Normal Distribution is important for Cpk & SPC | How to use non-normality to your advantage

[@framaxlo]

Tom, thank you for your insight. I like what you said at 7:45: If your process has no reason to behave logarithmically, really go into what the process is doing. Do not simply make the transformation just to make your Cpk work... wow.

[@TomMentink]

Nice to hear your feedback and very happy to see you liked it and my video contained a learning nugget for you.

[@domenicoscarpino3715]

Hi Tom, thanks as always for the brilliant lesson. How does central limit theorem would affect this topic? Wouldn't be able to get away with anything as normal if sample size is large?

[@TomMentink]

Hey Dom, that’s right - with larger us sample groups you don’t have to worry much about the normality of the underlying data anymore. But it can be an expensive way to deal with the problem 😅

[@domenicoscarpino3715]

@@TomMentink thank you very much as always. Yeah statistically large sample makes sense, pratically it's probably not always realistic 😃

[@surnamegm]

Thank You!

[@samarmushtaq3979]

How to calculate mean of Cpk and standard deviation when u are not having normal distribution

[@TomMentink]

Hi Samar, thanks for your question. The mean and StDev calculations themselves don't change. But if your data is too far from normally distributed, the statistics aren't correct anymore. If you know that your process behaves logarithmically, however, the way to use Cps-type of statistics is to first take the Log of your datapoints and then use all the usual Cpk calculations on that transformed data. The most important point is to get a stable process - so before doing too much statistics or data transformation, take a good look at why your measurements show non-normal data: it's very often 'special cause variation', aka a problem in your process. If that's the case: fix the problems (in fact, the data has given you a clear signal that you've got deviations - so it did it's job 😉)

[@TomMentink]

Does that help answer your question?

[@samarmushtaq3979]

@@TomMentink yes it helped alot ! Can you explain When the process is not centered how we can calculate its the value ok K for Cpk and plz make a video on process capability indices for control charts

[@TomMentink]

@@samarmushtaq3979 you are right to sense that a non-centered process has less wiggle room and cannot therefore handle as much extra variation from your measurement system. Generally, you want to center the process and not over-engineer your measurement system; but if the non-centering of the process is by design, I suggest that you use 2x the distance from mean to specification limit as your process’ range for purposes of the R&R calculation. On the question of capability indices in control charts - I decided to make that video: you’re not the first to hint to this and the short answer is “no, there are no capability indices in control charts”. But there is a link, of course, and that is much easier to explain with video than text, so hang tight, ‘your’ video is coming up in a couple of days.

[@nagarajuchanti8720]

❤

[@TomMentink]

This topic was indeed inspired by your question. I had something similar (spotting a special cause in a histogram), but it didn’t really feel complete. After our chat, things fell into place for me 😃

[@nagarajuchanti8720]

@@TomMentink That's my pleasure