Hey all,
AQL is a fairly new concept for me that is used in my industry. Is it possible to have segmented or progressive acceptance for lot acceptance? In other words can I release part of a lot before before the whole lot has been run and inspected?
Our work orders generally run about 100,000+ parts and we can run about 15,000-30,000 parts per day per machine. We run 7 machines and I have one quality coordinator on the floor per shift. To break the runs into smaller "lots" ups my inspection quantities to an unmanageble quantity and it is highly unlikely that I will get more manpower alotted to my department. But product ships as truckloads are completed, if released by quality. Defects are not generally sporadic in our processes once they begin they continue until corrected. they for the most part (except for contamination on the product) contained to one or two pallets and if the product is looked at according to schedule and criteria should be caught in a single pallet(1200-15000 parts). If when a defect is found the product is sorted and the defects removed back to the origin and at the same time the process is corrected so more defects are not being produced cna the product be released as we go?
I do not like AQL it is hard for me to get my mind around the idea that certain defects are acceptable if kept under a certain number. I am still trying to figure it out. I wonder if there is a "AQL for dummies" book.:confused:

"AQL" is something of an outmoded concept, and is generally misunderstood anyway. The popular notion is that it precluded rejecting anything so long as there were fewer than x defectives in a sample. What it actually means is that a lot may be rejected and returned to its source if there are x or more defects in the sample, but will not be returned if there are fewer. The receiving entity still has the right to reject bad parts, however. It usually boiled down to who had to do the sorting. Today, the idea of shipping a customer a lot known to include defectives is verboten.

You may treat any homogeneous quantity as a "lot" for sampling purposes. Avail yourself of a statistically sound sampling plan, do random sampling, and off you go.

From what I can see the industry I am in is way behind say automotive. AQL is still "the way it is done". I am not a statistics genius in any sense of the word. I am in quality because I am resourceful and have a fair amount of common sense and learn fairly quickly. With my limited knowledge of statistics I am not sure how to convince my company nor my customers for the need of more "modern" quality criteria, especially when the depth of understanding for our processes is so seemingly shallow. In order to run statistical sampling on our product I need an in control process, this I know. I also know that we are not even sure what in our process contributes significantly to control, we are still trying to find the control points. The product shows no evidence of control at this time.
Is there anybody with extrusion blow molding experience that can help with this.
And until then is there an answer to my original questions? I know some of them show my lack of understanding for AQL. I need clarity.

It will help if you can tell us what AQL plan you are using now or alternately, what defect rate you want to protect against.

A simple answer to your question was given by Jim - but allow me to elaborate on it with a few more specifics. Consider each portion of the grand lot you want to release as a "sub-lot". perform a full AQL sample size on that grand lot - randomly - and you can reelase that sublot of materials. Of course all of that sub lot must be present and available for inspection for the sample to 'cover' the sublot. each sublot of material is an 'inspection lot' for purposes of your AQL plan.

YOU define the inspection lot, so it is quite acceptable to have multiple inspection lots wihtin a grand lot or large batch or large setup. In fact, it woudl be to your advantage to break up the inspection inot smaller subsets to get quicker feedback product quality...

There is no great mystery about AQL sampling plans. basically, you pick the "acceptabe Quality level" that you are willing to ship to your Customer. Say .1% defective is OK, but no lot should have more than .1% defective. Then once you know the lot size you pick the sample size for the lot based on general inspection levels (there are tightened and loosened levels but they are really just confusing at this point and many people don't use them. You also need to determine if you will reject the lot on 1 defect or not. Generally accepted practice is to reject on 1 defect. This really only makes people feel better - it doesn't improve the quality level in shipped lots. The important thing to understand is the AQL level. The AQL level means that you will accept 95% of the lots that have that level defective in them...it's exactly the opposite of what many people think the AQL level means. (hence the name 'acceptable' quality level.)

As I've written many times in the forums, I never use AQL plans anymore. They are too complicated. I simply use the Binomial or Poisson to selct my sample size depending on how much confidence I want in detecting a defect level. Which is what most people think the AQL plan is doing. It also lets me easily change my confidence level, instead of being stuck with just 95%.

For example using the Binomial, if I want to detect a .1% defect rate 95% of the time I would have to sample n= 2995. (2996 for the Poisson). If I choose and AQL of .1% and I guess at a lot size of 35,000-150,000 then my sample size is 500. but with that sample size I will REJECT a lot that is .1% defective only ~40% of the time!

Also picking the sample size by lot size is not statistically necessary. it only adds to the confusion.

Hi All,
To throw something else out here for discussion. We are a nonprofit that provides work to handicapped persons to improve/increase their skills on all levels. We have a customer that manufactures bullets that we have been asked to do a gross defect inspection of their product. I have been asked to develop a sampling plan to eliminate the requirement for 100% visual inspection after the intitial sort for defects. My question is this. If I run 6 stations with a variable output of 1,000 to 3,000 bullets per shift, on a lot of 160,000 bullet, would ANSI 1.4 be a good place to start for a sampling plan?
Thanks for all the great inputs and knowledge on this forum.
Regards,
Keith:thanx:

Also see these related existing discussion threads related to AQL - Acceptable Quality Level: AQL - Acceptable Quality Level (http://Elsmar.com/Forums/search.php?do=process&titleonly=1&query=aql)