TS clause 7.1.2 states, “For attribute data sampling, the acceptance level shall be zero defects.”

I always think zero defects means no defect, but I see some people use the following sampling plan. Isn’t it strange that on the one hand you say zero defect, on the other hand, you say AQL?

Perhaps the phrase “attribute data” should be redefined, so that minor attribute characteristics that customers don’t care much don’t count, and limit the major defects as attribute data?

How do you think

I am guessing but this could mean zero defects in the sample - I have seen some 0-1 sampling plans, dont know how well they work in real life though.

You can also of course define what you call a defect as you suggest, only to those things that will cause an actual complaint/return.

The meaning is that C=0, C is the reject number.

What is missing or it might be cut of in the chinese is that if 1 defect is found the lot is rejected.

The AQL is needed to decide on the amount of risk that you want to take.

The page that you show looks like it is from Nicholas L. Squelia, 4th edition : Zero Acceptance Number Sampling Plans

see here http://elsmar.com/Forums/showthread.php?t=4880 and other related threads quoted there.

As to Perhaps the phrase “attribute data” should be redefined, so that minor attribute characteristics that customers don’t care much don’t count, and limit the major defects as attribute data?
This always begs the question who defines the importance of an attribute

Zero defects means one cannot use an AQL and if you find any "defects" in your sample, the lot is rejected. It doesn't matter which sampling plan you use, no "defects" are acceptable.

TS clause 7.1.2 states, “For attribute data sampling, the acceptance level shall be zero defects.”

I always think zero defects means no defect, but I see some people use the following sampling plan. Isn’t it strange that on the one hand you say zero defect, on the other hand, you say AQL?

Perhaps the phrase “attribute data” should be redefined, so that minor attribute characteristics that customers don’t care much don’t count, and limit the major defects as attribute data?

How do you think

A little history might help to explain the apparent contradiction. The widespread use of statistical sampling plans originated with U.S. military requirements at a time when mass inspection was the order of the day. In hopes of expediting the receipt of material and still having a reasonable level of confidence in product quality, it was decided that statistical sampling was the way to go. Contrary to popular belief, the "acceptance" numbers associated with AQLs were not intended to convey the notion that the customer would "accept" (pay for) defective product; it was another method of expediting shipments. The idea was that if a properly sampled lot resulted in finding x defectives in the sample, the lot could be shipped and the customer would assume the responsibility for whatever further weeding out of defects might have to occur. In most cases, customers reserved the right to debit suppliers for the defective product.

At some point, with no little influence from Phil Crosby, the idea that defects were "acceptable" became anathema, and customers generally put the onus of culling defectives back on the suppliers. Thus the C=0 concept was born. This didn't mean (and still doesn't today) mean that the familiar old sampling plans were discarded; it just meant that the acceptance numbers were. The confidence levels derived from the sampling plan AQLs were/are used as a way to assess the risks associated with sampling. While this might not make much sense, and there are C=0 sampling plans available, the old standards are still very much in use.

Now--as for attibutes inspection, the general automotive expectation is that attributes gages will always detect "bad" product and always accept the "good" ones. This differs from variables gaging wherein some defined level of error is considered acceptable. So when the TS 16949 standard refers to attributes data sampling in 7.1.2, it's referring (tangentially) to the measurement system. It could have been stated more clearly :mad: but it's not intended to mean that sampling plans for variables shouldn't also be based on the C=0 criterion.

As an aside, a basic tenet of statistical sampling--that of randomness of the sample--is widely abused, rendering the reliability of the plans generally useless. Most people who do the sampling aren't even aware of the meaning of "random" in it statistical sense--that each member of the lot being sampled must have an equal chance of being chosen. They think that "random" consists of closing your eyes (figuratively) and taking x number of pieces from the top of the pile. If you don't give your sample a fighting chance to be actually representative of the lot, it probably won't be.

I reconsidered this requirement. I think the trick may lay in the word " defect".

ISO9000:2005 clause 3.6.3 define "defect" as "non-fulfillment of a requirement related to an intended or specified use."

So if the customer doesn't put a requirement for an attribute characteristic, this attribute charactristic is born zero-defect.

For example:

Assume the product is a panel, the manufacturing process may add some black dots on the panel, and the black dots are not good-looking.

If the customer doesn't have a requirement on black dots, then the panel is zero-defect in terms of black dots.

If the customer says the number of black dots on one panel must not exceed 5. And we supply this customer 100 panels. Every panels have some black dots, but no panel has more than 5 black dots, then this 100 panels are zero-defect in terms of black dots.

How do you think?

I reconsidered this requirement. I think the trick may lay in the word " defect".

ISO9000:2005 clause 3.6.3 define "defect" as "non-fulfillment of a requirement related to an intended or specified use."

So if the customer doesn't put a requirement for an attribute characteristic, this attribute charactristic is born zero-defect.

If the customer doesn't define requirements for a given characteristic or feature, then the the possibility of defects doesn't exist, so "zero defects" in this context makes no sense. In other words, in order to achieve a level of zero defects, there has to be a possibility for defects to exist in the first place.

For example:

Assume the product is a panel, the manufacturing process may add some black dots on the panel, and the black dots are not good-looking.

If the customer doesn't have a requirement on black dots, then the panel is zero-defect in terms of black dots.

If the customer says the number of black dots on one panel must not exceed 5. And we supply this customer 100 panels. Every panels have some black dots, but no panel has more than 5 black dots, then this 100 panels are zero-defect in terms of black dots.

If you look at it only from the customer's perspective, then it's fair to characterize your example as a zero-defects situation. But it's also possible that in order to avoid having any black dots whatsoever, the supplier might define a "defective" as any panel having even one black dot. It's not unusual for suppliers to have production requirements that are more stringent than the customer's, so it depends on whose definition of "defect" you're dealing with.

How do you think?
This is a question I often get from my wife, and the answer is a closely-guarded secret. :tg: