Calibrating a 100A DC shunt

[enut11]

I want to characterise a second hand 100A shunt that I purchased on eBay. Turned out to be a bit more expensive than a new Chinese one but hopefully better quality.
https://www.ebay.com.au/itm/255796427831?var=555732568878
I have on loan a calibrated 100A shunt. I opted to test at around 20A using a HP Server 12v/600w supply and a stable 300w fan cooled programmable resistor load set to 0.45ohm.
According to the calibration sheet that came with the reference shunt, testing was done at the 4 minute mark at 20A, 30A and 50A DC current.
The biggest challenge with high current testing is the impact of heat on all components – power supply, cables, connectors, shunt and load.

[enut11]

The setup that I started with was to connect everything in series:
PS  >  DUTshunt  >  REFshunt  >  LOAD
and simultaneously measure the voltage across the shunts using two mains operated bench meters. The power supply negative terminal was not earth grounded.

Attached chart shows the results of my first test, DUT on the left and Reference shunt on the right. The test lasted 5 minutes. I have scaled the Y-axis steps to be the same on both graphs.

[enut11]

My initial analysis is that the DUT settles down after about 1.5 min and has a lower slope thereafter compared to the Ref which seems to continue upwards.
At the 4min mark, the DUT drops 14.056mV compared to Ref at 23.29mV.

Calibrated resistance of the Ref at 4 minutes is 0.999025mOhm. This computes to 23.31A test current.
At the test current, DUT resistance computes to 0.603 mOhm, ie 0.5% high.

A picture of the monster 300w load below. 3 fans, one for each tube (not shown) and one above the hotter part of the 2 stage resistive element. The DUT is the white shunt.

Comments on test setup and results welcome.
enut11

[enut11]

I tried to measure the change in the 300w resistor during the test using an isolated handheld DMM.
EDIT: calculated change over 4 minutes is 0.125%

[beanflying]

Maybe time for a glycerine or oil bath to keep it more stable?

Either that or buy one of these bad boys   eBay auction: #325785913564

[PartialDischarge]

My impression is that the use of that kind of load resistor totally invalidates your results. Simply put its resistance is too dependant on temperature (or power consumed too high) for you to have a precise constant current setup

[enut11]

Thanks @beanflying but the Valhalla is a bit above my $50 budget!

@PartialDischarge
Yes, the small change in voltage (~0.13%) across the 300w resistor will affect the test current but, as both shunts are in series, would not the relationship between DUT and Ref still be valid?
enut11

[bdunham7]

would not the relationship between DUT and Ref still be valid?

It would, and you should be plotting that ratio rather than the expanded plots of their actual values.  The overall curve that they both exhibit is probably mostly the current actually changing.

[PartialDischarge]

Ok, I see your idea now. Although the reference shunt voltage does not seem to be stable, contrary to your DUT.

[trobbins]

The reference shunt likely has a subtle varying tempco difference than the DUT, due to construction of the sensing resistance element and the resultant thermal conduction to the ends, versus to ambient.

Normally, shunts like the reference are turned on their side so that the resistance element has a lower Rth to ambient.  You could also include some airflow to further suppress any tempco contribution.

For most power applications, commercial shunts I've come across like those two would typically have a 1% tolerance from the manufacturer. 

I recently tested a batch of ten 5A 100mV marked "class 0.2" shunts from A.J.Williams (Australian) and was able to 'confirm' that shunt resistance varied between all units by less than 0.17% max.  Those shunts were circa 40 years old, and of unknown use history, but given the class rating and serial numbering and no outlier measurement results, it does appear that they were originally made with a 0.2% tolerance.  I could also conjecture that the three shunts closest to the median value are at worst <0.1% tolerance from the original manufacturer reference, and likely to have a lower tolerance, but haven't as yet been able to further that view.

PS, thanks for copying the shunt calibration report.

[enut11]

Ok, I see your idea now. Although the reference shunt voltage does not seem to be stable, contrary to your DUT.

I guess this is why the cal lab specified a time (4min) for the test to be taken.

[enut11]

would not the relationship between DUT and Ref still be valid?

It would, and you should be plotting that ratio rather than the expanded plots of their actual values.  The overall curve that they both exhibit is probably mostly the current actually changing.

Thanks @bdunham7, I will have a look at that...

[mendip_discovery]

IIRC the Fluke Bible has a method for measuring shunts. I haven't tried it yet.

I have mostly relied on the calibrator and its specifications for keeping a steady current but that is for 5min on and 5min off which I do for 5 forward reading and then 5 reverse. Taking the average of all the readings to get a resistance value.

The method my ex lab manager had was use a known shunt to confirm the current going through the unknown shunt and use that to correct the current and then do the maths. But he had two multimeters, broke one before he left.

Talk of shunt measurments before have suggested and idea to measure the temperature, and possibly using a small diy oven to change the temperature to simulate the effects of heating on the shunt. Using the temperature of the shunt during measurments when in use to correct the resistance measurments.

[Berni]

I did testing on 100A 300A even 1000A shunts before.

The thing you want to mostly test for is temp co. Some of the cheep shunts have huge variations due to temperature. The fact that these things are run hot also causes issues with thermocouple voltages at the sense terminals.

The way i did testing was to use a 500A lab PSU to test the current capability. This was used to heat up the shunt, if it didn't get hot enough it was wrapped in a cloth to help it get there. Then once that is done, the setup is switched over to a programable lab PSU that sources 9A trough it, a DMM with a 10A range checking the current and then another DMM reading the sense voltage. Another DMM is used to monitor temperature via a thermocouple. A script was used to toggle the PSU between 0A and 9A, continuously collecting measurements once every few seconds.

At the end the whole thing is put into a graph to show the results.

The other method i also tried is using a Danfysik Ultrastab flux gate current transducer (very accurate and linear) to directly measure a 300A current pushed trough the shunt. Also works well

[enut11]

The reference shunt likely has a subtle varying tempco difference than the DUT, due to construction of the sensing resistance element and the resultant thermal conduction to the ends, versus to ambient.

Normally, shunts like the reference are turned on their side so that the resistance element has a lower Rth to ambient.  You could also include some airflow to further suppress any tempco contribution.

For most power applications, commercial shunts I've come across like those two would typically have a 1% tolerance from the manufacturer. 

I recently tested a batch of ten 5A 100mV marked "class 0.2" shunts from A.J.Williams (Australian) and was able to 'confirm' that shunt resistance varied between all units by less than 0.17% max.  Those shunts were circa 40 years old, and of unknown use history, but given the class rating and serial numbering and no outlier measurement results, it does appear that they were originally made with a 0.2% tolerance.  I could also conjecture that the three shunts closest to the median value are at worst <0.1% tolerance from the original manufacturer reference, and likely to have a lower tolerance, but haven't as yet been able to further that view.

PS, thanks for copying the shunt calibration report.

@trobbins
Here are some pix of the Aussie cal kit. The second shunt is 25A

[enut11]

Cal certificate for the 25A shunt

[enut11]

would not the relationship between DUT and Ref still be valid?

It would, and you should be plotting that ratio rather than the expanded plots of their actual values.  The overall curve that they both exhibit is probably mostly the current actually changing.

This is the ratio of REF/DUT shunt voltage at 23.3A. The X scale spans 5 minutes. Apart from an initial dip at the beginning, the DUT still has a lower temperature coefficient.

[Smokey]

from the ebay listing description:

"Rated current: 100A or 150A
Voltage drop: 60mV"

If it's 60mV drop at 100A, that means about 0.6mOhms.  You measured 0.603 mOhm.
Seems pretty close to me.

[trobbins]

Perhaps the ratio plot in post #16 is highlighting the rise of temp of the element in the REF shunt, and it having a changing tempco just above ambient temp, as it moves from ambient temp to stable temp.  You may get different characteristic curves if, for example, the REF shunt only was kept cooled with a fan, or in the alternative the DUT shunt was kept forced-air cooled.  You may also be able to measure the temp rise on each element surface using a small thermocouple, or IR spot meter.

Without stress during life, I guess one could expect quite low aging impact on the 2003 cal of the REF shunts.

[beanflying]

If you are interested @enut11 drop me a PM and I will send you my little set of Shunts to play with. I have nothing planned for them for the next few months and have other options if I need to measure something anyway.

10, 100 and 500A 0.25% Murata's with Manganin elements so the tempco is low.



Interesting read for continuous operation they suggest derating them by a chunk https://www.farnell.com/datasheets/2237929.pdf

[bdunham7]

Apart from an initial dip at the beginning, the DUT still has a lower temperature coefficient.

The maximum difference in tempco is on the order of 0.05%, whereas the absolute difference between the two appears to be about 0.5%.  I don't see how you can conclude that one or the other has the lower tempco, just that they both seem close.  That seems a pretty good result unless you are hoping to characterize the DUT to an accuracy much better than its 0.5% spec.

[trobbins]

I also have a penchant for shunts ranging up to 500A 50mV, and am working towards some absolute accuracy tests, but that may take a few years and a step up in other calibrated equipment.  The big vintage Tinsley has a 1984 measurement to 0.01 micro-ohm (and uV) resolution at 1A but I don't know the origin of that, and I know of an even larger Tinsley that has a similar vintage measurement, so the hope is to one day see what age-related variation may have occurred.

Ciao, Tim from Melbourne

[beanflying]

Williams made some high grade stuff. Not really a shunt but it is 10A rated  I added this one to my collection a year or so back.

[mzzj]

Your setup looks quite horrible thermal emf-wise. Ditch those alligator clamps and wire the sense cables directly to current shunts without any connectors.
10uV thermoelectric voltage will ruin your whole measurement in this case!

Would be also good idea to measure the thermal EMF before and right  after the test to get idea how good the setup is.
Continue the logging while you disconnect the measurement current to see how many uV's of thermoelectric error you have after the measurement.

Also use enough heavy gauge wiring for the current connections and crimp proper terminals on the wires.

Having said that your results look about as good as you can expect with this level of shunts. Calibration certificates for the reference shunts leave some open questions.
After improving your wiring you could add forced air cooling alternately to DUT or REF shunt and plot the results.

Your measurement drifts about 0.1% and  20 degree temperature rise with 50ppm difference in resistor tempco's would be already enough for this.

Also read this if not already familiar: http://ohm-lab.com/pdfs/Shunt%20Calibration.pdf

[enut11]

Thanks @mzzj, that 'ohm' article is great and quite an eye opener. Currently away from home but I will certainly have a good look at my setup.
@beanflying is sending me his super Murata shunts to play with and I certainly need to up my game to do them justice.

What a minefield!
enut11

[enut11]

Apart from an initial dip at the beginning, the DUT still has a lower temperature coefficient.

The maximum difference in tempco is on the order of 0.05%, whereas the absolute difference between the two appears to be about 0.5%.  I don't see how you can conclude that one or the other has the lower tempco, just that they both seem close.  That seems a pretty good result unless you are hoping to characterize the DUT to an accuracy much better than its 0.5% spec.

Thanks @bdunham7. If the two shunts had similar tempco, I would expect the ratio curve to be flat?
enut11

[bdunham7]

Thanks @bdunham7. If the two shunts had similar tempco, I would expect the ratio curve to be flat?

The closer the flatter, although the scaling you choose for the graph affects how it looks--that would look pretty flat on a graph that went from 0 to 2, for example.

Keep in mind that we're using the term "tempco" as a rough approximation for how the shunts behave as they warm up from a fixed current.  In fact, it is possible that one is quite a bit warmer than the other and that temperature gradients and thermal voltages are part of the equation.  It might be interesting to compare them if your DUT was aggressively cooled, for example, while the one with the certificate was treated just as stated and everything was measured after 4 minutes as stated on the certificate.

[PartialDischarge]

The other method i also tried is using a Danfysik Ultrastab flux gate current transducer (very accurate and linear) to directly measure a 300A current pushed trough the shunt. Also works well

Thanks for the idea, just bought a couple of ULTRASTAB 867 at a very low price: https://www.ebay.com/itm/235153418291

Also didn't know but there are Murata 0.25% shunts on digikey until supplies last:

[beanflying]

Likely similar pricing but I got mine via E14/Farnells https://au.element14.com/c/automation-process-control/panel-displays-instrumentation/current-shunts?brand=murata-power-solutions which might suit others.

[Chance92]

Thanks @mzzj, that 'ohm' article is great and quite an eye opener. Currently away from home but I will certainly have a good look at my setup.
@beanflying is sending me his super Murata shunts to play with and I certainly need to up my game to do them justice.

What a minefield!
enut11

I'm going to calibrate the same 100A shunt with a calibrator and a current amplifier. Maybe we can compare results.

[Berni]

The absolute precision of shunt resistors does not really matter that much.

You can calibrate your shunt as long as you have an accurate way of measuring voltage and current. These are both things that bench DMMs can be very good at. Metal resistors like this don't suffer from any weird non linear effects, so you only need one calibration point to be confident in the whole range. So no problem in calibrating a shunt at 10A then extrapolating your calibration out to measure a unknown 100A current. But they do have a temp co! So the calibration is only valid at that temperature! Waiting for it to heat up is no excuse as it won't heat up the same every time.

The actual problem is that shunts get less repeatable once scaled up to 100s of amps. These shunts have very very low resistances, so you get very tiny voltages. This means thermocouple voltages matter a lot and can wreck your accuracy. At these currents things get hot, making the problem worse. The big block of brass that feeds the resistive element with current also has resistance and tempco of its own, so it can add some phantom voltages. To make things even worse the way you connect to the brass block even makes a difference, since turning a cable lug 90 degrees makes the current enter at a slightly different spot, taking a different path trough the brass block as a result.

I had shunts that show considerable differences in performance from just changing how the sense wire is connected to the screw.

So unless you have a really dialed in setup don't expect your measurements to be reliably within 1% by just buying a fancy precision resistive shunt. If you are after precision the LEM ultrastab transducers are way superior, and aren't even all that expensive on the used market.

[alm]

Funny how this is the third thread about the subject of high current in a short time. They are distinct topics, like one was about AC, but still interesting that this is apparently the high current season:

• Masuring 50A DC
• Calibration of Rogowski Coil at ~400 A AC RMS 50 Hz

So unless you have a really dialed in setup don't expect your measurements to be reliably within 1% by just buying a fancy precision resistive shunt. If you are after precision the LEM ultrastab transducers are way superior, and aren't even all that expensive on the used market.

I agree with everything you say. The document mzzj linked to earlier also describes testing multiple ways of attention sense wires. I also like DCCTs like LEM Ultrastab.

Although to some degree making shunts bigger or adding active cooling can increase the voltage drop. You often see that with lab-grade shunts that the shunts get bigger the higher the current so they can dissipate more. See for example this 15A shunt designed for dissipating up to 22.5W. Or the Fluke A40B series of shunts. Compare the size of the 100A shunt to the 1A shunt.

Another option is integrating an amplifier with the shunt so any uncertainty due to connection to the shunt and thermal EMF is handled within the box and can be measured as part of the calibration. This is how the Valhalla 2575A active current shunt works: both active cooling and built-in amplification.

[mendip_discovery]

That's part of the fun of the metrology side of things. TEAS is the beginning then you get into metrology and then they fun begins.

The main thing is to find a way to get consistent and repeatable results. Document what you do and have fun.

[enut11]

The other method i also tried is using a Danfysik Ultrastab flux gate current transducer (very accurate and linear) to directly measure a 300A current pushed trough the shunt. Also works well

Thanks for the idea, just bought a couple of ULTRASTAB 867 at a very low price: https://www.ebay.com/itm/235153418291

Also didn't know but there are Murata 0.25% shunts on digikey until supplies last:

@Berni, @PartialDischarge
Can you tell me a bit more about the Ultrastab devices? How do they work?  Are they useful under 50A?
enut11

[Berni]

Yep they are incredibly linear and stable once powered on.

I forgot the exact model i have, but i think it is designed for 500A, yet it can still easily measure a 10mA current trough it accurately.

The big difference that makes them so good is the flux nulling method. The magnetic sensor inside is only looking for zero, adjusting the current in the coil to get it there. This current in the coil is what is sent out as the output, so your output is the very same current made from the very same electrons that flown trough the coil to null out the measured current. So the conversion factor is purely dependent on the number of turns around the core that forms a transformer. Tho yes the zero stability of the magnetic sensor still matters, but they are specially designed sensors for that exact task of looking for 0 flux, designed to be as stable as possible at that point.

The 1 problem is the stay magnetization in the core and the earths magnetic field. So they do need the 0A to be calibrated on first power on. But 0A calibration standards are really cheap, it is simply leaving the center of the transformer empty.

[guenthert]

[..]
I forgot the exact model i have, but i think it is designed for 500A, yet it can still easily measure a 10mA current trough it accurately.
[..]

Interesting.  Funny that you picked 10mA as example.  I couldn't help but to think of Cern's 10mA reference [1].

Would it then also be possible to measure (i.e. compare) very small currents?  That is, if the induced (very small) current is compared to the unknown current, while a known (e.g. aforementioned 10mA or much less) is fed through the main port?

Tho yes the zero stability of the magnetic sensor still matters, but they are specially designed sensors for that exact task of looking for 0 flux, designed to be as stable as possible at that point.

So the question is, how precisely can those transducers determine 0 flux, but perhaps this belongs in another thread  .


[1] http://cds.cern.ch/record/643294/files/cer-002399331.pdf

[Berni]

I can give it a try at exactly how good the Ultrastab flux gate is at low currents.

I never actually used it for measuring small currents because the main reason why i bought one is to give me the ability to measure in the range of 10A to 300A with a reasonable amount of confidence. I just tried putting a few miliamps trough it once to see how usable it is down low (as this is what currents clamps are bad at) and i was very impressed with its performance.

The whole unit does heat up a fair bit as it runs (since the current being fed into the coils is provided inside by a linear AB class amplifier) and i never had to re-zero it once it warms up, so id assume the temp co of the flux sensor is also impressively low. The datasheet for a LEM IT 200-S ULTRASTAB claims a initial offset of 80ppm with 2 ppm/°C temp co and 1 ppm per month stability(for a 200mA test current). Those are really impressive figures. You will have a hard time finding a resistor this stable.

[mzzj]

I can give it a try at exactly how good the Ultrastab flux gate is at low currents.

I never actually used it for measuring small currents because the main reason why i bought one is to give me the ability to measure in the range of 10A to 300A with a reasonable amount of confidence. I just tried putting a few miliamps trough it once to see how usable it is down low (as this is what currents clamps are bad at) and i was very impressed with its performance.

The whole unit does heat up a fair bit as it runs (since the current being fed into the coils is provided inside by a linear AB class amplifier) and i never had to re-zero it once it warms up, so id assume the temp co of the flux sensor is also impressively low. The datasheet for a LEM IT 200-S ULTRASTAB claims a initial offset of 80ppm with 2 ppm/°C temp co and 1 ppm per month stability(for a 200mA test current). Those are really impressive figures. You will have a hard time finding a resistor this stable.

If you use the Ultrastab fluxgate sensor in fixed position and zero out the initial offset you can get better than 0.5ppm of full scale output in the low end.
For example typical 700A Ultrastab sensor has enough good short-term stability and linearity to measure with 0.35mA accuracy at lower end of the range.   
(IIRC the 600A unit I tested was as good as 0.15mA offset stability/repeatability.)
In practice this means that you can measure 350mA with 0.1% accuracy. Or 3.5A with 0.01% accuracy.
Loop the primary wire 10 turns trough the hole in the fluxgate and you get 0.1% accuracy down to 35mA with a maximum range of 70A.
100 turns primary would be still feasible if you really want and then you can get measure down to microamps with 7A maximum range.  1uA resolution with 7A maximum would be still feasible with these.
 

[enut11]

It occurred to me that the physical mount of a conventional current shunt may affect the stability of the output signal.
If both end blocks are rigidly mounted, would heat cause stress at the junction of the resistive element and the blocks.
If so, would this affect the output signal?

[mzzj]

It occurred to me that the physical mount of a conventional current shunt may affect the stability of the output signal.
If both end blocks are rigidly mounted, would heat cause stress at the junction of the resistive element and the blocks.
If so, would this affect the output signal?

For sure it is going to have some effect. Just look at the trouble and expense standard resistor manufacturers have gone trough to avoid stressing the resistive element.
BUT the effect is probably ballpark of tens to hundred of ppm, ie less than 0.01%

edit:
Best I could do right now in home lab was two HP 6632B's supplying 10A total to a 60mV 150A shunt and 34401A measuring the shunt voltage.
Pressing with plastic rod as hard as I can(about 200N give or take) on the middle of the shunts resistive element I see no change(less than the 0.1uV resolution of the 34401A) so that would be less than 0.0025% change from physical stress.

6632B Power supplies and 34401A are remarkably stable for short perioid of time, I have no problem staying within 1 digit (0.1uV) over a short time span like 2 minutes.

[tszaboo]

Pro tip, once you calibrated your shunt, flip it around and see if it measures the same value the other way.

[Irv1n]

 This photo of our cal bench: dci source, 2182, standard shunt 100A, 500A, 1kA, 2kA

[Berni]

That is a rather fancy looking setup. Thanks for sharing.

[PartialDischarge]

is that a pneumatic contactor to disconnect the shunt?

[Irv1n]

is that a pneumatic contactor to disconnect the shunt?

Yes

[enut11]

After a couple of weeks off with health issues and other distractions I am now looking to push on with this project.
The Murata 0.25% shunts from @beanflying have arrived, thanks.

Learnings so far:
1) The 4W setups were poorly implemented. Docs supplied by Forum members will help here.
2) Too much heat in the load resistors. The test voltage will be halved to around 6v.
3) Orientation of shunts is important. Stability is improved if the shunts are mounted to facilitate natural air flow.
4) Tests will be carried out over 1 min instead of 4 min. (EDIT: 2min to allow for the PS  and load to stabilize)
5) Shunt element temperature measurements and forced cooling will be considered if necessary.

The contenders below, from L to R:
- The shunt that started all this, 100A 0.5% form Bulgaria (white base)
- My home made 50A shunt (wood base)
- Murata 0.25% 10A, 100A and 500A (nice)
- 25A and 100A with 2002 calibration certificates
enut11

[joeqsmith]

Top left Weston 500A, Wade Allen 200A, Weston 10A.  Top right, Weston 150, 100, 50A.  Front center,  GE 100A 0.2%. 

No idea about any of these as I really have no way to measure them.  They would need to be sent in for calibration.   After I was given the old GE shunt, I tried a few tests with it.

[enut11]

@joeqsmith, nice collection of old shunts. They were not only practical but works of art. Love those old knurled and tabbed knobs.
With your GE 100A 0.2% shunt and a decent multimeter you can calibrate your own shunts. BTW, what is your power source?
enut11

[joeqsmith]

@joeqsmith, nice collection of old shunts. They were not only practical but works of art. Love those old knurled and tabbed knobs.
With your GE 100A 0.2% shunt and a decent multimeter you can calibrate your own shunts. BTW, what is your power source?
enut11

There is nothing I can do to calibrate anything.  Nothing I own is in current calibration.  Most of my experiments are making relative measurements and I can tolerate a fair bit of slop, so I run some cross checks and call it a day.  If I required absolute accuracy, the only thing I could do would be to send them in and request a report.  I've never looked into it.

The power supply shown in that last video has changed over the years.  Watch the first few minutes for details on the current (no pun intended) setup.

[enut11]

@joeqsmith. Wow, impressive setup! I see you test at very low voltages. Are you testing with DC or AC?

[joeqsmith]

DC only,  limited by the power supplies response time.  That setup was used to test several handheld multimeter leads and shunts to failure.  It was also used when testing a cheap UNI-T UT61E that I had modified to measure 20A with a custom made shunt I designed for it.  These are all very low resistance, or as Dave would say, until they are not.      

I used it to heat up some water once.       

[enut11]

OK, for DC the PS must have some very impressive rectifiers. Can you tell me more about your power supply.

[beanflying]

You might get some tips out of the Valhalla Calibrator manual. While it is only 10A max it only uses 1-2V in the test path. Circuit diagrams are at the last few pages. Its bigger brother might have a similar manual out there but I have never looked.

[joeqsmith]

OK, for DC the PS must have some very impressive rectifiers. Can you tell me more about your power supply.

It's just a couple of OTS supplies in parallel with a LEM sensor.  A small embedded computer controls the current/voltage and talks to the PC which is running a simple LabView program to automate it.

If you want to buy something, I would start with posting your requirements.  If you are just wanting to know which brands to look at, I suggest Cotek.  They offer some with a PC interface. 

I worked for a company where we had a few large Sorenson  power supplies.  Large meaning four eyehooks on the corners to hook the chains to in order to lift it.  I would run these from a PC as well.  I looked on eBay to see if I could find one.  No luck but maybe something like this:

https://www.ebay.com/itm/303119807724

Just depends what you are after. 

[enut11]

The ones I saw were 12v. How do you get the voltage down to a few volts? Is this an internal PS mod?

[joeqsmith]

The ones I saw were 12v. How do you get the voltage down to a few volts? Is this an internal PS mod?

No mods.  They are programmable.   If you wanted 12v for example, the Cotek AEK-3000-LV-12  can supply up to roughly 200A.  Have a look: 

https://www.amazon.com/AEK-3000-12-Switching-Supply-Programmable-Single/dp/B00YO4ZFSA

Like most supplies you can run CC or CV.  All programmed over the serial bus.  Software interface for this supply is trivial.  This would be a much better setup than what I have with my old stacked supplies. 

[joeqsmith]

Ours was something like this but rather than high voltage low current, seems it was maybe 20V and several hundred amps.   We are going back 40 some years ago so memory may be off. 

https://www.ebay.com/itm/153853602854

[enut11]

@joeqsmith, love the exercise bars on the BIG one.

Anyway, very high current supplies are way out of my hobby budget but it is interesting to see what is out there.
Thanks
enut11

[enut11]

Still working on my setup for the 20A tests. At 6v my programmable 300w load was still overheating as only a quarter of the elements were on. Now testing to see if 3v compliance is enough.

[enut11]

Been working on a stable low voltage high current supply and I think I have it.
https://www.eevblog.com/forum/projects/gigabyte-psu-as-a-high-current-source/

Next step is to improve the cooling for the 300W load resistor. With a 5v PC supply the resistor will have to handle about 100W which is a lot less than when I used a 12v supply.

[enut11]

OK, for DC the PS must have some very impressive rectifiers. Can you tell me more about your power supply.

It's just a couple of OTS supplies in parallel with a LEM sensor.  A small embedded computer controls the current/voltage and talks to the PC which is running a simple LabView program to automate it.

If you want to buy something, I would start with posting your requirements.  If you are just wanting to know which brands to look at, I suggest Cotek.  They offer some with a PC interface. 

I worked for a company where we had a few large Sorenson  power supplies.  Large meaning four eyehooks on the corners to hook the chains to in order to lift it.  I would run these from a PC as well.  I looked on eBay to see if I could find one.  No luck but maybe something like this:

https://www.ebay.com/itm/303119807724

Just depends what you are after.

@joeqsmith, how does the LEM current sensor work. Is it stand alone into a voltmeter or does it need interface electronics?

[Berni]

The LEM sensors mimic a current transformer, except that it works down to DC.

So what you get out of it is a smaller current that is related to the measured current by an exact factor like say 1:500 (depends on the model). So the only thing is does is extend the measurement range of your existing current measurement equipment like a bench DMM.

Tho since it is an active device it does need power. These things run from split rail +/- 15V and needs a fair bit of current, so it might make sense to build a power supply box for them.

[joeqsmith]

OK, for DC the PS must have some very impressive rectifiers. Can you tell me more about your power supply.

It's just a couple of OTS supplies in parallel with a LEM sensor.  A small embedded computer controls the current/voltage and talks to the PC which is running a simple LabView program to automate it.

If you want to buy something, I would start with posting your requirements.  If you are just wanting to know which brands to look at, I suggest Cotek.  They offer some with a PC interface. 

I worked for a company where we had a few large Sorenson  power supplies.  Large meaning four eyehooks on the corners to hook the chains to in order to lift it.  I would run these from a PC as well.  I looked on eBay to see if I could find one.  No luck but maybe something like this:

https://www.ebay.com/itm/303119807724

Just depends what you are after.

@joeqsmith, how does the LEM current sensor work. Is it stand alone into a voltmeter or does it need interface electronics?

I have a friend who designed sensors for them.   I used a similar nulling technique to design a totally different type of sensor.  Basically a hall type sensor is used to measure the flux in a gap.  Current is controlled to null the field so the hall always works in the same region.  Improves temperature stability.   The nulling current is then proportional to the signal we are measuring.   The Have a look at LEM's closed loop description
 
https://www.lem.com/en/hall-effect-current-sensors

[enut11]

I see the LEM sensors come in a variety of packages and up to several hundred dollars. As with everything, there is a cost/benefit choice. What do you get for the extra money, linearity, accuracy, range? Which one would suit a 0-50A setup if aiming for a 0.1% measurement?

[joeqsmith]

I see the LEM sensors come in a variety of packages and up to several hundred dollars. As with everything, there is a cost/benefit choice. What do you get for the extra money, linearity, accuracy, range? Which one would suit a 0-50A setup if aiming for a 0.1% measurement?

Assuming you didn't mean 1.0% or even 10%, good luck.  I think that one I use is +/-1% but that's just the sensor, not the system.   

Someone did post that 0.05% shunt early on.  You could try finding something that wasn't traceable and guess.  I don't see the point.   Personally if I had a need,  I would send off the parts for proper cal with a report.  Far cheaper and traceable.   I would have about as much confidence in something I constructed with OTS parts as I would with the guys buying these cheap standards to align their DMMs.   I would need to send off my custom calibrator to have calibrated.  Even then.... 

[Kleinstein]

A main difference between the good one and the cheap ones is the linearity and accuracy. To a small part also the range, when going really high, like > 200A.  The cheap ones are mostly hall effect based and thus mainly a thing for relatively high currents. They are also often direct reading and not using compensation. This limits the linearity and stability but also means the get away with low power.

The Ultrastab ones use the fluxgate principle and are much more accurate and linear, but usually also more expensice and they need significant more power. If one is lucky one may get a used one for a relatively good price, but the new ones are expensive. As kind of current-transformer the ratio is very close to an exact integer set by the turns ratio. So very little drift and accuracy problem.

At least in Germany there are 2 types from VAC offered very cheap ( < $2) that also use the flux gate principle, but for a smaller range (e.g. 16 A RMS, 50 A peak) and limited accuracy of 0.7%. The more normal price for these is likely in the $10-$20 range. I have some and looking at the internals I was a bit disappointed. Still impressive what the reach with the limited power and simple contruction. As a closed system they can be a bit more sensitive than most DC clamp meters.

[joeqsmith]

For $1500 USD maybe the FNPZR0100A,  0.01 ohm, 0.05%, 500W, 1ppm/degC.   Should be able to keep the temperature fairly stable with only 25W.   At a half volt,  signal may be high enough to directly read it.   Maybe then send it off for calibration and use as a standard?   Of course, that doesn't cover the cost of the power supply.... 

https://mm.digikey.com/Volume0/opasdata/d220001/medias/docus/760/FNP_Series.pdf

[Berni]

You can easily find the LEM Untrastab flux gate transformers on ebay for under 50 bucks:
https://www.ebay.com/itm/235153418291

The main benefit of them being that the conversion ratio is exactly und exactly the turns ratio. So even if you buy one of these used in unknown condition you don't have to worry about it being out of spec. It is either going to be broken and give no sensible output (or give output that is so wrong it is easy to tell) or it will work fine and be accurate. You won't suddenly have one of the turns of wire fall off the core and change your ratio from 1:1750 to 1:1751

This is also why they are so stable during operation since warming up a transformer also won't change the turns ratio (tho the flux null detector can, but it is specifically designed to minimize it). Since the entire thing operates using currents and not voltages, this means there is no thermocouple voltages to worry about. If you ever tried to measure things in the low microvolts you will see just how difficult it is to keep things stable when your own body heat creates stray voltages all over the place.

[joeqsmith]

You can easily find the LEM Untrastab flux gate transformers on ebay for under 50 bucks:
...

I've never used one but thought I would have a quick look.  OP was asking for 50A.  Maybe something like the it-60-s would be a good fit.  First thing I notice, it's discontinued according to Digikey.  Looking at the datasheet, there is nothing about the accuracy. 

https://www.lem.com/sites/default/files/products_datasheets/it_60-s_ultrastab.pdf

Their website makes a few claims but there's no meat on the bones.   
https://www.lem.com/en/product-list/it-60s-ultrastab

[Berni]

Sure there is.

Linearity error is 3 ppm max and temp co. is 2.5 ppm/K max. Initial offset is 250 ppm max and it might drift by 2.5 ppm/month.

Sure they tend to measure a lot more than 50A, but they are so stable that they are perfectly usable even down into miliamps.

What other accuracy stats you need to know?

[PartialDischarge]

  Looking at the datasheet, there is nothing about the accuracy. 

I believe the original manufacturer is Danfisyk not LEM and their specs are strange too.
The offset is given in ppm and only linearity not accuracy is given.

https://xdevs.com/doc/LEM/doc/867-700I_Installation_Package.pdf

[alm]

I guess absolute accuracy depends on the value of the burden resistor (and stability / tempco, obviously) and needs to be calibrated for the specific system. See the other threads on DCCTs and TiN's article on xdevs.com that were probably linked before in this thread or one of the other recent high current threads.

[JohnG]

For practical purposes, these behave as current sensing transformers that work down to DC. They have a fixed number of turns and a closed-loop feedback from the flux-gate sensor. They drive a current through the turns to cancel the flux from the conductor being sensed, and this current is run through an external burden resistor. The sensed current is reduced by the  turns ratio, which is an exact integer. The ppm and linearity readings refer to the current conversion ratio of the sensor.

IIRC, the flux gate sensor is quite good at determining the zero flux point with low offset, and the turns ratio is pretty close to an exact integer. This leaves the burden resistor as the key component determining the accuracy. However, the burden resistor is measuring a current reduced by the turns ratio, and is no longer in the main current path. Thus, you can use a much larger resistor value than you would for a shunt, making it much easier to get a high accuracy resistor, and giving a much larger sense signal than would be practical with a standard shunt due to the need to avoid dissipating too much power in the shunt.

Normally, these sensors are quite expensive in the $1k to $3k range. I purchased two brand new ones on ebay for $100 each, the ITN 600-S: https://www.lem.com/sites/default/files/products_datasheets/itn_600-s_ultrastab.pdf

These are 600 A, with a turns ratio of 1500. If you have the luxury of being able to wrap multiple turns through the sensor, I recommend getting the higher current ones. The initial offset is less, perhaps due to reduced sensitivity to the Earth's magnetic field, and if you need more sensitivity to lower currents, you can wrap multiple turns through the sensor. For example, with ten turns, the max effective range becomes 60 A.

I use these in a high accuracy efficiency measurement system for power converters. One of the great things about this system is that I can run a single current through an extra winding that I wrap through both current sensors and use a single current run simultaneously through both sensors to account for differences in the two sensors, thereby eliminating a major error source (efficiency is a ratio-metric measurement). The working windings used can then be optimized to take advantage of the full range of the sensor. For example, in a 2.5 kW, 48-12 V converter measurement, I can put 8 turns on the input sensor and 2 on the output sensor and use most of the sensor range.

You can also find some older ones sold under Danphysik. They are good as well, but the newer ones have some protection against user error, like sending a large current through the sensor when it is not powered.

Another company that makes these is Danisense. Here are some prices for new ones: https://gmw.com/product/ds/#productPricing.

Hope you find this useful,
John

[enut11]

Thanks @JohnG, @alm, @Berni, @joeqsmith and @Kleinstein
The magnetic current sensors are a fascinating technology that I can explore at a later date. Meanwhile, I will persist with improvements to my present setup to see what can be achieved.

The load resistors consist of 4 windings, a pair of 0.3R + 0.67R. Various series/parallel combinations are possible by changing links. I have improved the cooling and there are now 4 fans, 2 through the tubes and 2 overhead.

For the power supply, I am using a modified PC PSU. The nominal 5v output settles on ~4.6v at 20A but is very stable under load.
https://www.eevblog.com/forum/projects/gigabyte-psu-as-a-high-current-source/msg5164392/#new (see Reply #8).

Running the load resistors at ~4v/20A (instead of 12v previously) means they are much more stable for the few minutes of the test.

Picture of the new test setup below incorporates suggestions from Forum members, shunts on edge for better cooling and improved sensor wire connections to the logging DMMs.

As before, I will be comparing the voltage from a DUT to a calibrated unit. For these tests the Cal unit will be a 100A 0.25% Murata (small black). The large black unit was also calibrated (2002). The white unit is the shunt I purchased from Bulgaria and the one I need to be characterised as best I can.

I have a further 2 instruments, one monitoring the stability of the power supply and a clampmeter monitoring the test current.

[JohnG]

Make sure you don't melt the fans stick in the end of what looks like copper tubes.

John

[enut11]

Hi @JohnG. The fans blow fresh air into the tubes and work well.

Setting up for the first test with the new setup. Decided to duplicate results from Reply #1. DUT is 100A from Bulgaria and Ref is the 100A (MP, 2002 Cal).
enut11

[alm]

I don't remember if this was discussed already, but you could build up to 100 A. Say you had a DMM that you trust to 10A. Take x identical shunts, put them in series with DMM, and measure voltage drop at 10A. Then put them all in parallel, if necessary compensate differences between shunts with a series resistor, and send X*10A through it. Measure voltage across each shunt add and up for total current. Use this to calibrate one or more X*10A shunts or DCCT. Repeat until you're at 100A.

[enut11]

@alm, it has some merit but outside the capabilities of my equipment at this time.
enut11

[enut11]

First test with new setup. Time was limited to 2min so may not be directly comparable with previous results. DUT (BUL100A) is on 34401A DMM.

Load resistor was much cooler (~50C cf 100C previously). I don't see any real difference now between the REF (MP100A) and the DUT (BUL100A).

The ratio chart is also close to flat indicating similar tempcos.

Need to think about these results.
enut11

EDIT: Fundamental error here. I need to run these tests for at least 4 min in order to use the 2002 calibration data to compute the test current and hence the DUT resistance. 

[joeqsmith]

Sure there is.

Linearity error is 3 ppm max and temp co. is 2.5 ppm/K max. Initial offset is 250 ppm max and it might drift by 2.5 ppm/month.

Sure they tend to measure a lot more than 50A, but they are so stable that they are perfectly usable even down into miliamps.

What other accuracy stats you need to know?

I went to the PDF, typed "accuracy" in the search.  Came up short and called it a day. 

[HighVoltage]

Isabellenhütte makes some very good shunt resistors, the RUG-Z
They can be loaded to 250W and they are very stable.
I have a few of their 1 mOhm versions and I am very happy with them.

At 100A and 1 mOhm you just have 10W power dissipation in the resistor.
The connection is done with M6 screws.


 

[MegaVolt]

I don't remember if this was discussed already, but you could build up to 100 A. Say you had a DMM that you trust to 10A. Take x identical shunts, put them in series with DMM, and measure voltage drop at 10A. Then put them all in parallel, if necessary compensate differences between shunts with a series resistor, and send X*10A through it. Measure voltage across each shunt add and up for total current. Use this to calibrate one or more X*10A shunts or DCCT. Repeat until you're at 100A.

At high currents, the resistance of the wires that help connect the shunts in parallel will kill the entire metrology

[alm]

At high currents, the resistance of the wires that help connect the shunts in parallel will kill the entire metrology

Why? Each shunt is carrying 10A and using the same 4w sensing. Measure voltage across each parallel shunt and add up currents. Leakage current is probably not a concern at these voltages/currents. The only concern I'd have is shunts not sharing current equally thus being far away from the point at which they were calibrated, but that's why I suggested series resistors to compensate.

[MegaVolt]

Why? Each shunt is carrying 10A and using the same 4w sensing. Measure voltage across each parallel shunt and add up currents. Leakage current is probably not a concern at these voltages/currents. The only concern I'd have is shunts not sharing current equally thus being far away from the point at which they were calibrated, but that's why I suggested series resistors to compensate.

Yes. If you take measurements on each shunt, this should work if the current at this point in time is considered constant (the noise is not great).

[enut11]

Found this listing on eBay, $33AUD. Anyone care to comment on suitability for shunt current testing? What else do I need to set this up?

https://www.ebay.com.au/itm/134745360474?_trkparms=amclksrc%3DITM%26aid%3D777008%26algo%3DPERSONAL.TOPIC%26ao%3D1%26asc%3D20230811123856%26meid%3Da633a8ca142a48568297bdf8b264e1cf%26pid%3D101770%26rk%3D1%26rkt%3D1%26itm%3D134745360474%26pmt%3D0%26noa%3D1%26pg%3D4375194%26algv%3DRecentlyViewedItemsV2%26brand%3DUnbranded&_trksid=p4375194.c101770.m146925&_trkparms=parentrq%3Ad4b84d1718b0a8dae88e6978ffff78aa%7Cpageci%3A72d9534e-83f7-11ee-acdd-a6db99b57149%7Ciid%3A1%7Cvlpname%3Avlp_homepage

[alm]

Found this listing on eBay, $33AUD. Anyone care to comment on suitability for shunt current testing? What else do I need to set this up?

https://www.ebay.com.au/itm/134745360474 [snip tracking parameters that were making the URL ridiculously long]

Here is the datasheet. I don't quite see the point of a device with 1% accuracy and linearity error specs. Don't you have shunts that are better than this?

From the datasheet, it looks like the sensor just needs symmetric power rails from +/- 12V to +/- 15V, and has a voltage output of +/- 4V. But I don't see the point unless you need galvanic isolation or the wide bandwidth.

[enut11]

Thanks @alm. Poor accuracy and poor linearity. No go. Cheap for a reason.

[enut11]

Test results to duplicate Reply #1, MP100A (2002 calibrated Ref) vs 100A from Bulgaria (DUT).

Leap of faith here: I am assuming that a 21 year old calibration certificate is still valid!

Calculated resistance of the DUT is now 0.607 mOhm, ie 1.2% high?

REF/DUT ratio  chart looks good with a small downward slope.

Still, not sure if I'm making progress here or just running around in circles.  Maybe the next test, 2002 Cal shunt vs Murata, will shed some light.
enut11

[enut11]

This test compares the 2002 Calibrated shunt with the Murata 100A specified at better than 0.25%.

Calculated resistance of the DUT (Murata 100A) is 1.001 mOhm. Spot on, but is this just a coincidence?

REF/DUT ratio chart looks OK but with a downward slope. Also, been getting a weird oscillation during the tests??

[enut11]

Third of a trio of comparisons, this one MUR100A vs BUL100A at 20A.

Calculated DUT (BUL100A) resistance is .606 mOhm, ie 1% high.

Ratio of REF/DUT is a rising plot. May be due to the small size/mass of the Murata 100A which is less than half of the BUL100A.

[enut11]

Can we draw any conclusions from this project?

It is based on the assumption that the 2002 100A shunt is still in cal.

One result does seem encouraging, the 2002 MP100A vs MUR100A test. They seem to validate each other.

Until I get a better reference, I think I have done as much as I can. I played with test current reversals but the results were inconclusive. None of the shunts heated up by more than a couple of degrees - not surprising as they only had to deal with less than 0.5W.
enut11

[beanflying]

This test compares the 2002 Calibrated shunt with the Murata 100A specified at better than 0.25%.

Calculated resistance of the DUT (Murata 100A) is 1.001 mOhm. Spot on, but is this just a coincidence?

REF/DUT ratio chart looks OK but with a downward slope. Also, been getting a weird oscillation during the tests??

Interesting play time  Like any metrology you are now left wondering what about if? Time to chase the next level of certainty because 'settling' on what you have just isn't ok

I should rig up the 100A one when you get it back to me (NO Rush I am booked until the New Year) and hook it up to my 3-15V 60A bench supply and see what sort of comparative numbers I can get.

[enut11]

Hi @beanflying. Quite right. There is never an end to it. This can be good or bad but always something new to discover. Now curious about those damn LEM sensors!

A bonus for me was discovering how stable PC PSUs can be. At 20A on the 5v outlet, after 2min settling in time, the Gigabyte voltage stayed within a millivolt or so for 5 min at least. Not bad for $5.
enut11

[mzzj]

BUL100 and MP100 seem to have smaller drift or they drift in a same way after applying current. I'd be inclined to guess that MUR100 has highest drift since it is  physically lot smaller. But impossible to say without some extra equipment and measurements.

MP100 vs MUR100 ratio changes 0.08% during the test and it is more than I'd have expected as the shunts are operated at only 1/5th of nominal current.

What you might be able to do is add cooling fan right on top of the MUR100 and see if the BUL100/MUR100 ratio would be more stable.

[trobbins]

Another way to possibly get some more insight is to initially have two test shunts in series but with one shunt 'shorted' whilst the other shunt warms up and stabilises, and then unshort the first shunt, and monitor its change against the other shunt.  Then vice-versa, as a way to characterise each shunt's warm-up period.

[Berni]

The easy way to test for shunt temp co. is to wrap it up in some cloth for insulation, stick a thermocpule inside then log the thermocpuple as you run the test. That way you can produce a chart for the shunt resistance versus temperature.

This way you can extract the ppm/°C temp co out of it. These characteristics are often not quite flat, so it also gives you an idea of how much variation in temp co there is. In order to check that you got the right measurement you can repeat the procedure with different conditions and they should all follow the same curve, regardless of the test current.

The way i did this test is i used a big boatanchor 500A lab PSU to send a large test current trough the shunt, measuring it as it heats up (Using a Danfysik flux gate to accurately measure the current). Getting it up to about 100°C. Then i switch the cables over to a Rigol DP832 where a script over SCIPI toggles the PSU between 9A and 1A while a 8.5 digit DMM mesures the shunt voltage. This toggling lets me null stray thermal voltages. Since the test current is much smaller the shunt cools down and i am continously mesuring it while it does so. Towards the end i unwrap the insulation about it to let it cool faster towards room temp.

So this way i get a test at different currents and temperatures in one run. At the same time also testing out that the shunt can survive the full rated current.

Also do use cable shoes to connect to the large screw terminals. For low resistance shunts the resistance of the block on the end matters, so you will get different results depending on where exactly on the block the current enters. Similar for where the sense wires connect (the block doesn't have the same voltage everywhere on it).

[alm]

Can we draw any conclusions from this project?

Coming up with any absolute value beyond what you already know is impossible since you don't know for sure how much each resistor has drifted. But you could try to estimate the uncertainty of your relative measurements based on things you can measure, like the noise in the setup (based on standard deviation of the voltage values), temperature coefficient (see post above), accuracy of the DMMs you use to measure the voltage (probably negligible contribution) and repeatability of the comparison between the shunts. Try reversing the current to see the effect of this, and varying connections (see the Ohm-labs white paper someone posted early in the thread). The result of this should be that you can see the ratio between the resistors is 1.xxxx +/- 0.xxx as 95% confidence interval (k=2). That should tell you how accurately you know this ratio. A lot of metrology is about ratios, not absolute values. Absolute values are just ratios relative to some higher standard.

[PartialDischarge]

Finally received the Ultrastabs from ebay. They work great.
Used 6 x 10.5Ohm parallel resistors (YR1B10R5CC) as load, to get a total of 100mV/100A at the output. Used all coaxial cables to the sensor (+15, -15, output).

Offset was about 5uV ( 5mA), so could easily measure 100mA as 100uV after nulling this offset.

[tszaboo]

Isabellenhütte makes some very good shunt resistors, the RUG-Z
They can be loaded to 250W and they are very stable.
I have a few of their 1 mOhm versions and I am very happy with them.

At 100A and 1 mOhm you just have 10W power dissipation in the resistor.
The connection is done with M6 screws.

These shunts are great. How did you get them? I wanted to acquire one at home since I used them about a decade ago for calibrating things as the resistance reference.

[HighVoltage]

Isabellenhütte makes some very good shunt resistors, the RUG-Z
They can be loaded to 250W and they are very stable.
I have a few of their 1 mOhm versions and I am very happy with them.

At 100A and 1 mOhm you just have 10W power dissipation in the resistor.
The connection is done with M6 screws.

These shunts are great. How did you get them? I wanted to acquire one at home since I used them about a decade ago for calibrating things as the resistance reference.

They are usually very expensive:
https://www.distrelec.de/de/leistungswiderstand-1mohm-250w-isabellenhuette-rug-r001-tk1/p/16057580?queryFromSuggest=true&itemList=suggested_search
(EUR 1.251,88)

I was very lucky a few years ago, when they were offered at a huge discount at Distrelec

[enut11]

A lot of discussion above is about the tempco effects of different shunts, being related to either size or construction material.
In my tests, even at 20A, the shunt elements rarely changed by more than a couple of degrees C. Dissipation power= 20x20x0.001=0.4W
Seems to me there are other effects at present unknown.
enut11

[PartialDischarge]

I have a gwinstek mili ohm meter and in my experience the real problem with low valued shunts is where you take the measurement, errors become bigger wrt to the nominal value of the shunt.

[Berni]

That is because you are using a very low resistance shunt at low currents. So you also get a tiny signal out of the shunt that is more difficult to accurately measure (especially due to thermocpuple voltages)

Once you go >300A the pesky square in P= R * I^2 starts to become a problem since even a very low resistance shunt will be dissipating 10 to 100W of power, so things will certainly get rather hot even if it is a huge shunt.

Well made shunts can have impressively low temp co. but i have also seen some cheep chinese shunts that had horrifically bad temp co. like 1000ppm/°C or something. The alloy mix for the resistive material has to be carefully tuned just right to get that low temp co. This is why good shunts can be quite expensive.

[tszaboo]

A lot of discussion above is about the tempco effects of different shunts, being related to either size or construction material.
In my tests, even at 20A, the shunt elements rarely changed by more than a couple of degrees C. Dissipation power= 20x20x0.001=0.4W
Seems to me there are other effects at present unknown.
enut11

A few things. Dissipation goes up with square of the current. So 100A is 25 times as much dissipation. And nonlinearity comes with this.
Often times for four wire shunts, the actual value of the high current path is a lot higher than the nominal value. You have the nominal value between the 4 leads and some extra for the leads leading there.
When measuring 20A*0.001mOhm = 20mV, thermocouple effects will effect matter. I had a design where the leads were connected to different sized PCB planes, and the shunt was measuring current differently in either direction.
The better ones also have a voltage coefficient defined, that I'm not even sure what is the mechanism for that.

[Kleinstein]

Ther thermal EMF of the shunts can be quite an issue, if both sides of the shunt see different heat loads. A fuse in the path can be an issue here as it may produce significant heat on one side.
Getting low thermal EMF is a quite important parameter for the shunt at high precision, high power.

Many shunts are made from manganin and this material usually has a relatively large 2nd order TC that is not easy to trim / adjust by heat treatment. So even of low TC at room temperature the TC can get higher when going really hot. So the problem can be getting worse even faster than I²

[mzzj]

Some further reading: https://tsapps.nist.gov/publication/get_pdf.cfm?pub_id=32286

[Smokey]

Another thread here I just stumbled upon talking about calibrating a 100A shunt.
https://www.eevblog.com/forum/metrology/low-ohm-precision-resistor-standard-and-testing/