Capability Study Required as a Condition to Adopt a Sampling Plan?

Caster · Oct 14, 2007

Now here is an odd situation.

For years we have done a capability study, and submitted that as part of a package to our Big 3 customer to allow us to adopt a sampling plan.

I get a lot of flak for suggesting that the one does not need the other.

The rationale for this is lost in the mists of history. It seems to have been, first prove you're capable then we'll let you go to sampling.

As far as I know, a capbility study is not a prerequisite for sampling. Any plan can be selected (and it will of course have risk) <as will SPC>.

Any thoughts?

e006823 · Oct 15, 2007

I believe the capability study would help you select an appropriate sampling plan, see: http://www.variation.com/techlib/as-7.html

Regards,

Geoff Withnell · Oct 15, 2007

Caster said:
Now here is an odd situation.

For years we have done a capability study, and submitted that as part of a package to our Big 3 customer to allow us to adopt a sampling plan.

I get a lot of flak for suggesting that the one does not need the other.

The rationale for this is lost in the mists of history. It seems to have been, first prove you're capable then we'll let you go to sampling.

As far as I know, a capbility study is not a prerequisite for sampling. Any plan can be selected (and it will of course have risk) <as will SPC>.

Any thoughts?

Caster,
The connection is fairly straightforward. It is not the function of sampling in a capable process to detect defects! In a capable process, the function of sampling is to verify (obviously with some degree of risk) that the process remains capable. Given that the process remains capable, the level of defects will, by the very definition of capable, remain acceptably low. If the process is already known to be not capable, then we do not need to sample to detect the level of defects, we already know from the process incapability that the level of defects is probably unacceptably high. Sampling is inappropriate, and we must therefore take other measures, such as 100% inspection, to reduce the level of defects.

You can see this idea in Z1.4, which requires the halt of sampling if tightened inspection continues too long, since this indicates the capability of the process is degraded.

Geoff Withnell

Caster · Oct 16, 2007

Geoff Withnell & e006823

OK, thanks to you both. I think I have it now.

The cap study is to determine what kind of defect level the process is operating at.

From that, I can then pick a sample plan with an AQL that makes sense for the process defect level.

Thanks

Caster

Bev D · Oct 17, 2007

Capability is not necessary for sampling plans to work - this is a myth.

Sampling plans are based on the probability of detection given any defect rate. The stability - or capability - of the process has nothing to do with the probability of a sample detecting a specified level of defects. The random selection of the samples and the definition of the "lot" from which one is sampling do affect the effectiveness of the sampling plan.

The 'need' for a capability study prior to implementing a sampling plan is a belief of the Customer or regulatory agency that if the process is incapable or unstable, then they are better protected from receiving defects if you use 100% inspection/testing rather than a sample plan (no sample plan is 100% effective in detecting defects). So if your process is incapable (known to produce defects) or unstable (could prodcue defects at any time) then you should use 100% inspection/testing. If it is stable AND capable (yes - I know you cant' truly be capable if you aren't stable but many people forget this) then you can institute a sample plan at final acceptance...at least that's the theory...

This is a result of the "zero defects" philosophy as opposed the "acceptable defect rate" philosophy that the Mil Std plans were originally developed for.

Geoff Withnell · Oct 17, 2007

Bev D said:
Capability is not necessary for sampling plans to work - this is a myth.

Sampling plans are based on the probability of detection given any defect rate. The stability - or capability - of the process has nothing to do with the probability of a sample detecting a specified level of defects. The random selection of the samples and the definition of the "lot" from which one is sampling do affect the effectiveness of the sampling plan.

The 'need' for a capability study prior to implementing a sampling plan is a belief of the Customer or regulatory agency that if the process is incapable or unstable, then they are better protected from receiving defects if you use 100% inspection/testing rather than a sample plan (no sample plan is 100% effective in detecting defects). So if your process is incapable (known to produce defects) or unstable (could prodcue defects at any time) then you should use 100% inspection/testing. If it is stable AND capable (yes - I know you cant' truly be capable if you aren't stable but many people forget this) then you can institute a sample plan at final acceptance...at least that's the theory...

This is a result of the "zero defects" philosophy as opposed the "acceptable defect rate" philosophy that the Mil Std plans were originally developed for.

It is equally a myth that the Mil Std plans were originally developed for an "acceptable defect rate". The AQL is NOT permission to ship defective product. It is a convenient label for how much sampling inspection is warrented based on the consequences of not discovering a negative shift in quality. If you read the entire standard, it fairly clear the intent of MIL-STD-105 and its successor standards is not so much to accept/reject specific lots (although it obviously does that) as to detect a negative change in the process quality.

Geoff Withnell

Caster · Oct 17, 2007

Bev D said:
Capability is not necessary for sampling plans to work - this is a myth.

Bev D said:
Sampling plans are based on the probability of detection given any defect rate. The stability - or capability - of the process has nothing to do with the probability of a sample detecting a specified level of defects. The random selection of the samples and the definition of the "lot" from which one is sampling do affect the effectiveness of the sampling plan..

Bev D

This is more like the answer I was fishing for.

As I understand sampling, the OC curve tells me the producer and consumer risks for any given defect level in a lot. All in all done.

I did however like the idea of doing a cap study to understand my process defect rate and then selecting a sampling plan based on that. Very elegant.

In this case I think you are correct, the customer is looking for 100% protection from sampling so has come up with this mixed marriage of related cousins.

Cheers

Caster

Capability Study Required as a Condition to Adopt a Sampling Plan?

Caster

An Early Cover

e006823

Geoff Withnell

Caster

An Early Cover

Bev D

Heretical Statistician

Geoff Withnell

Caster

An Early Cover

Similar threads