Acceptance Sampling and AQL Plans

Nobody can check every feature on every part in every lot. Acceptance sampling is the shop's answer. You inspect a set number of parts picked from a lot, count the defects, and let a plan decide whether to accept the whole lot. It is a tested method for making that call, as long as you read the plan correctly and pull a fair sample.

What You Will Be Able to Do

• Explain why shops sample and what the risks are.
• Say what AQL means and what it does not mean.
• Look up a sample size and accept and reject numbers in a Z1.4 table.
• Pull a random sample from a lot.
• Know what to do when a lot passes or fails.

Why Sample

Shops use sampling when:

• The lot is too large to check everything at a fair cost.
• The test destroys the part, such as a pull test or a burst test.
• Inspecting by hand all day gets tiring, and tired people miss defects.

A lot is a batch of parts made under the same conditions. Keep lots separate. Mixing parts from two heats, two shifts, or two suppliers hides the difference between them.

Sampling is not for every feature. Critical and safety features are usually checked on every part, or held to a plan that accepts no defects.

What AQL Means

AQL stands for acceptance quality limit. Older documents say acceptable quality level. It is the worst average defect rate that the plan treats as acceptable over a series of lots. An AQL of 1.0 means a process that averages 1 percent defective or better should have nearly all of its lots accepted.

AQL describes how a plan behaves over many lots. It is not a promise about any one lot, and it does not give anyone permission to make defective parts.

Any sampling plan carries two risks:

• Producer's risk: a good lot gets rejected because the sample happened to catch several bad parts.
• Consumer's risk: a bad lot gets accepted because the sample happened to miss them.

Bigger samples lower both risks, and they cost more. The plan is a trade-off.

Reading a Z1.4 Plan

ANSI/ASQ Z1.4 is the common standard for attribute sampling, where each part passes or fails. It grew out of a US military standard, MIL-STD-105. Your inspection plan or the customer sets the AQL and level.

1. Find the lot size in the first table.
2. Choose the inspection level. General Level II is the normal default.
3. Read the code letter for that lot size and level.
4. Go to the table for your inspection type. Start with normal.
5. Find the row for your code letter to get the sample size.
6. Find the column for your AQL. That cell gives Ac (accept) and Re (reject).

If the sample has defects equal to or fewer than Ac, accept the lot. If it has defects equal to or more than Re, reject it.

Here is part of the first table for General Level II:

Worked example: a lot of 2,000 parts arrives. The plan says General Level II, AQL 1.0, normal inspection.

1. A lot of 2,000 falls in 1,201 to 3,200.
2. The code letter is K.
3. The sample size is 125.
4. In the AQL 1.0 column, row K shows Ac 3, Re 4.

If you find 3 or fewer defective parts in the 125, accept the lot. If you find 4 or more, reject it.

Arrows: some cells show an arrow instead of numbers. Follow it to the first cell with numbers, and use that row's sample size as well as its Ac and Re. For example, code H at AQL 0.65 points you down to a sample size of 80. If you keep the sample at 50, you are not following the plan.

Always read from your shop's copy of the table. Do not work from memory.

Switching Between Plans

Z1.4 is built to run over a series of lots, and the rules react to a supplier's record.

Tightened inspection is stricter. Reduced inspection uses smaller samples. If you always use the same plan and ignore these rules, you lose some of the protection that Z1.4 was designed to provide.

Pulling a Fair Sample

Everything depends on this step. A sample must give every part in the lot a fair chance of being chosen.

1. Set the sample size before you open anything.
2. Number the boxes, trays, or layers.
3. Use random numbers to pick where to draw from. A spreadsheet or a random number table works.
4. Draw from every container, and from the top, middle, and bottom.
5. Inspect every part you drew. Do not swap one out because it looks odd.

Worked example: the 2,000 parts come in 10 boxes of 200, and you need 125. That is 12.5 per box, so take 13 from five boxes and 12 from the other five. Pick the parts in each box using random numbers.

The easy sample is the top layer of the nearest box. That is often the last part made, or the part a supplier packed with care. It doesn't tell you much about the lot.

Defect Classes

Plans often sort defects by how serious they are.

Use the class and AQL written in your plan. If a defect does not clearly fit, ask quality.

Zero-Acceptance Plans

Many aerospace, medical, and automotive customers require C=0 plans. C is the acceptance number, so one defect in the sample rejects the lot. Ac is 0 and Re is 1, however large the sample.

These plans come from a separate table. Do not simply take the sample size from a Z1.4 table and change the accept number. Use what the customer requires.

What the Result Means

Worked example: the same lot of 2,000 runs at exactly 1 percent defective, so it holds about 20 bad parts. A sample of 125 will catch about 125 x 0.01 = 1.25 of them on average. That is under the Ac of 3, so the lot usually passes, and the other 19 or so bad parts go on.

So an accepted lot means it met the plan. It does not mean that every part in it is good.

If the lot passes: any defective parts you found in the sample are still bad. Tag them and pull them out. Release the rest.

If the lot fails: hold the entire lot and tag it. Follow the nonconformance process. Common outcomes are a 100 percent sort, a return to the supplier, or scrap. Do not take another sample in the hope of a better answer, unless your procedure says so.

Common Mistakes

Check Your Understanding

1. Your plan shows a sample of 125, Ac 3, Re 4. You find 4 defective parts. What happens?

Reject the lot. Four reaches the reject number. Hold and tag the lot, and start the nonconformance process.

2. A coworker says, "AQL 1.0 means 1 percent of what we ship is allowed to be bad." What is wrong with that?

AQL describes how a sampling plan treats lots over time. It is not a permission to make bad parts, and it is not a promise about any one lot.

3. A batch of 500 parts is under an AQL plan at Level II. What decides the sample size?

The lot size and the inspection level give code letter H, with a sample size of 50, unless an arrow in the AQL column sends you to another row.

4. Your lot passes with 2 defects in the sample. What do you do with those 2 parts?

Tag them and pull them out. A passing lot does not make them acceptable.

5. A customer's plan says C=0. You find 1 scratch that the plan classes as a defect. What happens?

The lot is rejected, because the accept number is zero. Follow the customer's requirements for what comes next.

Key Takeaways

• A sampling plan lets a sample decide the fate of a whole lot.
• AQL describes how a plan behaves over many lots. It does not excuse defects.
• Z1.4 runs from lot size to level to code letter, then to sample size, Ac, and Re.
• Follow the arrows, and read from the real table.
• Pull a random sample from the whole lot.
• An accepted lot can still contain bad parts, and defects in the sample are still pulled.