Sampling Plan Justification and Rationale

[jlshipley]

I have been tasked to create a sampling rationale for my company. Over the years no one has been able to justify the sampling plan we are using and would like for me to develop that plan. Along with our our written procedures we follow Z14 and Z19 sample plans. We currently test 2 samples -according to AQL of 0.65 (which is our standard) we should be pulling 32 samples. If the AQL were 6.5 we could pull 2 samples and be OK. My boss does not want to change the sample size we are using he wants me to justify what we are doing. I have calculated this stuff 500 different ways - our procedure would allow us to use capablity instead of AQL but we are not capable either.

My question is does anyone have an example of a rationale they have made or have any advice or examples of how I can justify something that I currently can't. The facts I have that seem to make it OK is the fact we have been running this product for 5 years with (0) customer complaints for this particular test.

Please help!!!

Thanks in advance,

 

[Bev D]

Re: Sampling Plan Rationale

well, I thought you were in the proverbial no-win space until your last statement. (I can't justify the unjustified unless I make up stories).

If your customer hasn't complained, perhaps your specification is too tight and you are in fact capable to the limits that your customer actaully needs.
I woudl pursue that avenue.

 

[jlshipley]

Re: Sampling Plan Rationale

Bev D.,

Thanks for your response - that is kinda what I figured - I think my 'Boss' is apprehensive that he does not have a better explaination then that, LOL.

Again - Thank you.

 

[btive]

Re: Sampling Plan Rationale

I would like to take this a step further and ask how you picked the AQL's you picked. I am being asked to justify our AQL levels and I am not sure which direction to go. With thousands of products how do you pick out a risk level for each one. It seems that an AQL is mostly a business decision. Any thoughts on how to approach this?

 

[Bev D]

Re: Sampling Plan Rationale

well it's certainly an informed business decision.
the AQL and RQl shodul be chosen based on the severity of the defect. severer defects that woudl cost a lot of money, turn the customer against you and/or injure people should have very low rates while hardly noticable cosmetic defects might have fairly high rates.

remember that the AQL is the defect rate that you will ACCEPT and ship 95% of the time
the RQL is the defect rate you will detect and REJECT 95% of the time.

I set my RQLs based on severity of the defect and my AQL at the process capability level - as long as the process is cabable to a level below the RQL; if it's not I only use the RQL until the process becomes capable.

 

[jlshipley]

Re: Sampling Plan Rationale

There are several methods you could use - if you already have some chosen AQL's that exist you could do a rick analysis on those to determine if your higher risk items have a higher AQL and the lower risks a lower AQL. If not I would suggest using an AIAG type of PFMEA rating to determine your risk level. Once you determine your risk - or rather risk potential then you can get a better idea of what level to set - there is no exact way to say I make widget 'A' so I use an AQL or 1 - it needs to be a logical based decision - also look at the amount of returns that and or complaints for each of your products - a quick simple way of creating AQL's are the number of complaints/defect divided by the number produced. Then determine how many parts you would have to test to have at least a minimum of 95% confidence that you have tested enough parts. Hope this helps?

 

[Bev D]

Re: Sampling Plan Rationale

...also look at the amount of returns that and or complaints for each of your products - a quick simple way of creating AQL's are the number of complaints/defect divided by the number produced. Then determine how many parts you would have to test to have at least a minimum of 95% confidence that you have tested enough parts. Hope this helps?

My caution here would be that if you set your AQL equal to your field defect rate, remeber that your inspection will accept and ship that rate 95% of the time - your inspection will not improve the field defect rate. If you set your RQL equal to the field defect rate you will reject that rate 95% of the time and will then have some improvement in the field defect rate. To achieve substantial improvement (through inspection) in the defect rate in the field you would need to set your RQL at a rate lower than teh field rate...

 

[jlshipley]

Re: Sampling Plan Rationale

Bev D,

Absolutely correct - I should have clarified better - it would only at a minimum give you the same result you have today.

 

[dohertgj]

It is our understanding that FDA is looking for a link between your PFMEA and your inspection levels (i.e. based on severity rating on the PFMEA). Can anyone speak to this and explain in a little more detail how you have gone about this. Thanks so much.

 

[Bev D]

It is our understanding that FDA is looking for a link between your PFMEA and your inspection levels (i.e. based on severity rating on the PFMEA). Can anyone speak to this and explain in a little more detail how you have gone about this. Thanks so much.

It's really fairly straightforward. Severe effects - or hazards- would require very small RQLs as you wouldn't want many of those to escape. Higher RQLs would be appropriate for Less severe effects like those created by cosmetic defects.

Technically the phrase 'inspection level' is a mil-std or ANSI phrase that refers to loosened (smaller) or tightened(larger) sample sizes based on past history.

A smaller sample size will increase the probability that you might miss a defect rate equal to your RQL if it were present so If your process has been very capable and stable then a loosened sample size is warranted as you can't miss what isn't there. If your process has a slight shift in the bad direction it might take you 2 or even 3 lots to detect it (assuming you are not using statistical process control) and if tech process has a large negative shift even the smaller sample size will catch it...

If your process is marginal to the RQL you probably want a tightened sample plan - larger sample size- to ensure that you catch a bad lot.

The statistical driver of loosened and tightened plans is the 'confidence level' of the plan. The confidence level for normal or general plans is 95%, meaning that you will detect a lot that has a defect rate equal to your RQL 95% of the time when it is present. Your probability of detection increases as the defect rate increases for the same sample size. however this choice is also based in large part on the severity of the defect. 95% is only chosen because that's what we've always done...

So if you need a tightened plan you would go to 99% confidence or even greater. For really severe defects I always choose a higher confidence level as well as a low RQL...but be ready for these to have very large sample sizes even with continuous data plans.