During a recent AS9100 audit, the auditor asked if we use sampling as a means of product acceptance. We stated that we do not use sampling as a means of product acceptance, we use First Article only. Since we are a CNC machine shop, we focus on the process and prevention of defects instead of spending money trying to detect defects. As a result, dimensional defects to customers for the last 2 years has been 0. However, the auditor stated that if we do First Article, we just sampled 1 part and that is considered sampling and this would not be a statistically valid sample. I disagree with this logic. The standard states, "WHEN the organization uses sampling inspection as a means of product acceptance...". The standard does not state the organization "SHALL". Can an organization not use sampling and still perform First Article?

During a recent AS9100 audit, the auditor asked if we use sampling as a means of product acceptance. We stated that we do not use sampling as a means of product acceptance, we use First Article only. Since we are a CNC machine shop, we focus on the process and prevention of defects instead of spending money trying to detect defects. As a result, dimensional defects to customers for the last 2 years has been 0. However, the auditor stated that if we do First Article, we just sampled 1 part and that is considered sampling and this would not be a statistically valid sample. I disagree with this logic. The standard states, "WHEN the organization uses sampling inspection as a means of product acceptance...". The standard does not state the organization "SHALL". Can an organization not use sampling and still perform First Article?Your post is somewhat confusing.
Firstly, you are correct, and the auditor wrong, when it comes to the FAI vs. sampling discussion. FAI is not what AS9100 means by sampling. FAI serves the purpose to show that the product can be produced within the tolerances for the measured characteristics.

What is confusing is the part where you state that we do not use sampling as a means of product acceptance, we use First Article only.Don't you inspect the parts from the production lots? Either 100% or sampling, I have never seen an aerospace product realization process which would not require some type of product verification process. FAI is no substitute for product verification activities. That is the part that I am confused with. Can you please clarify it?

FWIW - your intial item approval isn't the same thing as sampling. However, you should be checking parts during the production run, either to perform 'sampling' of the product or to perform SPC. Unless you have done some extensive machine capability studies and your cnc machines are very capable and the process is in control, you must perform some intermediate checks on part dimensions - not everything, but those which can be co-related with others. If you are performing control charting for SPC, then that is sampling, but it's normal to take 5 items periodically.

Oh, stating that your customer never had any reported dimensional defects, is inappropriate on your part, because you don't know if they've the capability to detect them........

Since it sounds like you don't do either of these, the auditor was 'on to something', but reported it incorrectly.

Your post is somewhat confusing.
Firstly, you are correct, and the auditor wrong, when it comes to the FAI vs. sampling discussion. FAI is not what AS9100 means by sampling. FAI serves the purpose to show that the product can be produced within the tolerances for the measured characteristics.

What is confusing is the part where you state Don't you inspect the parts from the production lots? Either 100% or sampling, I have never seen an aerospace product realization process which would not require some type of product verification process. FAI is no substitute for product verification activities. That is the part that I am confused with. Can you please clarify it?Perhaps deanda is performing product verification, but not recognizing the process by that term.

Typically, high tech machine shops have a Control Plan where they have several "critical characteristics" they check while the work piece is "in process." There are two categories of "critical characteristics":

Customer dictated functional "critical characteristics" which relate to the finished product being able to meet form, fit, and function requirements.
Process verification "critical characteristics" which may have little or no relation to the final requirements of the customer, but are indicative of whether the manufacturing process is performing according to the Control Plan. For example of "process verification": my own shop once made a part where we had an automatic probe check a diameter on every part before the automatic tool changer cycled to cut a thread on that diameter. If the diameter prior to thread cutting was off from our Control Plan, it hinted that several other dimensions could be nonconforming and the entire process would have been shut down to verify whether the process was in control before producing a pile of scrap and attendant wear and tear on machines and tooling.

We also performed SPC sampling, but rarely provided those SPC samples to anyone for verification outside of our own shop. We did NOT perform "traditional final inspection" to detect nonconforming parts, since our method was "prevention, not detection."

As addressed by the others the wording being used is open to misinterpertation but here is how I see the situation in overview:

Since you state you are opperating under a AS9100 system I will base my observations on you opperating in aerospace and ultimaty under the FAA since it appears you are US based.

The FAA has very clear expectations on product verification. If the orginization that they have granted production approval to is not inspecting products conformance to design 100% then they will allow the use of c=0 based statisticly valid sample plans. This is why the FAA was a part of the ARP9013 development. They know that it is uneconimical to perform 100% inspection that in reality is no better than 85% accurate. The responsabilty for verification is typicaly passed down the food chain as there are many things that can not be verified on items that are received and that is the customers requirement for you to perform and certify.

The FAI you have suports that your process can produce one part to design. What it does not do is suport continued verification of conformance accounting for variation, failure and error impacting the process and the output. You can not use your customer as the method of determining acceptability/conformity to design. Their PO/contract to you states that it is your responsability( if it does not then you are off the hook depending on what you state on your C of C).

If you choose to not verify your product 100% then you need to have a statisticly valid sampleing plan and for aerospace is has to be based on c=0. If your customers feedback is an accurat review of all features conformance to the design than it should be fairly easy to get to very low number of actual verifications minimizing inspection cost. What you need to have is the objective data on file to suport this that the FAI by its self currently does not support.
:2cents:

We are a small CNC machine shop that is currently beginning to incorporate "In Process Inspection Criteria" data sheets along with every job. I am wondering what your feelings are as far as Final Inspection being needed along with all of the "In Process". We definitely want to stick to "prevention, not detection". Currently not all of our parts go thru Final Inspection, unless a customer requires it. But upper management is leaning towards adding inspectors so all parts can go thru Final Insp. I feel that if the operators are checking and recording their dimensions that final insp. should not be needed.

Another question, do you think we should use a sampling plan to set the inspection frequency? If so, which one? Currently we are setting the frequency per part and not sticking to any one sampling plan.

I look forward to hearing your advice!

Thanks!

We are a small CNC machine shop that is currently beginning to incorporate "In Process Inspection Criteria" data sheets along with every job. I am wondering what your feelings are as far as Final Inspection being needed along with all of the "In Process". We definitely want to stick to "prevention, not detection". Currently not all of our parts go thru Final Inspection, unless a customer requires it. But upper management is leaning towards adding inspectors so all parts can go thru Final Insp. I feel that if the operators are checking and recording their dimensions that final insp. should not be needed.

If upper management wants to spend money on inspection, it's usually because they're afraid of something bad happening. Do they have reason to be afraid? Do you have significant numbers of escapes? In general it's best to reduce dedicated inspection as much as possible, and it's a little unusual these days to hear about a company wanting to add inspectors, especially at the back door.

Another question, do you think we should use a sampling plan to set the inspection frequency? If so, which one? Currently we are setting the frequency per part and not sticking to any one sampling plan.

What sort of inspection? Final? In-process? In general, sampling plans are used to determine the size of samples (how many parts out of a lot of x parts to inspect) and not the frequency.

The Sampling Plan would be for In Process Inspection. More like a sampling frequency, Ex: Inspect 1st Piece then every 25th part.

Thanks!

The Sampling Plan would be for In Process Inspection. More like a sampling frequency, Ex: Inspect 1st Piece then every 25th part.Since this thread is under AS9100, I will assume you are manufacturing parts for aerospace applications. What you describe would violate the AS9100 requirement for sampling to be statistically valid. Arbitrarily setting the inspection to 4% of the lot is not statistically sound. Further, if parts contain key characteristics, those characteristics are expected to be inspected 100%, and often, actual readings are supposed to be recorded.

Since this thread is under AS9100, I will assume you are manufacturing parts for aerospace applications. What you describe would violate the AS9100 requirement for sampling to be statistically valid. Arbitrarily setting the inspection to 4% of the lot is not statistically sound. Further, if parts contain key characteristics, those characteristics are expected to be inspected 100%, and often, actual readings are supposed to be recorded.

Keeping in mind my ignorance of aerospace requirements, it seems to me that the idea of "statistically valid" sampling doesn't mean the same thing in production as in final (or incoming) inspection. Statistical validity in production is usually verified by control charting (using either individuals or rational subgroups) and can't be known for certain until things are underway. Also, "statistical validity" (i.e., verification of stability) is subject to change as time passes, so the idea of selecting a statistically valid sampling plan as an a priori decision makes no sense.

Our company is actually not AS9100, we are ISO9001;2000. So I guess I should of asked my question elsewhere. Sorry for any inconveinence and thank you for your replies!

Our company is actually not AS9100, we are ISO9001;2000. So I guess I should of asked my question elsewhere. Sorry for any inconveinence and thank you for your replies!

Tonya:

Actually, the advice you have here has little or nothing to do with AS9100, per se. You asked a valid questionand you got really valid answers too. Neither have anything to do with a QMS standard, per se. It's all about the correct way to go about sampling for (statistical) process control - What Dr Deming was always on about!

It's just not appropriate to pull a number out of the air to set sampling plans around. You have to do the work to determine the process capability, first, even if you use a zero defect sampling table (like Mil 105E) etc.
:2cents: