Hi All,
I work for a company who have bought a significant amount of product, and I visited to find the reliability data. I ended up doing a search and came across this thread.
I have dowqnloaded and viewed the document mentioned, and have many issues with the data in it in terms of MTBF.
I've been performing as an SME in reliability for over a decade, and in the RE field for over 20 years, and I've seen, i think most of everything and lots of the same thing.
The issue i have with the data shown here is that it is, well ridiculous to be frank. As we all know, from experience, we rarely see electronics of any level of complexity still running after 30 years.. and after 50 years... well, not many people have been around long enough to see it, and in practice, most of it is long gone.
Whilst there are ways to predict failure rates, its the responsibility to outline how the predictions are calculated, and if there is any resemblence of what the field data shows. Most organisations who are really interested in improving reliability put their products under stress testing, and are very interested to see how it performs in the field, and have a focus on getting good data and performing useful, reliable analyses.
Thats why when i see mtbfs of 200,000 hrs, 400,000 hours, and 2,000,000 hours, i think, "hang on"... thats 22, 44 and 200+ years!!
now, perhaps 22 years is understandable, for redundant systems..but for plan old MTBF..its hard to believe..and 200 years for anything is, well, even harder to believe.
Now keep in mind that MTBF, if based on an exponential distribution, is the time when 63.2% of a population has failed..so for the 200 year case, nearly 40% of the population is still going strong after 200 years..
the more you look into it, the crazier it seems.
To provide another anecdote, ive many times spoken to companies who make such claims, and its usually something like this.. we've had 1000 items in the field for 2 years, and have had 3 returns. Now, ignoring that some items wont get returned, and working from the data, thats 17.5 million hours... and since we have 3 returns, then 17.5 million divided by 3 equals about 6million..right?...
well the math is right, but the logic is flawed, and its a very common mistake. Consider this.. since 3 of a thousand have failed, then 0.3% of the population have not made it to 18000 hours mtbf. That on its own points towards some clues in the reliability. And those werent just failures- they were failures that caused the product to be non functional... such failures are LESS COMMON than non critical failures..so the actual MTBF failure proportion would be higher than 3% of the population, by inference.
Now, the issue is that we dont have enough data to make a conclusion using averages, so the method is wrong. Fortunately the mathematical community has a way to effectively deal with this. Its called the chi squared method. It takes the number of samples into account and gives a confidence limit where the true MTBF lies between.
For conscientious companies, the usual approach is to err on the side of caution, so they take the position that their data has enough evidence to prove that the product has an MTBF of GREATER THAN the x% lower confidence limit.
In the example, the lower confidence limit is calculated as ( Reliability Analytics Toolkit) 2.2 million hours.
The kicker here is that this approach relies on the assumption that the failure rate is constant, throughout the life of the equipment- and although convenient, we know that it just isnt true. There is a gradual acceleration, and we usually ignore the initial teething failures also. Compounding this is that due to the relatively low period of observation, the calculation can still be very inaccurate.
So the message is- what has netcomm got in place to give accurate mtbf data, and if nothing else, on what basis can netvanta support the very high mtbf claims?..
