Calculating Sample Size for very low AQL

[nate_9]

Hi,

This is my first time posting in the cove. I have a unique situation where my AQL is 0.002% and I am trying to calculate a sampling size from a lot which is above 500,000 cups. My Ac and Re will be 0 and 1, respectively (since these numbers cannot go below anymore). I do not want to do 100% inspection and the Mil-Std-105E's chart does not go below 0.01% on AQL.

Any suggestion on my options? Is Mil-Std-105E the best option?

Thanks
Nate

[apestate]

Who set the AQL at .002%?

Also, are you inspecting all features or certain critical ones? Are there variables data or is it all attributes?

Is there anyone demanding that you do acceptance sampling instead of other tools that maybe don't rely on inspection to provide quality?

[Bev D]

well your sample size is pretty large (~2560, but I dont' have my all my spreadsheets so I could be wrong) however it wil lbe pretty large.

20 ppm is really small for categorical data plans. will your customer accept other approaches such as continuous data sampling coupled with SPC or perhaps mistake proofing? inspecting for this quality level is not a good approach.

[Statistical Steven]

Bev...2543 to be exact

I am also weary of such large sample sizes since attribute sampling plans are not very effective (i.e. inspector fatigue, etc.). Reducing the sample size to 500 still gives you an AQL of 0.01%

[nate_9]

Thanks for all the suggestions so far.

The situation is like this:-

We are planning to modify our product to contain more volume of liquid. However by doing this, we may jeopardize the seal integrity of the package, potentially inviting more consumer complaints relating to leaking product. A trial has already been ran with a lot size of 500,000 containers, but waiting on sampling and testing.

Our current product (control) has a historical leaking defect rate of 0.002% (1 defect per 50,000 containers), which I am considering as the AQL for the above product. My approach to this situation is to compare the new product with the existing and measure pass / fail by attribute sampling. I do not know if this is the best approach.

Bev / Steve, how did you calculate '2543' as the sample size? Talking about inspector fatigue and other issue with testing this sample size, I do not think will be an issue since our containers are small enough that we can test ~200 cups in 2 minutes under a vacuum chamber.

More clarification / suggestions are appreciated.

[Statistical Steven]

In Reply to Parent Post by nate_9

Thanks for all the suggestions so far.

Our current product (control) has a historical leaking defect rate of 0.002% (1 defect per 50,000 containers), which I am considering as the AQL for the above product. My approach to this situation is to compare the new product with the existing and measure pass / fail by attribute sampling. I do not know if this is the best approach.

The AQL does not always equal the defect rate of the "control" product. You have to ask what is the maximum defective that the customer finds acceptable.

In Reply to Parent Post by nate_9

Bev / Steve, how did you calculate '2543' as the sample size? Talking about inspector fatigue and other issue with testing this sample size, I do not think will be an issue since our containers are small enough that we can test ~200 cups in 2 minutes under a vacuum chamber.

More clarification / suggestions are appreciated.

Solve the binomial equation for n setting alpha equal to 0.05 and p=0.00002 (0.002%) with c=0.

[Bev D]

a clarification question: are you looking for the sample size for qulaifyign that the new process is no more defective than the old process? (a one time event)

or are you looking for the proper sample size for release testing of the new process (an ongoing, lot to lot process)

these are two different things and will yield two different answers...


Steven - that goodness I can still math in my head!

[nate_9]

Bev,

I am looking for the sample size for qualifying that the new process is no more defective than the old process.

Thanks
Nate