Ford's answers to the NHTSA 6.7 Investigation
There was a request for a link to Ford's answer's to the NHTSA investigation posted on a previous thread, since closed. Here is the link: Ford's NHTSA Answers to the 6.7 investigation This PDF i...
Forum Discussion
NewsW wrote:
Waaaait!
The immediate problem with using the statistical approach you have chosen is you have assumed that it is a normal distribution.
Concur with you that the sample size you have is at the lower range of acceptability to presume normal distribution --- but the issue that keeps coming to mind is... can it be non normal.
A different FMEA approach is to not concentrate on the central tendency, but on the outliers and ask yourself, why are the outliers there?
I once came across an issue where something coming off the line kept failing --- and the people on the line actually KNEW why each one that came off was a dud --- because they can sense it "felt" different.
Another example was from a case where operators A B and C were all making the same parts, but Operator C constantly had his dies last a lot longer. Reason: Operator C came in 30 min early, turned on the heaters to warm the die before he started his shift.
Hence, statistically, in this case, the central tendency (A and B) told you nothing useful --- but the answer was in detailed observation of C.
I didn't assume the normal distribution. I plotted the failure data and matched it to the distribution with the best fit using a regression calculation. This happens to be the Weibull distribution with a beta slope of ~2.5 and eta (shape parameter) of ~150,000 R^2 coefficient of correlation is 96.26%
That is a pretty good fit. See a simple explanation of the analysis here http://www.barringer1.com/nov07prb.htm#Weibull_Analysis
