Acceptance sampling
Lot acceptance by attributes and variables, AQL, OC curves, and why sampling cannot inspect quality in.
Acceptance sampling decides whether to accept or reject a lot based on a sample. It is used for incoming material and final lots, not to control a process.
Key terms
- AQL (acceptable quality level): the worst average quality that is still acceptable for the plan; lots at the AQL are accepted most of the time.
- LTPD / RQL (lot tolerance percent defective): a quality level the plan should reject most of the time.
- Producer's risk (α): chance of rejecting a good lot (at AQL), often 5%.
- Consumer's risk (β): chance of accepting a bad lot (at LTPD), often 10%.
- OC curve (operating characteristic): probability of acceptance plotted against lot quality. Every plan has one; read it before trusting the plan.
Plans
- Single sampling by attributes: inspect n units; accept if defectives . Standards: ANSI/ASQ Z1.4 (successor to MIL-STD-105E).
- Zero acceptance (c = 0) plans: reject on any defective; popular where defects are critical.
- Variables sampling: uses measurements (mean and standard deviation) and needs far smaller samples for the same protection. Standard: ANSI/ASQ Z1.9.
- Double and sequential plans: take a second sample only when the first is inconclusive.
Limitations
Sampling cannot make a lot better; it only sorts lots. A stable, capable process with SPC makes most incoming inspection unnecessary, which is why many customers ask suppliers for capability data instead.
Questions and answers
Our plan is "inspect 32, accept on 1 defective". How good must a lot be?
Read the OC curve. For this plan, a lot with 1% defective is accepted about 96% of the time, and a lot with 10% defective is still accepted about 16% of the time.
