this time it's not a RIFA

[TheMG]

It sure burnt up like a RIFA though!

The capacitor that burnt up was identical to the one immediately to the left of it. EPOC E68 polyester metallized film capacitor, 250V 2.2uF.

It's not clear in the picture but the capacitor that is not burnt does have visible signs of stress, the ends have cracks and a bit of melted plastic "oozing" out, so it obviously got very hot.

I haven't yet determined if this was just a result of heat from the adjacent capacitor on fire, or from electrical over-stress.

This is a power supply with a half-bridge converter topology where one end of the single primary winding is capacitively coupled between the positive and negative high voltage rail. The capacitors in the picture are the transformer coupling ones (these caps are not across AC mains).

Interestingly, neither of the switching MOSFETs have failed.

I'll update this once I've done more troubleshooting, but this is definitely unlike any SMPS failure mode I've ever come across to date.

[Bud]

Holy cheese, there WAS a small fire there...

[rsjsouza]

I have seen flames inside power supplies, control boards, etc. caused by environmental contamination in high humidity areas - especially near salt water. The thin layer of high conductive salt water can make current flow through it and degrade the insulation due to temperature and corrosion.

I see some white powdery substance on the brown capacitor at the bottom left, but I can't tell if this was deposited by the burning of the capacitor or by the environment.

[TheMG]

The white powdery deposits appear to be from a chunk of the silpad near the capacitor that burnt. The white stuff is all over the inside of the unit but thickest near the burnt area.

Power supply is from a clean dry environment, and actually were installed in redundant pairs, the other power supply still functioning and perfectly clean.

Actually the PCB is surprisingly undamaged. So the capacitor must have gone up quite rapidly as opposed to overheating slowly smouldering.

[fourtytwo42]

Cheap as chips capacitor overstressed by application fails 
You don't say how many watts the psu is or what the line voltage is but in that place it can see full rectified line during startup and it handles half the primary current so it needs a decent capacitor with a puka data sheet Not always easy to find 

[EHT]

Isn't that a more dangerous failure than with RIFAs? I mean they explode and immediately blow a fuse whereas this seems to have caught fire.

[rsjsouza]

Isn't that a more dangerous failure than with RIFAs? I mean they explode and immediately blow a fuse whereas this seems to have caught fire.

I have seen different Rifa failure modes, including evidence of flames. I think there is more than one report and video around here that covers different failure modes.

Thst reminds me that I need to source some X1/X2 capacitors for my HP equipment.

[TheMG]

Cheap as chips capacitor overstressed by application fails 
You don't say how many watts the psu is or what the line voltage is but in that place it can see full rectified line during startup and it handles half the primary current so it needs a decent capacitor with a puka data sheet Not always easy to find 

Dozens of these power supplies in the field powered on 24/7 for 10+ years and this is the first one I've seen this failure on.

It's a 400W power supply. There is a PFC regulator preceding this, 380VDC (from 120VAC mains input). Each of these capacitors sees half of the total voltage, well within the 250V rating.

I think I'm going to just get a couple replacement capacitors and see if anything unusual happens.

[floobydust]

For high pulse ripple current such as a bridge SMPS, I would say it needs to be a polypropylene cap, not polyester.
Some manufacturers give a spec for the peak current. Unfortunately smaller physical size = less capable and the mylar part might have just aged near its stress limit.
Example compact Panasonic ECWFD "... for DC applications that do not require a significant amount of pulse current " but they have an internal fuse to prevent bonfire. But no good other than DC bus.

Kemet R75 series dV/dt 100V/usec or ECWF-F2225 I think is good. Don't go near chinese CBB22 lol resist the urge they are $0.15 for a reason