From: Stan Hilliard
Date: 29 Aug 1999
Time: 20:03:30
Here are some comments related to your question about what population the sample size refers to.
First Strategy -- you can set up an acceptance sampling plan where a sample size, n, is taken randomly from the whole order of 200K parts. This would be done if the sample were taken after the entire lot is manufactured -- like from a warehouse.
Second Strategy -- you can develop an acceptance sampling plan for application during manufacture. Data collected under this plan can serve the purposes of making both acceptance and process adjustment decisions.
This plan must honor the natural homogenous subgroups like days, shifts, runs, RM lots, etc. Any type of change that has a reasonable probability of effecting the dimension of the parts could trigger the beginning of a new subgroup. Use your knowledge of the manufacturing process here.
The goal is to define subgroups so as to maximize the homogeneity of the quality characteristic within subgroups. There is a very thorough discussion of this on the Deja.com thread in Hubin's MSG #8.
Use the operating characteristic curve to evaluate the decision rule that you use for manufacturing process sublots. Read about it at http://www.samplingplans.com/usingoccurves.htm.
With the in-process strategy, you could sample each sublot as if you were sampling the whole order. If the sublot samples pass the in-process plan, you know that the overall lot plan would pass, so you don't have to implement that plan.
Again, acceptance sampling -- not SPC -- is the right quality engineering tool to verify that a particular lot meets the specification.
Sincerely, Stan Hilliard, CQE,CRE,CQA,PE shilliard@samplingplans.com