11.2 Acceptance Sampling & ANSI/ASQ Z1.4

11.2 Acceptance Sampling & ANSI/ASQ Z1.4

Acceptance Sampling Principles and Economic Trade-offs

Acceptance sampling is a quality audit procedure in which a randomized sample of $n$ units is drawn from an inspection lot of $N$ units, inspected against predetermined acceptance criteria, and used to sentence the entire lot as either accepted or rejected (non-accepted).

ACCEPTANCE SAMPLING CONCEPTUAL FRAMEWORK:

   Incoming Lot (Size N)  ===>  Random Sample Drawn (Size n)  ===>  Inspect Sample
                                                                           |
                                                                           v
   Accept Entire Lot (Released) <=== Count Defectives (d) <= Ac <=== Count Defectives (d)
                                                                           |
   Reject Entire Lot (Disposition) <=== Count Defectives (d) >= Re <-------+

Acceptance Sampling versus Statistical Process Control (SPC)

A fundamental distinction tested on the ASQ CQT exam is the operational boundary between acceptance sampling and Statistical Process Control (SPC):

• Acceptance Sampling: An acceptance/screening tool applied after production. It evaluates past quality, accepts or rejects completed lots, and does not provide active process feedback or reduce process variability.
• Statistical Process Control (SPC): A prevention tool applied during production. It monitors process stability in real time using control charts (such as $\bar{X}$ and $R$ charts) to detect assignable causes of variation before nonconforming parts are manufactured.

Acceptance sampling is an audit tool; it does not control the process.

100% Inspection versus Sampling Inspection

Quality technicians must evaluate whether 100% inspection, sampling inspection, or no inspection is appropriate for a given manufacturing situation.

INSPECTION METHOD SELECTION SPECTRUM:

   No Inspection           Acceptance Sampling          100% Screening
   +-----------------------+----------------------------+-----------------------+
   | Extremely capable     | Moderate defect risk       | Critical safety flaw  |
   | supplier (Cpk > 2.0)  | Destructive testing        | Inexpensive screening |
   | Low defect consequence| High screening labor cost  | High liability cost   |
   | Standard COTS hardware| Visual fatigue risk in 100%| Non-destructive test  |
   +-----------------------+----------------------------+-----------------------+

1. The Fallacy of 100% Inspection: Shop-floor supervisors frequently assume that 100% screening guarantees zero defect escapes. In reality, human 100% inspection is subject to visual fatigue, boredom, distraction, and repetitive handling errors, typically achieving only 80% to 85% defect detection efficiency. Re-inspecting the same batch multiple times is often required to approach 99% defect removal.
2. The Mandate of Destructive Testing: When testing alters, damages, or permanently destroys the product—such as tensile weld pull tests, burst pressure testing of hydraulic hoses, ballistic armor penetration, or destructive metallographic cross-sectioning—100% inspection would consume the entire production inventory. Sampling inspection is mathematically and operationally mandatory.
3. Economic Balance (Cost of Inspection vs. Cost of Defect Escapes): Total quality cost under acceptance sampling balances the cost of inspecting sample units against the expected economic damage of passing defective units. If the cost of inspecting a unit ($C_i$) is very low compared to the cost of an escaping defect entering assembly ($C_d$), 100% inspection is economically indicated. The break-even quality fraction ($p^*$) is: $p^* = \frac{C_i}{C_d}$
   • If incoming defect rate $p > p^*$: 100% inspection is economically optimal.
   • If incoming defect rate $p < p^*$: Acceptance sampling (or no inspection) is economically optimal.

---

Acceptance Sampling Terminology

To design and interpret sampling plans, quality technicians utilize standardized metrological parameters:

• Lot Size ($N$): The collection of identical units produced under essentially uniform manufacturing conditions (same machine, tooling, material heat, and operator shift) presented for lot sentencing.
• Sample Size ($n$): The number of units randomly drawn from the lot without bias, representing the lot population.
• Acceptance Number ($Ac$ or $c$): The maximum allowable number of nonconforming units (or defects) observed in the sample that still permits the lot to be accepted.
• Rejection Number ($Re$): The minimum number of nonconforming units observed in the sample that mandates the lot be rejected (in single normal sampling, $Re = Ac + 1$).
• Nonconforming Unit vs. Nonconformity (Defect):
  • Nonconforming Unit (Defective): A unit of product containing one or more nonconformities.
  • Nonconformity (Defect): An individual departure of a quality characteristic from its specified requirement. A single nonconforming unit may contain multiple nonconformities.

---

The Operating Characteristic (OC) Curve

The Operating Characteristic (OC) curve is the mathematical foundation of acceptance sampling. It graphically portrays the performance of a sampling plan by displaying the probability of accepting a lot ($P_a$) as a function of the actual lot fraction nonconforming ($p$).

OPERATING CHARACTERISTIC (OC) CURVE ANATOMY:

   Probability of 
   Acceptance (Pa)
     1.00 +-------------+ 
          | *           |
     0.95 |    *        | <--- 1 - alpha = 0.95 (Producer's Risk alpha = 0.05 at AQL)
          |      *      |
          |        *    |
          |          *  |
     0.50 |            *|
          |             |
          |              *
     0.10 |               *  <--- Consumer's Risk beta = 0.10 at LTPD / LQ
          |                * 
     0.00 +--------+--------+--------+---------------->
                  AQL      LTPD     Incoming Fraction Defective (p)

Type A versus Type B OC Curves

• Type A OC Curve: Calculated for a finite lot of size $N$ where samples are drawn without replacement. Exact probabilities are computed using the Hypergeometric distribution: $P_a = \sum_{d=0}^{Ac} \frac{\binom{D}{d} \binom{N-D}{n-d}}{\binom{N}{n}}$ Where $D = N \times p$ is the total number of defective units in the finite lot.
• Type B OC Curve: Calculated for an infinite lot size or a continuous production stream where the lot size is large relative to sample size ($N \ge 10n$). Probabilities are computed using the Binomial distribution: $P_a = \sum_{d=0}^{Ac} \binom{n}{d} p^d (1 - p)^{n-d}$ When $n$ is large ($n \ge 16$) and $p$ is small ($p \le 0.10$), the Poisson distribution ($ P_a = \sum_{d=0}^{Ac} \frac{e^{-\mu} \mu^d}{d!}$, where $\mu = n \cdot p$) is used as a convenient and accurate approximation.

Key Quality Reference Points on the OC Curve

Every OC curve balances two conflicting risks across two critical quality benchmarks:

The Effect of Sample Size ($n$) and Discrimination

• The Ideal OC Curve: If an inspection plan could achieve perfection, the OC curve would be a vertical step function: $P_a = 1.0$ for all quality levels better than AQL, dropping vertically to $P_a = 0.0$ for all quality levels worse than AQL. This would eliminate both Producer's and Consumer's risks.
• The Sample Size Effect: In reality, because sampling relies on a finite $n$, the OC curve is S-shaped. As sample size $n$ increases (while keeping the ratio $Ac/n$ approximately constant), the OC curve steepens, approaching the vertical ideal. A steeper OC curve provides superior discriminating power, sharply distinguishing between good lots and bad lots.

EFFECT OF SAMPLE SIZE (n) ON OC CURVE DISCRIMINATION:

   Pa 1.0 +--------+
          | \   |  |   Steeper curve (Larger n = 125, Ac = 3):
          |  \  |  |   - Sharply discriminates good vs bad lots!
          |   \ |  |
          |    \|  |
          |     \  |   Flatter curve (Smaller n = 13, Ac = 0):
          |     |\ |   - Poor discrimination; rejects good lots and accepts bad lots!
      0.0 +-----+--+---------> Fraction Defective (p)

Average Outgoing Quality (AOQ) and AOQL

When non-accepted lots undergo rectifying inspection (rejected lots are 100% screened, all defective items removed, repaired or replaced with good units, and returned to the lot), the long-term quality escaping downstream is termed the Average Outgoing Quality (AOQ): $AOQ = \frac{P_a \cdot p \cdot (N - n)}{N} \approx P_a \cdot p$ As incoming fraction defective $p$ increases from zero, AOQ initially rises, reaches a maximum peak, and then decreases back toward zero (because severely bad lots are constantly rejected and 100% sorted). The maximum peak of the AOQ curve is the Average Outgoing Quality Limit (AOQL), representing the worst possible average quality escaping into the factory over the long run under rectifying inspection.

---

ANSI/ASQ Z1.4 (MIL-STD-105E) Sampling System

ANSI/ASQ Z1.4 (the civilian successor to military standard MIL-STD-105E, also harmonized as ISO 2859-1) is the most widely utilized acceptance sampling system for attribute inspection (pass/fail, go/no-go, defect counts) worldwide.

Structure and Core Philosophy of Z1.4

• AQL-Indexed: Z1.4 is indexed by the Acceptable Quality Limit (AQL). The user specifies an agreed-upon AQL (e.g., AQL 0.65%, 1.0%, 2.5%, 4.0%), and the standard provides sampling plans designed to protect the producer by ensuring that lots conforming to AQL have a high probability of acceptance ($P_a \approx 0.88$ to $0.99$, typically $0.95$).
• Lot-by-Lot Continual Production: Z1.4 is designed specifically for a continuous series of lots from an ongoing production line. It is not intended for isolated, one-off lots (where LTPD-indexed standards like Dodge-Romig or ANSI/ASQ Z1.4 Section 11 limiting quality tables should be used).

General Inspection Levels versus Special Inspection Levels

Z1.4 provides seven distinct inspection levels, which dictate the relationship between the lot size ($N$) and the sample size ($n$):

Z1.4 INSPECTION LEVELS HIERARCHY:

   SPECIAL LEVELS (Small Samples)       GENERAL LEVELS (Standard Discrimination)
   +-----+-----+-----+-----+            +-----------+-----------+-----------+
   | S-1 | S-2 | S-3 | S-4 |            |  Level I  | Level II  | Level III |
   +-----+-----+-----+-----+            +-----------+-----------+-----------+
   <--- Destructive / High Cost         Reduced      STANDARD    Tightened
        Large Consumer Risk (beta)      Discrim.     DEFAULT     Discrim.

1. General Inspection Levels (I, II, III):
   • Level II (Normal Default): The standard baseline level. Unless an engineering drawing, contract, or specification explicitly mandates otherwise, Level II must always be used.
   • Level I (Reduced Discrimination): Requires approximately half the sample size of Level II (~40% to 50%). Utilized when less discrimination can be tolerated, or when a supplier has a proven history of exceptional quality.
   • Level III (Tightened Discrimination): Requires approximately 1.5 times the sample size of Level II (~150%). Used when the cost of an escaping defect is high, or when supplier quality is suspect.
2. Special Inspection Levels (S-1, S-2, S-3, S-4):
   • Provide very small sample sizes.
   • Z1.4 states that they may be used where relatively small sample sizes are necessary and large sampling risks can or must be tolerated. In practice that means destructive testing, highly expensive laboratory testing, and tests requiring extended environmental exposure, where large sample sizes are economically prohibitive. These levels accept substantial consumer risk ($\beta$) in exchange for manageable testing costs, and the standard warns against pairing them with AQLs that are inconsistent with such small samples.

Table I: Determining Sample Size Code Letters

To use ANSI/ASQ Z1.4, a quality technician cross-references the Lot Size ($N$) with the chosen Inspection Level in Table I to find the Sample Size Code Letter (letters A through R).

Table II-A Excerpt: Single Sampling Plans for Normal Inspection

Once the Code Letter is determined, the technician opens Table II-A, which cross-references the Code Letter (and its associated sample size $n$) against the designated AQL to locate the Acceptance ($Ac$) and Rejection ($Re$) numbers.

[!CAUTION] The Mandatory Arrow Rule in Z1.4 Tables:

• Downward Arrow ($\downarrow$): If the intersection of the Code Letter row and AQL column contains a downward arrow, follow the arrow down to the first plan below. You must adopt both the new sample size ($n$) and the new $Ac/Re$ numbers associated with that lower row!
• Upward Arrow ($\uparrow$): If the intersection contains an upward arrow, follow the arrow up to the first plan above. You must adopt both the new sample size ($n$) and the new $Ac/Re$ numbers associated with that upper row!
• Common ASQ CQT Exam Trap: Candidates often look up a downward arrow, adopt the new $Ac/Re$ values, but keep their original sample size. This completely invalidates the statistical plan!

---

Single, Double, and Multiple Sampling Plans

ANSI/ASQ Z1.4 provides three primary sampling plan architectures, each presenting operational trade-offs:

SAMPLING SCHEME DECISION PATHWAYS:

   SINGLE SAMPLING:       [ Draw n ] ---------> d <= Ac (Accept) OR d >= Re (Reject)

   DOUBLE SAMPLING:       [ Draw n1 ] --------> d1 <= Ac1 (Accept) OR d1 >= Re1 (Reject)
                               |
                               v  (If Ac1 < d1 < Re1)
                          [ Draw n2 ] --------> (d1 + d2) <= Ac2 (Accept) OR >= Re2 (Reject)

   MULTIPLE SAMPLING:     Successive small samples across up to 7 progressive stages

Operational Advantages of Double Sampling

Beyond reducing the Average Sample Number (ASN), double sampling provides a distinct psychological advantage on the shop floor. When a vendor's lot is marginally defective on the first sample ($Ac_1 < d_1 < Re_1$), the vendor is granted a "second chance" via sample $n_2$. This alleviates the contentious disputes often triggered when a large lot is rejected based on a single isolated failure in single sampling.