Chain Sampling Plan explanation and help - Reducing sample sizes (compared to Z1.4)

[Richard72]

Hi, I was hoping that someone might know of a resource that could offer some assistance with the concept of chain sampling.

We have a process that requires 100% visual inspection on every unit. All units can potentially suffer from 1 to 20 visual defects of differing root cause. On average, the defective percentage is 5% i.e. 5 % of all production has at least 1 of the 20 visual defects. The important criteria for us that the AOQL of any plan is 0.7%.

In trying to develop a plan(s) based on ANSI Z1.4, I used AQL's of 0.65, 0.4 & 0.25. The AOQL reduced as the AQL reduced of course but the problem is that at 0.25 AQL the AQOL was about 0.7, but here sample size exceeded Lot size so....still on 100% sampling (PS I was using sampling plan analyzer 2.0 for this).

I have now been asked to look at chain sampling as a method of reducing sample sizes (compared to Z1.4), but still maintaining an AOQL of 0.7%.

So, I have three questions I suppose;

1) If a plans AOQL = 2% at 5% defect rate, & you produce 1,000,000 units per year, does this mean that if you continue to input 5% defective parts on average (50,000), your plan will effectively filter 3% of them, thus outputting 2%? ...so AOQL = 2%

2) Are we trying to chase the rainbow here? If our product contains 5% defective parts, is there any plan that avoids 100% inspection while managing to deliver such a low AOQL (relative to incoming defects)

& Finally,

3) Does anyone know of any resource which can help me with chain sampling plans? I have a basic understanding of how they work but there does not seem to be as many examples or resources available on the web when compared to AQL based plans etc. from what i've read it looks like all the OC curves have to be manually calculated - worrying!

I'd be grateful for any assistance.

[harry]

Any help will be appreciated.

[Steve Prevette]

Do you mean "sequential" sampling? Basically you start with a small sample, and if it is really bad you immediately reject, if really good you accept. If in a gray area in the middle, you sample more, compare to thresholds and if not met, keep going.

There is the hope of a reduced amount of samples required, but if the rate is on the borderline it can take longer than a simple sampling plan. Plus some processes may not lend themselves to sample some, check the results, go back and sample more, etc.

[Mike S.]

Take a look at a C=0 sampling plan (i.e. Squeglia) and you might find the AOQL you want with fewer samples.