How to best test Electrolytic capacitors for premature failure reduced life?

[userx2]

Hello all,

We have a problem where switch mode power supply output capacitors have started failing prematurely.
It is a random capacitor batch issue that has arisen fairly recently.

This is a 12V power supply module we use from a manufacturer. It is solder mounted onto our PCB.
The capacitors are inside the module 1000uF/16V and there are 2 in parallel.

These output smoothing caps degrade in that their value moves from 1000uF to about zero in less than one year in the field. The power supply then fails by the voltage dropping as soon a the load is switched on.

I am currently at a loss of what we can do about this except for trying to identify a bad batch of supplies before production.

I am currently thinking of cycling the output load (a 12V pump drawing 530mA) ) on and off but that may not be the best way to test this.

I know these issues with the electrolyte quality have been around for a long time .
Has anyone had any experience with bad quality electrolytic capacitors and how to best test for this?


Best regards
X

 

[24vtingle]

Im always told that Rubycon are the best...but thats just companys where ive worked insisting on Rubycon.

 

[userx2]

Just for interest, I now have a datasheet for this capacitor.

What is special here is the size 8x16 as the caps are normally 10x16 or 8x20. I also note, it is a 3000h 105dC capacitor, which is not bad.

Interestingly, the cap manufacturer has come back to the PSU manufacturer stating they make 30k capacitors in one batch and there is no batch code on the caps, only on the packaging. So there are a heap more of these caps out there...

 

[scopeprobe]

Just for interest, I now have a datasheet for this capacitor.

View attachment 199473
View attachment 199474

What is special here is the size 8x16 as the caps are normally 10x16. I also note, it is a 3000h 105dC capacitor, which is not bad.

Interestingly, the cap manufacturer has come back to the PSU manufacturer stating they make 30k capacitors in one batch and there is no bach code on the caps, only on the packaging. So there are a heap more of these caps out there...

Pretty Large ESR on those. Not great for the output of a switcher. You've not posted the schematic of details of the switcher itself but you would likely need some decent ceramics around those electrolytics if operating above 100khz to limit their disipation during operation and prolong their life. Just as a comparison, a quick search of mouser i found 1000uF 16V electrolytic capacitor with 10x16mm package, quoting impedance of 28mR 1.8A Ripple @ 100Khz with 8-10k hours life which would perform significantly better given the same conditions of use

 

[userx2]

Pretty Large ESR on those. Not great for the output of a switcher. You've not posted the schematic of details of the switcher itself but you would likely need some decent ceramics around those electrolytics if operating above 100khz to limit their disipation during operation and prolong their life. Just as a comparison, a quick search of mouser i found 1000uF 16V electrolytic capacitor with 10x16mm package, quoting impedance of 28mR 1.8A Ripple @ 100Khz with 8-10k hours life which would perform significantly better given the same conditions of use

I suspect the smaller diameter of 8mm puts a limit on the ESR that can be achieved. They have said that they are using these special caps because of space/form factor constraints. 10mm diameter cannot fit apparently.

--- Updated May 7, 2025 ---

I also have a snippet of the schematic I can show. It looks like there is another cap, likely ceramic in parallel with the 2 problematic ones.

 

[24vtingle]

If you can put a sense res going off to the pump.....then measure the RMS current going to the pump....then the AC rms ripple current from that, that will appear in the cap bank, is SQRT(iRMS^2 - i(av)^2)
Where i(AC) is just the average current going to the pump.

Obviously the ripple from the switcher will affect the true ripple in the caps, but to get the above initial calc done would be good.

Is it possible you can tell us the details of the switcher so we can calc the ripple from that that will go into the cap bank?

Also do you know the rough ambeint temp around the DCDC module when its in the equipment?

Its deffo a point of interest that a 6W load situation has seemingly over-ripple currented 2mF worth of caps, which the datasheet says each have >1.2A of ripple current rating.

 

[userx2]

I do not have any further details of the schematic nor of what parts are used in the supply. They only shared this little part.
What is also noteworthy is that the 5V rail gets generated from the 12, which adds to the load on the 12V. and ripple. The 5V rail is rated at 1.2A, another 6W in addition the the 12V 1A rating. So the whole power supply is 18Watt.


EDIT: The ambient temperature can be up to 50dC but will mostly sit at 35ish.

EDIT2: The pump cycles about 20 -30 times a day to on for less than 15 seconds and more likely about 7 seconds.

 

[24vtingle]

OK thanks, we can of course only guess as we dont know details of Buck (assumed), Flyback (assumed) and type of Pump Motor and its Drive (if any).
But if i may assume the 12V is generated by an offline flyback(?)
..Could you put a probe (bearing in mind isolation) across the flyback secondary at all?....then we can see the conduction time of the flyback diode, and see if it looks right...if too short then RMS current will be unfortunately higher....offline flyback diode secondary conduction time should really at the very least be half of the period.

Is it one of those Innoswitch flybacks.?....choosing the wrong synch FET with them could cause mayhem with the downstream cap bank.

Of course i could be missing the obvious here....the assemblers may only have fitted one of the two el caps in each board.(?)

It sounds like your co get these built overseas and they go straight from the overseas assemblers to your customers, with no-one in your co actually handling them?

 

[userx2]

OK thanks, we can of course only guess as we dont know details of Buck (assumed), Flyback (assumed) and type of Pump Motor and its Drive (if any).
But if i may assume the 12V is generated by an offline flyback(?)
..Could you put a probe (bearing in mind isolation) across the flyback secondary at all?....then we can see the conduction time of the flyback diode, and see if it looks right...if too short then RMS current will be unfortunately higher....offline flyback diode secondary conduction time should really at the very least be half of the period.

Is it one of those Innoswitch flybacks.?....choosing the wrong synch FET with them could cause mayhem with the downstream cap bank.

Of course i could be missing the obvious here....the assemblers may only have fitted one of the two el caps in each board.(?)

It sounds like your co get these built overseas and they go straight from the overseas assemblers to your customers, with no-one in your co actually handling them?

The PSU maker is a fairly well known brand and one would expect that they at least know how to design their power supplies. I can't mention too much here as I risk getting into trouble.
The power supply gets supplied directly to one of our contract manufacturers which then assembles it with the rest of the components onto one of our pc boards. The boards then get shipped to our production factory and assembled with other components boards pumps etc into the units we produce.

 

[userx2]

I have since constructed / set up a test box with several PCBs cycling pumps on and off. We calculated that in the field these would cycle about 6000 times per year. The cycle count has already passed 10500 by now and nothing has failed. The cycle is set to 4s on and 10s off.
I have also added a heater in the form of large resistors on a metal plate, driven by a temperature controller, into the box to run the temperature at ~40dC ambient. So far so good.
If this problem is time related because these capacitors are not sealed properly and the electrolyte evaporates/dries slowly over time, then this test may not replicate the failures we get.

 

[crutschow]

The cycle is set to 4s on and 10s off.

I think the on time should be much greater than the off time.
Is it seeing the maximum output load?

 

[userx2]

I think the on time should be much greater than the off time.
Is it seeing the maximum output load?

Well, yes and no. I decided to test it like this first because:

The pump has a high inrush (not startup) current when left to discharge for a while.
In fact, I measured up to 16A for 55us (short spike) = some low ESR capacitor charging inside the pump. I measure this with a current clamp probe for the oscilloscope.
I wanted to test if that current spike contributes to this problem.

If the pump remains off for a short time, then that current drops significantly. Hence 10s off.
In the field, this pump will run up to 16 seconds.

But yes, I am aware that the overall accumulative loading of the PSU could be an issue and I will repeat this test with a 20s on time after this one has completed. I think will run this for 150k cycles.

To answer your question:
Yes, that pump is the maximum output load for the PSU on 12V, plus the 5V electronics, which are connected and operating as well.

Best regards
X

 

[D.A.(Tony)Stewart]

I was able to supply my design to Avaya (nee Lucent) in 8 wks from P.O. to 1st prototype which included 1 of 2 qualified OEM PSUs which took me 2 months of reliability and dynamic testing on 10 units, start/stop MTBF cycle etc, installed in a 1U 19 " Rack including sacrificed units for the infamous UL Coke Spill test and sledge hammer test to ensure, no flames. Since I had a lot of prior DVT experience , it wasn't too painful or costly. But Avaya had an inhouse UL guy to do the global documentation from my submissions.

 

[24vtingle]

Hi
The attached shows the worst case ripple current in the 12v rail caps from the buck and the flyback acting together.
It assumes the buck and flyback were designed in a "standard" way.
(ie, BCM flyback with D' = 0.8 and CCM buck with approx 20% (or so ) of i(ripple) in inductor.
As such, The worst case I(rms) ac ripple in the caps would be 1.09A

..Thats worst case as in reality some of the AC from the flyback may go into the buck and not into the 12V El caps....

However This does not consider ripple or surges from the pump.
So things may be worse than this. Though i doubt much worse as you say that the pump has a pretty big input cap bank of its own.(?)

Maybe the pump has a front end BLDC driver which puts loads of ripple in the caps we don't know.

Lots of other possibles of course.
-eg flyback regulation components wrong and flyback putting 20V onto the 16V caps, etc etc.
Or putting say just 6V on the caps and then the pump drawing lots more current ripple from them etc etc

 

[userx2]

The PSU manufacturer has now suggested changing these caps form 2x1000V/16V to 2x680uF/25V Solid state conductive polymer capacitors.
Maybe they have realised a problem in their design.

The smaller value must be because because of size limitations.

This is form their failure analysis:

 

[24vtingle]

The PSU manufacturer has now suggested changing these caps form 2x1000V/16V to 2x680uF/25V Solid state conductive polymer capacitors.

They are going to custom change your particular PSU units, or all of the range? Sounds costly if you get it done custom.
As you say you have units in the field since 2014 that havent failed, its just this batch from 2024 or so.

So initially we thought it was a duff batch of el caps...but now they are suggesting its poor design.....though for 8 years+ you had no failures with this design.
If those caps werent duff and they failed within the first few months then it makes you wonder if even solid polymer caps will also fail as the temperature must have been well hot. Has the customers use case changed at all?..ie higher ambient temperature now.

So, also, you are (potentially) having capacitors changed out due to the fear that in future bad electrolytics may be fitted(?).....essentially what we are saying here is that nobody should buy any PSU's that have electrolytic capacitors in them. You say that your PSU was made by a reputable manufacturer, and yet still this happened, in this post we are essentially condemning the entire offTheShelf PSU market...at least , the majority of those, which of course , use El caps.

Also, you are doing your tests at 40degc, but say that internal ambient may be 50degc in the field. You almost wonder if its worth putting a small micro and temp sensor in some of the products so that if they fail, then you can get them back and check what was the max temperature that they experienced.
On maybe just one in 100 PCBs, fit the temperature data logger...just say a little PCB that gets connected in or soldered in if you dont want the expense of a connector on all the PCBs.
Is it a plant watering system (the pump coming on regularly)?....sitting in a really hot greenhouse?

Because on a ripple current basis, those 2x 1000uF capacitors should have been fine.

Of course, the upstream flyback may have failed and was slamming huge surge currents repeatedly into the rail, rather than operating normally.
And maybe it failed because its input electrolytic capacitors failed.

Also, you say the pump turns on some 20 times a day for only about 15seconds each time.
You say there is a surge into the pumps cap bank when it gets turned on....16A you measured.
We should remember that the wet electrolytic cap is the master of surge handling.
Solid polymer lytics aren't so good at surges.

Ripple on the 12V rail shouldnt be a problem, it may be better to just go for a polypropylene film output cap and adjust the flyback so its loop could handle the reduced capacitance.
But to be honest, things could start getting expensive, and i think we all would like to be at the customer installation with a unit rigged up with internal thermocouples to see just how hot those el cap cases are getting.

 

[userx2]

Hello and thanks for the detail @vtingle.

Answering a few points.
Yes, they are suggesting the cap change, not us.
Yes, they are a well know manufacturer
I am not sure if they are doing this just for our specific power supply or the whole range.
I am not sure if they do or do not know exactly what is happening with the caps. The cap manufacturer is still analysing the failed caps for the PSU manufacturer

No, our customer use case has not changed since the introduction of this product.
Yes we do know what the internal temperature of the unit can rise to and we have collected plenty data in environmental chamber and field testing. Yes, with datalogger and thermocouples.
The test I am doing here at the moment is a reasonable approximation. I may have to tweak the cycle ON time a bit but right now, let it cycle.

The power supply manufacturer has confirmed that in all cases so far, changing the capacitors has fixed the power supply to pass all their load and production testing again.

These type of problems are a real pain in the butt.

Best regards
X

 

[24vtingle]

The power supply manufacturer has confirmed that in all cases so far, changing the capacitors has fixed the power supply to pass all their load and production testing again.

Thanks, and as you know, from ~2014 to 2024, these units showed none of this kind of failure for you, so are the cap manufacturers admitting a bad batch?...though as you said they are not batch marked so no-one would know which ones they are anyway.

So presumably the PSU manufacturer is changing from electrolytic to solid polymer because they think there are less "bad batch" occurrences with solid polymer?

Did you actually get a failed board back, then change out the caps to fresh 1000uF,16V 'lytics and do testing on that board?

The power supply manufacturer has confirmed that in all cases so far, changing the capacitors has fixed the power supply to pass all their load and production testing again.

As you know, Maybe The PSU manuf may have gaffed something else up, which in turn made the 'lytics fail...("bad batch victim playing") then they blame the 'lytics because it hides their other , more serious gaff

 

[userx2]

Comparison of the caps:

The original capacitor
1000uF 16V 1.28A ripple and 1.2Ohm ESR 3000h

The proposed polymer cap they propose "
680uF 25V 4.5A ripple and 0.16 Ohm ESR 5000h

From the sepcs, the polymer cap seems much better but of course, the value is lower.

 

[24vtingle]

Yes though sol pol's have less surge withstand than wet lytics. (eg your regular 16 Amp turn on surge to the pumps input caps)

 

[userx2]

The PSU maker has just confirmed that the original design was fine but they cannot guarantee that the EL caps they fit are always reliable.
They have since approached the official distributors or genuine Rubycon as well as 2 another reputable brands and none of them can supply 1000uF in 8x16.

The Polymer caps are so far the only ones that can fit.

They have also not reported that on one failed unit, the 5V/10V smoothing capacitor 470uF also went faulty.

Regards
X