Hacker News Reader

Comment by dereg | Hacker News Reader

How do you measure/test the life of something that's supposed to last for a really long time?

For example, we refit our house with some (beautiful) CREE LED downlamps that are supposed to last about 25 years. While that's nice to know, is it possible to test that claim, or is it purely theoretical? It's not like they have been keeping the same bulb on since 1986.

View on HN

Former Reliability Engineer here. Not sure for bulbs, since my experience was automotive, but I imaging it is similar. In general, we tried to determine all the factors that lead to failure - vibration, force, impurities, sunlight etc, then you try to accelerate these. So for sunlight, you might raise the intensity of UV etc. There are settings so X hours of acceleration == to X years normal use. These settings are developed in a variety of ways. After you run the test, you apply some stats analysis and a bit of handwaving and proclaim 25 years.

interesting. are those ever applied all together?

Yes. In those experiments goal is to figure out the most likely failure mechanism, not the estimated lifetime.

"..a bit of handwaving..."

legal disclaimers and marketing?

The other article given in comments here ( http://www.kk.org/thetechnium/archives/2011/06/the_clock_in_... ) has a lot more detail on this.

The paras around the picture "Testing the longevity of materials in space on the MISSE" are quite informative. Testing materials in space... I think they're taking it quite seriously.

well you can let 1000 of them run for a year and from the ratio of those that failed calculate (or rather guesstimate) for how many years can more than 500 be still running.

It's not perfect, but for a perfect solution you would need something like a time machine, I guess.

edit: this article linked from wiki has nice explanations http://db.usenix.org/events/fast07/tech/schroeder/schroeder_...

Some failures are at a random distribution - running 1000 for a year will help you estimate that.

But any moving part has a finite lifetime due to wear, and you want to know how long this is. Running lots of them for 1/1000 of the expected lifespan shouldn't produce any at all of this kind of failure.

The normal speed of the clock is very slow, so it should be easy to run it at an accelerated rate to test wear. (There are limits, though; apparently the Difference Engine jams frequently if you overclock it.)

Reply on news.ycombinator.com