2k]

Hum... to capture an actual change, if there is one? (yeah, there shouldn't be one)

Check out these spec sheets:

Class A PRTD element
http://www.omega.com/Temperature/pdf...N_2PT100KN.pdf

Class B PRTD element
http://www.omega.com/Temperature/pdf...D_ELEMENTS.pdf


Response time:

Class A
Water velocity: 0.4 m/s
50% response: 0.2 second (best element)
90% response: 0.6 seconds

Class B
Water velocity: 0.4 m/s
50% response: 0.7 second (best element)
90% response: 2.0 seconds


I suppose that the response time isn't important, given my setup, but it would be, under the original idea of using a controller for a fan-based cooling solution.

[bigben2k]

Something else I'm finding out the hard way...

http://cgi.ebay.com/ws/eBayISAPI.dll...827584260&rd=1

Is a transmitter for a thermocouple. Looks fine, until you pull up the specs ( http://www.prelectronics.com/ ): it's actually a 16 bit ADC, which limits the output resolution significantly (nothing I didn't suspect, but always good to check!).

[Incoherent]

Originally Posted by bigben2k

Something else I'm finding out the hard way...

http://cgi.ebay.com/ws/eBayISAPI.dll...827584260&rd=1

Is a transmitter for a thermocouple. Looks fine, until you pull up the specs ( http://www.prelectronics.com/ ): it's actually a 16 bit ADC, which limits the output resolution significantly (nothing I didn't suspect, but always good to check!).

What resolution are you looking for Ben? 16 bits is not low in my little world. Haven't really studied the spec sheet closely but it looks to me like you'd get at least 0.0091C resolution for the full range with a type T thermocouple. More if you limit the range, the analog resolution seems to be 18bits, wouldn't that be 0.0022C over a range of 150C? Accuracy seems OK too.

My quick take.

[talcum]

Originally Posted by bigben2k

RTDs suck!

I was set on using RTDs for my testbench, and I'm finding out that thermocouples would be more appropriate, for a number of reasons:

1-the response slope of an RTD is almost flat, making it virtually impossible to detect temperatures with a resolution of 0.01 or smaller. Thermocouples have a nice slope, but it's not quite linear. Omega does however carry "linear response" TCs...

2-Going over Omega.com 's PRTD elements, class A or class B, the response time is really poor, compared to thermocouples.

PRTDs have an advantage where their linearity varies very little, over a very wide temperature range. Useless to me, I'll be measuring temps within a very narrow range, and response time is actually important!!!

I might get one PRTD, but just for the sake of getting an accurate reference point, checking temps alongside thermocouples. BTW, a class A PRTD has an accuracy of +/- 0.15 deg C (element alone, not counting the temp reader).


Anyone want links to the PRTD info from Omega?

As a long time user of RTDs, they do have their problems, but by calibration, they can be used. Try this link for a better precision product http://www.airpaxtsp.com/tspsite/3000.html

You will want a 4 wire reader and you will want to cross calibrate the probes, but should be able to get +/-0.03 C pretty easily after cross-calibration.

[bigben2k]

I'm looking to get a measurement of the power drawn out by the block.

pHaestus has 0.01 resolution, and the error margin becomes quite high at 2 gpm. I'm looking for the magical 0.001 .

Talcum, any idea which RTD reader has a 0.01 resolution? Got any cheap options?

Otherwise, I'm leaning heavily towards a development kit for a 24 bit ADC chip by Texas Instruments:
http://focus.ti.com/docs/prod/folder...t/ads1217.html

[pHaestus]

Nothing both cheap and 0.001C for CERTAIN

[BillA]

no, Ben's mastery of temperature sensing sucks

how Ben, could you come to a conclusion totally at odds with science, technology, instrument mfgrs, and the sensor mfgrs as well ?

did you never question your 'insight' ?
show me a TC thermometer with more than 0.1°C resolution
did you find one ?
why not ?
after all, if you can endlessly divide a number then the resolution is whatever you wish, no ?
no

you apparently understand nothing about calibration
suggest some intensive investigation
before you make another asinine pronouncement

[bigben2k]

...and I finally catch up to this thread!

Granted that it's not a normal application. I suppose that I assumed that most products would not have a resolution of 0.001 because such a requirement rarely exists in real world, hence not being able to find such a product.

I realize that calibration for an absolute temp went out the door quite a few steps back at this point, but the target here is to obtain a temperature differential: different setup, different calibration (still fuzzy, but easier to do, I believe).


Someone please point out my flaw in logic:
I'm looking at a 24 bit ADC from TI, model 1217. Linearity is ~0.0012%, one bit accuracy (worse at higher sampling rates).

If I use a transmitter (RTD or Thermocouple), I'm going to add another linearity error, so I want to avoid that. The Texas Instruments ADC (Analog to Digital Converter) has a built-in PGA (Programmable Gain Amplifier), but it also induces a linearity error.

The TI ADC wasn't designed to be used to read temps, it's just made to accept a general input, within a certain range.

In order to read a voltage, I'll need a reference supply voltage, which will also induce an error.

If I go with an RTD, the voltage would probably have to be quite high, because the RTD's resistance changes so little over a temp range (low slope), where an TC changes more radically, but not linearly (if I remember correctly).

The high voltage would exceed the ADC's input, so it would have to be offset, which would probably add another error. I'd have to calculate the TC's voltage range, to see if it needs an offset, but I have to assume that that's the case here too.

Throughout all this, I'm out in the dark about what resolution can actually be extracted out of the temp probe, i.e. the output limitations of the probes, which may extend into the response time. At this moment, I'm assuming that either an RTD or TC temp probe is capable of actually capturing a temperature within a resolution of 0.001, the trick is to get that information. Line noise would be a significant factor at this point, requiring some shielding. Only a test setup (or a knowledgeable person) would tell me what kind of noise I'd be picking up, with various shielding solutions. a 4wire RTD would probably be most appropriate here.


That aside, calibration of a differential temp would be a challenge in itself, I agree. The styrofoam cup-o-water only goes so far.

[DrMemory]

The flaw in your logic is you can't increase the accuracy of the sensor itself by increasing the accuracy at which you measure the output of the sensor. All sensors have physical limitations to their accuracy. In the case of thermocouples, electrical noise and thermal noise in the sensor itself coupled with very low voltage output changes due to temperature limits its accuracy. At some point, increasing the accuracy at which you measure the output only increases the accuracy of measuring the electrical noise of the sensor itself.

Since thermocouples are non-linear, you must calibrate it over the entire range of the temperatures you intend to measure. This is the only way to get a one to one - temperature/ADC input voltage conversion with extremly high accuracy over the entire operating range. If you change thermocouples, you must calibrate it again. Most thermocouple circuits use a compensation circuit that "linearizes" the output of the thermocouple, but also reduces the accuracy. The compensation circuit can be built using the precision Op-amps needed to amplify the thermocouple output (see next paragraph).

Finally, ADC devices can't directly handle the low output voltage of thermocouples. They have their own accuracy limitations due to internal electrical noise. The thermocouple output must be amplified with a precision Op-amp which also have accuracy limitations due to internal electrical and thermal noise.

[BillA]

Ben
with each post you demonstrate your inability to integrate information

http://www.azonix.com/Catalog_MC/ins...ndex_body.html

you do not know what is out there, what its capabilities are, and have the laughable effrontery to presume that you are going to make a 0.001°C resolution anything, let alone a TC thermometer

have you read the 'theory of operation' section of a good digital thermometer ?
- its obvious

[bigben2k]

Thanks DrMemory, I suspected as much, but could not find any info to that effect.

Otherwise, I've always suspected that a 4 wire pRTD would more than likely be my best/only choice, for reducing the noise, short of working with candlelight... (which is still not out!) but again, I can find no info about the actual resolution that I might be able to extract from an RTD probe or element.

Again, I'm not looking to measure an absolute temperature, just a temperature differential. I fully realize that I have no hope of getting to an absolute reading with a resolution of 0.001 deg C, at least not without spending an insane amount of money.

I'm not looking to use the full bit range of the 24 bit ADC, just enough to be able to pick up the differential temperature between two probes with a resolution of 0.001 deg C. The temp range would be between 20 and 50 deg C. (differential would be water temp to heat die temp, maybe 20 deg C max?) Incoherent's 0.002 (18 bit) is the closest suggestion I've seen yet.


Does anyone (other than M. Grumpy ) have a feel about if this can be done, say with 2 * pRTD 4-wire probes?

[talcum]

Originally Posted by bigben2k

Thanks DrMemory, I suspected as much, but could not find any info to that effect.

Otherwise, I've always suspected that a 4 wire pRTD would more than likely be my best/only choice, for reducing the noise, short of working with candlelight... (which is still not out!) but again, I can find no info about the actual resolution that I might be able to extract from an RTD probe or element.

Again, I'm not looking to measure an absolute temperature, just a temperature differential. I fully realize that I have no hope of getting to an absolute reading with a resolution of 0.001 deg C, at least not without spending an insane amount of money.

I'm not looking to use the full bit range of the 24 bit ADC, just enough to be able to pick up the differential temperature between two probes with a resolution of 0.001 deg C. The temp range would be between 20 and 50 deg C. (differential would be water temp to heat die temp, maybe 20 deg C max?) Incoherent's 0.002 (18 bit) is the closest suggestion I've seen yet.


Does anyone (other than M. Grumpy ) have a feel about if this can be done, say with 2 * pRTD 4-wire probes?

Ben, I'll try to answer without sounding like BillA. Think of your error budget. First, remember that errors add in quadrature, then think of all your errors. How stable/precise is the current source you use to read the RTD's? How much is the self heating effect in your RTD's,? What is the precision/accuracy of your DAQ? (and then read the Analog Devices tech notes on accuracy vs bits of resolution)
And then do a few test calibrations with a crock pot and 4 or 5 sensors to get a feel for the scatter. Perform fits to ALL the sensors and then figure the errors for each sensor from the fit. Then see how you fit your error budget of .001 degree. I told you that .03 was very doable, .01 is hard, .001 and you are a god.

And if you want to skip to the chase Cole Palmer does sell matched 0.01 RTD's for $209 each.

Cheers

[BillA]

no talcum, Ben cannot "skip to the chase" because he wants also to build the bloody instrument himself !

".001 and you are a god"
godlike indeed, I cannot imagine building such an instrument from scratch

Ben, calibration is your friend
buy some junk, practice with hardware (stop posting words, ONLY data), learn with your hands

how do you intend to calibrate 'in house' ?
got a 'reference std' ? procedures ?
over what range ? intervals ?

and you want to know about AD converters, crap

[pHaestus]

buy some junk, practice with hardware

This is the best advice. You'll learn a lot more about what is possible/practical from just getting your feet wet than from technical documents at first. My setup works now reasonably well and it's almost 100% equipment bought from Bill because he needed better (better accuracy or GPIB-capable). I now am big on moving to labview a couple of years after I started for a simple fact:

I am testing almost all the time and it's getting to be a chore to sit in the basement and watch the clock and write down numbers/twiddle valves

But it took me YEARS to get to that point.

[bigben2k]

Funny, I was just going over AD's website... It makes the TI solution look grim.

Analog has ADC specificly for RTD inputs, which makes things more interesting but not necessarily easier, except for the current supply.

I'll try out something and advise later. Thanks for the tip Talcum!

In the mean time:
http://www.analog.com/UploadedFiles/...792_93_prf.pdf
(PDF: AD7793, 18 pages)

[bigben2k]

Wouldn't you know it...
http://www.temperatures.com/

and

http://www.temperatureworld.com/

[bigben2k]

...and...
http://www.sensorsmag.com/articles/0102/rtd/main.shtml

[BillA]

Originally Posted by bigben2k

...and...
http://www.sensorsmag.com/articles/0102/rtd/main.shtml

nice links Ben, did you read them ?

"If you use a PGA = 128, you would have an rms resolution of about 104 nV and a noise-free resolution of about 660 nV. The full-scale range of the A/D converter with a PGA = 128 would be limited to about ±20 mV, which exceeds the input signal range of 27 mV. But you could use the offset D/A converter to subtract 10 mV from the input signal, which would move the RTD signal back into the full-scale range. You would then have rms temperature resolution of about 0.003ºC and flicker-free temperature resolution to about 0.02ºC."

now I get flicker free resolution of 0.01°C with a Fluke 2180A, on eBay for $80 or so, with an instrument accuracy of ±0.02° after cal

but no, clever Ben is going to re-define low-cost high-resolution temperature measurement devices

Ben, why do I make THESE kinds of posts ?
So that others, less informed, do not mistake your commentary for useful, productive, or realistic

[DeadEye]

Originally Posted by unregistered

nice links Ben, did you read them ?

now I get flicker free resolution of 0.01°C with a Fluke 2180A, on eBay for $80 or so, with an instrument accuracy of ±0.02° after cal

but no, clever Ben is going to re-define low-cost high-resolution temperature measurement devices

Ben, why do I make THESE kinds of posts ?
So that others, less informed, do not mistake your commentary for useful, productive, or realistic

Hi Bill, I decided to watch this thread rather than be part of the discussion, probably to steer clear of your grumpy mood. LOL

A couple of comments and questions if you dont mind.

How do you calibrate your Fluke and does it have any output that can be logged for example?

I agree totally with your reasoning on measurement obtaining and accuracy. Having been a design engineer on weather stations in the 'real world' rather than the theorists of professors, then obtaining reasonable accuracy in temperature measurements is difficult to obtain in practice, using seperate discrete modules and components. This is true even when only 0.5C accuracy is desired.

In my situation, where it was a requirement to have ultra low battery powered data logging <15uA sleep and 50mA for 100mS (whole weather station), we had no choice but to go discrete, but otherwise the pro's know far better than we do.

As a final comment, it is good to have people in these forums that are knowledgable on various subjects, however cooler responses from all would make it a much friendlier place to frequent, and maybe others would learn more from this.

[BillA]

DeadEye
Ben is a special case, years in the making
(read his thread on the Radius wb, you will understand what is special about Ben)

I have a number of Fluke 2180A and 2190A thermometers, for cal I walk them to a cal lab ~400' from us, with the manual
- I get a table from 0-100° in 10° increments giving me the deviation, and I have all meters set in the same part of the 'curve' (wrt deviation)
(I incorporate the corrections into a spreadsheet which, with the sensor cal corrections, are used to determine the temp)

I also have a ref RTD, which is calibrated similarly by the same lab
then I use the ref RTD to 'cal' my working sensors in a bath or chamber

data logging is possible, but a chore to find the right parts - given that this gear is 20 yrs old
you need a 2300A scanner with the TC or RTD card(s) installed, AND a 1120A (GPIB translator) with the correct personality card - very difficult to find
or
for logging single point readings a quasi-prototype upper board was made, option 004, which has a GPIB output

[bigben2k]

Here's a link for a downloadable program, to get the resistance value of an RTD at various temp points (as per standards).

http://www.isotech.co.uk/prtcalc/index.html

(which of course says nothing about the sensor itself).

[bigben2k]

For those who doubted the existence of it, here's a thermometer, with RTD 3 or 4 wire input, with accuracy of +/- 0.01 deg C and a resolution of +/- 0.001 deg C.

http://www.azonix.com/Catalog/instru_item1.html

No idea about the price, but I'm sure it's up there.

[Incoherent]

That is extremely impressive.

Buy one Ben.