Big Batch (Large Lot) Sampling - Label Printing - MIL-STD-105E

Hello everyone!

I'm looking for help in composing a sampling plan for AQL purposes. I must admit that statistics aren't my stong point, MIL-STD-105 is beyond my comprehension So i could use some help in establishing the samplig tabel needed. We already have a proposal for a sampling plan but it involves checking in a huge number of pieces, like 2500 for a 1.000.000 pieces lot. Wich means inspecting 10-15 thousands pieces per shift. And since we only have one inspector per shift ... My company is involved in label printing so there are no special legal requierments for quality control. We just need it to prevent customers from claiming nonconformity of product for ridiculous percentages of defects (we have experienced return of product for 300 defective pieces in 210.000 pieces lot. And we're speaking about labels, not aireplane parts).
Qur AQL should be between 1.5 - 2.
Lots are anything betwen 10.000 an 1.000.000 pcs.

Thanks

Re: Sampling help needed

Try the calculator from sqconline.

You need to input 4 items: batch size, AQL level you want, inspection level you want and type of inspection. There are self explanatory notes there. I think you'll get much lower figures than the one you quoted.

Re: Sampling help needed

Great help. Thanks.

Re: Sampling help needed - MIL-STD-105E for big batches

I think the specified AQL of 1.5%-2.0% is too loose. You should use AQL of 0.15% since your process average is 0.14% (300/21,000). This will give customers better protection of receiving lots with defective rate greater than 0.15%. Below are sampling plans worked out based on Harry's recommended SQCONLINE calculator.

AQL-0.15%, Single Sampling Plan, Normal, Level II:

1. Lot Size N=10,001~35,000, sample Size,n = 315, accept lot is defectives found is 1 or 0. Reject lot if defective found is 2 or more.

2. Lot Size N=35,001~150,000, Sample Size, n=500, accept lot if defectives found is 2 or less, and reject lot is defectives found is 3 or more.

3. Lot Size, N=15001~500,000, Sample Size, n=800, accept lot if defective found is 3 or less, and reject lot if defectives found is 4 or more.

4. Lot Size, N>500,000, n=1250, accept lot if defective found is 5 or less, and reject lot is defectives found is 6 or more.

The OC curves will tell you the probability of a given lot will be accepted at a specific quality level of a lot submitted for sampling inspection.

By the way, you should fix the lot size, so that you can use only one samplling plan.

Eric Ng

Re: Sampling help needed - MIL-STD-105E for big batches

Just out of interest. How are these labels printed? How are the samples checked?, guessing the label come in rolls - with the entire batch on one or more rolls. What type of defects is the inspector looking for? Do defects occur as one offs such as a smudge or could it be a substantial number e.g. ink running out etc.

Re: Sampling help needed - MIL-STD-105E for big batches

I posted a spreadsheet a while back that might help ...
http://elsmar.com/Forums/showthread.php?t=12836


It allows you to input AQL (the level that should be accepted), RQL (the level that should be rejected), alpha (the odds of rejecting a good lot) and beta (the odds of accepting a bad lot). [NOTE: the lot size is not included! As long as the lot size is considerably larger than the sample size, the odds will be right. And actually small lots will improve the alpha & beta, so you can't really go wrong).

Then you can adjust the sample size until you get a statistically sound sampling plan.


For example, if you are willing to reject lots with 0.15% defects 5% of the time and willing to accept lots with 1.5% defects 5% of the time, the spread sheets tells you
* sample size: 199-237; accept: 0; reject: 1
* sample size: 315-417; accept: 1; reject: 2

NOTE1: The second is about the same as the the first plan suggested by Eric in his post.
NOTE2: there are no acceptable plans where you accept 2 or more.


Of course, as you change any of the parameters, you will get different results.

Tim F

Re: Sampling help needed - MIL-STD-105E for big batches

wallen said:

Just out of interest. How are these labels printed? How are the samples checked?, guessing the label come in rolls - with the entire batch on one or more rolls. What type of defects is the inspector looking for? Do defects occur as one offs such as a smudge or could it be a substantial number e.g. ink running out etc.

Click to expand...

You are right, labels are printed in rolls. Very large rolls, 2-5000m. But customers request that they are not delivered to them in such big rolls, it would be impractical (they can weight 15-80kg). So we cut them after printind and roll them in smaller rolls, 50-200m.
We need to check each label to match the customer aproved proof, no missing or unmatching colors, spots, smudges, etc.
The checking rolls of label is done on inspecting harware. The difficult part is inspecting aluminium lids and plate labels (labels not rolled). When they get printed the printing machine doesn't do any checks, only the operator. Then they get cut in different shape an sizes depending on customer request. We don't have a machine that can inspect such a diversity of shape and sizes, so we have to relly on the quality inspector, and production people. Plus, the aluminium is reflective and fools the machine in wrongfully identifing defects. And this can only be done by sampling.
The worts aspect of this is that usualy defect don't occur on random labels/lids, but rather in groups. So when we sample the lots, we can miss such groups of bad products. on the other hand, we can't visualy inspect 4.000.000 labels/lids a day.

Re: Sampling help needed - MIL-STD-105E for big batches

The worts aspect of this is that usualy defect don't occur on random labels/lids, but rather in groups.

Click to expand...

That's what I guess and also what I know about this industry. If defects don't occur in random, then random sampling (AQL type of check) doesn't work.

'If defects occur in groups' - I think you mean batches or sub-batches if you want to put it that way then, first piece check with random check in between may help.

Re: Sampling help needed - MIL-STD-105E for big batches

harry said:

That's what I guess and also what I know about this industry. If defects don't occur in random, then random sampling (AQL type of check) doesn't work.

'If defects occur in groups' - I think you mean batches or sub-batches if you want to put it that way then, first piece check with random check in between may help.

Click to expand...

Ok. I understand the reasons why random sampling may not work. And I have a first piece control. But I can't stop a printing press running 100m/minut for product inspection. I must inspect after production is finished.
I'm really in need for suggestion on how to approach quality control without random sampling and without 100% inspection.

Re: Sampling help needed - MIL-STD-105E for big batches

HiroMatsu said:

Ok. I understand the reasons why random sampling may not work. And I have a first piece control. But I can't stop a printing press running 100m/minut for product inspection. I must inspect after production is finished.
I'm really in need for suggestion on how to approach quality control without random sampling and without 100% inspection.

Click to expand...

Have you checked or discussed with the supplier or manufacturer of your printing press? At such speeds, you'll need scanner or laser type of inspection gadgets and they are normally very reliable.