C=0 Sampling for Variable Characteristics

[Douglas E. Purdy]

I have been doing a lot of reading [internet searches] recently and I am embarrassed to even present this thread, but I need feedback from my knowledgeable and experienced Cove Buddies! [Please be Gentle!] Is there any validity in utilizing a C=0 Acceptance Sampling Plan [meant for attribute characteristics] for variable characteristics by QC Inspection of a lot of machined product that has been monitored and measured by the machine operator at some designated periodic inspection [e.g., Every 5 or 10 pcs] that is not controlled in the traditional SPC [X-Bar & R-Chart] manner?

 

[Michael_M]

I have always understood "C=0" to mean that during the AQL inspection, if any parts are found to have a bad (out of tolerance) attribute then you have to inspection 100% of all parts on that attribute and finish the actual AQL as well.

 

[ScottK]

Validity is in the eye of the customer. Who is the customer and what do they think?

I would say that since you are gathering in-process data, you should chart that rather than do an AQL inspection at the end. It's redundant if you can can show control. If the process is not in control then you may need both until you learn how to control the process.

As to using an attribute plan for variable characteristics - It my not be "correct" but in my experience in many different factories it is not at all unusual to treat a variable charactersitic as an attribute by saying if it's between x an y units it passes, if its out of x and y it fails.

Of course different industries may feel differently about this... I'm in medical device (class I very low risk) now and have also done this in drug and food packaging as well as general industry. When I was in a machining plant we did SPC charting and were able to eliminate "final" batch inspection of individual components altogether.

 

[Douglas E. Purdy]

I have always understood "C=0" to mean that during the AQL inspection, if any parts are found to have a bad (out of tolerance) attribute then you have to inspection 100% of all parts on that attribute and finish the actual AQL as well.

Michael,

I do not quite understand what you mean by "and finish the actual AQL as well", unless it means to complete the inspection for the sampled product for all the identified characteristics to be inspected. Yes, when we find a "defective" for the given attribute(s) / characteristic(s) we then perform 100% inspection of the lot for the defective attribute(s) / characteristic(s).

Doug

 

[Douglas E. Purdy]

Validity is in the eye of the customer. Who is the customer and what do they think?

I would say that since you are gathering in-process data, you should chart that rather than do an AQL inspection at the end. It's redundant if you can can show control. If the process is not in control then you may need both until you learn how to control the process.

As to using an attribute plan for variable characteristics - It my not be "correct" but in my experience in many different factories it is not at all unusual to treat a variable charactersitic as an attribute by saying if it's between x an y units it passes, if its out of x and y it fails.

Of course different industries may feel differently about this... I'm in medical device (class I very low risk) now and have also done this in drug and food packaging as well as general industry. When I was in a machining plant we did SPC charting and were able to eliminate "final" batch inspection of individual components altogether.

Scott,

You hit the proverbial 'nail on the head' in so many ways with your response.

Thanks!

Doug

 

[Michael_M]

Michael,

I do not quite understand what you mean by "and finish the actual AQL as well", unless it means to complete the inspection for the sampled product for all the identified characteristics to be inspected. Yes, when we find a "defective" for the given attribute(s) / characteristic(s) we then perform 100% inspection of the lot for the defective attribute(s) / characteristic(s).

Doug

Yes, I personally thought it was obvious but one of our old customers kept making a HUGE point that when you do AQL and find a bad dimension you have to finish doing the AQL as well (spent about 20 minutes on the phone and 3 separate e-mails telling me this).

The customer did not want sampling inspection, they wanted 100% inspection but were not willing to pay for it (thus they are not our customer any longer). Any time the word AQL or sample inspection came up they went off on a tirade.

The entire event sticks with me to this day so I think I was channeling those events badly.

 

[Bev D]

Like Scott I have been several industries from aerospace to medical devices. measuring a 'variables' characteristic usind an Attributes approach to sample size and converting the results to pass fail (rather than using distributional statistics to predict the number that may be out of tolerance) is very very common. in fact I'd say its typical. Your Custeomr may not accept it? but I can't udnerstand why - unless I've misinterpreted your situattion.

I would add one clarification to Scott's comments about demonstrating control: the requirement is to demonstrate control AND capabliity. remember processes can be horribly out of spec and yet still in statistical control...

 

[Mike S.]

I would say that since you are gathering in-process data, you should chart that rather than do an AQL inspection at the end. It's redundant if you can can show control. If the process is not in control then you may need both until you learn how to control the process.

I respectfully disagree. What if the process is in control but does not meet the customer's spec?

Ideally you are in control and meet spec. Some customers will demand you show via AQL-type sampling that you meet spec. That is no reason not to do control charting of course!

Many customers and industries may demand C=0 AQL type inspection.

 

[David-D]

I'll try to provide an answer in a different direction than the process control vs acceptance sampling debate (i believe in both... sometimes) and instead comment on attribute and/or variable sampling plans.

For any attribute sampling plan, you can create an equivalent variable sampling plan by comparing the equivalence of the Operating Characteristic (OC) Curves of the sampling plans. In doing so, you can create matched (or families of matched sampling plans). Many industrial sampling standards are already constructed this way; for example, ANSI/ASQ Z1.4 (for attributes) and Z1.9 (for variables) are matched (they were copied from the previous MIL-STD-105 and MIL-STD-414). Some of the plans within do happen to be c=0 plans. For pure c=0 plans, MIL-STD-1916 contains both accept on zero (AOZ) attribute plans and corresponding matched variable plans (as well as continuous plans); MIL-HDBK-1916 provides the underlying OC curves for reference. There is also an ISO spec that is basically equivalent to 1916.

The short answer is that for whatever your sampling plan (especially if it is an actual AQL) it should be pretty easy to look up or derive (via OC curves) an equivalent variable sampling plan.

As a further aside, most of the c=0 variable plans I know of require both meeting the variable analysis (Cpk by any other name) and not finding any nonconforming units.

David

 

[toffeeman]

Bump

We have been using the C=0 Sampling plan for all Lot By Lot incoming inspections, where variable dimensional data is recorded. Where a dimension is found to be out of specification on one sample, the batch is either rejected or 100% inspected.

Can anyone provide an acceptable rationale for using C=0 for variable date instead of a sampling plan such as ANSI Z1.9?

Thanks