Mains filters, do they have a limited sevice life?

[jitter]

Recently, I modified a fluorescent fixture from magnetic ballast to electronic ballast. Since the MKT X2-class capacitor between L and N was no longer needed, I removed it.
Out of curiosity, I measured it with an LCR-meter and found it was end of life. Of the 22 nF about 8 pF was left and the ESR was too high to be measured.
This cap had been in use for about 25 years or so, whenever the lamp was lit.

This made me wonder about the lifespan of mains filters. I don't worry about the inductor, but there are also X and Y class capacitors in there. Surely they have a similar limited life span.

I looked up a datasheet of a mains filter and all it gives is an MTBF of 550,000 hours (almost 63 years ). But to me MTBF is meaningless as it doesn't represent the expected service life.

So, if a mains filter is expected to work 24/7, what would be the actual service life in which the filtering characteristics are not compromised? And what are the effects of the aging capacitors?

[SeanB]

They should fail safely, but there are a lot of stories here about the failure of Schaeffner filters in older equipment burning up if powered on after being off for a few days. The most common failure is the class X mains capacitor failing as a short or partial short, causing excessive power draw and overheating. The main cause is mains transients causing the capacitor to self heal, each time reducing the capacitance a little and removing some active metal from the capacitor. Eventually they heal down to an open circuit, or the self healing fails and the material shorts out, or the dielectric gets moisture in it and then heats up and melts into a short.

Very common in all metallised capacitors, especially those used on motors as run capacitors, or those used for power factor correction or as mains filters, where they are exposed to all transients on the incoming mains. The 60 year life is based on the unit only being fed smooth spike free AC voltage, which is not something that you can expect off the mains unless you have a ferroresonant voltage stabiliser upstream, which kinds of defeats the reason you put in the capacitor in the first place.

[jitter]

Thanks for your reply. That makes sense, although I was under the impression that safety caps should never become dangerous (like fire or shock hazards).

The most common failure is the class X mains capacitor failing as a short or partial short, causing excessive power draw and overheating.

This reminds me of some GW Instek dual lab PSUs we have at work. One day we came back from a break and discovered a chamber used for burning in medical equipment at elevated temperatures (about 40-45 C) had filled with smoke. Cause: the cap on the mains input of the lab PSU had failed and was bellowing smoke each time a spark flashed over.
We didn't think too much of it because that PSU had seen quite some use and was also in a hot environment. However, two weeks later the same happened to another lab PSU, but  that one was operating at no higher than room temperature.

[T3sl4co1l]

Yeah, MTBF is meaningless without conditions.  Most calculations miss the frequency and weighting that transients deserve.

The AC line is expected to have a mean maximum of 127VAC (264VAC elsewhere), which would probably give the stated MTBF based on the usual calculation method (a single activation energy extrapolation from an accelerated life test).

But the real AC line is also expected to have transients, with roughly a 1/f distribution of frequency and severity.  Very frequent events (such as RFI and ESD) are high frequency AC, or short duration transients, easily filtered.  Infrequent events (surges, dips, swells) contain more energy, perhaps enough to stress all the capacitors in a typical circuit (filter followed by bridge rectifier and bulk cap).  Lightning is the largest, least frequent transient that gets tested, and is quite capable of exploding capacitors.

Y caps are made to, at worst, arc over from such operation.  The line-to-ground voltage (including transients) is expected to be quite large under normal circumstances (i.e., 2500VAC isolation), and mustn't cause increased ground leakage (a safety hazard if the device becomes ungrounded).

X caps are made to be self-healing and self-extinguishing.  They naturally die from high voltage operation, as more and more of the dielectric gets burned away.

Clearly, an MTBF calculation including transients will give a significantly shorter lifetime expectation.  Breakdown is exponential with voltage, so even if the expected voltage is doubly inverse with time, it's quite reasonable to expect failure within a modest time frame.  Of course, you'd have to do a lot of characterization and calculation to establish what that time frame is.

As far as I know, equipment is not expected to pass EMC after initial sale and installation, so I guess no one cares about this as a failure mode.  Seems kind of dumb, but it would be rather hard to address, after all.

Tim

[Seekonk]

I worked for a company that made EC snubbers.  We had some legacy products that got a file under UL 508.  Boss wanted CE.  Told him that I didn't think they would pass and besides paperwork wanted o know the file of the X type capacitor used.  Wouldn't listen to me and said to send in paperwork with it blank.  He would take care of it later.  So they started doing the CE test and 70% of samples failed and they ended the test early.  $40K down the tubes.  I thought some of the ones with 220 ohm resistors in series with the cap would have survived the pulse test. 

[SeanB]

Resistor probably flashed over before the capacitor. The pulse would have exceeded the resistor voltage rating by far, so bang.