Statistical Process Control and Process Capability

Introduction and context

Statistical Process Control (SPC) is a systematic method for using data to understand, monitor, and improve process performance. It distinguishes ordinary, expected variation from unusual variation that signals a change in the process. Process capability analysis complements SPC by asking a different question: once a process is stable, can it consistently meet customer or engineering specifications?

These two ideas must not be confused. A process may be statistically stable but incapable of meeting specifications, or apparently capable over a short period while being unstable because of special causes. Effective quality management therefore follows a logical sequence: define the characteristic and measurement method, establish statistical control, assess capability, and then select corrective or improvement actions.

SPC is especially important in production management because it supports prevention rather than inspection. Instead of discovering defects after production, managers use timely process information to identify changing conditions such as tool wear, incorrect machine settings, unsuitable materials, operator error, or environmental variation.

Learning objectives

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

1. Explain common-cause and special-cause variation.
2. Select and construct appropriate variable and attribute control charts.
3. Calculate control limits and interpret chart signals.
4. Assess process stability and recognize limitations of control charts.
5. Calculate and interpret C_p, C_pk, P_p, and P_pk.
6. Distinguish process capability from conformance to specifications.
7. Recommend corrective actions based on statistical evidence.

Variation, stability, and specifications

Every process produces variation. Variable data are measured on a continuous scale, such as diameter, mass, temperature, cycle time, or strength. Attribute data are counted or classified, such as the number of defective units, scratches, missing components, or defects per unit.

Common and special causes

Common-cause variation is the predictable, inherent variation produced by the process system. Examples include small differences in raw material, normal machine vibration, routine measurement noise, and ordinary temperature fluctuations. Reducing common causes normally requires changing the process design, equipment, materials, methods, or management system.

Special-cause variation is caused by an identifiable and unusual event. Examples include a broken tool, a wrongly adjusted machine, a new operator using an incorrect method, a power interruption, or a defective batch of material. Special causes require investigation and correction of the specific event.

A control chart is designed to detect special causes, not to force every observation to the target value. Tampering—making frequent adjustments in response to random common-cause variation—often increases variation. Operators should adjust a process only when there is reliable evidence of a real change or when an intentional, planned improvement is being introduced.

Control limits and specification limits

Control limits are statistically calculated boundaries describing the expected behaviour of a stable process. Specification limits are externally defined requirements established by customers, designers, regulators, or engineering authorities. The upper and lower control limits are denoted UCL and LCL; specification limits are USL and LSL.

Control limits answer: “Is the process behaving unusually?” Specification limits answer: “Is the output acceptable?” A point outside a control limit may be nonconforming, but it may also fall within specifications. Conversely, a stable process may produce many observations beyond a specification limit if its natural variation is too wide.

Designing an SPC study

A sound SPC study begins with a clearly defined quality characteristic and a rational sampling plan. The subgroup should contain observations produced under approximately the same conditions and close enough in time to reveal short-term variation. Subgrouping should reflect the process question. For example, five consecutive parts may estimate within-run variation, while one part from each of five machines may compare machines; these are not interchangeable designs.

Before calculating limits, verify that:

• the measurement system is suitable, calibrated, and sufficiently precise;
• the characteristic and sampling frequency are clearly defined;
• data are recorded in time order;
• the process conditions and subgroup structure are understood;
• specification limits are not substituted for control limits.

A measurement-system study, such as Gage Repeatability and Reproducibility, is often necessary. A process cannot be managed intelligently when measurement error is large relative to the process variation or tolerance.

Variable control charts

Variable charts are used when measurements are continuous. The most common choices are the \bar{x}–R chart for small subgroups and the \bar{x}–S chart for larger subgroups. An Individuals–Moving Range chart is used when rational subgroups of several observations cannot be collected.

X-bar and R chart

Suppose subgroup i contains n observations x_i1, x_i2, ..., x_in. The subgroup mean and range are:

Across k subgroups:

For subgroup sizes commonly between 2 and 10, the control limits are:

The constants A₂, D₃, and D₄ depend on subgroup size and are derived from the expected distribution of ranges under approximate normality. For n=5, commonly used values are A₂=0.577, D₃=0, and D₄=2.114. The R chart should be examined first: an unstable estimate of within-subgroup variation makes the X-bar limits unreliable.

Worked example: X-bar and R chart

Five subgroups of four shaft diameters, in millimetres, produce the following data:

Thus, \bar{\bar{x}=10.031 and \bar{R}=0.022. For illustration, use A₂=0.729, D₃=0, and D₄=2.282 for subgroup size four.

The X-bar limits are:

The R limits are:

The second subgroup mean, 10.060, is above the X-bar UCL and should trigger investigation. The ranges are within their limits, suggesting a shift in the subgroup centre rather than an increase in short-term dispersion. The correct response is not to delete the point automatically; the team should investigate settings, material, operator actions, and measurement records. If a documented special cause is found and removed, revised limits may be calculated using an approved procedure and sufficient remaining data.

X-bar and S chart

For larger subgroups, the sample standard deviation is generally a more efficient measure of dispersion than the range. The standard deviation for subgroup i is:

The average standard deviation is \bar{s}=(s₁+...+s_k)/k. Limits are commonly written:

The constants A₃, B₃, and B₄ depend on subgroup size. These formulas assume independent observations within appropriately formed subgroups and an approximately stable distribution.

Individuals and moving-range chart

When only one observation is available at each time point, use an Individuals–Moving Range chart. For consecutive observations, the moving range is:

With average moving range \overline{MR}, an estimate of process standard deviation is often:

For a moving range of two observations, d₂=1.128. Approximate individual-chart limits are:

This chart is sensitive to autocorrelation and ordering. If observations are strongly correlated, the usual limits may be misleading; the sampling model must be reconsidered.

Attribute control charts

Attribute charts are appropriate when observations are counts or classifications. The chart must match both the data type and the sampling opportunity.

A defective unit fails one or more requirements; a defect is an individual nonconformity. One unit can contain several defects, so defective-unit charts must not be confused with defect-count charts.

p chart

If subgroup i contains n_i units and d_i defectives:

The average fraction defective is \bar{p}=\sum d_i/\sum n_i. Under a binomial model, the three-sigma limits are:

Negative lower limits are set to zero. If sample sizes vary, limits vary by subgroup.

np, c, and u charts

For constant sample size n, the np chart uses:

The lower limit is n\bar{p}-3\sqrt{n\bar{p}(1-\bar{p})}, truncated at zero if necessary.

For a constant inspection unit or opportunity, if c_i defects are counted:

The c chart assumes a Poisson-type count. When the number of opportunities varies, use the u chart. If u_i=c_i/n_i and \bar{u}=\sum c_i/\sum n_i:

Again, negative lower limits are set to zero.

Interpreting control charts

A point beyond a control limit is strong evidence of a special cause, although the cause still requires investigation. Other signals may include a sustained run on one side of the centre line, a long monotonic trend, or an unusual cycle. Exact run rules should be defined by the organisation; commonly used supplementary rules include eight successive points on one side of the centre line or six successive points increasing or decreasing. Rules should be applied consistently, because adding many rules increases false alarms.

The interpretation sequence is:

1. Check the measurement and data-recording process.
2. Examine the chart in time order.
3. Confirm whether the signal is credible and operationally meaningful.
4. Search for assignable causes using process knowledge, maintenance records, material lots, staffing, and environmental data.
5. Contain potentially nonconforming product.
6. Correct the cause and verify effectiveness.
7. Recalculate limits only when the process has genuinely changed and the change is documented.

A stable chart does not prove that the process is good; it only indicates that the observed variation is consistent with the assumed statistical model. Conversely, an unstable chart does not identify the cause automatically.

Process capability analysis

Capability indices should be calculated only after the process is stable and the measurement system is adequate. They also require a meaningful specification interval and a defined distributional assumption. The traditional indices assume approximately normal, independent data and a process standard deviation representing within-process variation.

Potential capability: Cp

For a two-sided specification:

Here, USL and LSL are the upper and lower specification limits, and \sigma_within is the within-process standard deviation. 6\sigma represents the approximate natural spread from \mu-3\sigma to \mu+3\sigma under normality.

C_p measures potential capability if the process is perfectly centred. It does not account for the process mean.

Actual capability: Cpk

where \mu is the process mean. The two terms are the upper and lower capability indices. If the process is off-centre, C_pk<C_p. A one-sided specification requires only the relevant term.

Many organisations use a minimum criterion such as 1.33, but the required value depends on risk, industry, customer requirements, and the cost of failure. A capability index is not a universal guarantee of quality.

Long-term performance: Pp and Ppk

Performance indices use overall, long-term standard deviation \sigma_overall:

Because overall variation includes changes across time, shifts, and batches, P_pk is often lower than C_pk. The distinction must be reported clearly; substituting one for the other can lead to false confidence.

Worked capability example

A filling process has LSL=98 g and USL=102 g. A stable study estimates \mu=100.8 g and \sigma_within=0.5 g. Then:

The upper and lower components of C_pk are:

Therefore:

The process has adequate potential spread if centred, but it is poorly centred toward the USL. The first action is to recentre the process, not merely to reduce variation. If the mean were brought to 100 g while variation remained unchanged, C_pk would equal C_p=1.333. Further reduction in variation would then be needed if a higher capability target applied.

Capability calculations should also be supported by distributional checks. If the data are skewed, bounded, multimodal, autocorrelated, or non-normal, normal indices may be inappropriate. Alternatives include a transformation, a suitable non-normal distribution, percentile-based capability, or direct estimation of the proportion outside specifications. The chosen method must be documented.

Selecting corrective action

SPC results should lead to disciplined action rather than automatic adjustment.

Corrective action should use evidence such as stratification, Pareto analysis, cause-and-effect diagrams, designed experiments, maintenance data, and process mapping. Control charts monitor the result; they do not replace technical problem solving.

Applied case: illustrative injection-moulding process

Background

A plastics manufacturer produces housings with a critical wall thickness specified as 2.00 ± 0.05 mm. Four consecutive parts are measured every hour. The production team uses an X-bar and R chart to monitor the process.

Problem

For several hours, the R chart remains stable, but the X-bar chart shows a gradual upward trend. The last two subgroup means are close to the upper specification limit. A simple inspection report shows few defects, so production initially proposes continuing without intervention.

Analysis

The stable R chart indicates that short-term dispersion has not changed substantially. The upward movement in subgroup means indicates a change in process location. Investigation of the time sequence shows that the trend began after a mould-cooling circuit was partially blocked. Reduced cooling increased the average wall thickness as the cycle progressed. The absence of immediate defects was misleading because the process was approaching, rather than yet exceeding, the specification boundary.

Approach and solution

The team temporarily segregates parts produced after the likely change point. Maintenance restores the cooling flow, verifies water temperature and pressure, and confirms the measurement system using reference parts. New subgroups are collected. The X-bar returns near its established centre while the R chart remains stable. The team documents the special cause, adds cooling-flow checks to preventive maintenance, and establishes a reaction plan for future trends.

Outcome and learning points

The process is returned to its previous operating condition without unnecessary adjustment of mould settings. The case demonstrates that control charts provide earlier warning than final inspection, that a stable range chart does not imply a stable mean, and that corrective action should target the physical mechanism responsible for the signal. It also illustrates why production data must be retained in time order and connected with maintenance records.

Technology, governance, and limitations

Modern SPC software can automatically calculate limits, display dashboards, stratify data, and issue alerts. Sensors and manufacturing-execution systems can support higher-frequency monitoring, while machine-learning methods may detect complex patterns. However, automation does not remove the need to define rational subgroups, validate measurements, protect data integrity, and understand false alarms. An algorithm trained on unstable or biased historical data may reproduce poor decisions.

Governance requires traceable calculations, controlled revisions to limits, secure records, and clear responsibility for reaction plans. Operators should not be penalised for reporting special causes; otherwise data may be hidden or manipulated. Capability results should not be used to justify releasing known nonconforming product, changing specifications without engineering authority, or ranking employees without considering the process system.

Important limitations include autocorrelation, non-normal data, mixed populations, changing sample sizes, short studies, inadequate measurement systems, and inappropriate subgrouping. Control limits estimated from too little data are uncertain. Capability indices calculated before stability has been established are especially unreliable. Statistical evidence must therefore be combined with engineering knowledge, customer risk, and regulatory requirements.

Best-practice implementation sequence

1. Define the critical quality characteristic and specification.
2. Validate the measurement system.
3. Select a rational subgroup and sampling frequency.
4. Collect time-ordered baseline data.
5. Choose the chart type that matches the data.
6. Investigate special causes and establish stable limits.
7. Maintain a documented reaction plan.
8. Calculate capability only for a stable process.
9. Separate within-process capability from long-term performance.
10. Use improvement methods to centre the process and reduce variation.
11. Review the chart and capability method when the process, product, equipment, or measurement system changes.

Chapter Summary

Statistical Process Control uses time-ordered data to distinguish predictable common-cause variation from unusual special-cause variation. Variable charts, including X-bar–R, X-bar–S, and Individuals–Moving Range charts, monitor measured characteristics. Attribute charts—p, np, c, and u—monitor classifications or counts and must be selected according to the data structure.

Control limits are statistical boundaries, whereas specification limits are customer or engineering requirements. A process should be demonstrated to be stable before capability is assessed. C_p measures potential capability based on specification width and within-process spread; C_pk also considers centring. P_p and P_pk use overall long-term variation. Stable but incapable processes require reduction of common-cause variation or recentering; unstable processes require investigation of special causes. Effective SPC combines statistical calculation, measurement-system discipline, process knowledge, ethical governance, and documented corrective action.

Key Terms

Common-cause variation: Predictable variation inherent in the process system.

Special-cause variation: Unusual variation attributable to a specific, identifiable event.

Control chart: A time-ordered graphical method for monitoring process behaviour.

Control limit: A statistically calculated boundary used to identify unusual process behaviour.

Specification limit: An externally defined acceptance requirement.

Rational subgroup: A group of observations collected under conditions intended to estimate a particular source of variation.

Variable data: Quantitative measurements on a continuous or approximately continuous scale.

Attribute data: Counted or classified quality data.

Defective unit: A unit that fails one or more requirements.

Defect: An individual nonconformity within a unit.

X-bar chart: A chart monitoring subgroup means.

Range chart: A chart monitoring within-subgroup ranges.

p chart: A chart for the fraction defective, allowing variable sample sizes.

u chart: A chart for defects per unit when opportunities or unit sizes vary.

Process stability: Consistency of process behaviour over time under a defined statistical model.

Process capability: The ability of a stable process to meet specifications.

Cp: Potential capability based on specification width and within-process variation.

Cpk: Capability index incorporating process centring.

Ppk: Long-term performance index incorporating overall variation.

Measurement-system analysis: Evaluation of measurement bias, repeatability, reproducibility, and suitability.

Review Questions

Short Answer Questions

1. What is the difference between common-cause and special-cause variation?
2. Why should control limits not be confused with specification limits?
3. When is a p chart preferred to an np chart?
4. Why is process stability normally assessed before capability?
5. What does C_pk<C_p usually indicate?

Long Answer Questions

1. Explain the construction and interpretation of an X-bar and R chart, including the role of rational subgrouping.
2. Compare p, np, c, and u charts and provide an industrial example for each.
3. Explain the assumptions and managerial meaning of C_p, C_pk, P_p, and P_pk.
4. Discuss why tampering can increase process variation and describe an appropriate reaction plan for a control-chart signal.
5. Evaluate the limitations of normal capability indices when data are autocorrelated, skewed, or generated by multiple process populations.

Analytical / Application Questions

1. A stable process has LSL=48, USL=52, \mu=51, and \sigma_within=0.40. Calculate C_p and C_pk, then recommend the first improvement action.
2. A production team reports a stable R chart but a sequence of increasing X-bar values. Diagnose the most likely type of change, identify three data sources for investigation, and propose a containment action.
3. A hospital supplies department records the number of picking errors per 100 orders, but the number of orders inspected varies by day. Select an appropriate attribute chart, state its assumptions, and explain how you would respond to a point above the upper control limit.

Multiple Choice Questions with Answers

Question 1

Which statement best distinguishes control limits from specification limits?

A. Control limits are set by customers, while specification limits are calculated from process data.

B. Control limits describe expected process behaviour, while specification limits define required acceptance boundaries.

C. Both are always equal to three standard deviations from the target.

D. Specification limits are used only for attribute data.

Correct Answer: B. Control limits describe expected process behaviour, while specification limits define required acceptance boundaries.

Question 2

A process has stable variation but its mean is close to the USL. Which index will reflect this lack of centring?

A. Cp

B. Cpk

C. R-bar

D. D4

Correct Answer: B. Cpk

Question 3

Which chart is appropriate for the fraction defective when subgroup sizes vary?

A. np chart

B. c chart

C. p chart

D. X-bar chart

Correct Answer: C. p chart

Question 4

For a subgroup of five observations, which constants are commonly used for an X-bar and R chart?

A. A2 = 0.577, D3 = 0, D4 = 2.114

B. A2 = 0.729, D3 = 0, D4 = 2.282

C. A2 = 1.128, D3 = 0.729, D4 = 0

D. A2 = 2.282, D3 = 0.577, D4 = 0.729

Correct Answer: A. A2 = 0.577, D3 = 0, D4 = 2.114

Question 5

A process has USL = 102, LSL = 98, and within-process standard deviation = 0.5. What is Cp?

A. 0.800

B. 1.000

C. 1.333

D. 2.000

Correct Answer: C. 1.333

Question 6

If a control chart shows a point beyond its upper control limit, the most appropriate immediate response is to:

A. Delete the observation and recalculate the limits.

B. Adjust the machine immediately without investigation.

C. Contain affected output and investigate a possible special cause.

D. Change the specification limit.

Correct Answer: C. Contain affected output and investigate a possible special cause.

Question 7

Which chart monitors defects per unit when the number of opportunities varies?

A. u chart

B. np chart

C. R chart

D. p chart

Correct Answer: A. u chart

Question 8

In the filling example, Cp = 1.333 and Cpk = 0.800. What is the main interpretation?

A. The process is centred and has excessive random variation only.

B. The process has potential capability but is poorly centred.

C. The process is unstable because Cp is greater than Cpk.

D. The specification limits are control limits.

Correct Answer: B. The process has potential capability but is poorly centred.

Question 9

Why should the R chart generally be examined before interpreting the X-bar chart?

A. The R chart determines the customer specification.

B. The X-bar chart can be interpreted only for attribute data.

C. Unstable within-subgroup variation makes the estimated X-bar limits unreliable.

D. The R chart identifies the process target.

Correct Answer: C. Unstable within-subgroup variation makes the estimated X-bar limits unreliable.

Question 10

Which situation most directly calls for reducing common-cause variation?

A. A documented broken tool creates one isolated signal.

B. A stable process has excessive spread relative to its specifications.

C. A single data-entry error appears outside the limit.

D. A new product introduces a different specification.

Correct Answer: B. A stable process has excessive spread relative to its specifications.