Inspection, Acceptance Sampling and Nonconformity Control

Introduction

Quality management requires more than detecting defects at the end of production. An organisation must decide what to inspect, how much to inspect, when to accept or reject a lot, how to control nonconforming outputs, and how to prevent recurrence. These decisions affect customer protection, production cost, delivery performance, regulatory compliance and organisational reputation.

Inspection provides information about conformity. Acceptance sampling uses information from a sample to decide whether a lot should be accepted, rejected, or subjected to further investigation. Nonconformity control ensures that defective or otherwise unacceptable outputs are identified, contained, evaluated and prevented from reaching the customer. These activities should support process control rather than substitute for capable processes.

The central managerial challenge is balancing two opposing risks. Excessive inspection increases cost and may create a false sense of security, while inadequate inspection allows unacceptable products to escape. Sound practice therefore combines risk-based inspection, statistically designed sampling, traceability, disciplined disposition and effective corrective action.

Learning objectives

By the end of this chapter, the learner should be able to:

• distinguish inspection, verification, validation, defect and nonconformity;
• design an inspection strategy using risk, process capability and customer requirements;
• explain acceptance sampling by attributes and by variables;
• interpret acceptance-number plans and operating characteristic curves;
• calculate and interpret producer’s and consumer’s risks;
• control, segregate, trace and disposition nonconforming outputs;
• distinguish correction, containment, corrective action and preventive improvement; and
• evaluate the effectiveness of a nonconformity-control system in a production environment.

1. Foundations of inspection and conformity

1.1 Conformity and nonconformity

A requirement is a need or expectation that is stated, generally implied or obligatory. A product, service, process or system is conforming when it meets applicable requirements. A nonconformity is the non-fulfilment of a requirement. A defect is a particular type of nonconformity associated with intended or specified use; however, organisations should avoid assuming that every nonconformity has the same safety or functional significance.

Examples include:

• a shaft diameter outside its drawing tolerance;
• a missing safety label;
• an incorrect material certificate;
• a process record completed after the required time; and
• packaging that does not protect the product during transport.

A defect may be critical, major or minor according to its effect. Classification must be defined in advance. A cosmetic imperfection may be minor in one product and unacceptable in another if it affects sealing, safety or legal compliance.

1.2 Inspection, verification and validation

Inspection is the determination of conformity by observation and judgement, supported where appropriate by measurement, testing or gauging. Verification asks whether specified requirements have been met. Validation asks whether the resulting product or process is capable of fulfilling its intended use under specified conditions.

Inspection may occur at:

1. incoming material and supplier receipt;
2. first-off or setup approval;
3. in-process operations;
4. final release;
5. dispatch and packaging; and
6. customer return or field investigation.

Inspection is not identical to quality. Inspection detects information about quality; process design, capable equipment, trained personnel and effective controls create quality. If a process produces a nonconforming item at a high rate, inspecting every item may be less effective than correcting the process.

1.3 Inspection strategy

A practical inspection strategy answers five questions:

• What characteristic must be checked?
• At which stage can it be checked most effectively?
• What measurement method and equipment are suitable?
• How much evidence is required?
• What action follows a conforming or nonconforming result?

Inspection intensity should reflect risk. Safety-critical, legally regulated and high-cost characteristics generally require stronger controls than low-risk cosmetic characteristics. Risk assessment should consider severity, occurrence, detectability, process history and customer impact. Failure Mode and Effects Analysis (FMEA) can help prioritise controls, but its numerical rankings should support—not replace—engineering judgement and evidence.

A useful inspection plan identifies the characteristic, specification, measurement method, sample size or frequency, responsible person, equipment, acceptance criterion, records and reaction plan. Measurement systems must themselves be suitable. A poorly calibrated instrument can classify conforming product as nonconforming or conceal a real defect.

2. Inspection methods and measurement decisions

2.1 100% inspection

In 100% inspection, every unit is examined or tested. It is appropriate when failure consequences are severe, inspection is inexpensive and reliable, or the process has not yet demonstrated stability. It is not automatically perfect: inspectors can make errors, tests can be destructive, and repeated handling can damage product.

For a binary inspection in which each unit has an independent probability p of being defective, the probability of finding at least one defective unit in n inspected units is:

Here p and the result are probabilities, while n is the number of units. The expression assumes a constant defect probability and independent inspection outcomes. In real production, defects may cluster, so the independence assumption may be inappropriate.

2.2 Sampling inspection

Sampling inspection examines a subset of a lot. It reduces inspection cost and is useful when testing is destructive, when a supplier has demonstrated capability, or when a statistically defined decision is sufficient. Sampling never proves that every unit is conforming. It controls the probability of making an incorrect decision.

Sampling may be:

• random, giving each unit a known selection opportunity;
• systematic, selecting every kth unit after a random start;
• stratified, sampling separately from important subgroups; or
• convenience-based, which is easy but statistically weak and vulnerable to bias.

A sample should represent the lot. Selecting only the most accessible units, the first units produced or units chosen by an operator’s judgement can produce misleading assurance.

2.3 Measurement reliability

Before relying on inspection, the organisation should evaluate calibration, resolution, repeatability, reproducibility, bias, stability and measurement uncertainty. A measurement system that is too variable relative to the specification can cause unnecessary rejection or acceptance of unsuitable product.

For quantitative characteristics, the observed value can be represented conceptually as:

where e is measurement error in the same units as x. This is a measurement model, not a claim that error is always random or normally distributed. Bias, drift and operator effects must be considered separately.

3. Acceptance sampling

3.1 Purpose and terminology

Acceptance sampling is a procedure for deciding whether to accept a lot based on information from a sample. It is a decision tool, not a method for estimating the exact defect level of every lot.

Important terms are:

• lot: a defined quantity of product produced under specified conditions;
• sample size, n: number of units inspected;
• number of defectives, d: sampled units that fail requirements;
• number of defects: individual nonconformities, which may exceed the number of defective units;
• acceptance number, c: maximum permitted defects or defectives in the sample;
• rejection number: the count at which the lot is rejected; and
• acceptance quality limit (AQL): a process-quality reference used in designing a sampling plan, not a guarantee that every accepted lot has quality at or below that value.

A plan is often written as (n, c). The lot is accepted if d ≤ c and rejected if d > c. The plan must also specify what happens after rejection, such as 100% screening, rework, supplier investigation or formal concession.

3.2 Attributes and variables plans

An attributes plan records classifications such as conforming/nonconforming or counts defects. It is simple and practical when requirements are pass/fail or when measurement is expensive.

A variables plan uses measured values, such as diameter, mass or tensile strength. It can provide more information with a smaller sample when the distributional assumptions are justified and the measurement system is capable. Variables plans commonly require assumptions about distribution, independence and known or estimated variability. They are unsuitable if the characteristic is not meaningfully quantitative or if the distribution is seriously non-normal without an appropriate method.

3.3 Binomial model for an attributes plan

If a lot is treated as having a constant defective probability p and the n observations are independent, the probability of observing exactly d defectives is:

where D is the number of defectives, d = 0, 1, …, n, and \binom{n}{d} = \frac{n!}{d!(n-d)!}. For an acceptance number c, the probability of accepting a lot is:

This binomial form applies when sampling is effectively with replacement or when the lot is sufficiently large relative to n. For finite lots sampled without replacement, the hypergeometric distribution is more exact. If a lot contains N units, exactly K of which are defective, then:

The hypergeometric model requires a fixed finite lot and sampling without replacement. The binomial approximation should not be used casually for small lots or large sampling fractions.

Worked example: a single-sample plan

A plan has n = 50 and c = 1. Assume independent units and a defective probability p = 0.02. The probability of acceptance is:

P_a(0.02) = P(D=0)+P(D=1)

Thus, under the stated model, a lot with a 2% defective rate has approximately a 73.6% chance of acceptance. This does not mean that 73.6% of the lot is accepted; it is the long-run probability that the sampling decision accepts such lots. The plan may therefore be inappropriate if a 2% quality level is unacceptable.

4. Operating characteristic curves and sampling risks

4.1 The operating characteristic curve

An operating characteristic (OC) curve plots the probability of accepting a lot, P_a, against the lot’s quality level, often expressed as percentage defective or fraction defective. A good plan usually has a declining curve: acceptance probability is high for good quality and low for poor quality.

The curve is a decision profile, not a process-capability curve. It depends on n, c, the distributional model and the definition of quality. A steep curve discriminates more strongly between quality levels, usually requiring more inspection or a larger sample.

For the special case c = 0 under the binomial model:

For example, with n = 50 and c = 0, a 1% defective level gives P_a=0.99⁵⁰\approx0.605. Even a zero-acceptance plan does not guarantee rejection of every imperfect lot.

4.2 Producer’s and consumer’s risks

The producer’s risk, α, is the probability of rejecting a lot judged to be good at the producer’s quality level, commonly called the acceptable quality level (AQL). If the probability of acceptance at that level is 1 − α, then:

The consumer’s risk, β, is the probability of accepting a lot judged to be poor at the consumer’s quality level, commonly called the lot tolerance percent defective (LTPD) or rejectable quality level (RQL):

A plan should specify the quality points and risks rather than merely stating that it is “strict.” A plan with small α and small β generally needs more information than a plan with relaxed risks.

Worked risk interpretation

Suppose a sampling plan has P_a(0.01)=0.95 and P_a(0.10)=0.10. If 1% defective is defined as the producer’s good-quality point, then α = 1 − 0.95 = 0.05, or 5%. If 10% defective is the consumer’s poor-quality point, then β = 0.10, or 10%. The plan therefore rejects 5% of lots at the good-quality point and accepts 10% of lots at the poor-quality point, in the long run.

4.3 Limits of OC interpretation

An OC curve does not describe the consequences of a particular lot after it has been accepted. It also does not account automatically for inspection errors, switching rules, rectification, finite-lot effects, defect clustering or changing process conditions. When rejected lots are 100% screened and corrected, the outgoing quality may be substantially different from the incoming quality. Such plans should be evaluated using both acceptance probability and average outgoing quality.

5. Designing an acceptance-sampling approach

A defensible design sequence is:

1. define the lot and the quality characteristic;
2. classify consequences and regulatory requirements;
3. establish the good-quality and poor-quality points;
4. select attributes or variables inspection;
5. choose α and β in consultation with customer and supplier risk owners;
6. calculate or obtain a plan with suitable n and c;
7. examine the OC curve and expected inspection effort;
8. specify randomisation, test method and decision rules;
9. define treatment of rejected lots and records; and
10. review the plan using actual process performance.

Published standards, such as recognised ISO or national sampling standards, can provide indexed plans and switching rules. Their use requires correct identification of lot size, inspection level, AQL concept, type of characteristic and sampling scheme. A standard table should not be treated as a substitute for understanding the risks.

Sampling is inappropriate as the sole control when any nonconforming unit could create unacceptable safety or legal consequences, when the process is unstable, or when the customer requires zero escapes. In these cases, prevention, mistake-proofing, automated monitoring, validated testing or 100% inspection may be needed.

6. Control of nonconforming outputs

A nonconforming-output process should prevent unintended use or delivery. It normally includes the following controls:

• identification of the nonconforming item, lot or record;
• immediate segregation or electronic status control;
• protection against mixing, substitution or accidental release;
• documentation of the requirement that was not met;
• evaluation of safety, function, legal and customer impact;
• authorised disposition;
• verification of rework or repair; and
• records supporting traceability and release.

Typical dispositions are:

No employee should be able to release a rejected item merely by changing its status without authority. Electronic systems should restrict permissions, retain an audit trail and link disposition to the affected lot or serial number.

Containment, correction and corrective action

Containment limits the immediate effect, for example by stopping shipment and quarantining inventory. Correction removes a detected nonconformity, such as replacing a wrong label. Corrective action eliminates the cause of the nonconformity to prevent recurrence. These are distinct activities. Reworking every defective unit may correct the output while leaving the cause untouched.

Root-cause investigation should be proportionate to risk and evidence-based. Useful methods include the 5 Whys, cause-and-effect analysis, fault-tree analysis, process mapping and analysis of stratified data. A sound investigation asks why the existing control failed, not merely who made the error. Human error may indicate unclear instructions, poor interface design, inadequate training, excessive workload or weak error-proofing.

Corrective action should include a defined owner, due date, evidence, risk assessment and effectiveness check. Effectiveness may be demonstrated through sustained reduction in defect rate, audit evidence, capability data, complaint reduction or successful challenge testing. Closure based only on completion of an action is inadequate.

7. Traceability and containment systems

Traceability is the ability to follow the history, application, location or disposition of an item or activity. It may include supplier batch, raw-material heat number, machine, operator, date, inspection result, software version, packaging unit and customer shipment.

Traceability should be designed according to risk. Full serial-number traceability may be justified for aircraft components or medical devices, whereas batch traceability may be sufficient for a low-risk commodity. The organisation should define forward traceability—from material to customer—and backward traceability—from customer or product to material, process and records.

A robust containment procedure identifies:

1. the initiating signal, such as a failed test or complaint;
2. the time window and potentially affected lots;
3. inventory and customer locations;
4. physical and electronic quarantine controls;
5. risk evaluation and communication responsibilities;
6. inspection, recall or field action criteria; and
7. release criteria after investigation.

Digital manufacturing execution systems, barcode or RFID identification, vision inspection, automated test data and statistical alerts can improve speed and accuracy. Artificial intelligence may assist with image classification or anomaly detection, but it requires validated performance, representative training data, control of false negatives, cybersecurity and human accountability. An AI prediction should not silently override a mandatory safety inspection.

8. Illustrative case: containment of an out-of-tolerance shaft

Background

A fictional precision-component supplier produces shafts specified at 20.00 ± 0.05 mm. The shafts are shipped in batches of 1,000 units to an assembly plant. Final inspection uses a sampling plan, while the process is monitored by in-process gauging.

Problem

A customer reports that an assembly is difficult to fit. Investigation finds several shafts measuring 20.08 mm. The last accepted batch had been sampled under a plan with n = 80 and c = 1; one sampled unit was defective, so the lot was accepted.

Analysis

The sampling decision was consistent with the stated plan, but the event reveals a broader control weakness. The defect was concentrated after a tool change, violating the assumption that sampled units represented the entire lot. The sampling plan could not reliably detect a short-duration shift. The gauge also had no automatic lockout when the tool-life limit was reached. Traceability linked the customer complaint to two production hours, one machine and three pallets.

Approach and solution

The supplier immediately stopped shipment, quarantined finished goods and identified customer inventory using pallet and batch records. A 100% verified inspection was performed on the affected population. The tool-change procedure was revised, a first-off approval was introduced, and the machine was interlocked at the tool-life limit. Operators received revised instructions, and measurement-system performance was confirmed. The supplier retained sampling for routine lots but added intensified checks after setup and after abnormal signals.

Outcome

The affected customer stock was screened before use; no unsafe product entered service. Three consecutive production runs were monitored with enhanced inspection, and the corrective action was closed only after the process demonstrated stable results and no repeat escapes.

Learning points

The case illustrates that acceptance sampling is not a substitute for process control, that random sampling can miss time-related defect clusters, and that traceability determines the speed and scope of containment. It also demonstrates the need to connect inspection results with tool management, setup approval and corrective action.

9. Governance, ethics and professional judgement

Quality personnel have a duty to report results accurately, preserve records and prevent unauthorised release. Suppressing a failed result, altering a record, relabelling rejected material or accepting a concession without authority can endanger customers and expose the organisation to legal and contractual consequences.

Sampling plans should not be manipulated to produce a desired acceptance result. Customer requirements, statutory rules and safety obligations take precedence over cost reduction. Where evidence is uncertain, the decision should be conservative and escalated to an authorised technical authority. Confidentiality must also be balanced with the obligation to disclose safety or regulatory issues.

10. Best-practice checklist

Effective organisations generally:

• define critical characteristics and reaction plans before production;
• verify measurement-system suitability;
• use risk-based inspection rather than habitual inspection;
• understand the assumptions behind each sampling plan;
• review the OC curve and both producer and consumer risks;
• use random and representative sampling;
• maintain clear quarantine and status identification;
• link material, process and shipment records;
• investigate systemic causes rather than blaming individuals;
• verify rework and authorised concessions;
• measure corrective-action effectiveness; and
• reduce inspection as process capability and prevention improve, without weakening essential safeguards.

Chapter Summary

Inspection determines conformity, but it does not create quality by itself. Inspection strategies should place controls where risk can be detected or prevented most effectively and should rely on suitable measurement systems. Acceptance sampling makes a probabilistic decision about a lot using a defined sample size and acceptance number. Attributes plans classify units or count defects, while variables plans use measured characteristics under stronger distributional assumptions.

The binomial and hypergeometric models provide foundations for calculating acceptance probabilities. The OC curve shows how a plan behaves across quality levels. Producer’s risk is the probability of rejecting a good lot; consumer’s risk is the probability of accepting a poor lot. Neither sampling nor an OC curve guarantees that every accepted unit is conforming.

Nonconforming outputs must be identified, segregated, evaluated and given an authorised disposition. Containment limits immediate exposure, correction removes the detected problem, and corrective action addresses its cause. Traceability supports rapid, proportionate containment and reliable investigation. Effective systems combine prevention, process control, measurement assurance, sampling, ethical governance and continual improvement.

Key Terms

Inspection: Determination of conformity by observation, measurement, testing or judgement.

Conformity: Fulfilment of a specified requirement.

Nonconformity: Failure to fulfil a requirement.

Defect: A nonconformity associated with intended or specified use.

Lot: A defined quantity of product considered for a decision.

Sample size: Number of units selected for inspection.

Acceptance number: Maximum permitted sampled defectives under a plan.

Acceptance sampling: Lot decision based on sample evidence.

Attributes inspection: Inspection based on classifications or counts.

Variables inspection: Inspection based on measured numerical values.

Operating characteristic curve: Graph of acceptance probability against quality level.

Producer’s risk: Probability of rejecting a good-quality lot.

Consumer’s risk: Probability of accepting a poor-quality lot.

AQL: Reference good-quality level used in sampling-plan design.

LTPD/RQL: Poor-quality level used to express consumer protection.

Containment: Action that limits the spread or use of a nonconformity.

Correction: Action that removes a detected nonconformity.

Corrective action: Action that eliminates the cause of a nonconformity.

Traceability: Ability to follow an item’s history, application, location or disposition.

Disposition: Authorised decision concerning treatment of nonconforming output.

Review Questions

Short Answer Questions

1. Distinguish inspection from validation.
2. What is the purpose of an acceptance number?
3. Define producer’s risk and consumer’s risk.
4. Why must nonconforming outputs be identified and segregated?
5. Distinguish containment, correction and corrective action.

Long Answer Questions

1. Explain how risk, process capability and measurement-system suitability influence an inspection strategy.
2. Compare attributes and variables acceptance-sampling plans, including their assumptions and applications.
3. Explain how an OC curve supports selection and communication of a sampling plan.
4. Describe a complete process for controlling and disposing of nonconforming outputs.
5. Discuss the role of traceability in customer complaints, containment and corrective action.

Analytical / Application Questions

1. A plant accepts a lot using n = 50 and c = 1. Using the binomial model, calculate the probability of acceptance when p = 0.02 and explain whether this is adequate if 2% defective is unacceptable.
2. A defect occurs only during the first hour after machine setup. Explain why a simple random sample from the entire lot might fail to detect it and design a stronger inspection approach.
3. A supplier proposes reducing inspection because three lots have passed. Identify the evidence and risk analysis required before approving the change.

Multiple Choice Questions with Answers

Question 1

Which statement best describes acceptance sampling?

A. It proves that every unit in a lot is conforming.

B. It makes a probabilistic decision about a lot using sample evidence.

C. It eliminates the need for process control.

D. It is identical to 100% inspection.

Correct Answer: B. It makes a probabilistic decision about a lot using sample evidence.

Question 2

In a single-sample plan, what does the acceptance number c represent?

A. The total number of units in the lot.

B. The number of inspectors required.

C. The maximum permitted sampled defectives for acceptance.

D. The customer’s risk percentage.

Correct Answer: C. The maximum permitted sampled defectives for acceptance.

Question 3

For an attributes plan with independent observations, which expression gives the probability of exactly d defectives?

A. \binom{n}{d}p^d(1-p)^n-d

B. np(1-p)

C. 1-(1-p)ⁿ

D. pⁿ(1-p)^d

Correct Answer: A. \binom{n}{d}p^d(1-p)^n-d

Question 4

A plan has P_a(p_AQL)=0.95. What is the producer’s risk at the AQL point?

A. 0.05

B. 0.95

C. 5.95

D. 0.50

Correct Answer: A. 0.05

Question 5

Consumer’s risk is the probability of:

A. Rejecting a good lot.

B. Accepting a poor-quality lot.

C. Finding zero defects in a sample.

D. Repeating a measurement.

Correct Answer: B. Accepting a poor-quality lot.

Question 6

For c = 0 under the binomial model, the probability of accepting a lot is:

A. np

B. 1-pⁿ

C. (1-p)ⁿ

D. p(1-p)

Correct Answer: C. (1-p)ⁿ

Question 7

Which action is an example of containment?

A. Redesigning the process to eliminate the cause.

B. Quarantining potentially affected inventory.

C. Updating the annual budget.

D. Calibrating an unrelated instrument.

Correct Answer: B. Quarantining potentially affected inventory.

Question 8

Which disposition requires documented authorisation when the original requirement is not met?

A. Use as is.

B. Routine acceptance.

C. Normal dispatch.

D. Automatic release.

Correct Answer: A. Use as is.

Question 9

Why can a random sample fail to detect a setup-related defect?

A. Random samples always contain only conforming units.

B. The defect may be concentrated in a time segment not represented in the sample.

C. Sampling cannot be used in manufacturing.

D. Setup defects are never measurable.

Correct Answer: B. The defect may be concentrated in a time segment not represented in the sample.

Question 10

What is the primary purpose of corrective action?

A. To hide the nonconformity from the customer.

B. To eliminate the cause and prevent recurrence.

C. To increase the sample size only.

D. To replace all inspection records.

Correct Answer: B. To eliminate the cause and prevent recurrence.