Light bulb reliability

[ban-all-sheds]

BAS, as I'm sure you know, 'average' is just a general term which covers a whole host of summary statistics

As you say "average" has no defined meaning. It can refer to mean, median or mode. And mean can be arithmetic, geometric or harmonic.

Echoes, it is extremely unusual for survival times (or times to failure) to be anything like Normally distributed (exponential distributions, or ‘fancy’ ones like Weibull and other distributions, are much more common), either in engineering or biology. If they were Normally distributed, then, of course, the arithmetic mean and median would be identical.

A very common curve for things made by man is the bathtub.

I'm not sure what 'average lifespan' is quoted in relation to consumer products (like lamps) - does anyone know? It wouldn't surprise me if it was the same MTBF/MTTF that electrical engineers are used to using. The marketing people would probably like that, too - since the 'long upper tails' of most failure distributions (i.e. a few of the items tend to last 'for ever'!) is likely to result in mean failure times being appreciably longer than median ones!

IHANI.

My gut feel for lamps is that it's probably quite a complicated curve - I can envisage it starting like a bathtub, i.e. a few early life failures with the line rapidly dropping, and then a significant period where it is some kind of normal/poisson hybrid, and ending with something rapidly increasing - exponential-ish.


I do know that in reality the "average life " quoted by lamp manufacturers is hard to verify (funny, that), and that you'd need to have a statistically significant population of lamps of the same type and run under identical conditions to those which the manufacturer assumes to see if you agree.

Apparently trivial things can make a significant difference - I'm told that incandescent lamps which are overvolted by less than 5% have their life almost halved, so if you buy lamps which are rated for 230V and use them here where our voltage is actually more likely to be 240V, you're very unlikely to see the average life which the maker quotes.

Westie might be able to tell us how spiky our supply can be, but would anybody be surprised to learn that a single spike could knacker the electronics in a mass-market CFL?

 

[JohnW2]

As you say "average" has no defined meaning. It can refer to mean, median or mode. And mean can be arithmetic, geometric or harmonic.

Exactly - and a good few others. Hence I thought it was slightly misleading for you to write;

A 15,000 hour life does not mean it's the average - it means that by 15,000 hours half will have failed.

... since you seemed to be implying that 'average' meant arithmetic mean (and, anyway, as I said, I'm not sure that 15k hour life is meant to be a median, as you imply).

Echoes, it is extremely unusual for survival times (or times to failure) to be anything like Normally distributed (exponential distributions, or ‘fancy’ ones like Weibull and other distributions, are much more common), either in engineering or biology.

A very common curve for things made by man is the bathtub.

Yep - 'bathtub' is a term that engineers tend to you to describe a mixed Weibull distribution. In practice, the 'early/infant failures' initial bit of the curve is often ignored, since interest is often primarily in those components which survive that initial period (or, often, 'surviove the warranty period ) - in which case one has a simple Weibull, or other type of exponential distrution. Precise modelling of failures can be very complex (because of differing modes of failure at different product ages) - but, in a pragmatic sense, is often not required (except sometimes as a tool in relation to product improvement).

IHANI. My gut feel for lamps is that it's probably quite a complicated curve - I can envisage it starting like a bathtub, i.e. a few early life failures with the line rapidly dropping, and then a significant period where it is some kind of normal/poisson hybrid, and ending with something rapidly increasing - exponential-ish

Failure curves of lamps have had a lot of exposure, since they are probably second only to human survival curves in being used for teaching about such things! At least at the 'first level', there's probably no need to do as much 'Balkanising' of the curve as you suggest - since, with appropriate choice of parametsers, a mixed Weibull curve alone can get fairly close to most situations. However, what I was questioning was not the distribution but, rather, what statistic is quoted as the 'life expectancy', 'average life' or whatever of consumer items like lamps. As I said, in engineering (including electronics/computing), it will nearly always be MTBF, not median survival - but I just don't know in the case of consumer products, where marketing considerations may win over engineering ones!

Kind Regards, John..

 

[ericmark]

Time to return to earth. He is talking about 6 lamps. How do you expect to really work out any average with just 6 items? It's far more to do with luck.

If you use 100 lamps then selecting a manufacturer who claims 20,000 rather than 10,000 hours may pay but with 6 it's just luck. Also since it's a spot lamp rather than a plain bulb the look of the bulb will also be important.

This

looks very different to this

and unlike with my plain bulbous shaped lamp selecting a decorative one is not just a matter of fitting and wattage but also what it looks like.

At £10 per lamp they are not cheap 15,000 hours at 6 hours a day = 6 years but at 24 hours a day = 1.7 years and of course the seller has no way to know how long the lights are on per day. So I would not consider replacing any lamp bought in April 2009. And that does not take into account any voltage problems. Clearly lost of neutral with a PME supply claims would be made to electric supplier not bulb manufacturer although I am sure that has not happened running lamps at 253 volt would reduce their life yet would still be within spec. Also spikes could affect the electronics within the lamps and to replace is mainly a good will gesture rather than something that could be forced.

The standard filament bulb was rated at around 1000 hours. So approx 6 months but I seemed to replace on a week which is about right with 18 bulbs. But then only used 12 bulbs so should have been one every 2 weeks. I never considered taking a bulb back even if it failed after a day.

 

[JohnW2]

Time to return to earth. He is talking about 6 lamps. How do you expect to really work out any average with just 6 items? It's far more to do with luck.

Exactly - 'luck' (chance, probability) is what everything statistical is about. As you imply, one would have to 'test' far far more than 6 lamps to destruction before being able to question whether the stated 'life expectancy' was correct or not.

Even if, no matter what sort of 'average' the 15,000 hours is, and no matter what the technicalities of the 'failure curves', there would be a surprisingly (to most people) high probability that, by virtue of 'bad luck', one set of 6 lamps would all fail very much earlier than that average would suggest - even more so if all 6 came from the same batch.

Kind Regards, John.

 

[fathertobe]

Well, I've had 2 (of 4) fail in a hallway too in less than 2 years.

11W 141432 (now a powerlens model, I think).

They've been switched on and off an average of twice a day - so 1000 times (let's say generously).

Total hours of use: max 1000.

Perhaps they're normally distrubuted but with a standard deviation of around 5000 hours.

 

[JohnW2]

Well, I've had 2 (of 4) fail in a hallway too in less than 2 years. .... Total hours of use: max 1000.

As Eric said, you're up against the limitations of very small samples here - i.e. you could simply be the victim of 'bad luck'.

On the face of it, your experience might be taken to suggest that the proportion of lamps which fail in 1000h or less is 50%. However, with a sample of only 4 lamps (and with various assumptions), all you can actually say about the proportion failing by 1000h is that, rather than 50%, you can be 95% confident that the proportion is between 6.8% and 93%, 99% confident that it is between 2.9% and 97% and 99.9% confident that it is between 0.9% and 99%. Hence, if you were very unlucky, what appears to be 50% failure within 1000h might actually be 1% or less. Furthermore, if you bought all four lamps at the same time, from the same batch, once one has failed early it becomes very much more likely (than 'usual') that a second one from the same batch will also fail early.

Perhaps they're normally distrubuted but with a standard deviation of around 5000 hours.

As I've said before, suvival/failure times are very rarely anything like normally distributed, not the least because of the relative high rate of early/'infant' failures that usually/often occur, even when most items have a very much longer life expectancy. However, FWIW, if it were Normally distributed, with a mean of 15,000h and a SD of 5000h, one would expect about 0.25% of lamps (1 in 400) to have a life of 1000h or less.

Kind Regards, John.

 

[fathertobe]

Thanks John. That's what I was aiming for - about the bottom .25%

In fact, it's clear from 10 minutes googling that CFLs are not impressing people much - however their lifespan is distributed.

I'm gonna replace mine with some quality LEDs. They're not much more expensive. And they'll:
1) be warmer (CFLs @ 3000K are still like shop lights)
2) start up instantly
3) be cheaper

The hassle is going to be in getting something that measures well for my fittings (and is good quality). Mine are fire-rated so I can't change the depth of the fitting ...

Is there such a thing as a GU10 extender? Does anyone know what I mean?

 

[ban-all-sheds]

CFLs @ 3000K are still like shop lights

Don't buy 3000K ones then....

be cheaper

Not to buy, if they're to be any good.

 

[JohnW2]

Thanks John. That's what I was aiming for - about the bottom .25% In fact, it's clear from 10 minutes googling that CFLs are not impressing people much - however their lifespan is distributed. I'm gonna replace mine with some quality LEDs.

In the context of this thread, a couple of points about LED lamps...

LEDs very rarely 'die' in the 'sudden and catastrophic' sense that an incadescent lamp or CFL does - rather, their light output gradually diminishes over time. The 'life' of an LED is therefore not usually measured to the time of its 'death' but, rather, to the time when it's light output falls to 70% ('L70') or 50% ('L50') of its original brightness.

Whilst the life of an LED, if adequately heat-sinked, is extremely long, the same is not necessarily true of self-contained LED lamps (i.e. those which plug straight in to a mains lampholder or fitting), since compoinents other tha the LEDs themselves are far more likely to die (and, unlike the LEDs themselves, will usually 'completely die') - cheap LED lamps may therefore not have anything like the lifespan of an LED per se.

Kind Regards, John.

 

[peteuk]

Wow, thanks to all for your interest and opinion.

FWIW I believe the 15,000 hour life should be MTBF for consumer products as suggested by some above (I've heard it quoted for computer disc drives for example) but again it is purely speculative.

I also understand the point that 6 bulbs cannot be statistically significant. But being a simple user it does feel that the bulb life I have experienced is low. Then again the 6 original bulbs were bought at the same time so they could be from "the same batch" if that's how they are made.

I don't know if it's because I notice the life of these more but I think it was rare for me to change incandescent bulbs. The few we still have have not been changed since we moved here in 2005 because they did not need replacing and I feel they aren't on long enough to justify an expensive low energy replacement.

I have no problem with the 'slow' warm up time - it gives my eyes time to adjust in the morning - LOL

Taking the above comments on board and the possibility of a similar failure rate for the Genura I will probably still opt for Megaman over Genura because of the lower initial purchase price (~1/3).
Or can anyone recommend a long life low energy R80 LED equivalent that doesn't cost a fortune.

For info, the Megaman bulbs seem to have 2 flourescents inside, 1 above the other. It is the lower one that has failed in each case - it stops lighting and there is only a dim light from the other flourescent behind it.

Thanks again to all,

Pete

 

[fathertobe]

CFLs @ 3000K are still like shop lights

Don't buy 3000K ones then....

be cheaper

Not to buy, if they're to be any good.

Well 3000K from a good manufacturer (megaman), is advertised as 'Warm White' but it ain't!

And yes, you're right - only cheaper in the long run. I'm figuring on about £15 per LED - decent make, e.g. Aurora.

 

[peteuk]

Have googled looked for an Aurora R80 LED and can't find such a thing.
Although it looks like they do a conversion kit from an R80 lightfitting which would also mean rewiring the lights so I don't want to do that.

There do seem to be some R80 LEDs on offer in bulk from Alibaba.com (100 at a time from China, very cheap).

Link to a recommended R80 LED anyone?

Thanks,

Pete

 

[JohnW2]

FWIW I believe the 15,000 hour life should be MTBF for consumer products as suggested by some above (I've heard it quoted for computer disc drives for example) but again it is purely speculative.

As I said, MTBF (rather than median survival time) is totally standard for all 'industrial' electrical components - as you say, including things such as hard drives - my uncertainty being whether the same is true of consumer products. However, as I also said, I suspect that consumer products probably also quote MTBF - since, with the likely shapes of survival curves, MTBF figures are probably larger than median survival, which marketing folk would like!

However, as we've seen to some extent in this thread, the general public are probably much more likely to think that they are being told about median survival time - just as they usually do with human 'life expectancy' figures.

I also understand the point that 6 bulbs cannot be statistically significant. But being a simple user it does feel that the bulb life I have experienced is low. Then again the 6 original bulbs were bought at the same time so they could be from "the same batch" if that's how they are made.

That last point is definitely crucial with large-volume mass-produced products, since 'duff batches' (or sub-batches) are a very real possibility

I don't know if it's because I notice the life of these more but I think it was rare for me to change incandescent bulbs. The few we still have have not been changed since we moved here in 2005 because they did not need replacing and I feel they aren't on long enough to justify an expensive low energy replacement.

My experience has been different. For my sins, I have a house with a very large number of bulbs/lamps (80+ I think). In the incandescent days, I seemed to spend most of my life replacing them, and I also had to buy vast batches of them every few months. Since largely changing to CFLs, the amount of replacement has reduced dramatically. So, whilst I cannot tell you whetehr the CFLs are living up to their full claimed lifespan (I haven't kept any records), I'm quite sure that they are lasting very much longer than did the (1000h or whatever) incandescents.

I have no problem with the 'slow' warm up time - it gives my eyes time to adjust in the morning - LOL

I agree. It's only when one has PIR-operated lights (e.g.for security) that slow start-up can be a problem.

Kind Regards, John.

 

[scousespark]

Maybe timne for a little 'Statistics 101'?

Kind Regards, John

God I wish I was as clever as what you obviously is.

I sometimes get home when Eggheads is on and am staggered by their patronising attitude. Then I come in here and......

 

[JohnW2]

Maybe timne for a little 'Statistics 101'?

God I wish I was as clever as what you obviously is. I sometimes get home when Eggheads is on and am staggered by their patronising attitude. Then I come in here and......

I also wish that I was as clever as you seem to believe I think I am There is a difference between humour (at least you quoted my smiley!) and being patronising - and, if you look through my posts, I would hope you will come to realise that patronising is the very last thing I ever intend to be - in contrast to what one sees from a good few other posters! The reality is that I am educated/'qualified' in a rather ridiculous and diverse range of fields, and self-taught/'experienced' in a good few others, but you'll never find me trying to make any mileage out of (or even mentioning) that, and I certainly don't 'talk down' to people. I hope that people simply take my posts 'as they find them'.

Back to the specific ... potentially confusing/misleading things were being said about Statistics, so I thought it would be helpful to some if I tried to clarify things a little. That's all.

Kind Regards, John.