QCW Capacitor Torture Testing

[Weston]

After realizing that a lot of the datasheets for the film capacitors commonly used in QCW MMCs are inaccurate / incomplete I decided to create a test setup to measure the current capability of some of the capacitors I found that look promising.

The lab I work in has a 1KW linear amplifer that can handle ~ any load, so the test setup is based on that and pretty simple except for the amplifier. Its a 40:3 matching transformer and a series inductor to resonate out the capacitance. The inductor is sized to it resonates at ~ 400KHz where QCW coils operate. Current is measure with a Pearson 2878 current probe and I am measuring the peak capacitor temperature with a thermal camera. This should allow me to validate capacitors up to many 10's of amps.

I am starting out with two test conditions, CW and a 10ms long burst repeated 5 times a second. Allowable current is determined by adjusting the current until I exceed the maximum temperature at steady state. 

One issue with the test setup is that I assume capacitors dissipate heat through the PCB traces but right now everything is floating in air with no cooling. I might make a test PCB to see what impact copper traces have on the capacitor thermal capability.

My plan is to work my way through some of the more promising capacitors on the spreadsheet of MMC capacitors I compiled: https://docs.google.com/spreadsheets/d/142e_12Ojj_ahDvq699jDEijQQiF_3dN0BCUUxEYnwjM/edit?usp=sharing


The first capacitor I am testing is the capacitor I used in my MMC, B32642B0333J, which is a 33nf 1000V part. The datasheet gives a current limit of ~2.8Arms at 400KHz "for sinusoidal waveforms, TA ≤85 °C, ΔT ≤15 °C". At a RMS current of 3A I get a 10 degree rise above ambient, so this limit agrees with my measurements. The capacitor hits 50C above ambient at 8Arms, which is about what I would be comfortable running it at for extended periods in a project, with forced air cooling you could probably push it more.

Running at 5x 10ms pulses a second (QCW test case) I can bring the capacitor to ~50 amp peak before I hit 50C above ambient. Both these cases have the same RMS current value, which makes sense.

Going down to 1x 10ms pulse a second I can bring the capacitor to 80A peak, which seems to be the limit of my test setup without a better matching transformer. The capacitor reaches 37C above ambient in that test case.

Below are some pictures of the test setup, its pretty basic. If anyone wants any specific capacitors tested let me know. You can send them to me or if they are cheap I can add one to my next digikey / mouser order.



 





 

[klugesmith]

Nice work there, Weston.

You made a good point about conduction cooling. 
Capacitors like the one in picture normally depend on heat sinking to PC board.  Maybe that's detailed somewhere in the data sheet.  I bet those tab terminals are much thicker than necessary for RF current handling, ESR from skin effect, or tolerating mechanical vibration.  Your FLIR will tell.

Unless the capacitor terminals are meant to mate with tab terminals crimped onto wire ends.

[Hydron]

OK so in the interests of being lazy I'm gonna re-post my IRC comments on this:

<Hydron> Weston: nice, i like the tester
<Hydron> not sure i'm as comfortable as you seem to be about a 50C rise over ambient however
<Hydron> if it were uniform then fine, but we don't know how bad any hotspots might be inside
<Hydron> https://www.digikey.com/product-detail/en/epcos-tdk/B32672L1622J000/495-3211-ND/1532431
<Hydron> thats the cap i'm using btw
<Hydron> if you wanted another test article

Weston then suggested using polypropylene's temp coefficient of capacitance as a method for internal temperature measurement - turns out it's quite linear and could probably be calibrated before testing using external heating (in an oven or environmental chamber or similar).

If a stable test setup (especially the air-core inductor) is used then maybe the resonant frequency itself could be used to measure temperature in a self-oscillating test setup, but I suspect there would be some hidden complications and influences from delays and parasitics which could make it a bit iffy to measure a ~2% full-scale change. Food for thought though!

[Weston]

I have been thinking about this some more after the conversation IRC and what it comes down to is that the film capacitor current ratings given in datasheets is really vague. My allowable delta T of 50C might be too large for these capacitors but there are a few ways to find out.

The ultimate limiting factor is the temperature of the hottest spot inside the capacitor, which is determined by the heat generated, the thermal resistance from the hotspot to the case / leads, and the thermal resistance from the case/leads to the environment.

The datasheets provide a current rating based on a fixed temperature rise of the case above ambient, typically 15C. This means that some large assumptions about the thermal resistance from the case to the environment are being made. Additionally, the allowable temperature rise is given as a constant value across all case sizes. This ignores the change in thermal resistance between the hotspot and the capacitor case with varying capacitor size. The thermal resistance, and thus the allowable case temperature, should increase as capacitor size decreases due to the reduced distance between the hotspot (typically at the center of the capacitor) and the case.

There are two ways of determining the allowable current ratings for these capacitors. The first would be to test the capacitors to failure and then apply some amount of derating based on the lifetime charts provided in the datasheets. I can realistically test a few current values for ~ 10 hours, which is probably close to total QCW coil lifetime anyways, and then determine lifetime from that. A test fixture / a controlled environment would probably be crucial for this. 

The second method would be creating a thermal model of the capacitor in something like COMSOL or FEMM (open source) which would allow me to back out hotspot temperature directly. The film layers can be modeled as an anisotropic material due to the preferential conductivity along the metal electrodes. I need to do the math to verify, but I believe that at these pulse duration I can model the heat generation as happening uniformly within the capacitor. Modeling the anisotropic thermal properties of the capacitor volume requires the ratio of metal to dielectric. The actually thicknesses of each layer does not really matter. I think I should be able to back this out by measuring the density of the active volume of the capacitor, given the known densities of aluminium and polypropylene. Related to this, I sectioned a capacitor with some sandpaper today. Its interesting to see how thick the end electrodes are (~1mm). I have not seen any datasheets talk about the importance of heat dissipation through the wire leads, but it seems like it should be important. I guess thats something that a thermal simulation should be able to tell me!

I think I am going to explore both methods over the next weeks / months, but the initial conclusion seems to be that we can significantly push film capacitors past their datasheet ratings for QCW coils. There are no capacitor killing current pulses / voltage spikes like in a normal DRSSTC and it is mainly a thermal problem.

[Mads Barnkob]

I attached an interesting paper "Thermal Simulation for Geometric Optimization of Metallized Polypropylene Film Capacitors"

Cooling of capacitors by forced air can be a solution to get a longer life time. Approximately 2/3 generated heat rise moves out axial and 1/3 radial. So it is most important to cool a capacitor at its terminals as it does not radiate the heat evenly from all over its surface. So how about coming up with a heat sink connection construction for the MMC?

[ElectroXa]

Hi
I've an other solution for capacitor cooling :
this solution is to put the PCB with the  caps under mineral oil or under paraffin wax, in an ikea tupperware 

cheers

[Weston]

The paper takes the anisotropic thermal modeling approach I was talking about. Helpfully, they provide some values for the thickness of the film and the metal coating: "For a capacitor made of a film of 6um thickness and metallized with a 15nm thick zinc layer, the axial conductivity is approximately two times larger than the radial conductivity". This will vary with exact capacitor construction and breakdown voltage, but it's a good starting point.

The B32642B0333J capacitor I am using in my MMC is 7mm wide, 18mm long, and 12.5mm tall. Based on the anisotropic thermal resistance I should be getting roughly equal heat flux out the sides and the end.

Given how cheap the capacitors are and the high voltages involved any sort of heatsinking beyond optimizing the PCB traces is going to be more effort than its worth.

Refining my approach, I am most interested determining if pulsed operation is any more stressful than continuous operation with an equal RMS current. This will be determined by the specific heat of the dielectric and metal coating and the time it takes the heat generated in the electrodes to dissipate through the dielectric layer.

So far though, my testing seems to show that the B32642B0333J capacitor is a pretty good choice. My $20 30 capacitor MMC is almost overkill for a table top QCW coil.

[Mads Barnkob]

Refining my approach, I am most interested determining if pulsed operation is any more stressful than continuous operation with an equal RMS current. This will be determined by the specific heat of the dielectric and metal coating and the time it takes the heat generated in the electrodes to dissipate through the dielectric layer.

I would put my money on continues operation to have them live longer than pulsed operation, on the assumption that their solder junction behave like a IGBT silicon ageing problem.

The only real danger to a plastic film capacitor in terms of heat, would be that the ones in the middle of the MMC gets hotter than the others and thus experiences proximity thermal damages if it is being pushed to the limit in the first place.