Inspecting every part in a large shipment is expensive, slow, and — surprisingly — less effective than it sounds. Fatigue degrades inspector accuracy, and 100% inspection still misses defects. Acceptance sampling offers a statistically grounded alternative: inspect a defined sample, apply a decision rule, and accept or reject the entire lot based on the result.
At the center of this system is the AQL — Acceptable Quality Level. It appears on virtually every incoming inspection plan and supplier quality agreement. Yet it is widely misread, even by experienced quality engineers. Let's fix that.
What AQL Actually Means
AQL is defined as the quality level that is the worst tolerable process average when a continuing series of lots is submitted for acceptance sampling. In plain terms: it is the defect rate at which the sampling plan will accept the lot most of the time — specifically, about 95% of the time.
抜取検査において、「この不良率程度であれば概ね合格させてよい」と定めた品質水準。AQL=1.0%とは、不良率1.0%のロットを95%の確率で合格させる計画を意味する。AQLは目標不良率でも保証値でもない点に注意が必要。
Here is the critical misunderstanding: an AQL of 1.0% does not mean "we will reject any lot with more than 1% defectives." It means that if the true defect rate is 1.0%, the sampling plan will accept that lot roughly 95% of the time. Lots with higher defect rates will also sometimes pass. This is not a loophole — it is the statistical reality of sampling.
"Sampling plans do not guarantee the quality of accepted lots. They provide controlled risk for both producer and consumer." — ANSI/ASQ Z1.4 Standard
Producer Risk and Consumer Risk
Sampling always involves two types of error — rejecting a good lot, or accepting a bad one. Quality standards give these errors formal names and assign probability limits to each.
| Error Type | Also Called | What Happens | Typical Limit |
|---|---|---|---|
| Alpha (α) | Producer's Risk | A good lot is rejected. The supplier suffers. | 5% (at AQL) |
| Beta (β) | Consumer's Risk | A bad lot is accepted. The customer suffers. | 10% (at LTPD) |
The Lot Tolerance Percent Defective (LTPD) — sometimes called the Rejectable Quality Level (RQL) — is the defect rate at which the plan will accept only 10% of lots. It defines the "protection" side of the plan: at this defect level, 90% of lots will be rejected.
抜取検査に伴う2種類の誤判定リスク。生産者危険（α）は良品ロットを誤って不合格にする確率（通常5%）。消費者危険（β）は不良ロットを誤って合格させる確率（通常10%）。両者はトレードオフ関係にあり、一方を下げようとするともう一方が上がる。
How to Read ANSI/ASQ Z1.4 Tables
ANSI/ASQ Z1.4 (equivalent to ISO 2859-1) is the international standard for attribute sampling plans — where each item is classified as conforming or nonconforming. Using the table requires three inputs: lot size, inspection level, and AQL.
Step 1 — Find the sample size code letter. Look up your lot size in Table I. The standard offers three general inspection levels (I, II, III) and four special levels (S-1 through S-4). Level II is normal. A lot of 500 pieces at Level II gives code letter H.
Step 2 — Find the sample size and acceptance number. Take code letter H to the Single Sampling Plan table (Table II-A for normal inspection). At AQL = 1.0%, this gives: sample size n = 50, acceptance number Ac = 1, rejection number Re = 2.
Step 3 — Inspect and decide. Inspect 50 pieces. If you find 0 or 1 defective, accept the lot. If you find 2 or more, reject it. The decision rule is that simple — the complexity lives in the statistical design of the plan, not in its application.
抜取検査の合否判定基準。不良品数がAc以下なら合格、Re以上なら不合格。ReはAc+1になることが多い。たとえばAc=1、Re=2の場合、不良1個以下で合格、2個以上で不合格となる。この2つの数字が抜取検査の核心。
Normal, Tightened, and Reduced Inspection
Z1.4 includes a switching system that adjusts inspection stringency based on recent quality history. This is one of its most powerful — and most neglected — features.
Start on Normal inspection. If 2 out of 5 consecutive lots are rejected, switch to Tightened inspection — which uses a smaller acceptance number (stricter pass/fail), effectively demanding better quality before the supplier is trusted again. If inspection is tightened and 5 consecutive lots are still rejected, halt production or escalate.
In the other direction: if 10 consecutive lots pass on Normal inspection with no rejections, you may switch to Reduced inspection — a smaller sample size, which saves cost while maintaining statistical protection. The switching rules create a self-adjusting system that applies more scrutiny to unreliable suppliers and less to proven ones.
Z1.4の切替ルールによる3段階の検査厳しさ。品質実績が悪化するとなみ検査からきつい検査へ移行し、合格判定をより厳しくする。逆に品質実績が安定していればゆるい検査に移行してサンプル数を削減できる。過去の品質履歴を使った動的な検査管理の仕組み。
Common Mistakes in Applying AQL
Treating AQL as a defect allowance. The most common mistake. AQL is not permission to ship 1% defective parts. It is the parameter of a risk-based inspection plan. A supplier shipping at exactly AQL will pass inspection most of the time — but "most of the time" still means some rejected lots and unhappy customers. The real target is always zero defects.
Using the wrong inspection level. Level II is standard, but some characteristics warrant Level I (less discrimination) or Level III (more). Safety and critical dimensions often require tighter AQL values or 100% inspection regardless of the sampling plan result.
Forgetting that each sample must be random. A sampling plan is only as good as its sampling method. Selecting parts from the top of the box, or only from one shift's production, defeats the statistical validity of the plan entirely. True random sampling — or stratified random sampling across the lot — is essential.
Not classifying defects before choosing AQL. Most plans distinguish critical, major, and minor defects, each with a different AQL. Critical defects (safety, regulatory) typically use AQL = 0.065% or even 0% (100% inspection). Major defects (function) use 0.65% or 1.0%. Minor defects (cosmetics) use 2.5% or 4.0%. Applying a single AQL to all defect types is a common and costly shortcut.
不良品の分類。致命欠点は安全・法規に関わる重大な欠陥（AQL=0.065%以下または全数検査）。重欠点は機能・性能に影響する欠陥（AQL=0.65%〜1.0%程度）。軽欠点は外観など軽微な欠陥（AQL=2.5%〜4.0%程度）。欠点分類ごとにAQLを設定することが正しいアプローチ。
When Not to Use Sampling Plans
Acceptance sampling is appropriate when the cost of inspection is high relative to the cost of passing a defective unit, when 100% inspection is impossible or impractical (destructive testing, perishable goods), and when the supplier has a known and reasonably stable quality history.
Sampling is not appropriate as the primary defense for safety-critical characteristics, when the supplier's process is known to be unstable, or when the defect rate is near or above the AQL. In those cases, 100% inspection is required — and even then, it should be seen as a temporary measure while the root cause is addressed, not as a permanent quality system.
Japanese quality philosophy, shaped by Deming and later Taguchi, has always viewed acceptance sampling as a necessary tool — but never as a substitute for building quality into the process. The ideal state is a process so capable that sampling becomes a formality. AQL tables are a starting point, not a destination.