without addressing the issues of sample size and representative nature of the full process variation, *I* would and have simply plotted the data in time series against the spec limits.
THEN there are 2 ways to assess the 'capability': one is a 'prediction' of the defect rate. The other is the 'margin' to the specification.
The first assessment can be handled with a straightforward ACTUAL (not modeled or predicted) defect rate. (you could back calculate the Cpk/Ppk from that or simply report the long term defect rate. (that coupled with the time series plot are the real answers to the capability question. The second assessment is even simpler: calculate the range of the data and divide it into the tolerance range to get a Cpk/Ppk index. This is far more statistically correct than trying to use the SD when the process is non-normal or chunky.
THEN there are 2 ways to assess the 'capability': one is a 'prediction' of the defect rate. The other is the 'margin' to the specification.
The first assessment can be handled with a straightforward ACTUAL (not modeled or predicted) defect rate. (you could back calculate the Cpk/Ppk from that or simply report the long term defect rate. (that coupled with the time series plot are the real answers to the capability question. The second assessment is even simpler: calculate the range of the data and divide it into the tolerance range to get a Cpk/Ppk index. This is far more statistically correct than trying to use the SD when the process is non-normal or chunky.