I need a quick answer to this:
My lot size is 501-1200, I am using an AQL of 1% at c=0. My table says take 34 samples and at c=0 if 1 fails, reject the lot. How does this relate to the Six Sigma capability of my supplier? or does it not?
My boss asked me this question. How do I reply ?
I know how to calculate Process Capability of my supplier from control charts etc...I know what Six Sigma is about. But how do AQL sampling plans relate to Six Sigma? Is there a connection here. Another way of looking at this:
If I know my supplier is at about 3 or 4 sigmas (x amt of defectives per million units produced), would this tell me what sampling plan (AQL) i should use? please help. TY.

well, I'm a little tired right now but I'll take a quick shot at it. the fast answer is that an AQL sampling plan applied ot a single lot can give you rough idea if the supplier's SHIPPED quality level is occasionally better or worse than the corresponding LTPD - remember an AQL plan means you have a 95% probability of ACCEPTING a lot that si at the AQL level or better. It's one of the things I don't liek about AQL plans. I prefer straight forward plans that give me the probability of DETECTING specified qualty levels or worse...it's easier to explain and actually describes what I'm looking to do.

However, remember the idea of determining a sigma level - and even having the supplier report their capability indexes - is to determine the suppliers 'internal' capability of producing defect free parts.

There is not a direct correlation between the two.

The sampling plan that you cited, AQL = 1%, will routinely accept up to 1% nonconforming. 1% equates to 3.83 Sigma. Therefore, the suppliers sigma level may be as low as 3.83 Sigma and this sampling plan will routinely pass product.

What is your objective? To detect deterioration in the sigma level? This plan will do that. If you are trying to force improvement by reducing the AQL (say to 0.5%), you will initially see a big upswing in rejects, and the resulting upheaval. You would have better results by developing the supplier in other ways before reducing the AQL.

The answers so far seem pretty much on track to me. I'll add a few more hard numbers.

For c=0, n = 34 and AQL = 1, your plan will accept lots with 1% defective only 71% of the time. In other words, more than 1/4 of "good" lots would be rejected. Lots would have to be down to about 0.15% defective before 95% would be accepted.

On the other hand, lots will have to be as bad as about 8.5% defective before you reject 95% of the lots. That's a lot of leaway between 0.15% defective in order to almost always be accepted and 8.5% defective in order to almost always be rejected.

Sampling (especially with a pass/fail test) is effect for discovering gross shifts in the mean. It is poor for detecting small shifts or slightly bad lots. It is tehrefore poor for establishing or monitoring "sigma level".

You could improve the odds by going to a bigger sample size (but of course that costs more). The standard Z1.4 plan (i.e. normal, level II) would have a sample size of 80 and c=2. Then you move up to 95% chance of accepting 1% defective, and 95% chance of catching 5.8% defective.

You could also improve the situation by finding a continuous variable to measure rather than just a binary pass/fail variable.


Tim F

Excellent analysis....ironically to Bev's point of the shortcomings of sampling plans, if you decrease the sample size you will get closer to accepting 95% of the lots with no more than 1% defective, but you will also you will accept some lots that have very high defective rates.

I agree with Bev's comment that we should really see a movement to more consumer risk based sampling and not producer based risk sampling plans. But of course those plans have enormous sample sizes.


The answers so far seem pretty much on track to me. I'll add a few more hard numbers.

For c=0, n = 34 and AQL = 1, your plan will accept lots with 1% defective only 71% of the time. In other words, more than 1/4 of "good" lots would be rejected. Lots would have to be down to about 0.15% defective before 95% would be accepted.

On the other hand, lots will have to be as bad as about 8.5% defective before you reject 95% of the lots. That's a lot of leaway between 0.15% defective in order to almost always be accepted and 8.5% defective in order to almost always be rejected.

Sampling (especially with a pass/fail test) is effect for discovering gross shifts in the mean. It is poor for detecting small shifts or slightly bad lots. It is tehrefore poor for establishing or monitoring "sigma level".

You could improve the odds by going to a bigger sample size (but of course that costs more). The standard Z1.4 plan (i.e. normal, level II) would have a sample size of 80 and c=2. Then you move up to 95% chance of accepting 1% defective, and 95% chance of catching 5.8% defective.

You could also improve the situation by finding a continuous variable to measure rather than just a binary pass/fail variable.


Tim F

I agree with Bev's comment that we should really see a movement to more consumer risk based sampling and not producer based risk sampling plans. But of course those plans have enormous sample sizes.

What we should see is more process control, so we don't have to worry about sampling after the fact. :D

What we should see is more process control, so we don't have to worry about sampling after the fact. :D
Any buddy help me to make a AQL in Textile industries, where different knitting process involves,
Help me in this regard.
Thanks

Hi,

I think AQL or OC curve will be related with logical representation of risk associated with sampling and how it will be minimize?

So, their are always certain RISK associated with it.

I must say u canot associated with two thing or compare with two things.