60th Percentile with Vendor Data
What does this title mean? Chris and Fred discuss how some vendors make ‘startling’ claims regarding reliability from some small amounts of test data. So where does this 60th percentile some of us hear from time to time come from?
Join Chris and Fred as they discuss some of the statistical gymnastics vendors go through to make certain claims about component reliability from ‘really’ small tests. How do we deal with vendors who make claims?
- What is out there? There is a method (that we won’t go into here) where if we test something and observe zero failures, we then assume that a failure is imminent (i.e. a failure was going to occur the moment we stopped testing) and then we assume a constant hazard rate, and we can then come up with a 60th percentile lower confidence bound on the MTBF. This is awful.
- Frequentist versus Bayesian. Oh dear … here come the statistics. Most textbooks rely on the frequentist (or classical) school of statistics. That is … there is no information beyond that contained in the data. We make no assumptions. We have no previous knowledge. However … most ‘frequentist’ analyses involve prior knowledge or assumptions. Like the one above … where we assumed that a ‘failure is imminent.’
- So what do we do #1? Start with understanding the DECISION. Let’s say the component in question is a relatively ‘well-known’ component in terms of how it fails, robustness and so on … then do we really need to know more about it’s reliability? If there are other components that are new, developmental, or otherwise problematic … then you don’t. However, if this component is ‘centrally important’ to the reliability of your system … and the only data you have is a vendor’s test (which you didn’t observe or be part of) with no failures observed? If this is not important to you … then neither is reliability.
- So what do we do #2? If there is no scope for testing or simulation regarding the reliability of your main component … then you can’t wish the problem away. If the component in question is (for example) a capacitor … then free research should help you work out how it might fail. Do we need to put in a clean power supply? … ESD protection? … or something else that will protect the functionality of this very important component for which you have no ‘statistical’ understanding of its reliability.
Enjoy an episode of Speaking of Reliability. Where you can join friends as they discuss reliability topics. Join us as we discuss topics ranging from design for reliability techniques to field data analysis approaches.