Pete,
A couple of ideas…First, MTB is not real helpful for non-normal capability analysis, but is rather decent at distribution fitting, so this is not too difficult. If you have fit the data to a logistic model, you could use the cumulative density function to determine the actual area under the curve beyond your spec limit. That would be your effective ‘defect rate’, which enables you to get a DPM or perhaps DPMO and from there it’s a short hop to sigma level.
Of course, I would look at it all umped together, but anticipate needing to assess the 2 strains of subprocesses independently to better understand the root causes within each.
Hope this helps.
.
