From: Stan Hilliard
Date: 22 Sep 2002
Time: 00:31:01
Mike,
1) Since the samples are collected during manufacture of the batch & checked in line, and you cannot predict the batch size in advance, one sampling scheme would be to sub-sample at constant frequency, allowing the composit sample size per batch to vary. The sub-samples would have a constant sample size and therefore a constant oc curve.
2) If you want constant oc curve on a batch basis, in your case you need constant sample size per batch. Changing the batch size does not affect the producers and consumers risks -- or any part of the oc curve.
When you sample the small percentage of the batch that you are, you essentially have sampling with replacement. That is, the removal of the sample from the batch has negligible effect on the remaining percent defective of the batch. As a rule of thumb, this "infinite population" situation occurs whenever the sample size is 10 percent or less of the batch size. In that case, use a constant sample size to attain a constant oc curve. The variations in batch size do not influence the oc curve.
www.samplingplans.com/usingoccurves.htm
3) The attribute standards mil-std-105 and Z1.4 call for increasing the sample size as batch size increases, but that changes the oc curves. I think that they did it that way for psychological reasons, and because it is harder to get representative samples from larger batches, and because there are more likely to be process changes within larger batches.
4) With the software program TP105, you can use the oc curve to design sampling plans for defectives (Binomial distribution, go/no-go) or defects counts (Poisson distribution, multiple defects per item).
www.samplingplans.com/software_oc.htm