Statistical Quality Control and Process Capability

Introduction

Statistical quality control (SQC) applies statistical reasoning to monitor, control, and improve processes. It is based on a central proposition: observed variation is unavoidable, but its sources can be studied and managed. Quality professionals therefore do not merely inspect finished products; they evaluate process behaviour, measurement reliability, and the probability that a process will produce conforming output.

This chapter develops an integrated approach to sampling, measurement-system analysis, control charts, process capability, and acceptance sampling. It also distinguishes common-cause variation, which is inherent in the current process system, from special-cause variation, which arises from identifiable and abnormal events.

Statistical methods do not replace engineering knowledge. A control chart can signal that a process has changed, but process owners must investigate why. Similarly, a high capability index is meaningful only when the process is stable and the measurement system is adequate.

Learning Objectives

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

1. Explain sampling concepts and distinguish descriptive from inferential uses of data.
2. Distinguish common-cause from special-cause variation.
3. Assess the adequacy of a measurement system using accuracy, precision, bias, repeatability, reproducibility, and stability.
4. Select and interpret appropriate control charts.
5. Calculate and interpret process capability indices, including C_p and C_pk.
6. Explain the purpose and limitations of acceptance sampling.
7. Apply statistical quality tools to operational and managerial decisions.

1. Statistical Thinking in Quality Management

A population is the complete set of units, transactions, or opportunities about which a conclusion is required. A sample is a subset selected from that population. A parameter describes a population, whereas a statistic describes a sample.

For a measured characteristic X, common sample statistics are:

Here, n is sample size, xᵢ is observation i, x̄ is the sample mean, and s is the sample standard deviation. The divisor n−1 provides an unbiased estimator of population variance under the usual independent random-sampling assumptions.

Sampling is useful because complete inspection may be destructive, expensive, slow, or impossible. However, a sample must represent the process or population. Convenience sampling can produce seriously misleading conclusions because the observed units may systematically differ from unobserved units.

1.1 Sampling principles

Important sampling approaches include:

• Simple random sampling: every unit has a known and generally equal probability of selection.
• Systematic sampling: units are selected at a fixed interval after a random starting point.
• Stratified sampling: the population is divided into relevant groups, or strata, and samples are taken from each.
• Cluster sampling: naturally occurring groups are selected, and units within selected groups are studied.
• Rational subgrouping: observations are grouped so that variation within a subgroup reflects short-term process variation, while differences between subgroups can reveal process changes.

In statistical process control, rational subgrouping is especially important. For example, five consecutive parts produced under nearly identical conditions may form a subgroup. Combining parts from different machines, shifts, or raw-material lots may conceal meaningful changes.

2. Variation and Process Behaviour

Variation may be classified as follows:

A stable process is not necessarily a capable process. Stability concerns predictability over time; capability concerns the relationship between the stable process distribution and customer or engineering specifications.

Examples of common causes include normal machine-to-machine differences, ordinary temperature fluctuation, and routine operator variation. Examples of special causes include a broken tool, incorrect program setting, contaminated material, faulty sensor, or unusual power interruption.

Over-adjustment is a frequent management error. If every small fluctuation is treated as a special cause, operators may continually alter a process that is functioning normally. This “tampering” can increase variation rather than reduce it. Conversely, ignoring a genuine special cause permits defective output to continue.

3. Measurement-System Analysis

Statistical conclusions are only as trustworthy as the measurement system producing the data. A measurement system includes the instrument, software, standard, method, operator, environment, and decision rules.

3.1 Measurement properties

• Accuracy: closeness of a measurement to the accepted reference value.
• Precision: closeness of repeated measurements to one another.
• Bias: systematic difference between the average observed value and the reference value.
• Repeatability: variation when the same operator measures the same item repeatedly with the same instrument.
• Reproducibility: variation between operators, appraisers, laboratories, or measurement conditions.
• Stability: consistency of measurement performance over time.
• Resolution: smallest distinguishable measurement increment.

An instrument can be precise but biased: repeated readings may cluster tightly around an incorrect value. It can also be unbiased on average but imprecise, producing widely scattered readings.

3.2 Gauge repeatability and reproducibility

A gauge R\&R study estimates the proportion of observed measurement variation attributable to repeatability and reproducibility. A simplified variance model is:

where σ²_process is product or process-part variation and σ²_gauge is measurement-system variation. The model assumes the components are appropriately estimated and approximately independent.

A commonly used study statistic is:

where σ_gauge is the estimated standard deviation of measurement error and σ_total is total observed standard deviation. Interpretation depends on industry and risk, but a measurement system contributing less than approximately 10% of total variation is often considered acceptable; 10–30% may be conditionally acceptable; above 30% generally requires improvement. These are practical guidelines, not universal laws.

A measurement-system study should use parts spanning the relevant operating range, trained appraisers, repeated trials, and a balanced design. If all parts are nearly identical, the study may underestimate the measurement system’s contribution because true part variation is too small to separate reliably from measurement error.

3.3 Measurement-system improvement

Improvement actions may include calibration, better fixturing, improved resolution, standardized work instructions, operator training, environmental control, automated sensing, or redesign of the measurement method. Calibration alone does not eliminate repeatability or reproducibility problems.

4. Control Charts

A control chart displays a statistic in time order with a centre line and statistically calculated control limits. It is a process-monitoring tool, not a specification chart. Control limits describe expected process behaviour; specification limits describe customer, design, or regulatory requirements.

For a statistic W_t measured at time t, a general three-sigma chart has limits:

Here, μ_W and σ_W are the mean and standard deviation of the chart statistic when the process is stable. For bounded statistics, a limit may be below zero or above a natural maximum; such a limit is interpreted according to the chart’s statistical model and may be truncated where appropriate.

The three-sigma convention is widely used because, under a normal model, approximately 99.73% of observations lie within ±3 standard deviations. This probability is not exact for every distribution, and non-normal or autocorrelated data require care.

4.1 Variables charts

Variables charts monitor measured characteristics such as length, mass, temperature, or pressure.

For individual observations with known process mean and standard deviation, an individuals chart can use:

For subgroup means with subgroup size n, assuming independent normal observations and known σ, the standard error is σ/√n, so:

In practice, μ and σ are usually estimated from initial stable data. The common x̄−R chart uses subgroup averages and ranges. If R̄ is the average subgroup range and d₂ is a control-chart constant dependent on subgroup size, an estimate is:

The constants must match the subgroup size and chart design.

4.2 Attribute charts

Attribute charts are used when data are counts or classifications.

For the fraction nonconforming p in a sample of size n, the p-chart limits are:

where p̄ is the average observed fraction nonconforming. Limits are constrained to the interval 0 to 1. If sample sizes vary, limits vary with n.

For the number of nonconformities per constant inspection unit, a c-chart may be used:

This assumes a Poisson-type count with a constant area or opportunity for occurrence. A u-chart is preferable when the number of opportunities or inspection-unit sizes vary.

4.3 Interpreting control-chart signals

A point outside a control limit is evidence of a likely special cause, but it is not proof without investigation. Other useful signals include:

• a sustained run on one side of the centre line;
• a persistent upward or downward trend;
• cyclic or seasonal patterns;
• unusual clustering near a control limit;
• alternating high and low points;
• stratification, where points cluster close to the centre line.

Signal rules should be defined before data are reviewed, because changing rules after observing results increases the risk of false conclusions. Control limits should not be recalculated merely to hide an out-of-control signal. Limits may be revised after a documented process change or after a justified removal of special-cause data.

5. Process Capability

Capability analysis compares the natural spread and location of a stable process with specification limits. Let USL and LSL denote the upper and lower specification limits, μ the process mean, and σ the within-process standard deviation.

The potential capability index is:

This index compares specification width with six standard deviations. It assumes a stable process and an appropriate estimate of within-process variation. C_p does not account for process centring.

The one-sided indices are:

The centred capability index is:

For a two-sided specification, C_pk reflects the nearer specification limit. If C_p > C_pk, the process is not perfectly centred. For a one-sided specification, the relevant one-sided index should be used rather than forcing a two-sided calculation.

Capability indices are ratios and have no units. A commonly used organizational benchmark is 1.33 or higher, but required values depend on risk, cost, regulation, and customer expectations. A value above 1 does not guarantee zero defects, and capability calculations based on an unstable process are not reliable.

Worked Example: Capability

A shaft diameter has specifications 49.80 mm to 50.20 mm. A stable process has mean 50.08 mm and within-process standard deviation 0.04 mm.

Therefore:

The process spread is potentially good, as shown by C_p = 1.67, but the process is shifted toward the upper specification. Centring the process downward would improve C_pk without necessarily reducing variation.

Capability versus performance

Capability studies generally use short-term or within-subgroup variation. Performance indices such as P_p and P_pk commonly use overall long-term standard deviation. The exact estimation method must be stated because the indices answer different questions:

• capability: what the process can do under relatively homogeneous conditions;
• performance: what the process has actually done over a broader period.

6. Acceptance Sampling

Acceptance sampling is a decision procedure for accepting or rejecting a lot based on a sample. It is not a method for controlling or improving the production process. It is appropriate when 100% inspection is impractical, destructive, costly, or less reliable than sampling.

A sampling plan is commonly described by lot size, sample size n, and acceptance number c. In a single-sampling plan, a sample is inspected and the lot is accepted if the number of nonconforming units is no greater than c.

For a binomial model with independent sampling and a constant probability p of a nonconforming unit, the probability of observing exactly x nonconforming units is:

The probability of acceptance is therefore:

This formula assumes independent trials and sampling with replacement, or a sufficiently large lot relative to the sample. For finite lots sampled without replacement, the hypergeometric distribution may be more appropriate.

A sampling plan should consider:

• Acceptable quality level (AQL): a quality level regarded as satisfactory for the supplier process, used as a plan-design reference.
• Lot tolerance percent defective (LTPD): a poor quality level that the customer wants the plan to reject with high probability.
• Producer’s risk (α): probability of rejecting a lot at a quality level regarded as acceptable.
• Consumer’s risk (β): probability of accepting a lot at an unacceptable quality level.
• Operating characteristic (OC) curve: relationship between incoming quality and probability of lot acceptance.

No sampling plan can guarantee that every accepted lot is defect-free. Acceptance sampling balances inspection effort against risks of incorrect decisions. It should be linked to supplier development, process capability, traceability, and corrective action rather than used as a substitute for process control.

7. Integrated Implementation Framework

A disciplined SQC deployment can follow these steps:

1. Define the quality characteristic, customer requirement, unit of product, and decision purpose.
2. Confirm that the sampling method represents the process and that rational subgroups are meaningful.
3. Validate the measurement system before interpreting process variation.
4. Collect time-ordered baseline data under normal operating conditions.
5. Select a chart based on data type, subgrouping, sample size, and distributional assumptions.
6. Investigate special-cause signals using process knowledge, traceability, and structured root-cause analysis.
7. Establish revised control limits only when the process change is understood and documented.
8. Assess capability only after demonstrating statistical stability.
9. Compare capability with specifications, risk, and customer requirements.
10. Use acceptance sampling, where necessary, as a risk-management arrangement—not as the primary improvement method.

A useful escalation logic is: measurement adequacy → stability → capability → improvement. Reversing this order can lead to false confidence.

Applied Case: Illustrative Medical-Device Packaging Process

Background

A medical-device manufacturer seals sterile pouches. Seal width must be between 8.0 mm and 10.0 mm. The organization receives customer complaints concerning occasional weak seals and wishes to reduce final inspection.

Problem

The quality team initially calculates capability from inspection records and obtains C_pk = 1.45. Management proposes reducing inspection immediately. However, the data combine three sealing machines, four operators, and several material lots. The gauge study has not been completed.

Analysis

The team first performs a measurement-system study using representative pouches, trained operators, and repeated measurements. The measurement system contributes 34% of observed variation, mainly because the fixture allows inconsistent positioning. The result is not adequate for capability decisions.

After improving the fixture and standardizing the measurement method, the revised study estimates measurement variation at 12% of total variation, conditionally acceptable for preliminary control with further improvement planned. The team then separates data by machine and uses rational subgroups collected in time order.

An individuals–moving range chart for Machine 2 shows a step increase following a sealing-jaw replacement. Investigation finds that the replacement jaw has a different pressure setting. After correction, the process becomes stable. The stable data show mean seal width 9.42 mm and within-process standard deviation 0.18 mm.

Approach and outcome

The process has adequate potential spread but is too close to the upper side relative to the desired internal benchmark of 1.33. Engineers reduce the average seal width by adjusting temperature and pressure within validated operating limits. A follow-up study produces a mean of 9.05 mm with the same estimated standard deviation:

The organization retains the control chart, continues periodic measurement-system checks, and introduces reduced but not eliminated acceptance inspection based on documented risk analysis.

Learning points

1. An impressive capability result is invalid if the measurement system is inadequate.
2. Mixed machines and operators can conceal special causes.
3. Stability must precede capability analysis.
4. Centring can improve C_pk without changing process spread.
5. Reduced inspection should follow evidence, validation, and risk assessment.

8. Limitations, Governance, and Contemporary Practice

Statistical tools depend on assumptions. Non-normal distributions, autocorrelation, changing sample sizes, seasonality, rare events, and measurement bias can invalidate standard interpretations. Analysts should examine time plots, distributional shape, sampling conditions, and process context rather than applying formulas mechanically.

Data integrity is a governance issue. Records used for release, regulatory compliance, or supplier evaluation should be attributable, legible, contemporaneous, original or suitably controlled, accurate, complete, consistent, and available. Deleting inconvenient points without documented technical justification is unethical and can create safety risks.

Digital manufacturing systems increasingly automate data collection and charting. Artificial intelligence can assist anomaly detection, image inspection, root-cause prioritization, and predictive maintenance. However, AI models may learn from biased labels, confuse correlation with causation, or generate false alarms when process conditions change. Human review, model validation, version control, cybersecurity, and explainable decision rules remain essential.

Best practice is to combine statistical evidence with engineering knowledge, operator experience, customer risk, and regulatory requirements. A mathematically sophisticated chart cannot compensate for poor sampling, unstable measurement, or weak process ownership.

Chapter Summary

Statistical quality control uses data to understand and manage process variation. Sampling provides an efficient basis for inference, but representativeness and rational subgrouping are essential. Common-cause variation belongs to the current process system and is addressed through systemic improvement; special-cause variation is abnormal and requires investigation and removal.

Measurement-system analysis establishes whether observed differences are real or caused by the gauge, method, operator, or environment. Control charts distinguish stable from unstable process behaviour and must be selected according to data type and assumptions. Process capability indices compare stable process variation and centring with specifications; C_p measures potential spread, whereas C_pk also reflects location. Acceptance sampling makes lot decisions under explicit producer and consumer risks, but it does not improve the process itself.

A reliable sequence is measurement adequacy, process stability, capability assessment, and continuous improvement. Ethical data governance and appropriate human judgment are necessary whenever statistical results influence product release, safety, or customer protection.

Key Terms

Population: Complete set of units or outcomes of interest.

Sample: Subset selected from a population.

Statistic: Numerical summary calculated from a sample.

Parameter: Numerical characteristic of a population.

Rational subgroup: Group of observations formed to represent short-term process conditions.

Common-cause variation: Routine variation generated by the existing process system.

Special-cause variation: Unusual variation attributable to an identifiable event or condition.

Measurement-system analysis: Study of measurement accuracy, precision, and variation.

Repeatability: Variation from repeated measurements by the same operator and instrument.

Reproducibility: Variation between operators or measurement conditions.

Control chart: Time-ordered chart with a centre line and statistically derived control limits.

Control limit: Boundary describing expected process behaviour under a statistical model.

Specification limit: Customer, design, or regulatory requirement boundary.

Process capability: Degree to which a stable process can meet specifications.

C_p: Potential capability index based on specification width and process spread.

C_pk: Capability index accounting for both spread and process centring.

Acceptance sampling: Procedure for accepting or rejecting a lot using sample evidence.

AQL: Acceptable quality level used as a sampling-plan design reference.

Consumer’s risk: Probability of accepting an unacceptable lot.

Operating characteristic curve: Curve showing probability of acceptance at different quality levels.

Review Questions

Short Answer Questions

1. What is the difference between a population and a sample?
2. Define common-cause and special-cause variation.
3. Why must a measurement system be evaluated before process capability is calculated?
4. What is the difference between a control limit and a specification limit?
5. What does C_pk communicate that C_p does not?

Long Answer Questions

1. Explain how rational subgrouping influences the interpretation of a control chart.
2. Describe the principal components of a gauge repeatability and reproducibility study.
3. Compare variables control charts with attribute control charts, including suitable applications.
4. Explain the assumptions and managerial implications of process capability analysis.
5. Discuss the role of producer’s risk, consumer’s risk, AQL, LTPD, and the OC curve in acceptance sampling.

Analytical / Application Questions

1. A stable process has specifications of 100 ± 2 units, mean 100.8 units, and standard deviation 0.5 units. Calculate C_p and C_pk, then recommend whether centring or variation reduction should be prioritized.
2. A control chart shows seven consecutive points above the centre line, although none exceeds the control limits. Explain why this may be a signal and propose an investigation plan.
3. A supplier’s lot is accepted under a sampling plan, but subsequent use reveals defects. Analyse at least three possible explanations and explain why acceptance does not prove that the lot is defect-free.

Multiple Choice Questions with Answers

Question 1

Which condition should normally be established before calculating process capability indices?

A. The process has been shown to be statistically stable

B. The lot has passed 100% inspection

C. The control limits equal the specification limits

D. The sample contains only conforming units

Correct Answer: A. The process has been shown to be statistically stable

Question 2

Which type of variation is normally addressed by changing the overall process system?

A. Special-cause variation

B. Common-cause variation

C. Measurement transcription error only

D. Sampling-frame variation only

Correct Answer: B. Common-cause variation

Question 3

A measurement system gives readings that are tightly grouped but consistently above the reference value. This indicates primarily:

A. Low precision and no bias

B. High accuracy and low stability

C. Good precision with systematic bias

D. Special-cause process variation

Correct Answer: C. Good precision with systematic bias

Question 4

For a two-sided specification, which index accounts for both process spread and centring?

A. C_p

B. C_{pu} only

C. C_{pl} only

D. C_{pk}

Correct Answer: D. C_{pk}

Question 5

A stable process has USL = 50, LSL = 40, and σ = 1. What is C_p?

A. 0.60

B. 1.00

C. 1.67

D. 3.33

Correct Answer: C. 1.67

Question 6

If C_p is substantially greater than C_{pk}, the most likely interpretation is that the process is:

A. Perfectly centred

B. Shifted toward one specification limit

C. Necessarily unstable

D. Measured with zero gauge error

Correct Answer: B. Shifted toward one specification limit

Question 7

Which chart is generally appropriate for the fraction of nonconforming units when sample sizes may vary?

A. p-chart

B. c-chart

C. X-bar chart only

D. R-chart only

Correct Answer: A. p-chart

Question 8

What is the primary purpose of acceptance sampling?

A. To identify every defective unit

B. To improve machine capability directly

C. To make a lot acceptance decision using sample evidence

D. To replace all process-control activities

Correct Answer: C. To make a lot acceptance decision using sample evidence

Question 9

Consumer’s risk is the probability of:

A. Rejecting a lot that is acceptable

B. Accepting a lot that is unacceptable

C. Measuring a conforming unit incorrectly

D. Detecting a special cause immediately

Correct Answer: B. Accepting a lot that is unacceptable

Question 10

A process has USL = 10.0, LSL = 8.0, mean = 9.4, and σ = 0.2. What is C_{pk}?

A. 0.67

B. 1.00

C. 1.67

D. 2.00

Correct Answer: C. 1.67