Home › Forums › Forum Basics › Welcome › Mean Time To Failure (MTTF) for Weibull Analysis
This topic contains 10 replies, has 5 voices, and was last updated by Gregory Heine 2 weeks, 5 days ago.
I want to understand what is ‘mean time to failure (MTTF)realy means. Not looking for definations. Already have enough from websites. Need examples that shows and explain this MTTF so I can see if MTTF is applicable im my situation.
Mike – Not sure exactly what you’re looking for, but to state the obvious is the average point in which you expect parts to fail. So some examples would be:
– MTTF for light bulbs would be the average number of hours that bulbs burn before going out
– MTTF for balloons would be the average amount of internal air pressure at which point balloons pop
– MTTF for tires would be the average number of miles driven on them before the tread wears out
MTTF is of course only one measure of reliability and the distribution is even more important, especially since failure tends to be distributed asymmetrically. Usually you would have percentiles of interest – for example, the 50th percentile is the point in which 50% of parts are estimated to have failed (not the same as MTTF for skewed distributions). Or maybe you care about the point in which 1% of parts will have failed because you want your warranty to cover that period.
Is this the kind of information you’re looking for?
Weibull is a family of distributions.
For different slope values,you get different distributions.
You also get different variances and skews depending on the distribution.
I used to ask the OEMs or anyone who supplied me with an MTTF requirement -“What “flavor” of MTTF do you want?”
Do you want Min MTTF or Max MTTF? Components and Systems are serial models.
Do you want this with a specific failure mode/mechanism or across
all failure modes/mechanisms? What subsystem / component would you like this applied to?
If is is a specification you are addressing,DO prompt them to give you a Reliability level and confidence,along with some meaningful %-tile of environmental coverage and %-tile use at life.
Then go back to your discussion on Weibull sample size.
MTTF/MTBF discussions are vague,non-value add for the above reasons.
Just have them give you a meaningful goal.
@mikeruss – I’m not sure whether Joel or Gomez really provided the answer that you’re looking for.
MTTF is a statistical measure. It is calculated as the total number of “run” hours (cycles, minutes or whatever measurement applicable) divided by the total number of items tested. In the purest form, you test all items until each fails, add up all the values and divide by the total tested. This is not the “expected mean life” as many seem to believe. This is due to the different forms of failure rates that exist.
No doubt you’ve heard of the “bathtub” curve. This is a failure rate form that fits things like electronics. There is a (relatively) high failure rate at very low use levels (these are infant mortality failures and often electronics shops “burn-in” their product to identify these and cull them from the production lot), then there is a very long period where there is a very low but relatively constant failure rate (bottom of the bathtub, poisson distribution), and then at some point (although there are a lot of electronics that never get to it) there is an increasing and usually normally distributed end of life failure distribution. This is just one form of failure curve and there are several others.
The particular failure distribution will dictate the anticipated mean life. In my past, I’ve specified reliability with the following types of parameters: Infant mortality no higher than x% at x cycles (or hours, etc.), Useful life period to have no more than x% failures at x cycles, and Mean end of life at x cycles.
If you’re interested in determining whether the reliability from one set of data is different from another (different designs, vendors, etc.), I would not use MTTF. There are statistical measures on the Weibull failure curve that would be a better measure.
@spazwhatsup – components and/or systems may or may not be serial or parallel (depends on how you define component). And of course, if you have a parallel system, that is going to affect the reliability of the system differently than if it is a serial system. Which just goes further to suggesting to Mike that further understanding requires a course in reliability.
@spazwhatsup – merely wanted to convey that one person’s component is another person’s system. Depends on your level of interest in a system. For example, for a multi-vehicle taxi service, one car might be a component, but certainly folks understand that there are many components within a vehicle.
MTTF is the average time that something fails. No average is a valuable number by itself. The average will become meaningful when you know the standard deviation SD of the process, then you can visualize the shape of your process. When you have the process visualized, then you can make series of predictions in terms of when would your product fail on an average, assuming various confidence intervals. Weibull is a common distribution to apply for determining MTTF.
@rpirasteh – MTTF and MTBF have fallen out of favor for meaningful reliability description. Bx life predictions are more meaningful and relevant and are the prefered description.
@mikeruss and @MBBinWI Not sure if this will help http://www.weibull.com/hotwire/issue80/relbasics80.htm
It pretty well explains what several people have said.
MTTF=total downtime /no.of break down,
MTBR=total up time /no.of breakdown
Santosh – thanks for adding such indescribable value to a thread 5 yrs cold. You are an invaluable addition to the site.
Some posts have a life of their own
© Copyright iSixSigma 2000-2017. User Agreement. Any reproduction or other use of content without the express written consent of iSixSigma is prohibited. More »