Search the Elsmar Cove!
**Search ALL of** with DuckDuckGo Especially for content not in the forum
Such as files in the Cove "Members" Directory
Social Distancing - It's not just YOUR life - It's ALL of OUR lives!
Me <——————— 6 Feet ———————-> You

4.10 Inspection and Testing - Do we have to do AQL (MIL-Std-105) sample inspection?



4.10 Inspection and Testing

We have some confusion at our small company (<20 employees) in regard to inspection and testing requirements. We produce over 500 different types of widgets in lot orders. Orders range from 6 pieces to over 100,000. Most orders go through similar process steps - slit, punch press, ship. We have no formal documented inspection process at the moment, and we would like to establish in-process inspection as much as possible. How do we begin to determine the number of samples to inspect (i.e. dimensional checks, visual checks) without slowing down our production process to record inspection information? Do we have to institute AQL (MIL-105) which seems like an excessive number of sample inspections? Do we have to record every inspection?


Captain Nice
Staff member
You can use any scheme you want - it depends entirely upon your company and products. You decide what is approprite for you. Using an AQL is one way of doing it.


Two injection molded products. 1 is a small toy soldier for small kids to be packed in bags - not painted or otherwise 'special'. Another product is a 'special' version of the same toy soldier which you paint and individually package as 'collector's items'.

In the first case you may decide to inspect 10 from each mold cavity to start as you bring the process up and you may only chack to ensure the mold cavities are completely filling out. A second chack may be made at the end of the run to ensure consistency (filled out) with the first.

In the second case you may add an additional check like a post-paint check of a certain quantity and / or a post-packaging check of some sort.

But the number you check and the frequency would depend upon what you believe is important (and customer requiremetns, if any). You could use an AQL (Acceptable Quality Level - says you will accept a certain number of nonconforming items) or any other methed you can justify as approipriate to your product and business.

Not my forte - maybe someone else can give you advice on methodologies of determining sample size.

As far as common operations, if the 'critical characteristics' of each item to go thru a common process are the same you can use the same inspection and / or test. Where the critical characteristics are the same but the actual dimension is different, you can use a matrix appended to the inspection / test instruction.

Look at your history of nonconformances for preliminary info on sample methodology and size. Also ask yourself what customer requirements may include (which may depend upon the product).

Don Winton


Must go with Marc on this one. Would like to add my two cents worth, however. As Marc said, how you determine your IP inspection up to you. ISO 900x 4.10.3 states that you shall “inspect and test the product as required by the quality plan and/or documented procedures.” This, in essence, states that any IP inspection is up to your operation. You do not, as stated above, have to perform IP inspections AT ALL. But, if you do not, be sure to justify the reasons for this decision. This should be customer first, then company, driven. Marc’s soldier example is a very good example. Quality of product is determined by the customer.

As for sampling. Under the assumption that IP is to be performed by sampling, AQL is one choice, but not the only. AQL allows for an “accepted” number of nonconformities. As Marc also stated, the number and frequency are up to you and your “quality plan and/or documented procedures.” I would first, before determining a sampling plan, determine your process capability in relation to your customer specifications and your capability to meet these requirements. For example, if process capability exceeds 1.5 or greater, the sample size may be less than if the process capability is 1.0 or less.

Sample size (n) can also be calculated based upon your probability (p) of expecting to determine if a lot is acceptable and the size of the lot (N). It has been a while since my quality engineer hat was on, but I will try to remember. If your are confident that process capability is capable of producing to greater than 1.5, then your probability would be, say, 0.99 (example, I do not keep these equations in my head. If I am wrong, someone let me know.). If process capability is less than 1.0, then the probability may be 0.90 (again, example). I do not have the equations handy (and they may have been lost in my reference material, it has been a while), but if you are interested, I will try to reconstruct and forward.

Remember, do not make things more complicated than they have to be and do not read in what is not there.

As Francis Bacon said, “If a Man Will Begin With Certainties, He Shall End in Doubts; but If He Will Begin With Doubts, He Shall End in Certainties.”

Hope this helps. Don

Don Winton

Some additional info regarding sampling. Developed this years ago. Do not remember the source. Hope it helps.

There are three alternatives for evaluating lots:

No inspection at all;
100% inspection;
Inspect a sample.

An economic evaluation of these alternatives requires a comparison of total costs. Assume that:

N=Number of items/lot
n=Number of items/sample
p=Percent defective in a lot
A=Cost if a defective passes
C=Inspection cost/item
Pa=Probability of acceptance

The total costs are calculated by the formulas below. These costs reflect both inspection and damage costs and recognize the probability of accepting or rejecting a lot under sampling inspection. The expressions can be equated to determine a break-even point.

No Inspection: N*Pa
100% inspection: N*C
Sampling: (n*C) + ((N-n)*p*A*Pa)+(((N-n)*(1-Pa)*C))

If the percent defective is greater than 5%, then 100% inspection should generally be used (this assumes that 100% inspection is 100% effective, which it is not. Don). If the sample size is assumed to be small compared to the lot size, the break-even point is:


Another cost model for attribute plans is considered below:



TC=Total Cost
A=Overhead cost
nmax=Maximum sample size
nbar=Average sample size
B=Cost/Unit of sampling
C=Cost/Unit of inspecting

Also, I have developed several (nine, I think) training sessions regarding statistical techniques. One regards sampling. I could forward under separate cover if you are interested. Format is Powerpoint 4.0 (mac). BE ADVISED: These were “under construction” when I dropped the project (client refused to pay). Therefore, they have not been subjected to “final” proofing and “peer” review. Word of Warning. (Marc, I cannot, at this time convert these to Acrobat PDF. I could forward to you to convert, if you are interested, with the above disclaimer included, Don).

One last thing: If a sampling plan is developed, be sure to be in compliance with ISO 9000x, Section 4.20.

With that, I depart (for now).

Nam et ipsa scientia potestas est.

Best Regards,
Top Bottom