Policy For Calibrating Thermocouples and How To Calibrate a Thermocouple

[Graeme]

One who is continually learning is now on another quest for the knowledge of the assembled gurus of The Cove. This young fool (well, middle-aged at least) rashly overturned a long-dormant can of snakes, and I am now seeking a policy useful to return them to a state of control. Then I can retreat to the safety of the electronic calibration lab and hope all of the production managers forget who I am ...

The problem is temperature measurements in a number of repair, rebuilding or low-volume remanufacturing processes. For example, one process is something to do with curing adhesives in composite material lay-ups. Temperature is an important parameter, often a critical one. (I don't actually know much about this process or any of the other problem processes.) The operators monitor temperatures with an array of thermocouples placed on the work and connected to digital thermometers. Some thermocouples are purchased, but most are manufactured in the shop as needed, from bulk thermocouple wire. They are replaced when they "wear out" or if the readings "look funny".

The digital thermometers are calibrated. The thermocouples are not calibrated.

I did a back-of-the-envelope MSA and discovered that, using the standard conformance specification for the thermocouple type, their measurement system uncertainty is way more than what is specified in the process instructions. But, "that's the way we've always done it".

(Before you ask, this is NOT an ISO, QS, AS or any other type of 9001 organization!) (Except for their electronic calibration lab! )

I would like to discover some policy examples y'all may be able to share covering (a) local manufacture of thermocouples by the shop that uses them; and (b) calibration of thermocouples and other temperature sensors. Here are my initial thoughts for your gracious nit-picking.

( **** Definition: Ice Point = the freezing point of pure water at "normal" atmospheric pressure, defined as 0.0 degrees Celsius.)

A: Local fabrication of thermocouples is acceptable provided the performance is correctly verified before use. This means the thermocouple output should be measured at the ice point and at the expected working temperature of the process. The thermocouple should be tagged with the date is was made, the lot number of the wire spool it was made from, and the measurements recorded at the verification temperatures. If needed, the tag should have room to tally the number of times the thermocouple has been used, and the maximum temperature it has been exposed to.

B: Other types of thermocouples and temperature sensors should be regularly calibrated. Typical instruments include but are not limited to

• Thermocouples, either open-junction or sheathed probes
• Thermistor or PRT probes
• Liquid-in-glass thermometers
• Bimetal thermometers

If thermocouples are mounted in process equipment they must be calibrated in place. Sensors that are not permanently mounted should be calibrated in the calibration lab on a regular basis. They should be calibrated at the ice point, the fixed point closest to the lower of the highest process temperature or the maximum usable temperature, and at other fixed points sufficient to verify the linearity.


Let the learning begin! (Thanks in advance.)

 

[Marc]

It's been years since I worked in a lab environment but as I remember we never calibrated thermocouples - we calibrated the sensor/indicator at specific voltage inputs. Each type of thermocouple wire had a different voltage range so that had to be accounted for.

Jerry may be able to shed some light on this.

 

[Graeme]

More information (than you ever wanted?)

Marc,

I have seen both calibrated and uncalibrated thermocouples used, but most organinzations I have been with or visited require thermocouples to be calibrated if they are used in a process that determines product quality. But then, most of those organizations have been military, or defense contractors.

Here is an example of why it may be a problem if an uncalibrated thermocouple is used.

• The type conformance specification for a standard Type-J thermocouple (iron vs copper-nickel) is an accuracy of +/- 2.2 deg.C or 0.75% of the output, whichever is larger.
• The performance specification for a typical digital thermometer in wide use (Fluke 51 or 52, for example) is +/-(0.1% of reading + 0.8 deg.C). It has a best resolution of 0.1 degree.
• Since these values are obtained independently and are essentially random (any thermocouple can be connected to any meter) the values can be combined by the RSS method.
• Axiom: The total uncertainty can never be smaller than the largest component.
• Even though the digital thermometer can display to 1/10 degree, the system accuracy can never be less than +/- 2.3 deg.C when the numbers above are combined. (If you do a worst-case combination, it is +/- 3.0 degrees!)
• Even if the digital thermometer is the fabled "perfect" one with zero uncertainty, the system uncertainty will still be +/- 2.2 deg.C unless the thermocouple is calibrated.
• An inadeqautely trained operator will usually accept the thermometer display as "truth" and use that to control or monitor the process. This can result in an in-tolerance measurement display but an actual process temperature that is out of tolerance by as much as 2.2 degrees.

The meters are calibrated as you describe -- we do a lot of them. My concern is that the production shops are taking a calibrated high-accuracy meter, connecting it to an uncalibrated, relatively coarse sensor, and believing they are making quality measurements. In fact, in an effort to "improve the quality" some of them recently got a large batch of new meters with accuracy specifications of +/-(0.05% of reading + 0.3 deg.C)! I have plotted all of this several ways in Excel, and with an uncalibrated thermocouple this new meter makes virtually Zero difference. (The difference it does make is in the calibration lab! It takes 4 to 6 hours to calibrate the new ones manually, using our high-end DC standard (Datron 4808). The less accurate ones can be done in 30 minutes on our automated system using a Wavetek 9100.)

A thermocouple can easily be calibrated to a k=3 accuracy of +/-0.5 deg.C at the ice point and the process temperature. Combining that with the Fluke 51, the system uncertainty is now +/-0.9 deg.C -- a major improvement, and within the process requirements.

I need to beat them with the math (assuming my reasoning is correct) but I also want to present a viable solution that will make everyone happy. (Or at least equally miserable?) It's much better to tell the bosses "you have a problem and here's how to fix it" than to just slap them with the problem & leave! Thus the quest for knowledge.

Graeme

 

[Marc]

You're way ahead of me! I'll have to sit back and see if there are any other experts lurking about.

 

[M Greenaway]

Our calibration policy is only to calibrate measuring equipment that determines the conformity of the product.

We have thermocouples in our injection moulding machines, but these are not calibrated because we determine the conformity of the product at subsequent in-process inspection stages.

If we relied on temperature as an important process parameter that was critical to product quality, and we could not subsequently verify the product (i.e. a special process) then we would need calibrated thermocouples.

You might compare this to the lead screws and dials and/or NC code on a lathe - should these be calibrated ? No, because we verify the product that comes off the lathe using other calibrated equipment.

We do have calibrated thermocouples in our test lab, which is ISO17025 approved, as clearly temperature is an important parameter in environmental testing.

So you need to look at the application of the thermocouple.

Hope this helps.

 

[Mike S.]

Not too long ago we had an adhesive curing operation that was mildly temperature sensitive. Full cure could be achieved, for example, with 1 hour at 100 C, or 40 minutes at 120 C, or 20 minutes at 140 C, with a max. allowed temp of 150 C. precise temp measurement was not needed, but it had to be "in the ballpark". We decided to calibrate the thermocouple and instrument for this particular application together by using boiling DI water and looking for a readout of 100 +/- 2 degrees C. It was a fast, easy way to do an in-house cal that was suitable for this application. Once this was done, we cured some parts at 140 C for 20, 30, and 40 minutes and did tensile testing of the bond. Strength was the same on all 3 sets so we verified that the total process was working. To allow some room for Murphy, we set the process at 140 C for 30 minutes and every 3 months we re-calibrated the unit/thermocouple and watched the process output for anything strance and had no problems. Just remember, the clock starts once the bond line reaches temp, not when the oven reaches that temp.

Mike S.

 

[Jerry Eldred]

I just found out my wifef's car died, and so I am headed out. Not to mention there are a lot of replies already.

This happens to be one of my strong areas. I'll give some preliminaries for the moment... more later.

We calibrate all thermocouples. If you're using them in a non-quantitative diagnostic mode, you can probably get by without cal. But we use them in numerous applications from 50 deg C to about 1200 deg C. And they do have errors, and at higher temperatures they do degrade.

The typical low temp problems are the impurities in connector materials (which cause an induced secondary junction of unknown error). This needs to be compensated for with a good zero reference. If you use detachable thermocouples in handheld meters (such as Fluke 51/52), a good zero in crushed ice is important. Gain and linearity checks are important as well. We've had numerous applications with type T and K in small oven chambers with as much as 1 to 2 degrees error on a new TC at temps around 150 to 250 C. Type K has a nasty habit of changing significantly at certain temps, and has to be rejuvenated through an anneal process.

High temps with type R and S (I won't even talk about B - they need to be removed from existence) cause outgassing of the Rhodium and subsequent emf degradation, and also causes precipitation of oxides. This causes the linearity and gain to drift.

 

[lhiga]

Thermocouple Calibration

Say Graeme,

You are correct, the navy uses the standard thermocouple tolerances and adds them to the tolerance of the meters in all their procedures where a thermocouple is HOT checked with a meter. You can find the wire uncertainty in the Omega Temperature Catalog. 2.2 deg C for Type J sounds right.

 

[Tekakwitha]

Thermocouples

We use thermolcouples in our lab for monitoring our ovens and water baths and freezers. They are attached to a yearly calibrated Omega monogram. Any suggestions as to validity of these wires? They were purchased certified from Omega.
Also we use a Data Paq to measure our cure and bake ovens for paint. If anyone knows of an umcomplicated way of verifying them please advise.:ca:

 

[GWCollier]

Re: Calibrating Thermocouples - Policy For and How To

I could use some help on calibration of a temperature controller. I've always had an established Calibration dept to handle thermocouple and controller calibrations but now I"m faced with preparing a procedure for an new company.

I know that I can input a source to controller as for the monitoring thermocouple to check that portion, but to check the output to the controlling thermocouple; whats the best approach? Does anyone have an old procedure that they have used that could be shared?

As to thermocouples: in the past we would purchase a "calibrated" thermocouple and maintain it. We would used this to make a "Ref" calibration thermocouple that was checked against the purchased unit and use the Rerference unit for calibrating operating thermocouples. This prevented the degradation of the purchased calibrated unit.