Thermocouple Meter Calibration - when authorities collide ...

[Graeme]

Calibration procedures and metrological measurements are supposed to be based on authoritative references ... but what do you do when several references have conflicting information? I have recently seen an example of this, and I would like to throw it out for discussion.

A technician was calibrating a digital temperature indicator that can use either J or K thermocouples. The calibration method applies a DC voltage to the meter inputs to simulate a thermocouple. This technician is still new in the lab and is eager to learn, so she has been studying a lot - even reading some of the stuff I write! Her diligent study has prompted a question. We have already fixed the potential problem in our process, but the fundamental question remains.

(Note: all temperatures listed are in degrees Celsius.)

The manufacturer specification for the meter covers almost the full range of the NIST emf tables for a J thermocouple: -200 to 760 degrees. There are accuracy limits specified over this entire range: ±(0.1% reading + 1.2 deg). We also calibrate similar meters with even tighter specifications: ±(0.05% reading + 0.6 deg). The question relates to the fact there is little agreement among the references we have available about the usable range of the Type J thermocouple, and about the existence or magnitude of error limits.

• The NIST web site (see the references, below) has emf vs. temperature tables that go down to -210 degrees. These tables are reproduced in the other references.
• The book by Nicholas & White, and the Omega reference section in their catalog, have a note stating that J thermocouples "should not be used at low temperatures". Neither source defines what "low" means.
• The ASTM manual states that J thermocouples "should not be used below the ice point". (0.00 degrees)
• The ASTM manual also states that there are no established limits of error for J thermocouples below the ice point.
• The table in Nicholas & White has limit of error specifications, but only down to -40 deg.
• The table in the Omega reference has limit of error specifications for the entire range.
• The NIST table has nothing on limits of error.

To me, the net effect of these conflicting sources seems to be that if your stick a J thermocouple probe into something colder than the ice point, you will get an indication on the meter but it may or may not have any meaning. This leads to the question our technician raised – if a meter is outside its specification range at a point (for example, -100 deg), then is it really "out of tolerance"?

Have you seen this type of problem? How would you handle it? (Don’t cheat by reading the final paragraph to see what we have done for now.)

References

• ASTM: Manual on the use of thermocouples in temperature measurement, 4th edition. ASTM, 1993.
• Nicholas & White: Traceable Temperatures, 2nd Edition. Wiley, 2001.
• NIST: ITS-90 Thermocouple Database (http://srdata.nist.gov/its90/main/)
• Omega Engineering: Temperature reference section (http://www.omega.com/temperature/Z/zsection.asp)

We have addressed the situation by changing the calibration procedure so that J thermocouple inputs are not tested below the ice point. This does not affect any of our current customers because all of them are measuring processes above 30 degrees.

 

[tomvehoski]

My first/biggest question is what type of tolerance do you have on the measurement being taken? What exactly are you taking the temperature of.

I used to handle quality control for injection molding temperature controllers. We pretty much used J thermocouples. It has been years and it looks like you have researched this more that I ever did, but as I recall the J temperature to voltage curve is not linear. The controllers we designed were not sophisticated enough to handle the curve, so we created a linear model of temperature vs. voltage - basically 1 degree C change = .xx volts, regardless of location on the temperature curve. At some points we knew the reading was below the actual, at some points above. For injection molding it was not that big of a deal to worry about a couple degrees since there were so many other sources of measurement error (distance from cavity, cold junctions, etc.)

The other thing to consider is the thermocouple itself. In addition to calibrating the meter, you also have to use a calibrated TC for precise measurements. If the voltage the TC puts out is wrong, your measurement will be off regardless of the meter. I recall on of my TC manufacturers claimed they could certify the TC within 1/4 degree C, for a very expensive price.

The advantage of thermocouples is they are cheap, have a wide range and fast response rate. Disadvantage is they are not very accurate.

Tom

 

[Graeme]

tomvehoski said:
My first/biggest question is what type of tolerance do you have on the measurement being taken?

Typical performance specifications of the thermocouple indicating meters ("thermometers") are stated in the third paragraph of my message.

What exactly are you taking the temperature of.

I am not "taking the temperature" of anything. I am calibrating a digital thermocouple meter by applying a DC voltage, as stated in the second paragraph of my message.

the J temperature to voltage curve is not linear.

Neither is the curve of any other thermocouple. The curves are described by polynomials with anywhere from five to 12 terms. But it is irrelevant to my question.

The other thing to consider is the thermocouple itself. In addition to calibrating the meter, you also have to use a calibrated TC for precise measurements.

The thermocouple itself is irrelevant to my question of calibrating the meter. The only issue I have is that type J should not be used below the ice point, but typical meters will display values down to -200 degrees. Since the thermocouple should not be used in that range, and the question of whether it has established limits of error in that range is ambiguous, then does that number on the meter really mean anything?

Calibrating the thermocouples -- while irrelvant to this question -- is an issue I have been working with this company about for some time, and they are starting to begin to think about doing it. A hard part of convincing them was making the responsible people understand that (a) thermocouples are consumable items and not capital assets, (b) they are not a "precision" measuring instrument, (c) using them changes their characteristics, and (d) if they are not calibrated then the workers don't have any clue about that the measured temperature is no matter what accuracy of meter they hang on the measuring end.

 

[Jerry Eldred]

WHEW!! This one is making me a little dizzy (maybe because it's Monday morning - another story).

I'll try to make a few observations (as the topic seems to have gone a few differing directions).

1. If you are in a "compliance" context (have to comply with government requirements such as FAA, FDA, NRC, etc.), on measurements that are auditable I would recommend using the most stringent of the ambiguous standards, and government or government/industry recognized standards should be the overriding requirement (as long as they are the more stringent).

2. Calibration of the meter (which I think most of us know, but for those reading who may not) is a completely separate issue from the thermocouple itself. You are essentially calibrating accuracy of a millivolt meter. So temperature limits on the thermocouple material are treated somewhat differently than the meter accuracy. If a meter is rated to measure down to -200 degrees C on type J, to unmuddy the waters, you could certainly calibrate the meter to it's specified range regardless of the thermocouple probe.

3. If you are NEVER using type J below 0 deg. C., an easy solution is set up specs not to calibrate below that point, and make it well documented so it is completely unambiguous to the user and auditor.

After stopping to do some research on type J, I have a couple of conclusions, and a couple of educated hypotheses (if I may throw a big word in)...

a. According to NIST, and some other sources, type J is indeed rated to operated down to -210 degrees C. Therefore they are legitimate to use throughout that range.

b. HOWEVER (A big however, and will seem opposite to (a)). I looked at a plot of type J millivolts per degree C curve (Seebeck coefficient plot). Type K and J both have kind of sideways 'S' shaped Seebeck coefficients, making them two of the most nonlinear of thermocouple types. Additionally, for whatever reason of physics, their millivolts per degree C nosedives below zero. What this means is that not only is algorithm needed by the meters firmware to convert mV into Deg C more complex, but the millivolt output is essentially much lower at the bottom end of the range (less than 0 Deg C). Therefore, the meter by definition cannot discern the temperature readings nearly as accurately. Consequently, because of the reduction in accuracy due to the physics of the material (and although it is specified to operate at those low temperatures), many manufacturer's apparently prudently choose not to spec below zero degrees C. So although it apears ambiguous, there seems to be a straightforward answer. This is my hypothesis based on what I found.

Type J mV/Deg C from a little above zero to it's upper limit is around 50 to a little greater than 60 microvolts per Deg C. By somewhere around -100 Deg C it is less than 40 microvolts per degree C, and by -210 Deg C, it is a little more than 20 microvolts per degree C. So the increase in uncertainty is likely a combination of a few things:

1. Do the math for a millivolt meter. Half the millivolts increases bottom end offset error.

2. The same amount of inhomogeneity produces two to three times the temperature error in the thermocouple material.

Hope my rambling is of some help.

 

[Jerry Eldred]

One more thing.

I did see at the NIST website a plot of the increase in uncertainty for type J below 0 Deg C. So although I couldn't find their data, they apparently do have documentation on low temp uncertainty for type J. If anyone is really interested, you could probably email Dean Ripple in the Thermocouple department and request some of his information. I've corresponded a couple of times with NIST, and find them interested in being helpful.

 

[D_Wood]

This is interesting. When I was working for a company that manufactured solid-state relays, which were rated -30°C - +70°C, Underwriters Laboratories (UL) always required our submittal samples to be equipped with Type-J thermocouples. During testing, we would sometimes have problems of functionality around the -25° to -30° range. It would drive us nuts trying to get the right component combination to work at those temps.

Now I wonder just exactly what was the temperature where these problems occurred? I suppose it's possible the temperature was below -30°C, right?

Now, my question. Do you get better accuracy with the Type-K thermocouple at colder temperatures?
-Dan-

 

[Graeme]

D_Wood said:
Now, my question. Do you get better accuracy with the Type-K thermocouple at colder temperatures?
-Dan-

Dan,

Again, my main concern and experience is with the meters, not the thermocouples. I have used thermocouples of various types, but not in the last four years. Frankly, I did not know as much about them back then, either.

The "accuracy" specification of the thermocouple is determined first by its type. Each type appears to be available in at least two grades - standard, and either "special" or "laboratory" grade, depending on the manufacturer. I generally refer to the ASTM Manual on the use of thermocouples in temperature measurement for the "most authoritative" specifications of those I have available. Those tables and data (chapter 10) are quoted from ASTM standard E320 which, I believe, defines the performance requirements for thermocouple types. (If a manufacturer can give me better specifications, that's a bonus.) Based on that manual, the performance specifications are:

Type J, K or N, 0 to 1250 deg. C: Standard tolerance +/- 2.2 deg. C or 0.75% of the temperature being measured, whichever is greater. Special tolerance +/- 1.1 deg. C or 0.4% of the temperature being measured, whichever is greater.

Type K, -200 to 0 deg. C: Standard tolerance +/- 2.2 deg. C or 2% of the temperature being measured, whichever is greater. Type N is not listed, but a commercial catalog lists specifications identical to K.

Note that ASTM specifically states that type J should not be used below 0 degree C. The other sources I originally listed (except NIST) state that J thermocouples should not be used "at low temperatures" -- even while publishing emf tables going down to -200!

There is also a comment in the book by Nicholas & White that may have a bearing on your question.

Type K is the most common thermocouple type and the most irreproducible, showing spurious errors of up to 8 deg. C in the 300 to 500 deg. C range and steady drift above 700 deg. C. ... The main advantages of Type K are the wide range, the low cost and ready availability of instrumentation.

(emphasis added) Their note on Type N is that it is very similar to type K but has much better reproducibility.

The second determinant of the accuracy of a thermocouple is the manner in which it is installed and used. The third factor is its calibration status.

As for accuracy experience in actual use, I will have to defer to others with more and recent experience in using them in processes.

(On your relay tests, I assume that the sub-zero testing was a realistic operating requirement for them. I ask becasue some years ago I was doing environmental effects testing on a piece of equipment that was installed inside a submarine. The project engineer specified a generic mil-spec test that included the ability to start up and operate after a cold soak at -40. The box was failing becasue the (mechanical) relays started sticking at about -20 degrees C. The project engineer was finally persuaded to modify the low end temperature requirement after I and a few others pointed out to her that, if the temperature inside a submarine was anywhere near -40, then the crew had problems that were vastly more serious than if this communications box was working or not!)

Graeme

 

[Jerry Eldred]

As I read through and think about J and K thermocouples, one word comes to mind - "Oxymoron".

They are utility temperature sensors. They are so common because they are so cheap. For measurements close to zero celsius, and up to about 100 C (or more) , T is quite a bit better. The oxymoron is between accuracy and J or K. I can't remember whether it is J or K (or possibly both) that has some magic temperature (my recollection is that it was either 160 C or 260 C). When the thermocouple is exposed to that temperature for any significant length of time, it's molecular lattice structure goes crazy, and it's measurements develop pretty large errors. You correct the lattice problem (if any of you folks understand that - if not, let me know and I'll give you my backwoodsy explanation) by bringing them to a pretty high temperature to anneal the lattice back into quiescence.

I had a previous manager with some very "odd" views about thermocouples. The only way to convince him his views were wrong was to do a lot of homework. So I spent considerable time self-teaching myself about material properties and became fascinated with it.

Guess I'd better go get some coffee before I ramble any further.

In answer to a previous remark. Type K, although it is rated to go below zero celsius, has a similar drop off in thermal emf below zero. A number of the thermocouple types do the same thing. Type B even limits its "good" range to greater than about 400 C (I have some pretty strong views on type B - so don't get me started). Have a great week everyone.

 

[Graeme]

Jerry,

Thank you for your input. Obviously Monday was not a holiday for you.

I am in a regulated environment -- the FAA is the governing agency. However, we do have some freedom in that we can limit the calibration of a meter if it is not used in a particular range. I am going to recommend that the cal lab supervisor poll all of the customers holding meters with type J capability. The objective is to determine (1) do they use J thermocouples, and (2) if so, do they need to use them below the ice point. The answers will determine the changes needed to our calibration procedures, and the need for limited calibration labels on the meters. As long as the calibration procedure, labeling and other processes meet the requirements of the lab's quality management system and the company's repair station manual, then compliance is not an issue.

(For those who do not know, the only part of the US airline industry that was deregulated in 1978 is where we can go and how much is charged. Everything else about airline operation is still highly regulated. Actually, it's a good mix. You get the benefit of fares that are on average much lower that they used to be in 1978 dollars (even less after adjusting for inflation!), and the other regulations that help keep scheduled air travel the safest form of transportation in existence.)

The performance specification of the meter is typically, as you surely know, a small fraction of the specification of an uncalibrated thermocouple. (For example, a Fluke 50S or 50D is specified at +/-(0.1% reading + 0.8 degree C) for Type J. At the ice point, that is less than 1/3 of the specification of a standard grade uncalibrated thermocouple.) So, calibration of the thermocouple meter is normally not a large problem. The input is generated by a millivolt standard, scaled to represent Celsius degrees for the appropriate thermocouple type. (Virtually none of the meters in our workload can actually display millivolts.) We normally calibrate the meter over its full specified range. My original question was prompted by a "what if" where the meter is outside of its specification but in a temperature area where, regardless of the existence of emf tables and polynomial equations, the thermocouple is not recommended for use.

As for operation below the ice point, I am now more skeptical of thermocouples than I was before! As you mentioned, there are some very strange wiggles in the curves, and the lower Seebeck emf makes the meter calibration more critical. I think I will also correspond with Dean Ripple about some aspects of this question, but I wanted to hear the voice of the denizens of the Cove first. Thank you again.

Graeme

 

[Bubba]

If you want to measure temperature below 0° C, I would definitely suggest Type T. We use a cryogenic chamber for one of our manufacturing processes here with a Type T thermocouple for the temperature controller. We operate around -120° C and wouldn't even think about using Type J or K for this application, even though Type J dominates the rest of our facility.