Capacitor lifetime over varying temperature.

(aluminum electrolytics lifetime prediction)

why exactly do you think it's not good enough?

Oliver

--
Oliver Betz, Munich
despammed.com is broken, use Reply-To:

dbvanhorn wrote in news:454d8ae7-1a08-4ce4-8786- snipped-for-privacy@s9g2000yqd.googlegroups.com:

With temperature swings that great and frequent I think you'll have more to think about than lifetime of electrolytics alone. If the capacitance isn't more than a few hundreds of µF it might be better to pay for ceramics in X7R or Y5V dielectric, each one having 10, 22 or 47 µF and choosing your best compromise on cost, ease of ganging them into arraye, etc.. Easiest is if they have leads and epoxy coatings, they survive stresses better, but the costs can be horrible, and bags of SMT 10µF caps are cheap enough, but assembly time might not be. Some firms sell pre-made assemblies to help with that. This also allows a small board, which is cheaper to isolate against the faster temperature changes. The smaller it is, the easier and cheaper it is to protect it. Electrolytics are low density parts so I wouldn't even consider them if there was any way to avoid them in this sort of situation.

To get you started:

Click on tools, then select Life & Temperature calculators.

If that's insufficient, define precisely what "good enough here" means to you, then discuss that with the cap manufacturer(s).

My guess is that you won't get the kind of prediction you want, but at least you'll have the opportunity to explore your options more fully.

Meanwhile, design your gizmo for maximum cap life, since that seems to be your goal.

Ed

dbvanhorn schrieb:

Let's do an example calculation for a Sikorel

10000uF, 100V capacitor (case dimensions 64.3 mm x 80.7 mm). Let's also say that ambient temperatures at the capacitor location are distributed as follows over the span of one year: 25 entire days at 100°C, 50 entire days at 90°C, 110 entire days at 80°C, and the remainder at 70°C or below. We use the bottom diagram on page 98 to find expected lifetimes (for a ripple current of 16A at 100Hz) of roughly 30000, 65000, and 135000 hours for the first three temperature bins, corresponding roughly to 1250, 2500, and 5500 days of operation. Your yearly operating days at these temperatures therefore eat up about 2.0%, 2.0% and 2.0% of the capacitor lifetime budget. In other words, with this Sikorel capacitor, your gizmo can be expected to last about 17 years. Good enough?

Taking lower temperature bins into account will decrease the expected lifetime somewhat, albeit not dramatically so (one would first have to extrapolate the diagram on page 98 to lower temperatures for this, which is fairly easy to do).

Martin.

First off, it assumes that the temperature is constant.

I'm fighting a couple of issues here. We don't have a good handle on operating temperatures at the moment, but I know that the variation is large.

Given the application, it's reasonable to assume that the temperature won't be below 10C or above 40C in the room, but the high end is not so well defined, since these are located in windows where the insolation could bring the temps up substantially.

Assume that I could have a sine-wave-ish temperature curve over a 24 hour period. This is probably wrong, but it will do. To keep things simple, assume a 40C swing with 25% of the time at -20,

25% at -10, 25% at +10, and 25% at +20 around some nominal temperature. Does the cycling itself take away lifetime? I would assume that it does, but I've never seen anything to tell me how much. Do I calculate with the average temperature, or try to bin it, and assume that the effects are relatively linear?

These are caps in the 2000uF range, so I don't have the option of going to ceramics.

e

This is closer to what I'm looking for, but my temperature variations are fairly large, and will occur on a daily cycle.

I may just have to assume some derating number and wing it, it doesn't look like anyone has data for this sort of applicaiton. Maybe automotive would but their temperature swings would be larger than what I'm dealing with, so they would probably be judging the parts too harshly.

The good news is that my ripple currents are relatively low, about 1/4 of rated, so I have that working for me.

(aluminum electrolytics lifetime prediction)

no, it doesn't if you rephrase it "twice the aging speed every 10K". See below.

[...]

I don't think this slow cycle causes a problem for the aluminum electrolytics, but to be sure, you could ask the manufacturer.

It could also be that mechanical stress causes other failures.

So you have to integrate the aging over time, IOW average 2^(T/10).

Oliver

--
Oliver Betz, Munich
despammed.com is broken, use Reply-To:

dbvanhorn wrote in news:38482363-828b-4efa-a082- snipped-for-privacy@v23g2000vbi.googlegroups.com:

I'd bin it. Completely. As in, throw away the calculations outright. Most industries derive stats from observations, then optimise for longer life, better performance, whatever... If you're trying to model that kind of complexity in advance, forget it. As Sam Goldwasser said to me a couple of times when discussing SPICE, 'a simulated circuit gets a simulated degree'. (Was a professor's remark to him, badly transcribed my me).

You can do a very detailed set of calculations but I doubt they'll be worth the bytespace they occupy unless you start building and testing.

Could you use tantalum with a high voltage rating? If the answer is 'no because of cost', then I doubt the project is a money-is-no-object exercise in perfection so you might as well just get the best electrolytics you can find and be prepared to replace them, and note how frequently you end up having to do it. Choose some with higher voltage and temperature ratings than you need, that's an easy cheap way to extend safety margins generally. You can always try to economise later if tests show that you can risk it.

And even in a sunlit window you likely have a way to mount the caps in the space least likely to get solar heating, and to provide a grille for convection. There really ARE too many imponderables for pretty calculations here, I think you will just have to build one and start basing your assessments on observations.

dbvanhorn wrote in news:a92abc93-eb79-4061-92a5- snipped-for-privacy@u11g2000vbd.googlegroups.com:

In many cases they probably do, but that would be their trade secret, earned from a lot of testing. They won't easily be persuaded to sacrifice that to save others from doing the work required. Look at how laser diodes are spec'd, and you'll soon find out how much of the exalted specs are derived from mass longterm testing, and how few actual specs there are for new devices. None of these people calculate in advance when the time is better spent in setting up tests and gathering data.