What Inspection Level for Small Lot Size and Special Levels (S1-S4)?

D

dyeysi

Dear Covers,

I'd like to ask your help on this and will very much appreciate your suggestions.
I am trying to establish a sampling procedure for our Final Outgoing Inspection.
Our company is producing beacons (used in PAGA system) but still assembling in small quantity.
Production activity is still irregular since we focus more on R&D.
Production schedule just depends when we have customer orders.
If we have customer order, normally quantitiy per lot is around 20.
If I will be using MIL-STD-105E:

My questions are:
1. When do you use the Special Levels (S1-S4)? Will this be applicable to us?
(From my previous company where I also worked as QA, we use General Inspection Level II
since production is regular and manufactures in high volume)
2. Is it OK to reference the sampling table for MIL-STD-105E but not use the switching rules?
3. Is there any ideal AQL for our case?

Current situation is, I am the only QA in the company.
Owner for incoming, inprocess and final outgoing inspections.
Testing done from the final outgoing inspection is the same with production.
I want to make a sampling procedure that will not be a burden since I do all inspections in the company.
Thanks
 
G

Geoff Withnell

Dear Covers,

I'd like to ask your help on this and will very much appreciate your suggestions.
I am trying to establish a sampling procedure for our Final Outgoing Inspection.
Our company is producing beacons (used in PAGA system) but still assembling in small quantity.
Production activity is still irregular since we focus more on R&D.
Production schedule just depends when we have customer orders.
If we have customer order, normally quantitiy per lot is around 20.
If I will be using MIL-STD-105E:

My questions are:
1. When do you use the Special Levels (S1-S4)? Will this be applicable to us?
(From my previous company where I also worked as QA, we use General Inspection Level II
since production is regular and manufactures in high volume)
2. Is it OK to reference the sampling table for MIL-STD-105E but not use the switching rules?
3. Is there any ideal AQL for our case?

Current situation is, I am the only QA in the company.
Owner for incoming, inprocess and final outgoing inspections.
Testing done from the final outgoing inspection is the same with production.
I want to make a sampling procedure that will not be a burden since I do all inspections in the company.
Thanks


To answer your questions:

MIL-STD-105 (or ANSI/ASQ Z1.4 which replaces it) are really designed for on-going, steady processes. This really isn't what you have. For a set of better, IMHO, sampling plans for this situation, I would go to "Zero Acceptance Number Sampling Plans, Fourth Edition
Squeglia, Nicholas L.". These have wide acceptance and minimum sample sizes.

AQL is an often misunderstood term. The AQL is the quality level you are willing to "routinely" accept in your sampling plan. Why would you accept any level other than no defects? Obviously, you would prefer not to. The AQL is the level you will settle for, since tighter inspection would be more costly. An AQL of 1.0 means that my sampling plan will routinely accept a lot which is 1.0% defective. So setting an AQL depends on three numbers. How costly is the inspection process, and how costly/likely is a missed defective. So figuring an ideal AQL requires sirgnificant knowledge of the process.

Geoff Withnell
 
W

w_grunfeld

To answer your questions:

MIL-STD-105 (or ANSI/ASQ Z1.4 which replaces it) are really designed for on-going, steady processes. This really isn't what you have. For a set of better, IMHO, sampling plans for this situation, I would go to "Zero Acceptance Number Sampling Plans, Fourth Edition
Squeglia, Nicholas L.". These have wide acceptance and minimum sample sizes.

AQL is an often misunderstood term. The AQL is the quality level you are willing to "routinely" accept in your sampling plan. Why would you accept any level other than no defects? Obviously, you would prefer not to. The AQL is the level you will settle for, since tighter inspection would be more costly. An AQL of 1.0 means that my sampling plan will routinely accept a lot which is 1.0% defective. So setting an AQL depends on three numbers. How costly is the inspection process, and how costly/likely is a missed defective. So figuring an ideal AQL requires sirgnificant knowledge of the process.

Geoff Withnell
If your typical lot size is 20 forget about sampling , do 100% !! To reach an acceptable AQL most sampling plans will tell you the sample size is almost as big as the lot. If you have to compromise do sampling at receiving or even in-process (if defects that slip through can be catched at final) but never on final, unless your lots are in the hundreds or thousands and customers agree.
Willy Grunfeld
 
G

Geoff Withnell

If your typical lot size is 20 forget about sampling , do 100% !! To reach an acceptable AQL most sampling plans will tell you the sample size is almost as big as the lot. If you have to compromise do sampling at receiving or even in-process (if defects that slip through can be catched at final) but never on final, unless your lots are in the hundreds or thousands and customers agree.
Willy Grunfeld

You seem to be under the assumption that 100% inspection will catch all the defects. It will not. The only way to assure no defects is to produce no defectives. Once this is understood, then it is an economic decision to do 100% inspection, no inspection, or something in between, also known as sampling. Then it is a question of balancing the cost of inspection, probability of a missed defective (this probability is ALWAYS greater than zero if defectives have been produced) and the cost of a missed defective. If you have process controls (I don't mean inprocess inspection) that assures the product is defect free, than why would you do ANY final inspection? If you know that your process is incapable and has produced defects, then you will want to do 100% inspection to screen as many as possible out - you will NOT get them all! If your process is basically ok, but you might have excursions that cause defects - this is where sampling comes into play. How likely are the problems? How serious are the defects? How expensive is the inspection? The correct sampling plan is an economic balance of these factors.


Geoff Withnell
 

rmf180

Involved In Discussions
Great point Geoff,

Often people lose the importance of prevention vs. detection. It is also important to note that 100% inspection is not 100% effective. In most cases you will find that 100% inspection is only 80% accurate. You can effectively get to 96% only by implementing 200% inspection (1-.2^2). How many companies are willing to take on the cost of 200% inspection?
 
W

w_grunfeld

Great point Geoff,

Often people lose the importance of prevention vs. detection. It is also important to note that 100% inspection is not 100% effective. In most cases you will find that 100% inspection is only 80% accurate. You can effectively get to 96% only by implementing 200% inspection (1-.2^2). How many companies are willing to take on the cost of 200% inspection?
Gentlemen,
These are all very good points but irrelevant to the original question which wasn't inspection vs. prevention but what sampling plan to use. I only answered this question, namely that for irregular manufacturing lots of 20, sampling doesn't make sense. I did not say that 100% inspection is full proof. None of us has any idea what is the product , given the small lots my assumption was that it is some sort of equipment , in which case the term final inspection may refer to final test. I wouldn't sign off a piece of equipment to ship to a customer without being 100% tested regardiless what in-process or prevention methods are in place. From my background in semiconductors, the major players have the most advanced SPC and quality assurance procedures in place, yet it is unthinkable to ship an IC which is not 100% tested as the last step.
 
M

Muadh

I may be mistaken here (I usually am) but, IMHO, it does not seem to me that the original question has been answered. Below is an excerpt from the standard that might provide the answer.

"Four additional special levels: S-1, S-2, S-3, and S-4, are given in the same table and may be used where relatively small sample sizes are necessary and large sampling risks can or must be tolerated.

NOTE: In the selection of inspection levels S-1,to S-4, care must be exercised to avoid AQLs inconsistent with these inspection levels. In other words, the purpose of the special inspection levels is to keep samples small when necessary. For instance, the code letters under S-1 go no further than D, equivalent to a single sample of size 8, but it is of no use to choose S-1 if the AQL is 0.10 percent for which the minimum sample is 125."

Without knowing how much is involved in inspecting a single part, it is easy to say 100% inspect because the lot size is only 20 pieces. But, what if it takes 4 hours each? And, if 100%, as we know, does not guarantee that defective parts won't be accepted, what advantage is there to do it when, if you can tolerate the risk, an S-1, AQL 1.5, sample of 8 will be sufficient?
 
Last edited by a moderator:

Wolf.K

Quite Involved in Discussions
"Four additional special levels: S-1, S-2, S-3, and S-4, are given in the same table and may be used where relatively small sample sizes are necessary and large sampling risks can or must be tolerated.

NOTE: In the selection of inspection levels S-1,to S-4, care must be exercised to avoid AQLs inconsistent with these inspection levels. In other words, the purpose of the special inspection levels is to keep samples small when necessary. ?

I wonder if anyone has a hint for me where to read more about the special AQL levels 1-4. I wonder, what kind of criteria shall I use to decide to use S2 and not S1 or S3 or S4?

E.g. I have two good running processes.

Process one is running for years, we never had any troubles with quality. But we cannot do 100% inspections, as the test is destructive. We did 315 test samples (10.000 units in batch), found 0 to reject. But 315 is 0,040 level. So, According to AQL, there is no AQL of 0,0, the nearest is 0,010, and we would have to check 1250. So, AQL suggests S1 5 test samples, S2 8 test samples, S3 20 test samples, and s4 32 test samples.

Process two is new (new product), we did 100% visual inspection for the first few batches (not destructive), 10.000 each, found c=0. So, According to AQL, there is no AQL of 0,0, the nearest is 0,010, and we would have to check 1250 again.

Squeglia does not help much either, or? His "index values" are based on the AQL, but lower sample sizes required as he removed all other possibilities beside "0 1".

If I have a validated manufacturing process, how shall I calculate e.g. the in process sampling plans? Can anyone recommend good literature? Especially, I need numbers for comparision, e.g. what are typical AQLs in different industries?

Truly yours
Wolf
 

Mike S.

Happy to be Alive
Trusted Information Resource
ANSI Z1.4 has some explanation of S-1 to S-4.

All sampling is based on cost vs. risk tolerance. "Typical AQLs" vary from industry to industry, customer to customer, even product to product. What does your customer expect? What is the risk if a bad part (or 10) makes it to the customer? Is your process in control? Do you monitor it using a control chart? Is there a non-destructive test (other than visual which you mention) that can help you to monitor the lot in additional to the destructive test?
 

Wolf.K

Quite Involved in Discussions
ANSI Z1.4: "Four additional special levels: S-1, S-2, S-3, and S-4, are given in the same table and may be used where relatively small sample sizes are necessary and large sampling risks can or must be tolerated. Note: In the designation of insprection levels S-1 to S-4, care must be exercised to avoid AQLs inconsistent with these inspection levels.".

That is not much information. Probably, the information is passed on by hearsay...

So, I have two questions: 1. How to apply S-1 to S-4 for finished batches, and 2. how to calculate in-process sample number.

Our old process has proven to have a capability of zero defects, for years. So, we expect it to provide product without any defects, and our customers expect the same. The process is partially validated. The risk is high, as we have medical devices class III which remain within the patient. As the process steps are validated, I wonder how I should calculate sample sizes for in-process controls. On QualityDigest, Steve Wise writes in the article "How to determine in-process sampling strategies": "Generally, most textbooks use sample sizes of 1, 3, 5, and 10. In the industry these sizes have become common as well." But these numbers should be somehow be calculated by statistical methods, right?
 
Top Bottom