I've a question on AQL which I'll need help to decipher...

Let say I'm producing FMCG products, where within 1 day, I'll be producing quite a high number of products. Now, if my client imposed an AQL number, which requires me to do a sampling check of more than 1,000 pieces check per day, is there another way out, whereby I do not need to inspect 1000 pieces at one go, but still satisfying the AQL ?

What if I can prove my Cpk is good ? Can I escape from doing 1000 over pieces of inspection ?

My point is, is there a way of 'by-passing' the required number of samples in the sampling plan, for a given AQL ????

Good morning YKT,

In regard to your question, yes there is another way out. When determining sampling size based on a AQL you take into consideration the "levels" of inspection.

"When it is necessary to determine an inspection level, a number of factors must be considered in order to optimize the cost/risk relationship. These are:

1.The operating characteristic (OC) curves to evaluate the technical properties of various plans.
2.The supplier's risk and discrimination afforded by various inspection levels.
3.Knowledge of the production process.
4.Process capability knowledge and past quality performance history.
5.Item complexity.
6.Cost and importance of examination or test, particularly when testing is expensive, time consuming or destructive.
7.Importance of the quality characteristics to be examined, that is, critical, major and so forth.
8.Analysis of consumer's risk."

http://www.sqconline.com/military-standard-105e-ansi-z1.4.html

This will let you tinker with the numbers a bit to give you an idea, I'm somewhat strapped for time atm so I will leave you with this for now and you can return with more questions if you need.

Oh I should add this is based on MIL STD 105E which is obsolete, use it only as an example.

I've a question on AQL which I'll need help to decipher...

Let say I'm producing FMCG products, where within 1 day, I'll be producing quite a high number of products. Now, if my client imposed an AQL number, which requires me to do a sampling check of more than 1,000 pieces check per day, is there another way out, whereby I do not need to inspect 1000 pieces at one go, but still satisfying the AQL ?

What if I can prove my Cpk is good ? Can I escape from doing 1000 over pieces of inspection ?

My point is, is there a way of 'by-passing' the required number of samples in the sampling plan, for a given AQL ????

On top of the comments from Benjamin :
Maybe you can talk to your client and proof that a sample check of 1000pcs is overkill.
Anyway, If you want to change a customer required item - talk to the customer.

Best regards,

Antoine

YTK,

I can think of a few different approaches to your question.

As you say, if you have SPC data showing the process is in control and capable, you may well be able to get the customer to accept that data instead of sampling data. If fact, SPC is in many ways better than sampling.

As Benjamin said, adjusting the "level" is another option. To get a sample size over 1000, it looks like you might be at "Q" on the table - presumably 500,000+ pieces with the typical Level II inspection. If you went down just one step to Level I, the sample size drops from 1250 to 500. If you use one of the "special levels", you might get down under 100 for a sample size.

Another option would be to get the customer to agree to "reduced inspection". This would also drop you from 1250 pieces to 500 pieces for sample size.

One final option I can think of is to go to double or multiple sampling plans. Sticking with the typical "Level 2, normal inspection for 500,000+ parts = Q" plan, the multiple sampling plan lets you start with only 315 pieces. If the sample is good, you can quit there. The downside is you might end up having to check 2200 parts! Multiple sampling is good if the lots tend to be either quite good or quite bad.

Tim F

are you measuring continuous data or is your inspection criteria categorical (pass / fail)?

is continuous there are many mor eoptions for a reduced sample size.

talking to your customer is always the best first approach....

are you measuring continuous data or is your inspection criteria categorical (pass / fail)?

is continuous there are many mor eoptions for a reduced sample size.

talking to your customer is always the best first approach....


thanks for all the replies. I am measuring variable data, i guess, this is what you call continuous, right ?


Anyway, some customers are still very stubborn, and i believe their understanding of AQL and sampling is quite limited as well. When they look at the MIL-STD sampling plan, and cross check on the lot size and the required AQL, they will be adamant to use the sample size, regardless of the size.

When they come in to audit, and found out we are not doing as per required, (taking let say 2000 samples for every lot), they will make noise and want to issue SCAR ! OMG !

These guys are pretty straight forward and they just follow the table....

How do I convince them ??? MY QC is screaming for doing a high number of sampling !!

thanks for all the replies. I am measuring variable data, i guess, this is what you call continuous, right ?


Anyway, some customers are still very stubborn, and i believe their understanding of AQL and sampling is quite limited as well. When they look at the MIL-STD sampling plan, and cross check on the lot size and the required AQL, they will be adamant to use the sample size, regardless of the size.


If you have "variables" data and your custoemrs are adamant about AQLs direct them to Mil Std 414 (now ANSI Z1.9). this is for variables data and allows for much cmaller sample sizes than Mil Std 105 which is for data that can only be judged as pass/fail. Both satisfy an AQL approach.

The other alternative is to propose a cost change that woudl pay for all of the inspection. Having the data to show that you have a stable capable process backs up why the large sample size pass/fail testing is not needed to ensure that you are meeting the requirements and that in fact the request for a specific type of test is a requiremetn of the contract itself. Money and data talk: the buyer at your customer can be your friend when it comes to unreasonable supplier quality folks...I have resolved many similar situations this way. (when the SQE won't listen to sound quality theory)

if I have a process with a proven consistent Cpk of 1.33 and above for couple of months, how can I related the Cpk reading to AQL ?

Technically Cpk will tell you the process capability, which in turn, if I sue the z-value table, I'll be able to determine the estimated number of product out of specs (or defects per million opportunity -just like in 6-sigma)

AQL is also a concept of giving confidence to the customer, saying that our process is capable of meeting a maximum defect per lot as agreed.

Thus, if I have a good Cpk, (which means I'm confidently churning out good products), technically it will be a good measure to tell them (customer) that they should have more confidence with my products, even without going through a stringent inspection (with a large number of samples !)

Anyway, the question is, how do i link or relate the Cpk data to AQL ?

Perhaps you can convince your customer with money!

The sampling is an additional cost on top of the actual production costs. it is mostly a non-value-added cost -- especially if you already have good Cpk numbers.

Tell your customer you will drop 2% off the cost (or whatever you can justify) if they will accept a smaller sample accompanied by SPC data! And on future orders, quietly add that same 2% to the price if they don't accept your generous offer!


One other thought. Since you are already checking some parts for SPC, you might use the same parts for both their sampling requirements and for your own SPC calculations. For example, if you draw a subgroup of 5 every 15 minutes for 24 hr, you would have close to 500 pieces already tested. You could call those the first pieces of your sample and then just add enough to meet the required sample size.


Tim F

Anyway, the question is, how do i link or relate the Cpk data to AQL ?

OK first is your process Normally distributed?

If it's close, calculate your theoretical % defective using the Normal distribution Z Tables for BOTH tails. (In case you are of old Six Sigma training, do NOT utilize the 1.5 sigma shift it's fake and will over inflate your defect rate)

This defect rate is compared to the AQL. Remember now that an AQL is the ACCEPTABLE defect rate that will ACCEPTED and passed on to the Customer ~95% of the time it occurs. If your defect rate is equal to or less than the Customer's AQL you have a case for eliminating the AQL inspection.

Of course, if I were the Customer I would want to know how you are controlling your process to detect and correct any shifts or drifts in the process. (SPC with REAL reaction plans?) Remember past stability is no guarantee of future stability...

I just add you one more way to escape from checking 1000pcs regularly on all lots. There is a provision in Inspection methodology called skip lot system, widely used in Auto segment. For example, just by this if your lots are consistantly accepted for 5 times you can waive from all lot checking for another 3 times then check the 4th one. If fourth is accepted, then again waive 3 lots then continue check on 4th one like that.

I've a question on AQL which I'll need help to decipher...

Let say I'm producing FMCG products, where within 1 day, I'll be producing quite a high number of products. Now, if my client imposed an AQL number, which requires me to do a sampling check of more than 1,000 pieces check per day, is there another way out, whereby I do not need to inspect 1000 pieces at one go, but still satisfying the AQL ?

What if I can prove my Cpk is good ? Can I escape from doing 1000 over pieces of inspection ?

My point is, is there a way of 'by-passing' the required number of samples in the sampling plan, for a given AQL ????

The best way is to improve your processes so you make fewer defects, and so you don't ship defective parts to customers. When you have control over that, AQL ceases to matter. My clients do not ship defective parts, none of them follow any AQL less than C=0 to my knowledge, and I am not aware of any of them that manually inspect more than a few parts (however, many have automatic inline checks).

They have earned the right to do small inspections, based on excellent performance.

Helmut,
You wrote

My clients do not ship defective parts, none of them follow any AQL less than C=0 to my knowledge, and I am not aware of any of them that manually inspect more than a few parts (however, many have automatic inline checks).

Are you implying that c=0 plans protect against shipping defective product?

Helmut,
You wrote



Are you implying that c=0 plans protect against shipping defective product?

I am stating that c=0 performance ensures against shipping defective product. Plans alone don't do anything except drive you to improve performance.

However, c=0 plans require you to quarantine any sample groups where any defective parts are found. If you do that, it certainly should significantly reduce any defects being shipped.

The best way is to improve your processes so you make fewer defects, and so you don't ship defective parts to customers. When you have control over that, AQL ceases to matter. My clients do not ship defective parts, none of them follow any AQL less than C=0 to my knowledge, and I am not aware of any of them that manually inspect more than a few parts (however, many have automatic inline checks).

They have earned the right to do small inspections, based on excellent performance.

OK there is a significant misunderstanding of what AQL is, and how it should be used going on here. C=0 is not an AQL, and it is possible to have any AQL with a c=0 sampling plan. As Bev D noted, the AQL is the defective percentage that will be ACCEPTED approximately 95% of the time. This does not mean (and ANSI/ASQ Z1.4 is very clear about this) that the ANY defects are acceptable. The decision to do sampling is an economic decision, and the AQL level is that level at which it makes economic sense to sample. The risk of accepting a lot where the defect level has increased significantly (this is what Z1.4 sampling is designed to detect) and the costs associated with that risk need to be weighed against the cost of doing the inspection. The choices are to do 100% inspection, sampling at some level, or no inspection. The choice that makes the most economic sense for you and your customer should be used. Juran's Quality Handbook has, IIRC, a good discussion of how to do the calculation.

Geoff Withnell

OK there is a significant misunderstanding of what AQL is, and how it should be used going on here. C=0 is not an AQL, and it is possible to have any AQL with a c=0 sampling plan. As Bev D noted, the AQL is the defective percentage that will be ACCEPTED approximately 95% of the time. This does not mean (and ANSI/ASQ Z1.4 is very clear about this) that the ANY defects are acceptable. The decision to do sampling is an economic decision, and the AQL level is that level at which it makes economic sense to sample. The risk of accepting a lot where the defect level has increased significantly (this is what Z1.4 sampling is designed to detect) and the costs associated with that risk need to be weighed against the cost of doing the inspection. The choices are to do 100% inspection, sampling at some level, or no inspection. The choice that makes the most economic sense for you and your customer should be used. Juran's Quality Handbook has, IIRC, a good discussion of how to do the calculation.

Geoff Withnell

Thank you for the technical clarification. I was using c=0 in a generic sense.

My point is, many customer want zero defects shipped. That is the only acceptable quality level for many customers. This forum is the ONLY palce I hear AQL still tossed around. It is a rather antiquated concept - what defect level is acceptable to most of your customers.

The point was, improve your process robustness, then a sample plan is not an issue.

Thank you, Geoff.

My point is, many customer want zero defects shipped. That is the only acceptable quality level for many customers. This forum is the ONLY palce I hear AQL still tossed around. It is a rather antiquated concept - what defect level is acceptable to most of your customers. The point was, improve your process robustness, then a sample plan is not an issue.


AQL is not an "antiquated concept"; it's a misunderstood concept, which makes it seem antiquated. It's the misunderstanding that's passé. Sooner or later, regardless of "robustness" (or maybe "robustivity"? :tg:) lots will have to be sampled. If you don't know how to rationally determine how to do the sampling, you're probably going to waste time and get unexpected results.

AQL is not an "antiquated concept"; it's a misunderstood concept, which makes it seem antiquated. It's the misunderstanding that's passé. Sooner or later, regardless of "robustness" (or maybe "robustivity"? :tg:) lots will have to be sampled. If you don't know how to rationally determine how to do the sampling, you're probably going to waste time and get unexpected results.

Fair point. Sampling plans are not antiquated. Most suppliers still sample.

Perhaps I should have said the notion that there is an "acceptable" level of defects that can be shipped is antiquated. The various AQL levels will allow for known defect levels to be acceptable, and thus shipped, based on major and minor.

I know of no customers in my circles that would find that acceptable. Perhaps because my clients have to achieve those levels of effectiveness, is why they consistently achieve such high performance levels. Anything less is simply not accepted by their customers.

Playing off the OP's question, is there any point to sampling 1000 pcs., but still accepting a certain level of defects?

I know of no customers in my circles that would find that acceptable. Perhaps because my clients have to achieve those levels of effectiveness, is why they consistently achieve such high performance levels. Anything less is simply not accepted by their customers.
I know of many companies who set PPM levels >0 for suppliers. That doesn't mean that the defective units, should they be received, are "acceptable." Generally speaking, suppliers can expect to be debited for defective units, regardless of the prescribed PPM level. PPM levels are generally used as triggers for corrective action, and not to signify that some percentage x of defective product will be accepted.


Playing off the OP's question, is there any point to sampling 1000 pcs., but still accepting a certain level of defects?
Maybe yes, maybe no. If we accept that a sample of ~1000 pieces is actually necessary (and it probably isn't in most instances), sometimes "accepting" (read: accepting reality) a proportion of defective product is the most economical alternative. Granted, those instances should be rare, but they do happen.

One more time. The AQL levels do not "allow for known defect levels to be acceptable". This is not the intent of the standards, nor is it how they are written. Even in "accepted" lots, any defectives found in the sample, or otherwise identified must be removed and dispositioned. To use ANSI/ASQ Z1.4 and Z1.9 correctly, it is necessary to read the document, not just the sampling tables, and use them appropriately. The AQL is an indexing method for sampling plans, which allows one to estimate, when using a sampling plan at a given AQL, how large a percentage defective the process will need to be at before the sampling plan begins rejecting them. These plans assume that the samples are being taken from a stable, ongoing process, and so state in the standard. If the process is not stable or on-going, then these plans are not appropriate. In fact, the plans require the suspending of sampling if too many defectives are found, since the process is no longer stable. The sampling level is an economic decision. If the economic consequences of missing a negative shift in process output is low, and the cost of inspection is high, then a high AQL with an associated low sample size is justified. If the economic cost of the missed shift is high, and the cost of inspection is low, then a low AQL with the associated high sample size is justified. Notice, please that neither of the preceeding two sentences even MENTIONED an allowable defect rate, which in either case could be zero, 3.4 PPM, 10% or whatever to which the contracting parties agree.

To go back to the original poster's question, the best way to negotiate a reduction in AQL, is to show the customer the data. Proceed something like this:

Here is the data that shows our process is stable and capable.
Here is the data on the risk of negative capability excursions.
Here is the data on the cost of the current sampling scheme.
We believe the cost of a missed negative excursion is about X. It will be an estimate, since at least some, if not most of the cost is the customer's.
Based on this data, we believe an AQL of Y, leading to this new sampling scheme, is more appropriate and cost effective.


Geoff Withnell

Fair point. Sampling plans are not antiquated. Most suppliers still sample.

Perhaps I should have said the notion that there is an "acceptable" level of defects that can be shipped is antiquated. The various AQL levels will allow for known defect levels to be acceptable, and thus shipped, based on major and minor.

I know of no customers in my circles that would find that acceptable. Perhaps because my clients have to achieve those levels of effectiveness, is why they consistently achieve such high performance levels. Anything less is simply not accepted by their customers.

Playing off the OP's question, is there any point to sampling 1000 pcs., but still accepting a certain level of defects?

One more time. The AQL levels do not "allow for known defect levels to be acceptable". This is not the intent of the standards, nor is it how they are written. Even in "accepted" lots, any defectives found in the sample, or otherwise identified must be removed and dispositioned. To use ANSI/ASQ Z1.4 and Z1.9 correctly, it is necessary to read the document, not just the sampling tables, and use them appropriately. The AQL is an indexing method for sampling plans, which allows one to estimate, when using a sampling plan at a given AQL, how large a percentage defective the process will need to be at before the sampling plan begins rejecting them. These plans assume that the samples are being taken from a stable, ongoing process, and so state in the standard. If the process is not stable or on-going, then these plans are not appropriate. In fact, the plans require the suspending of sampling if too many defectives are found, since the process is no longer stable. The sampling level is an economic decision. If the economic consequences of missing a negative shift in process output is low, and the cost of inspection is high, then a high AQL with an associated low sample size is justified. If the economic cost of the missed shift is high, and the cost of inspection is low, then a low AQL with the associated high sample size is justified. Notice, please that neither of the preceeding two sentences even MENTIONED an allowable defect rate, which in either case could be zero, 3.4 PPM, 10% or whatever to which the contracting parties agree.

To go back to the original poster's question, the best way to negotiate a reduction in AQL, is to show the customer the data. Proceed something like this:

Here is the data that shows our process is stable and capable.
Here is the data on the risk of negative capability excursions.
Here is the data on the cost of the current sampling scheme.
We believe the cost of a missed negative excursion is about X. It will be an estimate, since at least some, if not most of the cost is the customer's.
Based on this data, we believe an AQL of Y, leading to this new sampling scheme, is more appropriate and cost effective.


Geoff Withnell

Geoff, I feel like we are saying a similar thing in a different manner.

In a c=0 plan, if one defect is found the lot must be quarantined and 100% inspected.

In an AQL plan, (going on memory, I no longer have any copies), finding a defect part or two in the sample, with an AQL of 4, the lot would be passed. The two bad parts found would be removed, but the remainder would likely still have more defects, based on the percentage identified in the relevant column. Isn't that how it works? That is deemed acceptable quality, even though there is a high liklihood there may be a few more defects in the lot. That is what I meant by sending defective parts.

I was making the argument that if one gets better results from the process by having higher standards, then the sampling sizes can go down tremendously. Of course, as you suggest, you can try to make an economic argument to a customer that more inspection is not economical. However, more and more customers are rejecting that argument and requiring 0 defects.

The last client I had that still practiced AQL levels, did not understand their customers were not accepting AQL. They were rejecting 50,000 pc lots with 1 or 2 defects.

I agree that c=0 plans make more sense. I prefer c=0 plans, and have instituted them whenever I had that power for the very reason you point out. However, even C=0 plans have associated AQLs! Deciding to go to c=0 plans does not eliminate the exercise of determining the appropriate AQL, it just means you will be using the lowest sample size for that AQL, and will not accept a lot if any defectives are found in the sample. Note that it is quite possible that the defective found may be the only defect in the entire lot. I have seen this more than once, where a set-up piece was inadvertently included, or similar "fliers". Too many organizations have sample sizes that are inappropriately large, or small, or which should be doing either no inspection or 100% inspection, because the sample size or AQL was just "pulled out of the air". And a rational c=0 program is not just using the c=0 sample plans from the ANSI standard. You are then side stepping a lot of the calculations involved, and aren't really getting the AQL coverage you think you are. I strongly recommend using "Zero Acceptance Number Sampling Plans, Fourth Edition, Squeglia, Nicholas L."

Geoff Withnell

Geoff, I feel like we are saying a similar thing in a different manner.

In a c=0 plan, if one defect is found the lot must be quarantined and 100% inspected.

In an AQL plan, (going on memory, I no longer have any copies), finding a defect part or two in the sample, with an AQL of 4, the lot would be passed. The two bad parts found would be removed, but the remainder would likely still have more defects, based on the percentage identified in the relevant column. Isn't that how it works? That is deemed acceptable quality, even though there is a high liklihood there may be a few more defects in the lot. That is what I meant by sending defective parts.

I was making the argument that if one gets better results from the process by having higher standards, then the sampling sizes can go down tremendously. Of course, as you suggest, you can try to make an economic argument to a customer that more inspection is not economical. However, more and more customers are rejecting that argument and requiring 0 defects.

The last client I had that still practiced AQL levels, did not understand their customers were not accepting AQL. They were rejecting 50,000 pc lots with 1 or 2 defects.

YKT,
If you have enough evidence to prove the production process is stable, such as good cpk value and control chart.
I think, your client can accept your requisition on reducing the inspection sample quantity.

Those information given above are quite helpful… you are really great guys!! :rolleyes:

…anyway, I’m not going to add more on “numerical” aspects but on the principle of continual improvement…

…if you can prove that you are improving (using quality tools that are acceptable from your customer), the very next thing to do is to reduce the inspection activities – that’s continual or continuous improvement afterall…

Of course, if I were the Customer I would want to know how you are controlling your process to detect and correct any shifts or drifts in the process. (SPC with REAL reaction plans?) Remember past stability is no guarantee of future stability...

I agree with Bev D on this. In my mind acceptance sampling should be used for incoming receiving, where the product is homogeneous and the goal is to verify that the upstream processes were controlled and generating good product.

For outgoing product, acceptance sampling may be done as a mandate for final inspection to duplicate how the customer will inspect the product. But the key is to control the process.

The worst thing you can do is use acceptance sample sizes to determine line inspection frequency. As the process owner, you have more information to work with, such as time sequence information on process performance that the customer does not have in their pile of parts in a box. I would hope that properly implemented SPC for the key characteristics in question would be an acceptable alternative to acceptance sampling in the light of this. Otherwise, the customer may be overcautious or lack a thorough understanding of process control.

Either way, and in any event, the cost of inspection is a cost, and should be a part of the piece price. If there is evidence that a more cost efficient and equally successful alternative is available - and the benefit can be shared - some customers will think harder on their requirements :cool: