Weibull analysis may _appear_ to be straightforward, but only until you start to interpret deviations from a 'nice' linear set of data. When the data, as plotted on properly transformed axes, is nearly linear, as few as 7 points will clearly make the point for an internal report. In my lab group, at one time we had the 'rule': 3 points - talk among ourselves; 5 points - talk (verbally) with the engineer who submitted the samples; 7 points - write the company internal report. Said report restricted itself to parts processed in a similar manner to these samples.
I have seen data on turbine blades - publicly displayed - that used 7 points. I believe this data applied to development work. For life estimates that are critical-to-mission, I sure hope that more data is applied.
The thing to watch out for in life testing is that we are playing here with physical fatigue (even chip failure
) most of the time. The interplay of multiple failure modes can really confuse the daylights out of mathematical weibull calculations until you figure out which failure goes in which mode's group. Maybe the lab tech could see the different mode in the failures. Maybe they didn't or couldn't notice. The analyst is now left with the problem of assigning data points to categories based on how the results look. Your sophomore stat instructor should have warned you not to do that, until you understand Bayes a _lot_ better.
I haven't plotted out the data offered for this thread, but I would watch out for a couple things. Early failures could well be a weak failure mode, with different MTTF and weibull slope. IFF the first 5 and the rest of the data more or less overlap (graphically), then I might argue to the engineers that 5 points in the next test will give them an early heads up to results. But any wild deviations from linear plots mean all bets are off until the failure mechanisms are understood better. That takes hard thinking; sorry 'bout that.
Please do not start another thread on the same topic. If you want to move a thread to another forum, please report the post and ask a moderator to move the thread.
