Inspection and Acceptance Sampling

Introduction and context

Inspection is the systematic examination, measurement, testing, or comparison of a product, service, material, or process against specified requirements. Acceptance sampling is a statistical decision procedure used to accept or reject a lot on the basis of information obtained from a sample. Together, they connect quality planning with operational control.

Inspection does not create quality; it detects evidence about quality. Quality must primarily be built into product and process design, supplier controls, capable processes, standard work, and prevention activities. Nevertheless, inspection remains essential when conformity is safety-critical, when incoming materials affect downstream performance, when process verification is required, or when the customer requires documented evidence of conformity.

A sound inspection system answers five questions:

1. What characteristic is to be inspected?
2. Against which specification or acceptance criterion?
3. At what stage should inspection occur?
4. How much should be inspected?
5. What action follows a conforming or nonconforming result?

Learning objectives

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

• distinguish incoming, in-process, and final inspection;
• select inspection characteristics, methods, equipment, and records;
• explain 100% inspection, sampling inspection, and their limitations;
• interpret sampling plans, AQL, producer’s risk, consumer’s risk, and operating characteristic curves;
• calculate simple acceptance probabilities using the binomial model;
• decide when full inspection, sampling, or process control is appropriate; and
• design a practical and ethically defensible inspection and acceptance plan.

1. Inspection as part of the quality system

Inspection may be classified by timing, purpose, and method. Timing determines where information enters the process; purpose determines what decision is made; method determines how evidence is obtained.

1.1 Incoming inspection

Incoming inspection evaluates materials, components, packaging, and services received from external suppliers. It may include document review, identity verification, visual examination, dimensional measurement, functional testing, or laboratory analysis. The extent of incoming inspection should reflect supplier capability, historical performance, material criticality, and the consequences of failure.

A mature organization does not automatically inspect everything received. It may use approved-supplier status, supplier certificates, second-party audits, statistical sampling, dock-to-stock release, and periodic verification. However, reliance on supplier information must be justified by evidence and periodically revalidated.

1.2 In-process inspection

In-process inspection verifies that work remains within requirements while transformation is occurring. Examples include checking machining dimensions after a setup, monitoring temperature during heat treatment, verifying torque during assembly, or testing software builds before integration. In-process inspection is generally more valuable than late detection because defects can be corrected before additional value is added.

Inspection should be positioned at control points identified through process mapping, failure mode and effects analysis, hazard analysis, or regulatory requirements. The best control point is often immediately after the operation that can create the defect and before the defect can be concealed or propagated.

1.3 Final inspection

Final inspection confirms that the completed product or service satisfies specified requirements before release. It may include a review of records, appearance, dimensions, performance, labeling, packaging, and required certificates. Final inspection is necessary when release requires an overall conformity decision, but it should not be the sole quality strategy. A final inspection can detect a defect without explaining its cause, and it cannot recover all costs of rework, delay, or reputational damage.

1.4 Inspection characteristics

A characteristic may be measured as a variable, such as length, mass, voltage, or strength, or classified as an attribute, such as conforming/nonconforming, present/absent, or pass/fail. The inspection method must be validated for accuracy, precision, resolution, repeatability, and reproducibility.

A measurement result should be traceable to a defined method and, where relevant, calibrated equipment. If a measured value is close to a specification limit, measurement uncertainty may affect the conformity decision. Organizations should therefore define decision rules, guard bands, and escalation procedures rather than allowing inspectors to make inconsistent personal judgments.

2. Inspection planning

An inspection plan is a controlled document linking requirements to methods, frequency, responsibility, sampling, and reaction. A useful inspection plan normally specifies:

Inspection planning begins with risk. Characteristics affecting safety, legal compliance, interchangeability, or essential performance generally receive stronger controls. Critical characteristics may require 100% verification, automated error-proofing, or independent testing. Less critical characteristics may be controlled through process capability and periodic audits.

The inspection method must be operationally clear. “Check quality” is inadequate; “measure bore diameter with calibrated gauge G-14 at 20 ± 2 °C; accept 24.98–25.02 mm” is actionable. Sampling instructions must specify how units are selected. A convenient or first-in-line sample is not statistically representative unless the plan explicitly makes that assumption defensible.

3. Full inspection and sampling inspection

3.1 100% inspection

Full inspection examines every unit in a defined population. It is appropriate when:

• failure could cause serious injury or regulatory violation;
• the inspection is rapid, reliable, and inexpensive;
• the lot is small or the cost of failure is very high;
• defects are highly detectable and inspection does not damage the product; or
• a customer or legal requirement explicitly demands it.

Full inspection is not automatically perfect. Inspectors can miss defects, gauges can be miscalibrated, repeated handling can cause damage, and visual fatigue can reduce detection performance. If the probability that an inspector correctly detects a defect is less than one, some defects may escape even under 100% inspection.

3.2 Sampling inspection

Sampling inspection examines fewer than all units and uses the sample result to make a decision about the lot. Its advantages include lower cost, shorter release time, less destructive testing, and reduced inspection fatigue. Its central disadvantage is risk: a good lot may be rejected and a poor lot may be accepted.

Sampling is appropriate when testing is destructive, the supplier or process is demonstrably stable, the lot is large, inspection cost is significant, or a quantified level of protection is acceptable. Sampling is inappropriate when every unit must be known to conform, when a single defect has unacceptable consequences, or when the process is unstable and the sample cannot represent future performance.

3.3 Prevention versus detection

Inspection should be integrated with prevention. Statistical process control, mistake-proofing, automated interlocks, supplier development, preventive maintenance, and standardized work often reduce defects more effectively than increasing sample size. A sampling plan is a decision mechanism, not a substitute for process improvement.

4. Acceptance sampling terminology and models

A lot is a defined quantity of product, material, or service presented for acceptance. A sample is a subset selected from that lot. A sampling plan specifies sample size and acceptance criteria. For an attributes plan, the plan is commonly written as n units sampled and the lot accepted if the number of nonconforming units is no greater than c.

A unit is nonconforming when it fails one or more specified requirements. A defect is a failure of a quality characteristic; terminology varies by industry, so the organization must define whether the decision concerns defective units, defects, or both.

The fraction nonconforming in a lot is:

where D is the number of nonconforming units and N is lot size. The value of p is usually unknown when a sample is taken.

For a simple attributes plan with sample size n, acceptance number c, and an effectively large lot, the number of nonconforming units in the sample can be modeled approximately by a binomial random variable X. The acceptance probability is:

Here, p is the lot fraction nonconforming, n is the sample size, c is the acceptance number, and P_a(p) is the probability of accepting a lot with quality level p. The binomial model assumes independent selections with a constant probability of nonconformity and is most appropriate when the sample is small relative to the lot and selection is random. For finite lots sampled without replacement, the hypergeometric distribution is more exact.

For c=0, the expression simplifies to:

For example, with n=50, c=1, and p=0.02:

Thus, a lot containing approximately 2% nonconforming units has about a 73.6% chance of acceptance under this plan. This result illustrates why a single sample cannot provide certainty about lot quality.

5. Operating characteristic curves and risks

The operating characteristic (OC) curve plots the probability of accepting a lot against its actual or assumed fraction nonconforming. A good plan has a high probability of acceptance for acceptable quality and a low probability of acceptance for unacceptable quality. The curve is not a statement that the lot has a particular quality; it describes the performance of the sampling rule.

Figure: An illustrative OC curve; actual curves depend on sample size and acceptance number.

The acceptable quality limit (AQL) is a designated quality level used as a reference in acceptance sampling systems. It is not a guarantee that every accepted lot will contain no defects, nor is it necessarily a target for the process. Its meaning depends on the applicable standard and the complete sampling scheme.

Producer’s risk, commonly denoted α, is the probability of rejecting a lot whose quality is considered acceptable. Consumer’s risk, commonly denoted β, is the probability of accepting a lot whose quality is considered unacceptable. A plan should identify quality points and risks, not merely quote an AQL.

AQL and the limiting quality level, sometimes called rejectable quality level in particular standards, are used to balance these risks. The producer wants reasonable protection against unjustified rejection; the consumer wants protection against receiving poor quality. Increasing sample size generally makes the OC curve steeper and can reduce both risks at selected quality points, but it increases inspection cost and time.

6. Attributes and variables sampling

Attributes sampling records counts or classifications. It is simple and suitable for pass/fail characteristics, but it may require larger samples to obtain strong discrimination.

Variables sampling uses measured values and a statistical model, often assuming an approximately normal distribution. It can provide more information per unit and may require smaller samples, but it requires reliable measurements, a justified distributional assumption, and control of parameters such as mean and standard deviation. A variables plan must not be used merely because measurements are available; its assumptions and standard must be appropriate.

When a characteristic has lower and upper specification limits L and U, a capable, approximately normal process is often summarized by:

and, allowing for process centering:

Here, μ is the process mean and σ is the within-process standard deviation. These indices assume meaningful specification limits, stable process behavior, and an appropriate estimate of variation. They describe process capability; they do not replace acceptance sampling or prove that every unit conforms.

7. Designing and implementing a sampling plan

A defensible plan follows these steps:

1. Define the lot, unit of product, characteristic, and decision purpose.
2. Classify the consequence of nonconformity and identify regulatory requirements.
3. Establish the quality levels requiring protection, including acceptable and unacceptable reference points.
4. Select attributes or variables inspection and verify measurement capability.
5. Select sample size and acceptance number using an approved standard or statistically justified design.
6. Define randomization and sampling timing to prevent selection bias.
7. Specify disposition of rejected lots: 100% screening, return to supplier, rework, concession, or investigation.
8. Define escalation rules for repeated failures and supplier corrective action.
9. Pilot the plan, review operating performance, and revise it when risk or process capability changes.

Sampling must be random or systematically representative. If a lot contains layers, shifts, machines, or pallets with different risk, stratified sampling may be required. A sample taken only from the most accessible cartons can create false confidence.

Rejected lots require a clear policy. Options include screening every unit, sorting and reworking, replacing material, returning it to the supplier, or accepting under a formally authorized concession. The selected disposition should preserve traceability and should not silently convert a failed requirement into an informal tolerance.

8. Worked application: incoming valve inspection

An industrial manufacturer receives 2,000 valves per lot. A leakage defect can cause equipment downtime but is not a direct safety hazard because a downstream pressure test is also performed. The supplier has stable performance, and destructive testing is unnecessary. The organization chooses an attributes sampling plan of n=80 and c=1 for an initial supplier-monitoring period.

If the true lot fraction nonconforming is 1%, the binomial approximation gives:

The lot therefore has approximately an 89.4% probability of acceptance under the plan. If the true fraction is 5%:

The probability of acceptance is then only about 8.6%. The plan discriminates substantially between the two quality levels, but it does not guarantee acceptance of every 1% lot or rejection of every 5% lot.

The inspection plan requires random selection across the delivery, calibrated leakage-test equipment, identification of the inspector and equipment, and immediate segregation when two or more nonconforming valves are found. A rejected lot is 100% tested, the supplier is notified, and three consecutive conforming deliveries are required before reduced monitoring is considered. The downstream pressure test remains in place because incoming acceptance is not the only protection against leakage.

This is an illustrative case, not a universal sampling recommendation. A safety-critical valve, unstable supplier, or legally controlled product would require a different risk analysis and possibly full inspection or independent certification.

9. Challenges, ethics, and governance

Sampling plans can be misused when managers select a convenient plan without understanding its risks, treat AQL as an allowed defect quota, or change acceptance criteria after seeing results. Such practices undermine informed customer protection and may violate contractual or regulatory obligations.

Quality records must be truthful, traceable, and protected from alteration. Inspectors need authority to stop release when evidence is inadequate. Conflicts of interest should be managed; production personnel should not be pressured to record conforming results to meet shipment targets. Personal data in inspection systems should be limited to legitimate operational needs.

Automation and machine vision can improve consistency and speed, but false positives and false negatives must be validated. Artificial intelligence may classify images or predict defect likelihood, yet the organization remains responsible for data quality, model drift, explainability appropriate to the risk, cybersecurity, and human escalation. AI should support—not conceal—the need for capable processes and accountable decisions.

10. Best-practice principles

• Inspect critical characteristics at the earliest reliable control point.
• Use prevention and process capability wherever they provide stronger protection than end-of-line inspection.
• Base sampling on risk, not on habit or the desire to minimize inspection cost.
• Use approved statistical standards and document the selected quality points and risks.
• Ensure random, representative selection and maintain lot traceability.
• Validate measurement systems before interpreting inspection results.
• Separate acceptance decisions from unauthorized concessions.
• Review plans using supplier performance, process stability, customer complaints, and changes in design or regulation.
• Treat a rejected lot as a signal for investigation, not merely as an isolated administrative event.

Chapter Summary

Inspection provides evidence of conformity at incoming, in-process, and final stages, but it cannot replace prevention and process control. Inspection plans define characteristics, requirements, methods, frequency, responsibilities, records, and reaction actions. Full inspection is justified by high consequence, small lots, reliable low-cost testing, or explicit requirements; sampling is useful when risk is acceptable and testing every unit is costly, destructive, or unnecessary.

Acceptance sampling uses a sample to decide about a lot and therefore involves uncertainty. Attributes plans use sample size and acceptance number, while variables plans use measured values and statistical assumptions. The OC curve shows the probability of acceptance at different lot quality levels. AQL is a reference quality level, not a promise of zero defects. Producer’s risk concerns rejection of acceptable quality, and consumer’s risk concerns acceptance of unacceptable quality. Ethical and effective systems require representative sampling, validated measurement, traceability, clear disposition, and continual review.

Key Terms

Lot: A defined quantity presented for acceptance.

Sample: A subset selected from a lot for examination.

Inspection: Examination or measurement against specified requirements.

Incoming inspection: Evaluation of externally supplied materials or components.

In-process inspection: Verification during production or service delivery.

Final inspection: Conformity verification before release.

Nonconforming unit: A unit failing one or more specified requirements.

Defect: A failure of a quality characteristic.

Acceptance sampling: Statistical decision-making based on a sample.

Sampling plan: Sample size and acceptance or rejection rules.

Acceptance number: Maximum sample nonconformities permitted for acceptance.

AQL: A designated reference quality level in an acceptance sampling system.

OC curve: Relationship between lot quality and probability of acceptance.

Producer’s risk: Probability of rejecting acceptable quality.

Consumer’s risk: Probability of accepting unacceptable quality.

Attributes sampling: Sampling based on counts or classifications.

Variables sampling: Sampling based on measured numerical values.

Measurement uncertainty: Doubt associated with a measured result.

Stratified sampling: Sampling from identified subgroups or layers.

Concession: Authorized acceptance of a specified nonconformity.

Review Questions

Short Answer Questions

1. Distinguish incoming, in-process, and final inspection.
2. What is the purpose of an acceptance number?
3. Define AQL and explain why it is not a zero-defect guarantee.
4. What does an OC curve represent?
5. State two conditions under which 100% inspection may be appropriate.

Long Answer Questions

1. Explain how risk analysis should influence an inspection plan.
2. Compare full inspection and sampling inspection, including their advantages and limitations.
3. Explain producer’s risk and consumer’s risk and their importance in plan selection.
4. Compare attributes and variables sampling and identify the assumptions of variables sampling.
5. Describe how a rejected lot should be controlled, investigated, and dispositioned.

Analytical / Application Questions

1. A plan samples 40 units and accepts a lot if no more than one is nonconforming. Using the binomial model, calculate the approximate acceptance probability when the lot fraction nonconforming is 2%.
2. A supplier’s process becomes unstable after a design change. Should the organization continue reduced incoming sampling? Justify your decision and propose controls.
3. Design an inspection strategy for a safety-critical component, explaining where prevention, in-process checks, final inspection, and acceptance sampling should be used.

Multiple Choice Questions with Answers

Question 1

Which inspection stage evaluates materials before they enter production?

A. Incoming inspection

B. Final inspection

C. Post-release inspection

D. Service inspection

Correct Answer: A. Incoming inspection

Question 2

For an attributes plan, what does the acceptance number specify?

A. The lot size

B. The maximum number of nonconforming units allowed in the sample

C. The supplier’s defect rate

D. The measurement uncertainty

Correct Answer: B. The maximum number of nonconforming units allowed in the sample

Question 3

What does the OC curve plot?

A. Inspection cost against labor hours

B. Process mean against standard deviation

C. Probability of acceptance against lot quality

D. Defects against production time only

Correct Answer: C. Probability of acceptance against lot quality

Question 4

Consumer’s risk is the probability of:

A. Rejecting acceptable quality

B. Accepting unacceptable quality

C. Finding a measurement error

D. Increasing sample size

Correct Answer: B. Accepting unacceptable quality

Question 5

Using the binomial model, for n=50, c=0, and p=0.02, the acceptance probability is approximately:

A. 0.02

B. 0.36

C. 0.64

D. 0.98

Correct Answer: C. 0.64

Question 6

Which condition most strongly supports 100% inspection?

A. The consequence of failure is extremely serious

B. The supplier has excellent recent performance

C. The lot is very large and testing is expensive

D. The characteristic is irrelevant to function

Correct Answer: A. The consequence of failure is extremely serious

Question 7

Variables sampling primarily uses:

A. Only visual judgments

B. Numerical measurements and statistical assumptions

C. Supplier reputation without data

D. The customer’s invoice value

Correct Answer: B. Numerical measurements and statistical assumptions

Question 8

Why must sampling be representative?

A. To ensure the sample reflects relevant variation in the lot

B. To make every lot automatically acceptable

C. To eliminate all producer’s risk

D. To avoid recording results

Correct Answer: A. To ensure the sample reflects relevant variation in the lot

Question 9

Which statement about AQL is correct?

A. It guarantees that every accepted lot has zero defects

B. It is a reference quality level used in a sampling system

C. It is always the legal maximum defect rate

D. It eliminates consumer’s risk

Correct Answer: B. It is a reference quality level used in a sampling system

Question 10

If two nonconforming units are found under a plan with c=1, the immediate sampling decision is normally to:

A. Accept the lot

B. Ignore the second unit

C. Reject or hold the lot according to the plan

D. Reduce the sample size

Correct Answer: C. Reject or hold the lot according to the plan