Using Models to Predict Capacitor Leakage Current

[joeqsmith]

Of the eight caps that were damaged, six exhibit very high leakages compared with the control.  At 20V, we said 1nA was the max and these certainly fail.   #2 (grn) and #4 (yel) are interesting. 

[joeqsmith]

I decided to retest #4 to see if would behave the same and the leakage profile changed dramatically.

[joeqsmith]

Of course, I decided to see how it would continue to change so I ran a total of 13 sweeps on it.

[joeqsmith]

Zooming into the vertical axis we can get a better idea what this trend is looking like.   And while it has greatly improved, keep in mind that the vertical axis is in uA.  The fail point is 1nA or 0.001 and even after 13 sweeps the part has failed every time. 

[joeqsmith]

If you are curious about if the 10 seconds is not long enough to settle. 

[oldway]

Of the eight caps that were damaged, six exhibit very high leakages compared with the control.  At 20V, we said 1nA was the max and these certainly fail.   #2 (grn) and #4 (yel) are interesting.

If only 6 of the 8 damaged capacitors have a leakage current much higher than the normal leakage current, it means that only 6 out of 8 faulty capacitors can be detected by an LV test

It's not acceptable.

As Dave said:

The problem with claiming that you "don't need to test at X" is that it only takes a single case to prove your theory wrong.
And given the countless multitude of chemical, material, and manufacturing science at play inside a capacitors, it's not a claim I'd be willing to make even if I had evidence for it.

Reply 213 page 9
https://www.eevblog.com/forum/testgear/mr-carlsons-lv-capacitor-leakage-tester!-it's-finally-out!/200/

It remains to be verified whether the test at rated voltage makes it possible identify 100% of the defective capacitors or not....

That is why I asked to make the measurement of the leakage current on these 8 capacitors at the rated voltage (here 1000V) and to compare them with the leakage current of the capacitors which were not damaged.

[joeqsmith]

The attached graph shows capacitor #4 with a one minute selling time (red) and a five minute settling time (green) when compared with the last sweep. 

[joeqsmith]

We are interested in if any parts fall below the manufacture's 1nA limit (again assuming linear which it is not).   In the original plot showing the 10 samples, we can see something is going on in the sub 5nA range. 

[bugi]

@Oldway

I think it is a bit premature to say "not acceptable" already, when the testing has just begun, and plenty of things remains to be tested and learned. Also, he has not yet revealed how they have been damaged; perhaps those two are not damaged in a way that would also cause problems at high voltage. (I do assume they have issues, but the information revealed so far does not reveal it, thus any conclusion for "not acceptable" is premature.)

Also, again, the interest of this topic is not, afaik, to find out if/how low voltage leakage allows detecting all possible higher voltage failures, but about studying the leakage model in general (and could include high voltage and/or breakdown, too, but not a necessity). For the leakage model point of view, those already shown results are quite awesome.

That said, I am also interested in any higher voltage results, but I have patience... I understood those results were just the "First attempt at automating the testing", the latter ones some quick additions for exceptionally interesting results.

[joeqsmith]

Deselecting all of the data sets except the five of interest, we can clearly see caps #7 & #8 deviate from the two control caps.

[joeqsmith]

As we continue to push into the noise, we have three caps left that fall under 1nA.   

All the capacitors were stressed using various high voltage transients, DC, various current limiting resistors ranging from 250K to 60M. 

Cap #6 is a little special.  In this case I had taken my old piezo ignitor that has damaged so many UNI-T meters and provided 50 pulses to the part. 

[oldway]

One does not need to know or find out the breakdown to model a limited range, especially if the goal is not to decide whether the cap is good or not. Note the OP's / topic's limited goal to just find more measurement results on the non-linear leakage behavior vs. some ideas that leakage would be caused by a simple (ideal) resistor with linear response to voltage. (And there is no demand on the voltage range of the measurements, though the larger the better).

Ignoring the breakdown will just result in a model that will not be as useful above the measured range (extrapolation of a model is always a risky thing). Even if the breakdown (or anything else for that matter) would start to affect leakage well below its voltage (or at any range), a simple model vs. measurements might reveal how the model starts to drift at higher voltages (or another range), leading naturally to limit models usable range. Just like every model should list the limitations. (E.g. "Note that this simple model is only valid upto 85% of breakdown voltage, whatever that voltage happens to be.")

Of course, it would make the model better to also study the behavior near and up to breakdown, so as to know how it would affect the model, and especially, how far below the breakdown it has an effect on the leakage current. This is a sort of win-win possibility: if the breakdown affects leakage only near the breakdown voltage, then the bad side is that it can not be predicted beforehand, but the good side is that the simpler model is valid for larger range.  On the other hand, if the breakdown affects leakage further below the breakdown voltage, the bad side is the simpler model has lesser useful range, but it may also become possible to predict the breakdown beforehand.  It would then depend on how the breakdown affects the leakage, if or how well it could be predicted (with a better model that includes breakdown effects).

EDIT: Personally I'd be happy just for a model that handles the non-linearity below breakdown effects better than the typical "parallel ideal resistor".

A tester must give reliable and reputable results ...

Otherwise, no need to waste time testing the capacitor, just check the circuit correctly, it is sufficient and more efficient than an unreliable test.

The tester must give 100% certainty or at least one result equal to the old high voltage testers ...

If we are not able in 2018, with all the technical means at our disposal, to do better than we did in 1950, we can eat our diploma and change our profession.

That is why I asked to do the tests both at low voltage and at the rated voltage on the damaged (they have been stressed, but are they really damaged ? For exemple, a X2 capacitor is not damaged by surges) capacitors and to see if the results are concordant or not.

Whether or not you can use a low voltage test to detect a faulty capacitor is well a Joeqsmith's goal in this topic, he wrote:
"I can't tell the the good from the bad with my RLC meter.   Time to see if my old equipment can detect which ones are damaged using only a 24V DC source."

Can a faulty capacitor be detected using just a 24V power supply?

If you want to use models to predict a capacitor leakage current:
- It is necessary to know up to how much voltage this models can be used , it is an essential information .... Clearly, it is essential to know the breakdown voltage of the capacitor.
In a faulty capacitor, there is no guarantee that the capacitor will be able to block the rated voltage.
- the different types of capacitors certainly will not follow the same model, and there are hundreds of different types of dielectrics.
If it is necessary to wait for the models of all types of capacitors, it will be too early for ever to draw conclusions.

[joeqsmith]

I am in the process of setting up to collect some data at higher voltages.   First I wanted to run a basic stability test.  Using a 10Meg MOX, power supply was set to 1KV and I collected data for 5 minutes.   The scale is mA now rather than uA.   I ended up running a second sweep just to make sure the supply would not continue the downward trend.

[GEOelectronics]

The visial recollection of "capacitors" to one group of technicians and to other groups of technicians can be vastly different.

Lest anyone doubt that capacitors, ICs and transistors are not used that much with 1000+ High Voltage, I attach this picture of a precision 2400V DC generator (HVG) that uses state-of-the art regulators and multipliers and controls, all on one PCB. 6V goes in, 2400V DC regulated comes out (or any other up to 2500V) in a variation of KFC multiplier.
Geo>K0FF

[joeqsmith]

Lest anyone doubt that capacitors, ICs and transistors are not used that much with 1000+ High Voltage, I attach this picture of a precision 2400V DC generator (HVG) that uses state-of-the art regulators and multipliers and controls, all on one PCB. 6V goes in, 2400V DC regulated comes out (or any other up to 2500V) in a variation of KFC multiplier.
Geo>K0FF

I don't think anyone would doubt that we are using high voltages in consumer products, or in general.  Did you build this supply for general hobby use or was it for a product? 

[oldway]

I just should like to remember that what was used in the old HV capacitors testers to detect a bad capacitor was not measuring the leakage current but well measuring the value of the breakdown voltage.

My CR analyser Model TE-46 has two ranges of leakage current , 1mA full scale and 10ma full scale.....This has nothing to do with the leakage current of nA's you are measuring even with known bad capacitors.

Trying to models capacitors is useless for detecting bad capacitors.

From the breakdown voltage, model changes and leakage current increase rapidly and reach mA range.

Such high leakage currents only occurs when the applied voltage is higher than the breakdown voltage.

If the leakage current increase rapidly at a voltage lower than the rated voltage, that means that the condensator is bad.

This is how these old testers worked.

 

[joeqsmith]

Looking again at the low voltage data, CAPS1_10, we can see the highest currents are found with #3, #2, #5, #4 and #1.   

1KV_CAPS1_10_10MEG shows the current for all 10 capacitors being swept from 250 volts up to 1KV.   The white flat line was our 10Meg MOX resistor with 1KV applied.   Note how capacitors #7, #1 and #3 exhibit a very sharp edge up to the current limit.

[joeqsmith]

Deselecting caps #7, #1 & #3 and zooming in on the remaining parts we can see #5, #2, #8 & #4 all deviate from the control. 
#4 continues to behave interesting compared with the others.    If we wanted to detect this fault at high voltages, we better have a way to detect these events.    #8 is a bit odd as well as the leakage drops with the higher voltages. 

[joeqsmith]

Looking closer into the noise floor we can see #4 actually has the lowest leakage at 1KV.  It would be difficult to tell at 1KV if #4, #8 ard #6 are any different than the control caps #9 & #10. 

[GEOelectronics]

Hobby.

Powers a photmultiplier tube form a 4 x D cells.

Geo>K0FF

[joeqsmith]

Comparing cap #6 (tan) which again was exposed to 50 strikes from the piezo grill ignitor,  it appears to have slightly better performance than our control parts.   

[joeqsmith]

Hobby.

Powers a photmultiplier tube form a 4 x D cells.

Geo>K0FF

What are you trying to detect with the PMT?  What wavelength?  Do you have a blog showing the entire project?

[bugi]

Soo, combining info from multiple images.

Low voltage:

• #1 to #5 were noticed to have very clearly too high leakage within 20V. #4 playing really weird stuff..
• #6 was hanging happily around with the controls (#9 and #10).
• #7 and #8 were noticed after zooming in, #8 just barely above the limit at 20V.

High voltage sweeps then seems to show that..

• #1 to #5 are still the highest leakers.
• #1, #3 and #7 apparently have a breakdown below the 1kV.
• #2, #4, #5, #8 all deviate from the control (in the zoomed view).
• #4 and #8 are especially interesting, having lower leakage (at high voltage or over measurements/time) after first having higher leakage.
• #6 still shows no signs of problems. (Is it really damaged after all?)
• #5 seems to have some up and down behavior, too.

For me, the annoying result is the changes over time/measurements (i.e. the kinds of changes where the leakage gets less). Apparently, measure long enough and bad leakage can vanish

EDIT: #8 started barely above the expected limit, had clearly more than control ones at higher voltages at first, but ended up with low leakage (and no breakdown).
EDIT2: the control caps' (and couple others) seem to have quite constant leakage over the 250V-1kV range; I'd have expected an increase (at least linear if not even growing). The #4 of course being the joker and getting even lower (is it about to break open circuit?)

[joeqsmith]

Soo, combining info from multiple images.

Wait, what?  You didn't expect me to make sense of this did you??     I still have at least one test I would like to run.   

Low voltage:

• #1 to #5 were noticed to have very clearly too high leakage within 20V. #4 playing really weird stuff..
• #6 was hanging happily around with the controls (#9 and #10).
• #7 and #8 were noticed after zooming in, #8 just barely above the limit at 20V.

Yes, this seems correct.   There is a reason I ran these low leakage ceramics.   I assume detecting a fault with them would prove more difficult.    Certainly with a good electrometer and test setup (which I don't have), every damaged device I tested could be detected at low voltages except for #6.   I am not suggesting that you could use a simple threshold.  Keep in mind, this was a very small and crude test of one type of capacitor all from one lot.     

High voltage sweeps then seems to show that..

• #1 to #5 are still the highest leakers.
• #1, #3 and #7 apparently have a breakdown below the 1kV.
• #2, #4, #5, #8 all deviate from the control (in the zoomed view).
• #4 and #8 are especially interesting, having lower leakage (at high voltage or over measurements/time) after first having higher leakage.
• #6 still shows no signs of problems. (Is it really damaged after all?)
• #5 seems to have some up and down behavior, too.

Lets ignore everything except #4, #6 and 8. 

#6 is a bit of a coin toss you may say.  If I gave you that cap and you had no idea what it was, would you use it?  If I told you I hit it with the grill starter 50 times, then would you?     We know that grill starter's energy is very low compared with the IEC standard which is why I designed that gun I now use. 

#8 and 4,  I am not too surprised after seeing a few other parts appear to improve as well. 

For me, the annoying result is the changes over time/measurements (i.e. the kinds of changes where the leakage gets less). Apparently, measure long enough and bad leakage can vanish

EDIT: #8 started barely above the expected limit, had clearly more than control ones at higher voltages at first, but ended up with low leakage (and no breakdown).

It would be next to impossible to come up with a simple model (or even a complex one) based on these parts.  They are not stable and would not make a good reference standard to evaluate other detectors.     

[bugi]

Soo, combining info from multiple images.

Wait, what?  You didn't expect me to make sense of this did you??     I still have at least one test I would like to run.   

It was more about me starting to lose track on which cap was doing what. Too many graphs etc.  Combined it all into one message so I don't have to scroll back and forth the whole page.