Sampling for Internal Test based upon ISO 2859-1, G-I

[nurhakim]

Hi Cove,

I have this scenario in the workplace, and I need your opinion:

We carry out internal test for a medical device, bag wise, based upon ISO 2859-1, G-I.
In one particular bag, we have 7200 pcs. According to G-1, we sampled 80 pcs. Our AQL is 1.0, Accept:2, Reject:3.

We split the test into two, visible defect test and water test. If, during first test i.e. visible defect test, we found 2 pcs of defect, we have 78 pcs left. For water test, our sampling plan is still 80 pcs. Do we need to add another 2 pcs for water test or can we use the 2 defective samples? And what if the defective samples are not fit to mount on the water test apparatus.

Thanks.

 

[Bev D]

technically as long as the visual defects can't effect the water test you could use them. However, its simpler to just add 2 more units that are not visually defective

 

[David-D]

I don't know the specifics of this ISO but since it sounds like your two defects aren't independent (you say they couldn't be leak tested which I'd generally classify as a defective part), I wouldn't be comfortable with pulling out some defects in your sample based upon the visual and then doing the leak.

Think of it this way: I have a sample of 80 parts with 4 that leak, two of which are so malformed that they'd also fail a visual inspection. If I do the visual first then I pull out the two worst ones (just passing the visual AQL) and then I do the leak test and get 2 more failures (just passing the leak test). If I switched the order of inspeftions though, you'd get 4 leak test failures, rejecting the lot/bag for leak issues. It sounds like your defects aren't independent so you are basically pre-screening your sample before you test it.

I think you need to leak test the visual rejects or just count them as leak failures automatically, otherwise you are in actuality tampering with your sample - it would be like (in an edtreme case) pulling a sample of 100 parts, sorting through them to get rid of any that look bad and then pulling your sample from that - your sample is not representative of the original population from which it was drawn.

David