Author Topic: Overloading Film capacitors with ripple current, how well they tolerate it?  (Read 1679 times)

0 Members and 1 Guest are viewing this topic.

Offline MiyukiTopic starter

  • Frequent Contributor
  • **
  • Posts: 908
  • Country: cz
    • Me on youtube
Hi,
I have a question, do you have experience how long and how film capacitors fail under high ripple current ?

I am building for my own use a big adjustable load up to 3kW
It is designed as 3 phase interleaved inverting buck-boost converter at 50kHz, with bank of light bulbs at output, with 24V or 72V settings

It should be able to work at 10-15 minute intervals with total working time in lower tens of hours
It have not to last for ages and should be cheap

Now to capacitor question.
On both sides are film capacitor banks which sees huge ripple current
Is here only problem of heating and resulted overheat or some other failure mechanism ?
 

Online moffy

  • Super Contributor
  • ***
  • Posts: 2217
  • Country: au
The losses are typically around 1% of load current (tan(d) metal polyester film) so it depends what their rated power disipation is.
 

Offline T3sl4co1l

  • Super Contributor
  • ***
  • Posts: 22436
  • Country: us
  • Expert, Analog Electronics, PCB Layout, EMC
    • Seven Transistor Labs
Hm, why film?  Electrolytic will be fine at that kind of frequency.  Interleave helps a lot too.

If you must --
1. Don't use PET, it's lossy.  It also has runaway losses; once the dielectric starts breaking down, it melts, carbonizes and pretty soon you have a hole spitting goo out of your ex-capacitor.  PP is often similar price, or more available anyways.
2. PP losses vary by type, of course.  I suppose partly construction, and also how the metallization is applied, how much and where.  EMI and DC link types tend to be very thin, so the current ratings are meager.

I've melted B32672s before, they tend to expand, cracking the epoxy fill around the base, extruding gray capacitor-goo that eventually breaks down, starting a spark and shorting the supply.  Astonishingly, the breakdown isn't instant; caps can be swollen and oozing without shorting out instantly.  I guess there isn't enough shear to mix the plates together, when this happens.

It's not like gooey goo, it's melted plastic, so it hardens once everything cools down again.

On-times of 15 minutes is probably enough for everything to come up to temperature, so I don't think you'll have any savings from thermal performance.  You'll need to rate it as good as continuous.

3. Prefer high ripple or snubber types, if possible.  More expensive, but more tolerant of abuse.  Not always rated for much RMS current, by the way -- CDE 935C series for example heats up pretty quickly despite apparently being of good construction.  I suspect some of these have a skin effect limitation, that high frequency currents only flow on the outside of the assembly, hence the current density is higher.  The losses may be good at 1 or 10kHz (and this is reflected in the tan d), but not so great at 100 or 1000kHz.

One of the best I've used is Illinois Capacitor PPB series.  Once used a bank of 10 x 0.33uF 630V's in parallel, as a coupling capacitor for a 70A, 400kHz load.  They got toasty, but handled it no problem.

And TDK has some lines with similar performance.  These are more expensive families, and also pretty bulky in large values (which somewhat defeats the current capacity as well -- it's a scaling problem), but can be worthwhile.

4. Consider a resonant topology.  Takes as much capacitor and inductor capacity as power output (i.e., about 3kVA worth of each), but can have savings in efficiency, size and ripple.  The higher operating frequency is the biggest bonus to size.

Tim
Seven Transistor Labs, LLC
Electronic design, from concept to prototype.
Bringing a project to life?  Send me a message!
 

Offline Conrad Hoffman

  • Super Contributor
  • ***
  • Posts: 2077
  • Country: us
    • The Messy Basement
+1 on the Illinois cap parts. You want the lowest dissipation factor you can get and that means polypropylene, but they wont' be small and/or you'll have to use several.
 

Offline schmitt trigger

  • Super Contributor
  • ***
  • Posts: 2431
  • Country: mx
Capacitor internal construction is paramount, and in this regard, not all manufacturers are equal.

The only failure I have thoroughly investigated, because it involved a very costly recall, that particular failure was related to the way the terminals were bonded to the film itself.
An uneven conductive epoxy application, which lead to current crowding, overheating and big sparks.

Afterwards, whenever we were considering a new capacitor in a high ripple situation, we would perform due diligence with a capacitor cross section.
 

Offline Weston

  • Regular Contributor
  • *
  • Posts: 221
  • Country: us
When using a film capacitor as a filter capacitor the reactive power is going to be low. In that situation most of the losses will occur in the electrodes, which is a function of the construction of the capacitor, and not (tan(d)) of the dielectric.

I spent a while a few months back looking at the ripple current rating of capacitors to minimize the cost of a resonant capacitor bank and got as far as testing some capacitors under a thermal camera with a big RF PA. The current ratings manufactures provide are typically an incoherent mess. The datasheet current ratings are typically given for a 10C temperature rise of the case over ambient for all case sizes. Given that you care about the internal hotspot temperature and larger capacitors have a lower surface area to volume ratio and a larger distance from the internal hotspot to the case, this implies that larger capacitors will have a higher internal temperature for the same 10C temperature rise of the case. At some high dv/dv you can destroy the electrodes, but for DC link applications you are basically going to be entirely limited by thermal dissipation which is a function of I^2.

Manufacturers also sometimes provide thermal derating curves which you can use to back out the thermal resistance from the hotspot to the case. If the derating curve stops at an ambient temperature above what you will operate at you can extrapolate it out and use it as a "over rating" curve.

I tested the B32642 series capacitors, which have lower capacitance than you want for DC link decoupling and are much smaller parts (so have a lower thermal resistance from the hotspot to the case) and found that I could run them at over 2x the rated current with no measurable loss in capacitance.

Most of the provided capacitor ratings also do not assume forced air cooling, you could also probably push the capacitors harder with forced air cooling.
 

Offline MiyukiTopic starter

  • Frequent Contributor
  • **
  • Posts: 908
  • Country: cz
    • Me on youtube
Outptu have about 30A ripple at resulting 150kHz at low voltages
At high input voltage it can have duty cycle shorter than 1/3 and input caps will see full ramps 0-70A what will get again 30A rms

I dont want to use electrolytic at least at input side to avoid huge input capacity to have possibility to measure dynamical load
Power stages are peak current controlled by MCU directly and I want to have just two stage LC filter at input and as small as possible to keep load appear resistive and not to affect measured source behavior

I also choose this topology to have simple control over it from software, I am not able to do this with resonant topology
« Last Edit: February 07, 2020, 05:50:52 pm by Miyuki »
 

Offline Conrad Hoffman

  • Super Contributor
  • ***
  • Posts: 2077
  • Country: us
    • The Messy Basement
Find out what induction heater people use. I built a very small one and was amazed at how easily it heated film capacitors. Polyester was useless, only those Illinois parts did well.
 

Offline T3sl4co1l

  • Super Contributor
  • ***
  • Posts: 22436
  • Country: us
  • Expert, Analog Electronics, PCB Layout, EMC
    • Seven Transistor Labs
Incidentally, induction heating caps are almost all metallized film, AFAIK.  Seems to be not so much the material (give or take how much metallization is applied, or whatever), but how it's put together (a floating electrode design is common IIRC).  Stonking great copper plates to sink the heat out of them is an obvious advantage.

The ratings aren't usually very useful for filtering, though.  2.4uF is a common "large" value.  You want 10-100 times that for filtering DC, even more at these low voltages.

You can get industrial caps in such values, with conduction or water cooling, but they're even more specialty, and you'll want to be sitting down before you see the price and availability / lead times...

Outptu have about 30A ripple at resulting 150kHz at low voltages
At high input voltage it can have duty cycle shorter than 1/3 and input caps will see full ramps 0-70A what will get again 30A rms

I dont want to use electrolytic at least at input side to avoid huge input capacity to have possibility to measure dynamical load
Power stages are peak current controlled by MCU directly and I want to have just two stage LC filter at input and as small as possible to keep load appear resistive and not to affect measured source behavior

What is this, DCM?  Yikes and a half!  At least do CCM.  For which, average current mode is easier: the filter takes up control loop bandwidth so your digital control is less strict.

Best case, the analog side is passively stable and safe: even if the MCU crashes, it's just stuck at whatever setpoint it was last given.  Can even add a missing pulse detector as a watchdog.

I don't know how "small as possible" you're expecting, as it sounds like you're planning on making it pretty big.  You need some pretty huge, ferrite cored inductors to handle that much current ripple, and both chokes and caps are huge at this frequency.  Small would be 200-1000kHz, SiC MOSFETs and schottkys, resonant as an option, phase interleave is still good, uhh ferrite cored chokes of course, maybe some beefy ceramic caps if cost is no object, but otherwise film caps will be fine.


Quote
I also choose this topology to have simple control over it from software, I am not able to do this with resonant topology

Resonant LLC is typically controlled by frequency modulation, and on/off keying at light loads or low output voltages.  Protection/limiting is often similar, i.e. peak current mode.  Even easier from an MCU, I suppose...

It's a rather mysterious and opaque topology on first look, but it turns out a surprisingly crude control works out pretty well.  The real magic seems to be correctly choosing L, L and C with respect to Fmin and Fmax, and source and load ranges; the source range at least is pretty narrow.  Or, I think, one or the other is narrow, since the output range is actually quite flexible, and reciprocity is a thing.  But you can't have both flexible, because the impedance of at least one side (and therefore the available current flow at a given voltage) is set by the tank impedance.

Tim
Seven Transistor Labs, LLC
Electronic design, from concept to prototype.
Bringing a project to life?  Send me a message!
 

Offline MiyukiTopic starter

  • Frequent Contributor
  • **
  • Posts: 908
  • Country: cz
    • Me on youtube
I want it as universal as possible to have input voltage from basically zero to 400V
And small inductor about 10uH are still reasonably small at currents up to 100A
And also it can keep switching losses low even with moderate speed Si diode with low Uf

"Just" that capacitors issue
 


Share me

Digg  Facebook  SlashDot  Delicious  Technorati  Twitter  Google  Yahoo
Smf