Tempco tests on power resistor current shunts

[enut11]

I have been working on an LCD monitor for my HP 6177C DC Current Source. It has 5 ranges from 20uA to 200mA full scale. It is based on a simple current-shunt principle using a 4.5 digit 200mV LCD readout. Required accuracy of the monitor is 0.01%. Low tempco resistors are needed to maintain cal over a reasonable temperature range.

The shunts are 1R, 10R, 100R, 1K and 10K. I happen to have low tempco 100R, 1K and 10K resistors so the challenge was to find  suitable 1R and 10R shunts.

[enut11]

I could simply buy some low tempco resistors but I thought it might be fun to study the behavior of 'garden variety' power resistors, like the one shown below, that are available at our local electronics parts store.

[enut11]

I built a mini oven based on the temperature controller I use in my LM399 Voltage Ref ovens. A 10K thermistor in a bridge configuration controlling a heater via an opamp/power transistor. This one I made variable from 20C to 35C.

[enut11]

First test was a 1ohm 5W 'white cement' power resistor and, I must admit, it produced a very peculiar result.

A HP 34401A DMM in 4-wire ohms mode was used to record resistance on the 100ohm scale. Logging program was from Forum member Ian Johnson.

The test was over 60 minutes. The first 5 min the oven was off with ambient temperature 25C. Over the next 55min the resistance initially dropped for about 6 min and then more or less stabilised at an intermediate value by the time the temperature reached 35C.

How do you explain this behavior? How do you work out the tempco for the 10 degree test? Is it max to min or max to final value?

Comments anyone?

[Andreas]

Comments anyone?

Had similar experience with unstable resistance value on a power resistor.
Maybe the contact between resistance element and terminals is only crimped and not welded.

with best regards

Andreas

[macaba]

I recently measured TCR on a batch of 2PPM 1206 resistors and encountered a similar effect to what you are seeing (overshoot then decay). Thermal EMF is my best guess, even with OHMF/OCOMP ON/DELAY 1 (on my 3458A). I found that a 30 minute settling time at each temperature setpoint was enough for the decay curve to settle to a flat line, you may find it requires 1-3 hours (I'm guessing) due to the larger thermal mass.

[jonpaul]

no need for this test, use a precision 4 terminal shunt resistor, available to 0.001 or 0.01 ohm 1% Isobellehutte in Germany or Tepco. At currents over 10 A use a meter shunt.

Jon

[mendip_discovery]

no need for this test, use a precision 4 terminal shunt resistor, available to 0.001 or 0.01 ohm 1% Isobellehutte in Germany or Tepco. At currents over 10 A use a meter shunt.

According to RUG-Z PDF which is the calibration standard one they manufacture. With 0.1% and a temp/co of 1uΩ/Ω per °C, it's worth a sniff.

[enut11]

no need for this test, use a precision 4 terminal shunt resistor, available to 0.001 or 0.01 ohm 1% Isobellehutte in Germany or Tepco. At currents over 10 A use a meter shunt.

Jon

Thanks @jonpaul. I am on a (steep) learning curve and resisting the temptation to buy my way out 

[enut11]

Noted: A longer time-frame has been suggested by members.

I recorded the resistance of the 1ohm/5W during a 120 min cooling cycle. More predictable trend but still large excursions in resistance. 
Calculated tempco was 153pp/C. is this typical?

This type of shunt is forever banned from my metrology experiments!

[enut11]

To explain a bit more, as I will be using a 200mV panel meter, the shunts will be:

1R for 200mA
10R for 20mA
100R for 2mA
1K for 200uA
10K for 20uA

Very low value shunts are not required as currents are low and burden voltage is not an issue for the current source.

[KE5FX]

If you don't want to go all the way to an expensive Kelvin resistor expressly designed for low tempco, the metal-finned Dale wirewound resistors make an excellent substitute.

[enut11]

I have an RS 10W finned resistor that I can test, along with a few 2R resistors that I can parallel if they are suitable.

Two of the 2R resistors are labeled: Micro-ohm and Pacific, both are 10W

[enut11]

With the mad holiday season over I have spent some time improving my tempco oven setup. I am using Ian Johnson's PDVS2 logging program, available on this Forum. The logger has provision for up to 2 meters as well as a temperature/humidity sensor. The temperature sensor I am using is the USB TnH SHT-10 V2.
https://www.eevblog.com/forum/metrology/3458a-logging-via-windows-app-revisited/

Below is a sample run on a Fluke 10.05K resistor logged from my HP 34401A. The chart shows a heating and cooling cycle. I am concentrating on the cooling cycle from approx record #400 over 2 hours.
10,050.449R at 36.7C
10,050.665R at 28.8C
Delta R=0.216 ohms
Delta temp = 7.9C

.216/10050x1000000=21.492 total ppm shift
21.492/7.9=2.72 ppm/C
Too good??

Feedback on my calcs would be appreciated
enut11

[berke]

Nice oven.

Do you feel like doing a test using say 100 ohms of copper wire?

I'm working on a driver for a small solenoid (~1.5W) and these buggers get quite hot quite fast.  The manufacturer suggested that the coil resistance can be used to measure its temperature.  Apparently copper has a tempco of roughly 0.4% per K.  I can't non-destructively put a thermocouple inside, but I found the resistance variation tracks the resistance predicted using that model reasonably well, using the case temperature.

I can check your tempcos if you provide CSVs.

[enut11]

Hi @berke. Sorry, but the oven is too small to test 100R of copper wire.

Below is the 10.05K test CSV. Next test will be a 2R 10W resistor.
BTW, I had to rename the file extension to TXT to upload it.

Columns are date/time, resistance, temperature and humidity.
thanks
enut11

[enut11]

Nice oven.

Do you feel like doing a test using say 100 ohms of copper wire?

I'm working on a driver for a small solenoid (~1.5W) and these buggers get quite hot quite fast.  The manufacturer suggested that the coil resistance can be used to measure its temperature.  Apparently copper has a tempco of roughly 0.4% per K.  I can't non-destructively put a thermocouple inside, but I found the resistance variation tracks the resistance predicted using that model reasonably well, using the case temperature.

I can check your tempcos if you provide CSVs.

@berke, I have two small relays, one has 70R coil and the other 170R. I could test the coil tempco of one of those if you want?
enut11

[jonpaul]

Worked at precision lab, Julie Research Laboratoire in 1966..1968.


Your plots look typicall of the setup, réf, meter warm-ups or contact R variations with warmup.

So, copper TC or R is well known,  as are all metals and common wire materials, just look up the TC.

Actual shunts and precision wirewound resistors use special zero TC wire or compensate TC with a PTC/NTC combo.

Enjoy

Jon

[berke]

Below is the 10.05K test CSV. Next test will be a 2R 10W resistor.
BTW, I had to rename the file extension to TXT to upload it.

Thanks, I find a temperature coefficient of minus 22.35e-3 ohm per degree when I throw out the first hour of data.  Otherwise you have two slopes (see bottom left figure.)

EDIT: Sign

[berke]

@berke, I have two small relays, one has 70R coil and the other 170R. I could test the coil tempco of one of those if you want?
enut11

Sure that would give an interesting comparison point, the 70R sounds comparable to what I have.

I can share the Octave script for the plots if you want.

Like jonpaul said the copper tempco is well-known but the practical information about the deviations is what is most interesting.  Also it's not as fun to just look up the value!

[enut11]

Unfortunately the temp/humidity sensor has become intermittent so no more tests until I sort that out

[enut11]

Thanks for the analysis @berke. Quite an impressive set of graphs. How did you generate these?

My simple approach to use delta max and min resistance and temperature yielded a tempco of 2.72 ppm/C. If I interpret your figure of -22.35e-3 correctly this translates to -2.235 ppm/C. If that is the case I was not too far off
enut11

[berke]

You're welcome.

Let's see, on the bottom right plot R goes from 10050 + 0.66 ohm to 10050 + 0.50 ohm from 28.6 to 36 degrees (the labels on the Y axis are messed up, insufficient precision), that gives (0.50-0.66)/(36-28.6)=-21e-3 ohm/degree, dividing by 10050 that gives -2.1 ppm/°C so yes we're in agreement.

I used GNU Octave, a program similar to Matlab except you don't have to fork out $4000 to get an fft() function and another $4000 to be able to load HDF5 files.  Script attached (paths need to be adjusted.)

No worries for the test, if/when you get back to it ping me.

PS. Looks like the forum doesn't like *.m files!  Zipped.

[enut11]

@berke, I have two small relays, one has 70R coil and the other 170R. I could test the coil tempco of one of those if you want?
enut11

Sure that would give an interesting comparison point, the 70R sounds comparable to what I have.

I can share the Octave script for the plots if you want.

Like jonpaul said the copper tempco is well-known but the practical information about the deviations is what is most interesting.  Also it's not as fun to just look up the value!

@berke, attached is a partial test of a 70R relay. The temperature sensor failed again during this test. I switched the oven on at record #150 and off at #550. Hopefully there is some useful data to analyse. During the test I did note a large change in resistance and it could be observed in real time! Note CSV changed to TXT.
enut11

[berke]

@enut11, this data doesn't seem to be good.  I can't find an interval with a decent one-to-one relation between temperature and resistance, see below.

[enut11]

The USB temperature sensor was not touching the relay so it is possible that, especially during the heating cycle, the resistance and temperature got out of sync. The temp sensor is rather bulky and there is not much room in the oven. Perhaps if you slide some of the data back and forward you will find a better correlation.

I am trying a different logger, one that can handle 2 DMMs, one for resistance and the other for temp. This time the temperature sensor will be a thermistor which is much more reliable.
enut11

[enut11]

@berke
Here is the 70R relay (TXT) file recorded with 2 DMMs, one for resistance and one temp. It is summer here and quite hot in my office and
 oven cooling takes a long time so I cut it short a bit.

I am still trying to get an improved setup for my current shunt tests. This latest setup uses a 34401A for 4W resistance and a 34461A for temperature via a 5K thermistor. Logger is the Agilent Digital Multimeter Connectivity Utility V1.0.2.0

The utility does not allow a timed sample rate so the best way to slow it down and taking too many samples is to select 100 NPLC. This also smooths out the data points.

[tszaboo]

For such a tempco test, I have the following suggestion: Have the starting point as the high/low temperature, and don't do temperature control. If you have a controller, it will do on/off control and upsets your measurement. Just isolate the setup, place a thermal mass in it if necessary, measure temperature at the beginning and switch of cooling/heating, and let it reach room temperature. The results will be much less noisy and easier to curve fit them.
And that being said, regular wirewound resistors will easily be ~200ppm/K, if accuracy matters you need high precision ones made exactly for this.

[enut11]

You are correct @tszaboo and it shows up on this run on a Microohm 2R 10W resistor. You can clearly see the oven influence up to about record #850 when the oven was turned off.

After that, the resistance change (0.00015 ohm) does not look right for the temperature drop (8.9C). I need to look closer at this.
I am using a 10mA CC from a HP6177C to boost the signal to the DMM. Measured directly, the 2R was only 2% of the lowest 100 ohm range on the meter (HP34401A).
enut11

[berke]

@berke
Here is the 70R relay (TXT) file recorded with 2 DMMs, one for resistance and one temp. It is summer here and quite hot in my office and oven cooling takes a long time so I cut it short a bit.

@enut11, you are right about the lag, if I take into account about 39 seconds of lag for the temperature (hand-optimized) I get a nice linear correlation.

But it comes to 3.79 ohm per degree (~11%) which way too high to just be the copper.  What's going on here?

[enut11]

With that data I attached the thermistor to the side of the relay so it should track better than the previous run.

May be the current source? I am using a HP6177C at 10mA. Here is the warmup plot fed directly into my 34461A DMM.
[EDIT] forget the above line the relay 70R was measured directly by the DMM 

[berke]

Did you put any thermal paste?  Otherwise it's just playing convection lottery.

[1audio]

How significant are thermocouple effects here? They could swamp the resistance changes if not accounted for. I believe some of the low resistance meters (HP for example) use AC excitation to avoid those effects.

[enut11]

I have no idea nor how to check it. I am using these Chinese Kelvin clips.

[enut11]

Did you put any thermal paste?  Otherwise it's just playing convection lottery.

No thermal paste. Too messy in an oven. I am going to rely on time as an equaliser.
enut11

[1audio]

both the resistors and the connections have a multitude of different metals and Seebeck potentials at every connection. Heating them and having the meter at room temperature ensures there will be voltages. They will be in the millivolts so not a big issue when the test voltage is 1V but very significant with a low Ohm resistor. Read more in the intro for the HP4328 here: http://ftb.ko4bb.com/getsimple/index.php?id=manuals&dir=HP_Agilent/HP_4328A

[enut11]

@berke
Here is the 70R relay (TXT) file recorded with 2 DMMs, one for resistance and one temp. It is summer here and quite hot in my office and oven cooling takes a long time so I cut it short a bit.

@enut11, you are right about the lag, if I take into account about 39 seconds of lag for the temperature (hand-optimized) I get a nice linear correlation.

But it comes to 3.79 ohm per degree (~11%) which way too high to just be the copper.  What's going on here?

Hi @berke. Pls check your 70R relay calcs again. I calculate about 1/10 of your results.
enut11

[Kleinstein]

There is some effect of thermal EMF. Some resistors are rather good in this respect and some are not. For example Konstantan has a thermal EMF of some 40 µV/K relative to copper, while good manganin is usually specified with < 1 µV/K. NiCr variants are usually also good.
Ideally there would be not temperature difference across the resistor, but with self heating of the resistor and asymmetric thermal design a temperature difference proportional to the power will develope. For DC operation this would result in a linear change of the apperant resistance with the voltage, that could become relevant for low value resistors.
As a small effect there is also the peltier effect at the contacts that causes an extra heat flow from one side to the other. With metals, especiall with low thermal EMF this effect should be relatively small though.


For the measurement of thermal effects I would not recommend using clips. The clips can mechanically move and this way change the resistance. For the thermal tests soldering of 2 wires to each end should be the better way. For the 4 wires it is than OK to use connectors (maybe avoid high thermal EMF or thermal gradients).

[enut11]

both the resistors and the connections have a multitude of different metals and Seebeck potentials at every connection. Heating them and having the meter at room temperature ensures there will be voltages. They will be in the millivolts so not a big issue when the test voltage is 1V but very significant with a low Ohm resistor. Read more in the intro for the HP4328 here: http://ftb.ko4bb.com/getsimple/index.php?id=manuals&dir=HP_Agilent/HP_4328A

Thanks @1audio. The HP4328A is an impressive instrument. Low level AC to measure resistance is an interesting approach as it avoids extraneous DC temperature effects. Pity a 4328A costs more than a 34401A second-hand.

Thanks @Kleinstein. Next I will solder the 4-wires to the DUT and see if I can measure the difference.

[berke]

Hi @berke. Pls check your 70R relay calcs again. I calculate about 1/10 of your results.
enut11

Indeed, what was going on is that I made a stupid mistake by swapping R and T while reading the file, and not noticing that the resistance is around 30 even though the file is named 70R.

That being fixed I get 0.371% per degree which is what to expect from copper.

[enut11]

@berke, good you found that. So, did the test tell you what you needed to know?
enut11

[berke]

@berke, good that you found that. So, did the test tell you what you needed to know?
enut11

Yes, thank you! It gives me a concrete data tempco data point and an idea of the uncertainty.  According to these guys https://www.engineeringtoolbox.com/resistivity-conductivity-d_418.html the official Cu R tempco is 0.429%.  This means that unless your relay is very atypical the practical value can be within 15% of that.  I don't think that kind of data is easy to come by.

[enut11]

Back to my current shunt tempco tests, here is a MicroOhm 2R 10W resistor.
Seems very good at 11 ppm/C.
Not sure what caused the reversal in resistance at around record #420. The oven heater was off for the whole test.

@berke, I would appreciate if you could verify this.
enut11

[Kleinstein]

The voltage measured for the 2 ohm resistor is quites small. The jump around reading #420  is only some 200 nV, so not really much. If not carefull this would be thermal EMF from the connectors with a heating effect (e.g. a person sitting next to it and than leaving the room). Also the DMM internal current source could show some LF noise. At least the 34401 uses an BJT based OP-amp for the current source, but there is still 1/f noise.

Another point is just a random jump in the actual resistance, e.g. from slip stick like movement between parts or a microscopic crack (e.g. in an oxide layer or paint) moving. Relative to the resistor the effect is not that small, but still possible to be a real effect of a not so stable resistor.
Overall it is not clear, if there is really a temperature effect, or more like random variations that

For the rather low resistors, especially when going below 1 ohm the test current (1 mA) from the DMM may not be enough to really get good results.

[enut11]

Thanks @Kleinstein. I am now using a soldered connection 4W resistance measurement.
Granted, the 2R resistor is only a small fraction of the lowest range on the 34401A and there appears to be a lot of noise in the plots.
I did try a 10mA constant current using a HP6177C but, as shown in Reply #30, the HP takes about 90min to warm up and adds another dimension to the results.
I am really just trying to find the best method available to me so I can study temperature dependence of low ohms current shunts.
enut11

[iMo]

[OT]: @Kleinstein: I wonder how the popcorn noise of the LM399 (in the above 34401A measurement setup) translates into a result. Imagine a 399 jump by, say, 4uV up or down during that measurement..

[Kleinstein]

The ohm measurement in the 34401 and most better DMMs is running ratiometric. So the same voltage reference is used to generate the current and in the ADC. So noise in the reference would ideally have no effect.  Because of different filtering / frequency response there could still be a short (e.g. 1 reading)  transisent effect on jumps from popcorn noise. As the noise is up and down this would be just a part of the noise, no longer time trend or such.

[bsdphk]

Power resistors encapsulated in ceramics suffer from significant mechanical stress as the temperature changes, because the ceramics and the resistance element have different thermal expansion coefficients.

[enut11]

As a sanity check on my methodology, I tested what I call a 'cement trough' power resistor available from the local electronics store.
The tempco of 144 for a cheap resistor looks OK to me although I do not have a specification.

[berke]

@berke, I would appreciate if you could verify this.

Hi,

The resistance varies too little to get a good estimate, but that is consistent with a very small tempco.  As it is the estimate is 16 µohm/°C or 8 ppm/°C.

[mendip_discovery]

no need for this test, use a precision 4 terminal shunt resistor, available to 0.001 or 0.01 ohm 1% Isobellehutte in Germany or Tepco. At currents over 10 A use a meter shunt.

Jon

So I thought I would look at the Isobellehutte ones and that is out of my price range. Wost is the min order.

Qty	Unit	Resistance	Price
5	RUG-Z-R001-0.1-TK1	0.001 Ω	£983.34 each
5	RUG-Z-R010-0.1-TK1	0.01 Ω	£748.84 each
5	RUG-Z-R100-0.1-TK1	0.1 Ω	£748.84 each
5	RUG-Z-1R00-0.1-TK1	1 Ω	£748.84 each
5	RUG-Z-10R0-0.1-TK1	10 Ω	£748.84 each

[enut11]

Wow @mendip_discovery, those prices are eye-watering.

Hi precision costs but in metrology low tempco costs even more!

[enut11]

This 10R test completes my set of suitable resistors for the precision 4.5D current monitor project.
If you have been at this game for a few years it is amazing what you can find in your spare parts box. 

BTW, I have not stopped testing my other resistors for tempco. I have found some surprises, both good and bad.
I used to think that all WW resistors were better than most but some had very high tempco (~300) and were probably wound with copper wire!
Testing is the only way.
enut11

[mzzj]

So I thought I would look at the Isobellehutte ones and that is out of my price range. Wost is the min order.

Wow @mendip_discovery, those prices are eye-watering.

Hi precision costs but in metrology low tempco costs even more!

And if I remember at all correctly the Isabellenhutte RUG is suprisingly unstable based on some measurements shown here on eevblog forums. Manufacturers 1 year spec is wide open but even low power room temperature use resulted in considerable drifts (order of 100ppm per year or something like that)

Still like comparing apples to rotten potatoes if you take  metrology lab primary standard like MI's 0.1 ohm 9210A in comparision: 0.2ppm per year and 0.05ppm/cel tempco.
I haven't priced one but I would be suprised if it was less than 5 to 10 grand. 
https://mintl.com/wp-content/uploads/9210A.pdf

[enut11]

Apart from having a good DMM, the most difficult part of resistor tempco testing might be the oven. There are many oven designs on the Forum.

I happen to have had a spare controller from my LM399 voltage standard tests but simple PID relay controllers are cheap and readily available on eBay. The eBay controllers do have a variable heating cycle but this is not important as you should only be testing during the cooling cycle when the controller is off.

You will need a heating plate. I used a piece of aluminium and glued a number of resistors to it totaling about 50 ohms. With a 12v plugpack the power was less than 3 watts. The low power allows time for the DUT to saturate at the max temp, usually just under 40C.

A thermometer is used to register the min and max temperatures or, as I did, use a second DMM to record temperature (not really needed). If you buy the controller below, it comes with a temperature sensor.

You will also need a proper insulated box in order for the DUT to stabilise prior to recording data. I use polystyrene foam from the local grocer and cut/glue to size the oven.
enut11

[enut11]

This was a very interesting exercise and for me an introduction to resistor temperature co-efficient testing.
I will continue looking for hidden gems in my spare parts box. There will be other projects waiting for them.

Learnings
There are many traps and careful technique is needed, especially when you are chasing parts per million:
Eg, using clips to connect to low value resistors can give variable results. Soldered connections are more reliable.
or, using data from the heating cycle which includes oven effects.
or, testing over too short a timeframe. Long cooling down tests are better.
or, using an unstable current source. Characterise your equipment first.

The list goes on...

Thanks to all the Forum members who contributed to the thread.
enut11