Hello-

I am new to the manufacturing field and have never had to work with sampling and inspection before. However, I am not new to the quality field. We are ISO 9001:2000 and pursuing AS9100. We are also audited by a company that would like us to be ISO 13485 compliant.

My question: For all the samples the Inspectors measure, do they need to show evidence of their sampling? For instance, if they have 100 items as a sample size, do they have to do 100 Final Inspections and show all measurements? The inspectors have never shown evidence of their sampling before. They will be very resistant! We now follow a C=0 plan with AQLs related to part or company.

Please let me know if you need more information. As a newbie, this particular question is weighing heavily on me. :frust:

Thank you, in advance, for your answers!

When I audit, I will ask for evidence that the sampling plan has been followed. For example, if the sampling plan calls for 5 samples at particular times of the day, well, show me that it was done.

I also look for someone to understand the 3 components of sampling: size of the sample, frequency of sampling and type of sample (consecutive, random, etc). I ask this because I want to know if someone there understands the effect of sampling on the statistics involved.

From the inspectors, I want to know if they know the sampling plan and are following it. If the sampling plan is written down somewhere, that just makes it easier, but in ISO9001 it is not required to be written down (although there may be a corporate or customer requirement for this). If the process is complex or involves many inspectors, I will ask several inspectors what the sampling plan is. If they all give me the same answer, there is no NC. If their answers vary, there is an NC.

Your post:
My question: For all the samples the Inspectors measure, do they need to show evidence of their sampling? For instance, if they have 100 items as a sample size, do they have to do 100 Final Inspections and show all measurements? The inspectors have never shown evidence of their sampling before. They will be very resistant! We now follow a C=0 plan with AQLs related to part or company.

I do not quite understand the situation. It sounds like they're taking measurements, determining that the product passes, and not recording anything other than perhaps a pass-fail.

Are they measuring all parts? I assume not. So I assume they are measuring a sample. So what kind of sample? Once a hour? 100 sequential parts? Whenever they feel like it?

The sampling plan can affect the outcome of your statistical analysis. As an auditor, I check sampling plans (for conformance) when I audit the process surrounding manufacturing of course, but also during the audit of data analysis and use of data in process monitoring.

Conformance to sampling plan: If the plan calls for 100 pieces, is that done? This can be many things, not necessarily a list of the 100 pieces. it can be a signature, for example.

Your post:
My question: For all the samples the Inspectors measure, do they need to show evidence of their sampling? For instance, if they have 100 items as a sample size, do they have to do 100 Final Inspections and show all measurements? The inspectors have never shown evidence of their sampling before. They will be very resistant! We now follow a C=0 plan with AQLs related to part or company.

I do not quite understand the situation. It sounds like they're taking measurements, determining that the product passes, and not recording anything other than perhaps a pass-fail.

Are they measuring all parts? I assume not. So I assume they are measuring a sample. So what kind of sample? Once a hour? 100 sequential parts? Whenever they feel like it?

The sampling plan can affect the outcome of your statistical analysis. As an auditor, I check sampling plans (for conformance) when I audit the process surrounding manufacturing of course, but also during the audit of data analysis and use of data in process monitoring.

Conformance to sampling plan: If the plan calls for 100 pieces, is that done? This can be many things, not necessarily a list of the 100 pieces. it can be a signature, for example.

AuditMonkey-

Thank you for both answers!

The sampling plan is written and visible in the Inspection Room. The Inspectors sample throughout the day and put their signature on the required line each time they visit. They were just recording it as PASS or FAIL, but I want them to record actual measurements. SPC is not in use yet--we are a very small shop. They randomly sample until the last part, which is for final inspection. Also, they seem to be checking different dimensions, not all, with each part they sample.

Going back to the 100 as a sample, are they required to 100% measure all dimensions for each sample part instead of some? If so, wouldn't they have to write down all dimensional measurements for evidence?:confused:

This might help you with the practice and actual audit.

Also, they seem to be checking different dimensions, not all, with each part they sample.

What dimensions are required is usually defined by the organization rather than being left to the inspector. I find it odd that they apparently decide for themselves what to measure and what not to measure. I would very probably give you a NC if I audited you - for non-robust effectiveness of the sampling plan.

Going back to the 100 as a sample, are they required to 100% measure all dimensions for each sample part instead of some? If so, wouldn't they have to write down all dimensional measurements for evidence?

There are 2 questions here.
1. What needs to be written down. If your internal procedures do not require the actual dimensions to be written down nor do your customer requirements, the pass-fail notation would be acceptable, although I strongly question the effectiveness of the choice. That seems on the surface to go back to the old goal-post mentality of SPC. (If it meets the spec, it is ok.)

2. I have rarely seen a part where all dimensions need to be checked. Usually the organization determines which dimensions are critical to control and has those measured. If you are going to implement SPC, you would need to determine those critical measurements required and yes, have those written down. Yes, I will audit to be sure they are written down. And yes, I do audit those measurements including checking the math :).

Then of course, you need to decide which statistical method to apply. You have variables so the usual suspects would be Shewhart charts or one of that family.

Hello-

I am new to the manufacturing field and have never had to work with sampling and inspection before. However, I am not new to the quality field. We are ISO 9001:2000 and pursuing AS9100. We are also audited by a company that would like us to be ISO 13485 compliant.

My question: For all the samples the Inspectors measure, do they need to show evidence of their sampling? For instance, if they have 100 items as a sample size, do they have to do 100 Final Inspections and show all measurements? The inspectors have never shown evidence of their sampling before. They will be very resistant! We now follow a C=0 plan with AQLs related to part or company.

Please let me know if you need more information. As a newbie, this particular question is weighing heavily on me. :frust:

Thank you, in advance, for your answers!

I thought sampling plans based on things like AQL values went out with the dodo (extinct) but perhaps not in the USA.? I honestly can't remember the last time I heard someone in Europe referring to AQL values. The fact that defects are "allowed" seems to go against everything I fight for. Of course mistakes are inevitable but NEVER should they be permissable. I'll bet no Japanese company or product would base their name and reputation on "allowing" defects.

To simplify (and at the same time complicate) matters a component with 20 dimensions could be out of tolerance on two relatively uncritical dimensions. The simple, logical thing to do is to note the deviation(s) and document it. The action to be taken can then be determined but at least then the error is documented. A deviation on a critical dimension would of course require more effort and action.

AQL is so old-fashioned that this is the first I have seen it or heard it some time. Even in the US :D

This might be more common in very small shops which have not had a well trained QM.

AQL is so old-fashioned that this is the first I have seen it or heard it some time. Even in the US :D

This might be more common in very small shops which have not had a well trained QM.

Whew :D my faith in the good old US of A has just been restored :applause:

If a manufacturer can't produce without defects then they should make the neccessary extra number and apply 100% inspection. That might make these suppliers wake up instead og forcing the "customer" to do an incoming inspection.

A company that needs an incoming inspection invariably has sub-standard suppliers. How about incoming inspection being placed directly under the purchasing dept.? That might make the buyer think more about total costs!

I know of one company that normally supplied their suppliers with measuring equipment (gauges etc.). At a meeting the suppliers were informed that this practice would stop as it involved many extra costs including calibration. One of the suppliers immediately stated that this would raise the price on their products. The prompt reply was, "Good, as we'll then finally know the real price for our deliveries".

I often wonder why I'm not Mr. Popular with all suppliers :notme:

Gordon and AuditMonkey,

Does that mean you would use SPC and 100% inspect all parts...no sampling?

Please elaborate about the alternatives and the way you would approach it. As many details as possible will really help me. :confused:

Thank you!

Gordon and AuditMonkey,

Does that mean you would use SPC and 100% inspect all parts...no sampling?

Please elaborate about the alternatives and the way you would approach it. As many details as possible will really help me. :confused:

Thank you!

I'll let auditmonkey speak for himself and let you know my opinion.

I'd try and avoid sampling whenever possible. To do this I'd try and get to know as much about the supplier as possible and pay a visit if the distance isn't an obstacle.

If things went by the book - which they rarely do - I'd have wanted a trial sample sent before the first delivery. Usually the supplier takes great care on this type of delivery but the catch 54 (for him) is that if he can do this once then why not every time!?

On the first delivery I'd do as thorough an inspection as I felt the product price, importance (criticallity) and time warranted. I'd assume not to find anything out of spec, but anything I did find (good and bad) would be noted and filed. I'd try to do as much measurement - SPC - (as opposed to gauge inspection) as possible to see how much of the tolerance on the various dimensions is used. If the inspection result was deemed unsatifactory then I'd return the delivery and let the supplier sort things out. Of course nowadays many companies have deliveries supplied and the supplier knows that production needs them here and now. In such a case where I had to arrange searching for and removing deviations/defects I'd inform the supplier that I'd be billing them for the work. They could of course send one of their own employees to do this work.

If you really want a medal with stars and stripes, get your purchaser to realise that what seems to be a "real bargain" (low price) usually comes at a cost. Buying (too) cheap can be very expensive :agree1:

Too many buyers think their job is to buy as cheaply as possible and that correct quality is a side issue for others to deal with. Reliable suppliers are worth their weight in gold, but are happy with cash :D

Bad designers/draftsmen are responsible for most manufacturing mistakes in production and bad purchasers reponsible for the extra costs receiving inspection incurr.

If you have any questions or comments then just let me have them. It's rare that two situations are identical (and I know it's not always as easy as I've just made it sound), so if you have something specific then let's get practical :biglaugh:

In essence, I agree.

For incoming receiving, it is best to have the SPC data from th supplier. It is even better if they are performing enough SPC to control the process (Yes, just because they dump out a chart or two, does not mean the are even controlling the correct parameters.) Attempt of SPC of an incoming distribution - especially if it is from a properly controlled precision machining operation (OK - that's not likely - but, still..) the sampling would never reveal the true distribution or its level of control. So getting it in correct time sequence is much better information.

100% inspection of all incoming parts is a very bad situation - should be avoided at all costs.

But in some industries - e.g. medical - some customers may still want to see a final inspection (c=0 sampling) of your product - which is considered equivalent to the customer's dock. Nonetheless, one of the biggest problems is failing that inspection, having to rework or remake (and typically, you are already late for the shipment.). Using SPC, understanding that production can not manufacture to full design specifications and other fundamental process controls, you can do a much better job of passing final inspection first time through.

100% inspection of all parts would only be applicable if you have a very low or very slow output or supremely critical application, I would think.

In essence, I agree.

100% inspection of all incoming parts is a very bad situation - should be avoided at all costs.
100% inspection of all parts would only be applicable if you have a very low or very slow output or supremely critical application, I would think.

I hope no-one thinks I've suggested 100% inspection in receiving inspection AT ANY TIME. :mg:

Bottom line - either you have a reliable supplier or you don't. If it's the latter find a new supplier asap!

Companies with receiving inspection usually have bad suppliers and this means lousy purchasers that aren't doing their job.

The primary function of any Quality Dept. should be to finds ways to eliminate the possibilty for faults being made. Look at the design dept. and the purchasing dept. as the main culprits for poor in-house quality. Incompetence from these two usually end up with production taking the heat - and they often have no-one to pass the buck to other than the Quality Dept.

While I'm firing my canons - why do sales people always try to sell non-standard products instead of what is usually produced?

I am always happy to see the level of intelligence about quality on this site. :applause:

I agree with Gordon on an approach to incoming inspection. I will add that I set up with each supplier a mutually agreed upon set of statistically based confirmations of their product that they need to do before shipment of that product to me. This includes sending me the statistical data with each shipment. Depending on the criticality of the product, I may or may not do my own incoming inspection. If I do, it is usually a mirror of the checks done by the supplier.

That agreement has a clearly defined set of steps to reduce their outgoing controls as their process demonstates long term statistical capability. If a product shows up that is not acceptable, that agreement also states that we go back to the beginning and start the acceptance process all over again. There have been a few instances where they have clearly demonstrated that they had a special cause that escaped their system, that they have implemented corrective action that eliminates that special cause and I have agreed not to go back to the beginning.

I am always happy to see the level of intelligence about quality on this site. :applause:

I agree with Gordon on an approach to incoming inspection. I will add that I set up with each supplier a mutually agreed upon set of statistically based confirmations of their product that they need to do before shipment of that product to me. This includes sending me the statistical data with each shipment. Depending on the criticality of the product, I may or may not do my own incoming inspection. If I do, it is usually a mirror of the checks done by the supplier.

That agreement has a clearly defined set of steps to reduce their outgoing controls as their process demonstates long term statistical capability. If a product shows up that is not acceptable, that agreement also states that we go back to the beginning and start the acceptance process all over again. There have been a few instances where they have clearly demonstrated that they had a special cause that escaped their system, that they have implemented corrective action that eliminates that special cause and I have agreed not to go back to the beginning.

At the risk of turning this particular thread into a mutually back-slapping gathering, I must admit that I agree with Audit Monkey and his approach. I can't think of a single, logical reason why any supplier would intentionally supply components that he KNEW the customer would object to. When it happens there is usually (always?) a reason.

I'll give an example from real life. I know of a (European) company the got an order from a Japanese company. On the lenght of the product they (the Japanese) specified a tolerance that was half that of standard and were willing to pay extra for this. The operators that made this particular product weren't exactly reknowned for keeping the length dimension within standard tolerance! Anyway, the first delivery was outside the Japanese specified requirement but still (miraculously) within the "old" standard tolerance.

The complete delivery was returned promptly by the Japanese incoming inspection and the company informed that they had one more chance to get it right. The production manager read the riot act to the machine operators. Now it gets interesting - it turned out that manufacturing to half the standard tolerance was no problem. It just required that the operators payed more attention and kept an eye on the process instead of "messing about" together.

"Quality is free" - in fact in this case it was profitable as it didn't cost the company extra (after they had learned their lesson) to produce to Japan who even payed more for the product.

If anyone should doubt this "story" I can give the name of the company, the name of the (well known) Japanese company and the product - but only as a PM.

Gordon and AuditMonkey,

Does that mean you would use SPC and 100% inspect all parts...no sampling?

Please elaborate about the alternatives and the way you would approach it. As many details as possible will really help me. :confused:

Thank you!

An after-thought to this. SPC can't be applied to all inspection, but I'd advocate use of measurement every time over, for example Go No-Go gauging. whenever possible.

My reason for this is that it requires a considerably smaller sample size and gives considerably more information on manufacturing capability. A simple histogram would quickly show the distribution of the delivery dimensions and how much of the tolerance was used. Two or more peaks on the histogram would probably indicate that when the batch was made, machine adjustments were necessary, or that everything wasn't manufactured at the same time. The important thing is - how within tolerance is the dimension! For example, threaded components (external and internal) should preferably be measured on the pitch diameter to see how much of the tolerance is used. The thread ring or plug gauge can still be used to determine correct thread profile and pitch. Knowing the thread was within pitch diameter tolerance before using a solid gauge would certainly reduce wear on the gauge by avoiding "force". The outside diameter of an external screw thread, and the internal hole diameter of an internal thread can always be easilly measured with a (digital) caliper.

Anyone requiring information on this should send me a PM as there are considerable cost benefits from thread measurement. I'll repeat something I wrote in another thread - manufacturing the perfect thread is a combination of measurement and solid gauge use.

All-

I'm sorry if I'm not getting my point across very well. It's not incoming inspection, but in-process inspection.

Going back to my original question: For all the samples the Inspectors measure, do they need to show evidence of their sampling? For instance, if they have 100 items as a sample size, do they have to do 100 Final Inspections and show all measurements? The inspectors have never shown evidence of their sampling before. They will be very resistant! We now follow a C=0 plan with AQLs related to part or company.

I was mainly concerned about the evidence needed to show what they measured before their final inspection. Do they write down each dimension measured, let's say for, 100 parts or as someone else mentioned, just the critical dimensions? Can we use an SPC-like sheet for them to jot down or does their evidence have to be more exact?

Gosh, I hope I'm making sense. Thank you!
:D

All-

I'm sorry if I'm not getting my point across very well. It's not incoming inspection, but in-process inspection.

Going back to my original question: For all the samples the Inspectors measure, do they need to show evidence of their sampling? For instance, if they have 100 items as a sample size, do they have to do 100 Final Inspections and show all measurements? The inspectors have never shown evidence of their sampling before. They will be very resistant! We now follow a C=0 plan with AQLs related to part or company.

I was mainly concerned about the evidence needed to show what they measured before their final inspection. Do they write down each dimension measured, let's say for, 100 parts or as someone else mentioned, just the critical dimensions? Can we use an SPC-like sheet for them to jot down or does their evidence have to be more exact?

Gosh, I hope I'm making sense. Thank you!
:D

Everything I've written about incoming inspection applies in principle to in-process inspection. A measurement sheet will probably be necessary, but keep it very simple and logical. If I was emailed a typical drawing I'd make some suggestions.

Why on earth would process inspectors require a (huge) sample size of 100 components??? Assuming the machine operators (or the guys setting up the machines) measure occasionally, get them to note down what measurement results they get and set these components to one side. The process inspectors can then check the components that are put aside.

I can only see advantages from this as the machine operators take more care and the process inspectors get more time to concentrate on the critical issues. I'm also sure that your sampling size and time will be drastically reduced. I'll bet that today it's the process inspectors that get blamed when discrepancies are found! As a guy once said to me, "Of course you can find mistakes if all you do is look for them"!

I could write much more on this as I've heard all the (feeble) excuses for doing it. When push comes to shove, make sure you're standing up-hill of the other person. :tg:

Hi - I read the posts. I would like to briefly add a couple of notes.

1st - I work for an older established copmany here in the USA. We still us AQL's and we use several mil-standard sampling plans. We do this because it works for us.

2nd - We are global. We have had fewer fights between global affiliates if we start a proccess with a mil-standard sampling plan.

3nd - If you have a new process, you need start somewhere. Even if you do not use one of the old mil-standard sampling plans, starting with something in which you have a working history. (The history piece is the important point.) Don't forgrt to ask material suppliers and equipment suppliers for their input as well as input from the production operations personnel.

4th - You can change a sampling plan, especially in the early days of a new process. The hole point of a sampling plan is to allow you to perform a cost effective inspection with a resonable use of resources. If you release items that should have been held or you continually prove that you are over inspecting, change your plan. As a general rule however, once you establish a sampling plan that works with a given process, unless your process changes, you almost never need to change the sampling plan. (This is important for doing your longterm data trend analysis.)

5th - Know why you chose to follow a given plan and be able to explain / defend it. I personally am never happy with the answer "well we've always done it that way". Operators should also know why they follow a plan. "Because my boss said so or because Quality said so..." is not an answer that holds much water.

The most important take away is this - any sampling plan must work for the organization, accomplish the objectives of catching non-conformances and releaasing good products with a resonable use of resources.

Best Wishes and Good Luck

It's not that I disagree with what you've written but I do have a couple of remarks.

"We still us AQL's and we use several mil-standard sampling plans. We do this because it works for us".

"I personally am never happy with the answer "well we've always done it that way".

Doesn't quote no. 2 tend to contradict your quote no. 1?

To me it's rather like using an old fashioned typewriter intead of a computer. I know the saying, "If it ain't broke don't fix it" but that isn't quite the same as staying in the past.

As you are global I could get the thought that maybe you are a good enough customer that foreign companies may go along with what you want just because "The customer is always right".

To me AQLs will always be associated with a relatively large allowable fault percentage or a very large time consuming sample. AQLs also support "good/bad" criteria as opposed to a "how good/how bad" criteria.

;)

It's not that I disagree with what you've written but I do have a couple of remarks.

"We still us AQL's and we use several mil-standard sampling plans. We do this because it works for us".

"I personally am never happy with the answer "well we've always done it that way".

Doesn't quote no. 2 tend to contradict your quote no. 1?

To me it's rather like using an old fashioned typewriter intead of a computer. I know the saying, "If it ain't broke don't fix it" but that isn't quite the same as staying in the past.

As you are global I could get the thought that maybe you are a good enough customer that foreign companies may go along with what you want just because "The customer is always right".

To me AQLs will always be associated with a relatively large allowable fault percentage or a very large time consuming sample. AQLs also support "good/bad" criteria as opposed to a "how good/how bad" criteria.

;)

OK - I am curious by nature. So let's play the game.
You are a new engineer, you work for a small start-up company. You are not aware of AQL's and OC curves. You have a new process to make your new product. Your threw out is too large to do a 100% inspection, where would you start to create a sampling plan? I know what I think, but I want to know what others think too.

OK - I am curious by nature. So let's play the game.
You are a new engineer, you work for a small start-up company. You are not aware of AQL's and OC curves. You have a new process to make your new product. Your threw out is too large to do a 100% inspection, where would you start to create a sampling plan? I know what I think, but I want to know what others think too.

First of all, sampling plans are specifically designed for lot sampling, that is incoming receiving and final inspection. They are of little value for in-process.

So, how do you determine sampling frequency? Process capability study. Determine what kind of variation you have (normal, non-normal, etc.) and what the rate of variation is. In process sampling is a time function, whereas lot sampling is random. That is a very important point.

Once you understand your process (and if you chose to control it correctly with SPC, then even better - you will reduce variation and overcontrol if you are machining especially), you can make sane sampling decisions. For example, you may find that the process will operate for a week until it needs adjustment or tool change. Take that time and divide by 7 (7 is a run, right?). Now look at that sampling frequency and assess the risk of a special cause (such as tool breakage). Weigh the risk of sorting and scrap versus cost of sampling. You may chose to sample every 2 hours.

Your capability study may tell you to sample every 5 minutes. Well, guess what? That is what you have to do.

But, rubber stamping a sampling plan designed to accept or reject lots as process verification? Not such a good idea at all. But, don't feel bad - lots of people do it "because it works for them" (translated: they don't know any better, and have been lucky) or "they don't know where else to start with a new process."

OK - I am curious by nature. So let's play the game.
You are a new engineer, you work for a small start-up company. You are not aware of AQL's and OC curves. You have a new process to make your new product. Your threw out is too large to do a 100% inspection, where would you start to create a sampling plan? I know what I think, but I want to know what others think too.

bobdoering has said it better than I could ;) :applause:

I'm not quite sure I got the jump from an old global company to a "small start-up company", but if I was new and/or dealing with a new process then AQLs would be about the last thing on my mind. Machine capability to manufacture to spec would be my priority and SPC is definately the road to take.

I know I'm repeating myself, but virtually alll sampling plans I know of, only reveal good or bad. If machine capability from SPC showed that the tolerance(s) couldn't be maintained, then no amount of sampling could ever solve the problem. "One measurement is better than a thousand opinions".

To put it another way, I wouldn't look for defects - I'd look for a method to avoid them even happening :evidence:

And there you go. :)

And there you go. :)

Yeps, shouting, screaming and kicking - but I'm definately going :lmao:

On a more serious note, can anyone give an example (from real life!) where an AQL sampling plan would be better than SPC?

With a tolerance of +/- 0.1 and a sample of say 25 items, a sampling plan would register an item with a dimension 0.001 above or below tolerance as a fault as "serious" as one of 0.05 and register it as a "4% error" - which is close to nonsense. A simple X-diagramme will beat an AQL sampling plan any time - in my opinion. Does anyone disagree?

On a more serious note, can anyone give an example (from real life!) where an AQL sampling plan would be better than SPC?

Actually, yes - for what is was designed for, final and incoming inspection. Some folks try to SPC that, and (especially for precision machining (http://elsmar.com/Forums/blog.php?b=111)) it does not work. They can not deterine the original distribution from the process - so their decisions can be wrong.

Actually, yes - for what is was designed for, final and incoming inspection. Some folks try to SPC that, and (especially for precision machining (http://elsmar.com/Forums/blog.php?b=111)) it does not work. They can not deterine the original distribution from the process - so their decisions can be wrong.

Oh boy! :D I'm not sure I agree but it'll probably depend on the items in question.
My main dislike towards AQLs is that they predetermine a number of faults as being acceptable. If the process is correct, then "only" human error is involved.
I know one company that has a a requirement wherein an approved process must never exceed +/- 2 sigma when manufacturing.
Re final inspection then I find (even although it can often be necessary) that it's rather like closing the stable door after the horse has bolted.
With incoming inspection then my belief is (unless it's used to evaluate a new supplier) that it's a sign that purchasing is only looking for a "cheap" deal and aren't held accountable for all "extra" expenses.
If I required/demanded documentation from a supplier I'd rather look at a simple histogramme (or something similar) showing the distribution. If I then measured a sample I could quickly see if theirs and mine tallied. I could probably also see from the distribution whether or not I was being given "selected" items from a larger quantity.
With precision machining I suppose it's all a question of how accurate measurement is possible.
I could write in much greater detail but I'm certain you get my drift Bob. If not just point out any weaknesses you find with my remarks ;)
As to whether you agree or not is another matter and I look forward to any comment you may have. :bigwave:

My main dislike towards AQLs is that they predetermine a number of faults as being acceptable. If the process is correct, then "only" human error is involved.

I'm not sure how the second statement supports the first, but AQL is a method of determining the probability that a lot of size x with y defectives will be accepted using a given random sample size. This doesn't mean that a certain number of defects is considered "acceptable." It means that one accepts the level of risk inherent in the sampling plan.

Here is a little guide that you can use from ZSI.com
For Inprocess inspection.
This is what the every one is talking about.

Oh boy! :D I'm not sure I agree but it'll probably depend on the items in question.
My main dislike towards AQLs is that they predetermine a number of faults as being acceptable. If the process is correct, then "only" human error is involved.

I guess my point was that AQL as a tool is better than SPC specifically for final inspection and incoming inspection. That should not be inferred that for overall product quality assurance it is a better tool. (It is only better when SPC is improperly applied - but that is really not the point I was making.)

Accepting an AQL where C>0 is an oddity these days, but it can be more of an issue where the process is not capable, and the rare bad part can be more cheaply pitched than to make the process and inspection changes. That is my best guess for using C>0.

In medical, where AQL was king, they are slowly buying into accepting SPC data rather than sampling as evidence, and aptly reducing the sampling plan.

If I required/demanded documentation from a supplier I'd rather look at a simple histogramme (or something similar) showing the distribution. If I then measured a sample I could quickly see if theirs and mine tallied. I could probably also see from the distribution whether or not I was being given "selected" items from a larger quantity.

I would rather see time sequenced data SPC - run chart at a minimum. Comparing histograms only works if the sampling was exactly the same - which is very unlikely if you did not receive the sample population of parts (other locations received some), and you did not sample the same way. I go through some of the sampling errors in my previous link that I have found that make that kind of analysis ineffective.

With incoming inspection then my belief is (unless it's used to evaluate a new supplier) that it's a sign that purchasing is only looking for a "cheap" deal and aren't held accountable for all "extra" expenses.


I agree.

I know one company that has a a requirement wherein an approved process must never exceed +/- 2 sigma when manufacturing.

Well, if there SPC is incorrectly applied, that is a good way to protect themselves. But if it is correct, then they are overcontrolling (wildly so, if it was precision machining), and it is certainly anti-lean, as they are wasting resources to maintain that control. But, to each their own.

Here is a little guide that you can use from ZSI.com
For Inprocess inspection.
This is what the every one is talking about.

Page 10 of the book is a good chart for documenting the process inspection - very similar to automotive control plan, except with the odd structure of identifying the plan and the data collection in the same form. But, the "bubble" stating to use AQL for in-process inspection is an incorrect use of the tool. Benefit of the doubt - maybe a bad language interpretation?

But, as I have said, many people have been implementing this incorrectly for a long time. No mystery there.

I'm not sure how the second statement supports the first, but AQL is a method of determining the probability that a lot of size x with y defectives will be accepted using a given random sample size. This doesn't mean that a certain number of defects is considered "acceptable." It means that one accepts the level of risk inherent in the sampling plan.

I'll include two others in this reply.

World Quality - it looks very interessting and I'll take the time to print and read it - thanks :applause:

bobdoering : I don't think we'd agree 100% but that certainly doesn't mean I think you're wrong in anything you write :D You'd probably win on a head to head :D :truce:

To Jim: either I'm misunderstanding something or I certainly disagree - from a practical (as opposed to theoretical) point of view.
Correct me if I'm wrong, but wouldn't specifying an AQL of say 1 mean that up to 1% defects would be accepted? Or, to put it an other way, the receiver couldn't return the lot as being out of spec?
I've seen (long, long ago) AQLs of 10 and 4 specified and I can't think of any company today that would "accept" anywhere near that level today.
Of course no-one wants defects but AQLs indicate that a number of defects will be accepted.

Someone once said to me, "If you are lucky you get what you pay for, it's rare you get more". Please, no-one beat me on the head with that - it's just my attempt at humour :notme:
I've never known if the following is true or not, but a Portugese supplier of hand grenade casings (and that is not what we bought from him :mg:) told me that that was their favourite delivery. When I asked him why, he replied that they'd NEVER received a complaint.

To Jim: either I'm misunderstanding something or I certainly disagree - from a practical (as opposed to theoretical) point of view.
Correct me if I'm wrong, but wouldn't specifying an AQL of say 1 mean that up to 1% defects would be accepted? Or, to put it an other way, the receiver couldn't return the lot as being out of spec?

That might have been true back in the day, especially where US MIL work was concerned, but these days I think most people understand that defectives are subject to return/debit regardless of the AQL. In fact, AQL is rarely (if ever) specified in contracts, at least in the US.

That might have been true back in the day, especially where US MIL work was concerned, but these days I think most people understand that defectives are subject to return/debit regardless of the AQL. In fact, AQL is rarely (if ever) specified in contracts, at least in the US.

I'm not quite sure how far "back in the day" was, but less than 25 years ago AQL was still popular. I have worked for a Danish company supplying to the US military and AQLs were prevalent.
Part of the problem may be as you write "but these days I think most people understand that defectives are subject to return/debit regardless of the AQL" as it may be that those that aren't "most" create problems for themselves.
Going out on a limb here, I'd be surprised at any company specifying to a supplier any AQL although it could maybe be used internally or sampling purposes where measurement was "difficult". For example the companies that prefer to use solid gauges as opposed to measurement.
Everything can't be perfect - but it'd be great if it was :lol: