21.3 Sampling Risks, AQL, LTPD, and Standard Sampling Plans

Designing an effective acceptance sampling system requires balancing statistical risks against inspection economics. Because sampling inspects only a fraction of a lot, two statistical risks are unavoidable: rejecting an acceptable lot (Producer's Risk, $\alpha$) and accepting an unacceptable lot (Consumer's Risk, $\beta$). To manage these risks across supply chains, quality engineers utilize standardized sampling systems—most notably ANSI/ASQ Z1.4 for attribute inspection and ANSI/ASQ Z1.9 for variable measurement—and model outgoing quality via rectifying inspection frameworks. On the FE exam, mastery of AQL, LTPD, AOQ, AOQL, ATI, ASN, and switching rules is mandatory.

---

1. The Four Key Operating Parameters on an OC Curve

Every sampling plan is designed around two designated coordinate points on its Operating Characteristic (OC) curve, representing the conflicting interests of the producer (seller) and consumer (buyer):

                 The Four Key Points on an OC Curve

       Pa
      1.0 ┼───────┐
          │        \ 
 1-alpha  │.........\....... Point 1: (AQL, 1 - alpha)
   (0.95) │          \       Producer's Risk: alpha = 0.05
          │           \
          │            \
          │             \  Indifference Zone
          │              \
          │               \
    beta  │................\... Point 2: (LTPD, beta)
   (0.10) │                 \__ Consumer's Risk: beta = 0.10
      0.0 ┴───────┬──────────┬────────►
          0      AQL        LTPD       Incoming Fraction Defective (p)

1. Acceptable Quality Level (AQL) and Producer's Risk ($\alpha$)

• Acceptable Quality Level (AQL): The poorest incoming quality level (expressed as percent defective or fraction defective $p$) that the consumer considers satisfactory as a long-term supplier process average.
• Producer's Risk ($\alpha$): The probability that a sampling plan will erroneously reject an acceptable lot having quality equal to the AQL: $\alpha = 1 - P_a(\text{AQL})$
• In statistical hypothesis testing, $\alpha$ represents a Type I error (rejecting the null hypothesis $H_0: p \le \text{AQL}$ when it is true).
• Industrial standard: $\alpha$ is conventionally set to $\alpha = 0.05$ ($5%$), meaning that a good lot meeting the AQL has a $95%$ probability of acceptance ($P_a = 0.95$).

2. Lot Tolerance Percent Defective (LTPD / RQL) and Consumer's Risk ($\beta$)

• Lot Tolerance Percent Defective (LTPD) (also designated Rejectable Quality Level, RQL): The unacceptable incoming quality level that the consumer wishes to reject consistently.
• Consumer's Risk ($\beta$): The probability that a sampling plan will erroneously accept an unsatisfactory lot having quality equal to the LTPD: $\beta = P_a(\text{LTPD})$
• In statistical hypothesis testing, $\beta$ represents a Type II error (failing to reject the null hypothesis when the alternative $H_1: p \ge \text{LTPD}$ is true).
• Industrial standard: $\beta$ is conventionally set to $\beta = 0.10$ ($10%$), meaning that a lot containing LTPD defectives has only a $10%$ chance of slipping through inspection ($P_a = 0.10$).

The Indifference Zone and Discrimination Ratio

The quality interval between AQL and LTPD is the indifference zone. Lots falling in this range are neither overwhelmingly good nor catastrophically bad. The ratio $\frac{\text{LTPD}}{\text{AQL}}$ is the discrimination ratio:

• A smaller ratio (approaching 1.0) requires a steeper OC curve and therefore a much larger sample size $n$.
• A larger ratio represents a looser filter requiring smaller, less expensive sample sizes.

---

2. Rectifying Inspection and Average Outgoing Quality (AOQ)

In many industrial manufacturing environments, rejected lots are not scrapped or returned to the supplier. Instead, they are subjected to rectifying inspection (also called screening inspection):

                    Rectifying Inspection Program

                       Incoming Lot of Size N
                                  │
                                  ▼
                        Sample of Size n Drawn
                                  │
                 ┌────────────────┴────────────────┐
                 ▼                                 ▼
          Lot is Accepted                   Lot is Rejected
          (Probability Pa)                 (Probability 1 - Pa)
                 │                                 │
                 ▼                                 ▼
        Uninspected (N - n) units         100% Screening Inspection:
        pass to production with           All N units are inspected;
        original fraction defective p.    ALL defectives replaced with
                 │                        conforming units.
                 │                        Defect rate = 0.000
                 │                                 │
                 └────────────────┬────────────────┘
                                  ▼
                       Average Outgoing Quality
                          AOQ = Pa * p * (N - n) / N

The Mathematical Formula for AOQ

Under rectifying inspection, any defectives discovered in the $n$ sample units are repaired or replaced with good units. If the lot is accepted (probability $P_a$), the remaining uninspected $(N - n)$ units still contain defectives at rate $p$. If the lot is rejected (probability $1 - P_a$), the entire lot undergoes $100%$ screening, leaving zero defectives.

The expected Average Outgoing Quality (AOQ) is:

$\text{AOQ} = \frac{P_a \cdot p \cdot (N - n)}{N}$

When the lot size $N$ is very large relative to sample size $n$ ($N \gg n$), the ratio $\frac{N - n}{N} \approx 1$, yielding the widely used engineering approximation:

$\text{AOQ} \approx P_a \cdot p$

               Average Outgoing Quality (AOQ) Curve and AOQL

       AOQ
        ▲
        │                 AOQL (Maximum Outgoing Defects)
   AOQL ┼..................┌───┐
        │                ┌─┘   └─┐
        │              ┌─┘       └─┐
        │            ┌─┘           └─┐
        │          ┌─┘               └─┐
        │        ┌─┘                   └─┐
        │      ┌─┘                       └─┐  As p -> 1.0, Pa -> 0,
        │    ┌─┘                           └── 100% screening cleans
      0 ┴────┴─────────────────────────────────► all rejected lots!
        0                   p_max            1.0   Incoming Quality (p)

The Average Outgoing Quality Limit (AOQL)

The behavior of the AOQ curve reveals a profound engineering insight:

1. When incoming quality is perfect ($p = 0$), outgoing quality is obviously perfect ($\text{AOQ} = 0$).
2. As incoming defect fraction $p$ rises moderately, $P_a$ remains close to 1.0, so defectives pass through uninspected and $\text{AOQ}$ increases.
3. As $p$ becomes very poor, the sampling plan rejects nearly every lot ($P_a \to 0$). Almost all lots undergo $100%$ screening, and defective units are completely replaced. Consequently, outgoing quality improves and $\text{AOQ}$ drops back toward zero!

Because the AOQ curve rises to a single peak and then declines, it possesses an absolute mathematical maximum called the Average Outgoing Quality Limit (AOQL):

$\text{AOQL} = \max_{0 \le p \le 1} \text{AOQ}(p)$

Strategic Meaning of AOQL: Regardless of how terrible incoming supplier quality becomes—even if the supplier sends $10%$, $20%$, or $50%$ defective lots—the average long-term outgoing defect rate entering your production facility can never exceed the AOQL! The rectifying sampling filter guarantees this upper bound.

---

3. Average Total Inspection (ATI)

Under rectifying inspection, the industrial engineer must budget for the inspection labor required to process incoming shipments. The Average Total Inspection (ATI) represents the expected total number of units inspected per lot:

$\text{ATI} = n \cdot P_a + N \cdot (1 - P_a)$

Factoring algebraically:

$\text{ATI} = n + (1 - P_a)(N - n)$

Interpretation of Limiting Cases

• If incoming quality is outstanding ($p \to 0, P_a \to 1.0$): $\text{ATI} = n + (0)(N - n) = n$ Only the initial sample of $n$ units is inspected. Inspection cost is minimized.
• If incoming quality is terrible ($p \to 1, P_a \to 0.0$): $\text{ATI} = n + (1.0)(N - n) = N$ Every single lot is rejected and $100%$ screened. Inspection cost reaches its absolute maximum.

---

4. Double and Multiple Sampling Plans & Average Sample Number (ASN)

Single sampling plans force a definitive sentence from one sample of size $n$. However, lots that are exceptionally high in quality (e.g., zero defectives) or catastrophically poor (e.g., dozens of defectives) do not require inspecting all $n$ units to reach an obvious conclusion. Double sampling plans capitalize on this efficiency.

                     Double Sampling Plan Protocol

                         Sample 1: Size n1
                         Count Defectives d1
                                  │
         ┌────────────────────────┼────────────────────────┐
         ▼                        ▼                        ▼
     d1 <= c1              c1 < d1 < r1                 d1 >= r1
  ┌──────────────┐         ┌──────────────┐         ┌──────────────┐
  │  ACCEPT LOT  │         │   SAMPLE 2   │         │  REJECT LOT  │
  └──────────────┘         └──────┬───────┘         └──────────────┘
                                  │
                                  ▼
                           Sample 2: Size n2
                           Count Defectives d2
                           Cumulative: d1 + d2
                                  │
                     ┌────────────┴────────────┐
                     ▼                         ▼
              d1 + d2 <= c2              d1 + d2 >= r2
             ┌──────────────┐          ┌──────────────┐
             │  ACCEPT LOT  │          │  REJECT LOT  │
             └──────────────┘          └──────────────┘

Double Sampling Plan Parameters

A double sampling plan is specified by five parameters: $(n_1, c_1, r_1, n_2, c_2, r_2)$, where typically $r_2 = c_2 + 1$:

1. Draw sample 1 of size $n_1$. Let $d_1$ be the observed defective count:
   • If $d_1 \le c_1$: Accept the lot immediately.
   • If $d_1 \ge r_1$: Reject the lot immediately.
   • If $c_1 < d_1 < r_1$: Decision is deferred; draw sample 2 of size $n_2$.
2. Inspect sample 2 of size $n_2$. Let $d_2$ be the observed defectives in the second sample:
   • If $d_1 + d_2 \le c_2$: Accept the lot.
   • If $d_1 + d_2 \ge r_2$: Reject the lot.

Average Sample Number (ASN)

The Average Sample Number (ASN) measures the expected number of units inspected per lot to reach a sentencing decision under the sampling plan (without rectifying inspection):

• For Single Sampling: $\text{ASN} = n$ (constant regardless of quality).
• For Double Sampling (without curtailment): $\text{ASN} = n_1 \cdot P(\text{decision on sample 1}) + (n_1 + n_2) \cdot P(\text{sample 2 required})$ Since $P(\text{decision on sample 1}) + P(\text{sample 2 required}) = 1$: $\text{ASN} = n_1 + n_2 \cdot P(c_1 < d_1 < r_1)$

Economic Advantage of Double Sampling

For lots of either very high quality ($p \ll \text{AQL}$) or very low quality ($p \gg \text{LTPD}$), double sampling plans reach a decision on the first sample ($n_1$). Because $n_1$ is typically only $50%$ to $65%$ of the equivalent single sample size $n_{\text{single}}$, double sampling reduces overall inspection labor by $25%$ to $50%$ for typical production lots while maintaining identical risk protection ($\alpha, \beta$).

---

5. Standard Attribute Sampling System: ANSI/ASQ Z1.4 (ISO 2859)

ANSI/ASQ Z1.4 (the civilian successor to military standard MIL-STD-105E and equivalent to international standard ISO 2859) is the world's most widely adopted attribute sampling system. It is an AQL-indexed system designed for continuous production streams.

1. Inspection Levels

The standard establishes three General Inspection Levels and four Special Inspection Levels:

• General Level II: The standard, default inspection level used unless otherwise specified.
• General Level I: Used when less discrimination can be tolerated. Sample sizes are roughly half that of Level II, cutting inspection cost at the expense of higher risk.
• General Level III: Used when high discrimination is required. Sample sizes are roughly $1.5\times$ that of Level II, steepening the OC curve.
• Special Levels (S-1, S-2, S-3, S-4): Employ extremely small sample sizes for destructive testing or expensive testing environments.

2. Sample Size Code Letters

To find a sampling plan:

1. The user locates the Lot Size $N$ and designated Inspection Level (e.g., Level II) in Table I of the standard to determine the Sample Size Code Letter (letters A through R).
2. The user cross-references the Code Letter and desired AQL in Table II-A to read the required sample size $n$, acceptance number $Ac$ ($c$), and rejection number $Re$ ($r$).

              ANSI/ASQ Z1.4 Dynamic Switching Rules Architecture

                  ┌─────────────────────────────────────┐
                  │         NORMAL INSPECTION           │◄────────────┐
                  │       (Default Starting Mode)       │             │
                  └───────┬─────────────────────▲───────┘             │
                          │                     │                     │
            2 of 5 lots   │                     │ 5 consecutive       │ 1 lot rejected;
             rejected     │                     │ lots accepted       │ or production
                          ▼                     │                     │ irregular
                  ┌─────────────────────┐       │                     │
                  │ TIGHTENED INSPECTION├───────┘                     │
                  └───────┬─────────────┘                             │
                          │                                           │
            10 lots stay  │                                           │
            on tightened  │                                           │
                          ▼                                           │
                  ┌─────────────────────┐                             │
                  │   STOP SHIPMENTS    │                             │
                  │ (Discontinue Insp.) │                             │
                  └─────────────────────┘                             │
                                                                      │
                  ┌─────────────────────────────────────┐             │
                  │         REDUCED INSPECTION          ├─────────────┘
                  │  (10 lots accepted + steady prod.)  │
                  └─────────────────────────────────────┘

3. Dynamic Switching Rules (Critical Exam Topic)

ANSI/ASQ Z1.4 enforces dynamic switching rules between three inspection severity modes: Normal, Tightened, and Reduced:

---

6. Standard Variable Sampling System: ANSI/ASQ Z1.9 (ISO 3951)

While Z1.4 evaluates pass/fail attributes, ANSI/ASQ Z1.9 (successor to MIL-STD-414, equivalent to ISO 3951) governs sampling by variables.

Operational Principles

Instead of counting defective units, variable sampling measures a continuous numerical dimension (e.g., shaft diameter in mm, tensile yield strength in MPa, resistance in ohms). The sample mean $\bar{X}$ and sample standard deviation $s$ are calculated, and standardized quality statistics are computed relative to specifications:

$Q_U = \frac{\text{USL} - \bar{X}}{s} \quad \text{and} \quad Q_L = \frac{\bar{X} - \text{LSL}}{s}$

The estimated percent defective beyond $\text{USL}$ and $\text{LSL}$ is determined from standard tables based on $Q_U$ and $Q_L$. If the total estimated percent defective is less than or equal to a maximum allowable limit $M$, the lot is accepted.

Variable Sampling (Z1.9) vs. Attribute Sampling (Z1.4)

Exam Key: Variable sampling plans provide equivalent $\alpha$ and $\beta$ protection with vastly smaller sample sizes than attribute plans because continuous measurements carry far more statistical information than binary tallies. However, variable sampling strictly requires normal distribution of data.

---

7. Step-by-Step Worked Engineering Calculations

Worked Example 21.3.1: Rectifying Inspection, AOQ, and ATI Analysis

Problem: A receiving inspection department evaluates incoming shipments of precision aluminum fasteners in lots of size $N = 4,000$. The established single sampling plan has a sample size of $n = 120$ and an acceptance number of $c = 2$. Rejected lots undergo $100%$ rectifying screening inspection, where all defective fasteners are replaced with conforming fasteners at vendor expense.

The supplier's quality level currently averages $p = 0.0125$ ($1.25%$ defective).

Required:

1. Using the Poisson approximation, calculate the probability of lot acceptance ($P_a$).
2. Compute the exact Average Outgoing Quality (AOQ) in parts per million (PPM).
3. Compute the approximate AOQ and calculate the percentage error introduced by the approximation.
4. Calculate the Average Total Inspection (ATI) per lot.
5. If the supplier's quality deteriorates to $p = 0.050$ ($5.0%$ defective), recalculate $P_a$, AOQ, and ATI. Discuss the economic consequence.

Solution:

Step 1: Compute Acceptance Probability at $p = 0.0125$

• Poisson parameter: $\lambda = n \cdot p = 120 \times 0.0125 = 1.50$
• Probability of observing $d \le 2$ defectives: $P_a = e^{-1.50} \left( 1 + 1.50 + \frac{1.50^2}{2} \right) = e^{-1.50} (1 + 1.50 + 1.125) = e^{-1.50} \times 3.625$ $P_a = 0.22313 \times 3.625 = 0.80884 \approx 0.8088$ Roughly $80.9%$ of lots are accepted on the sample. Producer's Risk is $\alpha = 1 - 0.8088 = 0.1912$ ($19.1%$).

Step 2: Compute Exact Average Outgoing Quality (AOQ) $\text{AOQ}_{\text{exact}} = \frac{P_a \cdot p \cdot (N - n)}{N} = \frac{0.80884 \times 0.0125 \times (4,000 - 120)}{4,000}$ $\text{AOQ}_{\text{exact}} = \frac{0.80884 \times 0.0125 \times 3,880}{4,000} = \frac{39.2287}{4,000} = 0.009807$ $\text{AOQ}_{\text{exact}} = 0.9807\% = 9,807\text{ PPM}$

Step 3: Compute Approximate AOQ and Error $\text{AOQ}_{\text{approx}} = P_a \cdot p = 0.80884 \times 0.0125 = 0.010111 = 1.0111\% = 10,111\text{ PPM}$

• Relative error: $\text{Error} = \frac{10,111 - 9,807}{9,807} \times 100\% = +3.10\%$ The approximation slightly overstates outgoing defects by $3.1%$ because it ignores the fact that the $n = 120$ sampled units were fully cleared of defectives.

Step 4: Compute Average Total Inspection (ATI) $\text{ATI} = n + (1 - P_a)(N - n) = 120 + (1 - 0.80884)(4,000 - 120)$ $\text{ATI} = 120 + (0.19116)(3,880) = 120 + 741.7 = 861.7 \approx 862\text{ units inspected per lot}$

Step 5: Impact of Quality Deterioration to $p = 0.050$

• New Poisson parameter: $\lambda_{\text{new}} = 120 \times 0.050 = 6.00$
• New acceptance probability: $P_a = e^{-6.0} \left( 1 + 6.0 + \frac{6.0^2}{2} \right) = e^{-6.0} (1 + 6.0 + 18.0) = 25.0 \times e^{-6.0}$ $P_a = 25.0 \times 0.0024788 = 0.06197 \approx 0.0620$ Acceptance probability crashes from $80.9%$ to just $6.2%$!
• New Outgoing Quality: $\text{AOQ} = \frac{0.06197 \times 0.050 \times 3,880}{4,000} = 0.003006 = 0.3006\% = 3,006\text{ PPM}$ Notice that AOQ actually improved from 9,807 PPM down to 3,006 PPM! Why? Because $93.8%$ of all shipments were rejected and $100%$ screened, purging defects!
• New Average Total Inspection: $\text{ATI} = 120 + (1 - 0.06197)(3,880) = 120 + (0.93803)(3,880) = 120 + 3,639.6 = 3,760\text{ units per lot}$
• Economic Conclusion: Outgoing quality remains protected at the cost of massive inspection workload. The facility must now inspect 3,760 out of 4,000 units per lot ($94%$ inspection workload), dramatically increasing screening costs and causing severe receiving dock gridlock.

---

8. NCEES Reference Handbook Tips & Realistic Exam Traps

• Trap: Producer's vs. Consumer's Risk Coordinates: On conceptual exam questions, keep these coordinates clear in your mind:
  • Producer's Risk ($\alpha$): Associated with AQL. $\alpha = 1 - P_a$. Good lots rejected. Hurting the supplier.
  • Consumer's Risk ($\beta$): Associated with LTPD / RQL. $\beta = P_a$. Bad lots accepted. Hurting the customer.
• Trap: Neglecting $(N - n)/N$ in AOQ: When $N$ is small or moderate ($N \le 1,000$) and $n$ is significant ($n \ge 100$), always use the full formula $\text{AOQ} = \frac{P_a \cdot p(N - n)}{N}$. For example, if $N = 500$ and $n = 100$, $(N - n)/N = 0.80$, so omitting the term introduces a massive $25%$ error.
• Trap: ANSI/ASQ Z1.4 Switching Rules Count: Memorize the exact switching rules for Z1.4:
  • Normal to Tightened: 2 out of 5 consecutive lots rejected.
  • Tightened to Normal: 5 consecutive lots accepted.
  • Discontinue Inspection: 10 consecutive lots stay on Tightened.
  • Normal to Reduced: 10 consecutive lots accepted with steady production. A common exam distractor states "Switch to tightened when 1 lot is rejected"—this is false; it requires 2 out of 5 consecutive rejections.
• Trap: Variable Sampling Normality Requirement: An exam question might ask: "Which sampling system should be selected when sample size must be minimized for a continuous dimensional tolerance, assuming the process is known to be non-normal?" If the distribution is non-normal, ANSI/ASQ Z1.9 cannot be used directly without transformation; attribute sampling (Z1.4) or non-parametric methods must be selected.