Nordic JMP Users Group
A typical example of a normal mixture distribution is seen in injection moulding where different cavities have different means. It is reasonable to assume that each cavity is normal distributed. In addition, it is also reasonable to assume (although not necessary) that the standard deviation within a cavity is the same across cavities. This obviously leads to a normal mixture distribution.
It should then be obvious than to use normal mixture distributions/quantiles in JMP, but this is not what we see being used. Instead, we see many ways of calculating Ppk on multimodal distributions. Most of the methods have serious issues. We present here the issues and their consequences and a method that solves the issues.
Process capability indices like Pp and Ppk are traditionally being used to describe quality. However, Ppk describe Out of Specification (OOS) rates more than quality. If Pp is very high (which is the case in injection moulding), the mean can be far away from target, where products work the best, still having a high Ppk. In addition, the relation between OOS rate and Ppk is distribution dependent, so when data are not normal distributed, Ppk is not a good estimator for OOS rate.
