Changing Sampling Accept Number to Zero, Arbitrarily

[WCHorn]

Paragraph 8.2.4 of AS9100C cites "When the organization uses sampling inspection as a means of product acceptance, the sampling plan shall be
justified on the basis of recognized statistical principles ... "

Some SAE AS specifications, which formerly cited 1.0 AQL sampling plans to MIL-STD-413 or its successor, now state to use that same sampling plan with the exception of changing the accept number to zero for all sample sizes.

To me, that invalidates the statistical basis of the sampling plan and creates a nonconformance to AS9100C. What do you all think? Does altering the accept number in a traditional sampling plan invalidate the statistical basis for the plan?

 

[MTProcessing]

My first thought is, what kind of production process views a nonconformance as an acceptable condition? I've never worked on any project where a customer said "We're OK with 2% of the products we're selling failing due to manufacturing defects."

On second read, this isn't some lone QA person changing this, the change is in a group of SAE specifications that were researched and composed by a committee of industry representatives? As a random internet poster, I wouldn't presume to judge that the majority of all the applicable committee members got their statistics wrong.

 

[Steve Prevette]

Not knowing the exact wording of the SAE AS specs, the impact of moving the sample size from a sampling plan where say I accept on 2 defects, reject on 3 to rejecting on 1 defect is:

1. The number of "false" rejections against the original AQL will go up. Obviously, lots that would have passed under the old plan that had 1 or 2 defects will now fail.

2. The number of "false" acceptances will go down.

Basically, there is a transfer of risk from the receiver to the supplier.

Now, I could point out that if you are willing to go with "c=0" sampling plans, you might as well just construct the c=0 sample size and save yourself some money on sampling.

 

[WCHorn]

Not knowing the exact wording of the SAE AS specs, the impact of moving the sample size from a sampling plan where say I accept on 2 defects, reject on 3 to rejecting on 1 defect is:

1. The number of "false" rejections against the original AQL will go up. Obviously, lots that would have passed under the old plan that had 1 or 2 defects will now fail.

2. The number of "false" acceptances will go down.

Basically, there is a transfer of risk from the receiver to the supplier.

Now, I could point out that if you are willing to go with "c=0" sampling plans, you might as well just construct the c=0 sample size and save yourself some money on sampling.

One of the specifications I?m reading stipulates that ?The sample size shall be in accordance with inspection level II of ANSI/ASQ Z1.4 with AQL of 1.0 except that the acceptance number shall be zero.? Before it was revised, it simply said ?sampling shall be in accordance with inspection level II of ANSI/ASQ Z1.4 with AQL of 1.0.?

I concur with your conclusion and recommendation. However, I don?t have enough leverage with the governing SAE committee (which is dominated by ?users?, I am from the ?supplier? side) to convince them of my belief that a) using this method of changing sampling plans is a poor decision compared to using a statistically valid C=0 sampling plan (like Squeglia?s C=0 plans or the SAE?s own ARP 9013/1), as you point out, and b) the change that they incorporated was a quantum shift in quality expectations (even though I demonstrated it with operating characteristics curves).

Thanks for your response.

 

[WCHorn]

My first thought is, what kind of production process views a nonconformance as an acceptable condition? I've never worked on any project where a customer said "We're OK with 2% of the products we're selling failing due to manufacturing defects."

I don't quibble with the superiority of C=0 sampling plans over AQL sampling plans and 100 percent conformance; I'm down with that. In high-volume manufacturing plants though, sampling is extremely important and must be well thought out.

On second read, this isn't some lone QA person changing this, the change is in a group of SAE specifications that were researched and composed by a committee of industry representatives? As a random internet poster, I wouldn't presume to judge that the majority of all the applicable committee members got their statistics wrong.

All governing bodies make mistakes occasionally or sometimes make decisions based on expedience. In this case, I believe there is an inconsistency between AS9100 and the actions this particular committee took.

 

[Mike S.]

Paragraph 8.2.4 of AS9100C cites "When the organization uses sampling inspection as a means of product acceptance, the sampling plan shall be
justified on the basis of recognized statistical principles ... "

Some SAE AS specifications, which formerly cited 1.0 AQL sampling plans to MIL-STD-413 or its successor, now state to use that same sampling plan with the exception of changing the accept number to zero for all sample sizes.

To me, that invalidates the statistical basis of the sampling plan and creates a nonconformance to AS9100C. What do you all think? Does altering the accept number in a traditional sampling plan invalidate the statistical basis for the plan?

To answer your question, IMO it does not "invalidate" the statistical basis for the plan, it does, as Steve says, change the statistical basis -- i.e. increases the roubstness of the plan -- makes it harder for the supplier to pass a lot and gives additional protection to the consumer (changes the OC curve).

If you are in aerospace, look at the SAE ARP 9013 series of standards for additional options.

 

[Steve Prevette]

To answer your question, IMO it does not "invalidate" the statistical basis for the plan, it does, as Steve says, change the statistical basis -- i.e. increases the roubstness of the plan -- makes it harder for the supplier to pass a lot and gives additional protection to the consumer (changes the OC curve).

If you are in aerospace, look at the SAE ARP 9013 series of standards for additional options.

I suggest be careful with the phrase "increases the roubstness of the plan". The action will increase the number of false rejects. Similar as to if you go home and change the internal settings of your smoke detectors to alarm on lower values. Will this make the smoke alarm function "more robust"? Perhaps not - you'll have false alarms daily and will pull the batteries from the detectors in disgust.

 

[Mike S.]

I suggest be careful with the phrase "increases the roubstness of the plan". The action will increase the number of false rejects. Similar as to if you go home and change the internal settings of your smoke detectors to alarm on lower values. Will this make the smoke alarm function "more robust"? Perhaps not - you'll have false alarms daily and will pull the batteries from the detectors in disgust.

In many cases you are doing what the example in the other thread mentions -- essentially changing to a lower AQL -- like from 1% top 0.15% etc.

You increase the number of possible "false alarms" as well as making it alarm earlier on true "fires".

Whether or not "robustness" is the most appropriate adjective, a look at the OC curve is always appropriate to see if the plan will do what you want it to do.

But if the customer is demanding a certain plan, like Z1.4 but C=0, you can do it or not take the business, regardless of what the OC curve looks like. What the customers typically will not accept is fewer samples or higher Ac/Re numbers -- whatever you want to call that.

 

[Statistical Steven]

Let's make something very clear, any sampling plan can be justified statistically as long as it has a sample size and accept/reject number. I can give you many sampling plans with an AQL of 1%, for example n=6,c=0 (LTPD=45%) or n=37,c=1 (LTPD=12.5%). Every sampling plan has an OC curve associated with it. The OC curve defines the probability of accepting/rejecting lots at a given defect rate. If you are required to use Z1.4 then you should use the switching rules.

 

[Mike S.]

AS9100C says "...justified on the basis of recognized statistical principles and appropriate for use (i.e. matching the sampling plan to the criticality of the product and to the process capability)"

That involves more than just showing an OC curve or quoting a quality parameter like LTPD, though. IMO sampling plans should be selected with input from several groups including the customer, QA, Engineering, etc.