What robust methods are common for Process Capability Indices for Non-Nornal distributions?
Jun 19, 2014 2:05 PM(2157 views)
In Europe, we found it common to use Cpkn as a non-normal Cpk metric, where mean was replaced with median, and stdev was replaced with IQR/1.35. For some "regulatory" or "customer" reports, this seems to be accepted rather than the usual "torture the data until it says its normal" approach with transformations. Does anyone have similar experience or better ideas? (Note: any use of capability metrics should include sample size as confidence intervals are wide for samples of 10 or even 100 raw data points). In addition, some parameters are trimmed by "sanity limits" when engineering knowledge is "profound" on that issue, cleaning up typo's and corrupted measurements from operator mistakes.
Other novel uses of the standard formula for Cpk (which requires USL and LSL) is to use same formula, but use current Control Limits instead of Spec limits...simply to check SPC charts over time to see if the limits are still roughly 3 sigma limits, which would give value of modified Cpk equation (Cpk-control) of 1.0 roughly rather than roughly 2.0 for spec limits on a really "six sigma" process parameter. This chart limits check method uses latest 25 points minimum we noticed.