Process capability platform is missinging function of analyzing data with different distribution types

[#1]

Hi there,

I recently discovered JMP 15 process capability Platform (Analyze -> Quality and Process -> Process Capability) is currently lacking function of calculating process capability based on the fitted distribution. Instead it always assume all data is normally distributed and it is quite misleading (at least this assumption should be noted somewhere so user can understand how the cpk / ppk is calculated). The issue for doing this is the estimated yield loss % (following sigma/ppm with normal distribution assumption)  would be far away from observed yield loss % and such cpk/ppk can also be misleading users think their process is capable and then get confused the non-conformal rate is always not accurate even with large dataset. 

Furthermore, even the data distribution (non-normal) is set by user manually in the launch window for just one column/process, the process capability platform is still showing exactly the same cpk/ppk values by using normal distribution assumption. 

However, the distribution test can be done under distribution platform (continuous fit -> fit all) and it is quite considerate that only the distribution type that fits the data is unfolded and usually by checking GOF, p value would tell if it is a good fit. Then user can use compare distribution function to see the calculated ppk (usually quite different than the one calculated using normal distribution) (why only long term capability is calculated with non-normal distribution?). Under the correct distribution, I did observe the estimated yield loss % is very close to the observed yield loss %. I guess this can be done by manual steps such as make combined data table then sort out the ppk per process with correct distribution, but it would be much slower and users would not be able to get the nice capability box plot with all processes included. 

Can anyone chime in if I am missing anything here? I really appreciate if JMP can optimize the process capability platform so it can make cpk/ppk/yield projection really useful for industry users. 

Online reference: Process Capability and Non-Normal Data

https://www.spcforexcel.com/knowledge/process-capability/process-capability-and-non-normal-data

Thanks in advance.

Harry

[#2]

Currently, you need to use the Distribution Platform to calculate non normal Cp and Cpk values.

[#3 · Accepted solution]

The dialog for Process Capability is a bit strange.  You can select the distribution type, or select best fit but you then need to make sure you click 'Set Process Distribution'.  This will be confirmed by the variable name having an &Dist suffix:

Here is the output if you select 'best fit'

[#4]

Hi Dave,

Thanks for the trick. It worked for me so that I can see the best fitted distribution and ppk in the summary page.
Is there a reason why only ppk was calculated or is it possible to calculate cpk as well for those non-normal distribution process?

Thanks,
Harry

[#5]

Capability indices like Cpk are based on estimates of short-term variation (sigma). Performance indices like Ppk are based on estimates of long-term variation. How would use estimate short-term sigma for a non-normal distribution?

[#6]

don't use sigma, use the fitted distribution and derived percentiles.

[#7 · Accepted solution]

Read your referenced article for their explanation.

[#8]

Thank you, find the answer by reading it again.