5. Statistical Process Control and Process Capability

Introduction and Context

Quality management requires more than inspecting finished products. Inspection identifies nonconforming output after value has already been added; statistical process control (SPC) seeks to understand and control the process that produces the output. This distinction is strategically important. A process that is stable and capable is more likely to produce consistent quality at low cost, whereas a process dependent on sorting, rework, and final inspection conceals sources of variation and transfers quality costs downstream.

SPC provides a disciplined method for distinguishing routine, inherent variation from unusual variation arising from identifiable causes. Process capability analysis then asks a different question: even if the process is stable, can it consistently meet customer or engineering requirements? Stability concerns predictability; capability concerns conformity to specifications. A process may be stable but incapable, capable for a short period but unstable, or both stable and capable.

For production managers, SPC is not merely a statistical technique. It is a management system involving measurement, sampling, operator response, maintenance, engineering decisions, and organizational learning. Effective use requires suitable data, rational subgrouping, clear control limits, technically meaningful specifications, and a reaction plan that prevents the chart from becoming a passive display.

Learning Objectives

After studying this chapter, the reader should be able to:

1. Explain process variation, common causes, special causes, stability, and capability.

2. Select suitable control charts for variable and attribute data.

3. Construct and interpret X-bar–R, X-bar–S, individuals–moving range, p, np, c, and u charts.

4. Recognize non-random patterns and investigate special causes without overreacting to natural variation.

5. Calculate and interpret Cp, Cpk, Pp, Ppk, and related capability measures.

6. Design practical reaction plans for production environments.

7. Identify assumptions, limitations, ethical issues, and digital opportunities associated with SPC.

Variation, Control, and the Logic of Statistical Evidence

Every production process exhibits variation. Even when machines, materials, methods, and people appear unchanged, measurements differ because of many small influences. These are commonly called common causes or natural causes. Examples include minor raw-material differences, normal tool wear, small temperature fluctuations, and measurement noise. A process affected only by common causes is statistically stable or in statistical control, meaning that its future behaviour is reasonably predictable within the limits of the existing system.

Special causes, also called assignable causes, produce variation that is not part of the stable process system. Examples include a broken tool, incorrect machine setting, a new operator using an unapproved method, contamination of material, a sensor failure, or a power interruption. Special causes may increase variation, shift the process mean, create trends, or generate isolated extreme observations.

The central SPC principle is that action should depend on the type of variation. If a common-cause problem is treated as though it were special, managers may make unnecessary adjustments, creating tampering and increasing variation. If a special cause is dismissed as ordinary noise, the process may continue producing defects. Control charts provide evidence for making this distinction, but they do not replace technical investigation.

A control chart contains a centre line representing the process average or expected level and statistically derived upper and lower control limits. These limits describe the range of variation expected when only common causes operate. They are not specification limits. Control limits are calculated from process data; specification limits are established by customers, designers, regulators, or engineering requirements. Confusing the two is one of the most serious errors in process analysis.

A point outside a control limit is strong evidence of a special cause, assuming the data and calculations are valid. However, points inside the limits may also indicate non-random behaviour. A long run on one side of the centre line, a sustained trend, cycles, or an unusual alternation pattern may reveal a process change. Interpretation should use an approved set of rules, such as those based on Western Electric or Nelson principles, while avoiding excessive rules that generate false alarms.

Data, Measurement, and Rational Subgroups

The quality of an SPC study depends first on the quality of the data. Before selecting a chart, the analyst should establish the quality characteristic, unit of observation, sampling frequency, measurement method, and process boundaries. A gauge that is biased, unstable, or excessively variable can create apparent process signals or conceal real ones. Measurement-system analysis, including gauge repeatability and reproducibility studies where appropriate, should therefore precede important capability decisions.

A rational subgroup consists of observations collected under conditions that are as similar as practical, so that within-subgroup variation reflects short-term or immediate process variation. For example, five consecutive parts produced on one machine during a short interval may form a rational subgroup. Differences between subgroup averages then provide information about changes over time. If a subgroup mixes parts from several machines, shifts, or lots, the resulting chart may conceal important sources of variation.

Sampling frequency should reflect process speed, risk, and the likely time scale of special causes. Frequent sampling is justified when a defect can develop quickly or has severe consequences. A slower sampling interval may be adequate for a stable chemical or long-cycle process. The sample size should balance statistical sensitivity, inspection cost, and operational feasibility.

Control Charts for Measured Characteristics

X-bar–R Charts

The X-bar–R chart is commonly used when a measurable characteristic is collected in rational subgroups, especially for subgroup sizes from about 2 to 10. The X-bar chart monitors changes in the subgroup mean, while the R chart monitors within-subgroup dispersion.

For subgroup i of size n, the mean is

x̄ᵢ = (xᵢ₁ + xᵢ₂ + ... + xᵢₙ)/n

and the range is

Rᵢ = maximum observation − minimum observation.

Let x̄̄ be the average of subgroup means and R̄ the average range. The usual limits are:

X-bar chart: UCL = x̄̄ + A₂R̄; CL = x̄̄; LCL = x̄̄ − A₂R̄.

R chart: UCL = D₄R̄; CL = R̄; LCL = D₃R̄.

The constants A₂, D₃, and D₄ depend on subgroup size and are obtained from standard statistical tables. The R chart should generally be examined before interpreting the X-bar chart because estimates of variation and the X-bar limits depend on within-subgroup variation. If the R chart is unstable, the X-bar chart may be misleading.

X-bar–S Charts

When subgroup sizes are larger, the sample standard deviation is a more efficient measure of dispersion than the range. The X-bar–S chart uses the subgroup mean and standard deviation. With s̄ as the average subgroup standard deviation, typical limits are:

X-bar chart: UCL = x̄̄ + A₃s̄; CL = x̄̄; LCL = x̄̄ − A₃s̄.

S chart: UCL = B₄s̄; CL = s̄; LCL = B₃s̄.

The constants depend on subgroup size. The same interpretive principle applies: investigate instability in the S chart before drawing conclusions from the X-bar chart.

Individuals–Moving Range Charts

When production is slow, testing is destructive, or only one observation is available at a time, an individuals–moving range (I-MR) chart is appropriate. The individuals chart tracks each observation; the moving range usually measures the absolute difference between consecutive observations. The average moving range estimates short-term variation. For two-point moving ranges, the individuals limits are often calculated as:

UCL = x̄ + 2.66MR̄; CL = x̄; LCL = x̄ − 2.66MR̄.

The moving-range chart has centre line MR̄ and an upper limit commonly equal to 3.267MR̄. I-MR charts require caution when data are autocorrelated, seasonal, or produced by a process with changing operating conditions. A single observation per time period may also make the chart less sensitive than subgroup charts.

Control Charts for Attribute Data

Attribute data classify output as conforming or nonconforming, or count defects. The chart must match the sampling model and the quantity being monitored.

p Chart

A p chart monitors the fraction or proportion of nonconforming units in samples. Sample sizes may vary. If p̄ is the average fraction nonconforming and nᵢ is the size of sample i, the limits are:

UCLᵢ = p̄ + 3√[p̄(1 − p̄)/nᵢ]

LCLᵢ = p̄ − 3√[p̄(1 − p̄)/nᵢ].

Negative lower limits are set to zero. A p chart is appropriate when each unit is classified as conforming or nonconforming, even if a unit can have several types of defects.

np Chart

An np chart monitors the number of nonconforming units rather than the proportion. It requires a constant sample size. Its centre line is np̄ and its limits are based on the binomial variance n p̄(1 − p̄). It is often easier for operators to understand when the inspection sample size is fixed.

c Chart

A c chart monitors the number of defects per inspection unit when the inspection opportunity or area is constant and the count is reasonably modelled by a Poisson distribution. One unit might be a length of cable, a panel, or a fixed document batch. The limits are c̄ ± 3√c̄.

u Chart

A u chart monitors defects per unit when the number of units or the inspection opportunity varies. If cᵢ defects are found in nᵢ units, uᵢ = cᵢ/nᵢ. The limits vary with sample size and are based on ū. A u chart should not be used simply because the data are counts; the denominator and inspection opportunity must be considered.

The distinction between a nonconforming unit and a defect is essential. A single product may contain several defects but is still one nonconforming unit. Using a c or u chart for defective units, or a p chart for defect counts, can produce invalid conclusions.

Control-Chart Interpretation and Reaction Plans

A control chart should be read as a time-ordered record, not as a collection of unrelated points. Common signals include:

• one point beyond a control limit;
• two of three successive points beyond a two-sigma warning zone on the same side;
• four of five successive points beyond one sigma on the same side;
• eight or more consecutive points on one side of the centre line;
• six or more points steadily increasing or decreasing;
• a repeating cycle or sudden change in dispersion.

The precise rules should be defined by the organization and applied consistently. A signal is a prompt for investigation, not automatic proof of a particular cause. The investigator should check the record of machine settings, material lots, operators, maintenance, environmental conditions, measurement equipment, and recent process changes. The suspected cause should be confirmed where possible, and the corrective action should address the process rather than merely remove defective units.

A practical reaction plan specifies: the signal that triggers action, the person responsible, immediate containment, required records, investigation steps, disposition of affected output, escalation criteria, and verification of effectiveness. For example, a control-limit violation on a critical dimension may require stopping the machine, segregating production since the last verified good check, checking the gauge, identifying the machine condition, correcting the cause, and restarting only after confirmation. A minor run-rule signal on a low-risk process may require increased sampling and supervisory review rather than an immediate shutdown.

Process Stability and Capability

Capability analysis should normally be performed only after the process is demonstrated to be stable. If special causes are present, the estimated mean and standard deviation combine fundamentally different operating conditions and do not represent a predictable process. Capability indices calculated from unstable data can create false confidence.

Assume a characteristic has lower and upper specification limits, LSL and USL, and a stable process with mean μ and standard deviation σ. The potential capability index is:

Cp = (USL − LSL)/(6σ).

Cp compares the width of the specification interval with the natural process spread. It does not account for centring. A process can have a high Cp but still produce defects if its mean is close to one specification limit.

Cpk incorporates centring:

Cpk = minimum[(USL − μ)/(3σ), (μ − LSL)/(3σ)].

The smaller component identifies the limiting side. When Cp is substantially larger than Cpk, the process has potential capability but is off-centre. When Cp and Cpk are both low, the process spread is too large, the specifications are too narrow, or both.

For a one-sided specification, the relevant CPU or CPL measure may be used. Capability indices are often compared with organizational criteria such as 1.00, 1.33, or higher, but no universal threshold is appropriate for every industry. Critical safety characteristics, high-volume products, and low-consequence characteristics may require different standards.

Short-term indices, commonly Cp and Cpk, use an estimate of within-subgroup variation. Long-term performance indices, commonly Pp and Ppk, use overall variation across the full data set:

Pp = (USL − LSL)/(6s),

Ppk = minimum[(USL − μ)/(3s), (μ − LSL)/(3s)],

where s is the overall standard deviation. A substantial gap between Cpk and Ppk may indicate time-dependent shifts, batch-to-batch variation, tool wear, or other instability. These indices should not be interpreted without understanding the sampling period, distribution, and measurement system.

Normality is often assumed in basic capability analysis, but the assumption should be evaluated rather than accepted automatically. Skewed, bounded, multimodal, or autocorrelated data may require transformation, a non-normal capability method, stratification, or a different model. A capability index is not a universal measure of quality; it is conditional on the data-generating assumptions and specification meaning.

Applied Example: A Machining Process

A machining cell produces shafts with a nominal diameter of 25.00 mm and specifications of 24.90 to 25.10 mm. Five consecutive shafts are measured every hour. The X-bar–R chart is constructed from 25 subgroups. The R chart shows all points within its limits, while the X-bar chart shows a steady upward movement over six consecutive subgroups. No point is beyond the control limit, but the pattern suggests tool wear or thermal drift. The process should not be declared stable merely because no point exceeds a limit.

After investigation, the team finds that the cutting insert is not being replaced at the planned interval. A preventive replacement standard is introduced, and the chart is restarted using data collected after the process has been returned to normal conditions. The revised process has a mean of 25.01 mm and a within-process standard deviation of 0.018 mm. Its potential capability is approximately Cp = 0.20/(6 × 0.018) = 1.85. Its centring capability is approximately Cpk = min[(25.10 − 25.01)/(0.054), (25.01 − 24.90)/(0.054)] = min(1.67, 2.04) = 1.67. The process is both reasonably centred and capable under the stated assumptions.

The example illustrates three points. First, pattern recognition can identify risk before a limit violation. Second, capability calculations should follow stabilization and corrective action. Third, a high capability result does not eliminate the need for continued monitoring; it describes performance under the conditions represented by the data.

Illustrative Case Study: Reducing Surface Defects in Powder Coating

Background

A fictional manufacturer of industrial enclosures experiences customer complaints about surface blemishes. The production manager initially proposes 100 percent final inspection. The quality engineer recommends an SPC-based investigation instead.

Situation and Analysis

The process contains several coating lines, and the inspection team records the number of blemishes per enclosure. Because enclosure area varies slightly, the team first considers a u chart rather than a c chart. Data are stratified by line and shift to avoid mixing distinct processes. A measurement review shows acceptable agreement among inspectors.

The initial u charts reveal that Line 2 has a sequence of increasing defect counts during the night shift. Investigation finds that the powder feed hopper is not being cleaned according to the standard, and humidity control is weaker during that shift. The issue is a special cause, not evidence that the entire coating process is inherently incapable.

Approach and Solution

The plant introduces a documented hopper-cleaning checklist, humidity alarms, operator training, and a rule requiring confirmation of the first three enclosures after a cleaning or material change. The reaction plan requires segregation of output produced after a signal until the last verified conforming inspection. The team does not recalculate limits immediately after every improvement; it first verifies that the revised process is stable over a suitable period. Only then are baseline limits updated.

Outcome

The number of blemishes falls, and the chart displays random variation around a lower centre line. Management reduces intensive inspection gradually rather than abruptly, retaining periodic audits and continuing the control chart. The improvement is sustained because the response addressed the process conditions and assigned ownership.

Learning Points

The case demonstrates the importance of selecting a chart according to the data structure, stratifying when multiple processes are mixed, validating measurement, and linking statistical signals to a defined reaction plan. It also shows why final inspection alone would have detected symptoms without correcting humidity and cleaning practices.

Managerial and Professional Implications

SPC affects production planning, maintenance, procurement, engineering, and workforce management. A process signal may be caused by incoming material, an unrealistic production schedule, inadequate preventive maintenance, or an engineering change rather than operator performance. Managers should therefore avoid blame-oriented investigations and focus on evidence and system conditions.

Control charts also support supplier development. A supplier may provide capability evidence for a critical dimension, but the customer should understand how the samples were selected, whether the process was stable, which measurement system was used, and whether the stated specifications are customer requirements or internal targets. Capability claims without traceable data are weak evidence.

Ethically, organizations must not remove unfavourable observations merely to improve an index, change sampling after seeing results, or adjust control limits to hide signals. Data integrity requires preserving original measurements, documenting exclusions, and explaining any rational basis for removing observations. Regulatory environments may require validated methods, audit trails, and retention of quality records. For safety-critical characteristics, statistical evidence must complement—not replace—engineering judgment and mandatory compliance controls.

Digital SPC, Automation, and Emerging Practice

Modern manufacturing execution systems and Internet of Things platforms can collect measurements directly from sensors, gauges, and inspection equipment. Automated charting can provide real-time alerts, electronic reaction plans, and links to machine histories. Analytics can help identify relationships between process signals and variables such as temperature, tool life, material lot, or energy consumption.

Artificial intelligence and machine learning may support anomaly detection and predictive maintenance, but these methods do not remove the need for sound SPC. A predictive model trained on biased or unstable data may generate false alarms. Automated systems should distinguish between an alert, a confirmed special cause, and an authorized process adjustment. Human review remains important where actions affect product release, safety, or regulatory compliance.

Digital systems should preserve timestamped data, calibration status, operator and equipment identity, revision history, and the reason for any intervention. Cybersecurity and access control are also quality concerns because unauthorized changes to limits, recipes, or records can compromise process integrity.

Common Errors and Practical Safeguards

Several mistakes repeatedly weaken SPC programmes:

1. Using specification limits as control limits. Specifications describe acceptability; control limits describe observed process behaviour.

2. Charting mixed populations. Combining machines, products, shifts, or materials can conceal instability.

3. Ignoring the measurement system. Poor measurement creates misleading signals and capability estimates.

4. Calculating capability from unstable data. Capability indices then describe no predictable process.

5. Choosing a chart without considering the denominator. Defects, defective units, and defects per unit require different charts.

6. Reacting to every fluctuation. Excessive adjustment creates tampering.

7. Failing to react to non-random patterns. A process can be statistically unusual without crossing a limit.

8. Updating limits after every signal. Limits should be revised only after a documented and sustained process change.

9. Treating indices as guarantees. Capability is conditional on the period, assumptions, and measurement method.

Best practice is to begin with a clearly defined process, validated measurement, rational subgroups, an appropriate chart, documented interpretation rules, and an assigned reaction plan. The chart should be visible to those who can act, and its effectiveness should be reviewed through defect reduction, response time, false-alarm rate, and sustained process performance.

Chapter Summary

Statistical process control uses time-ordered data to distinguish common-cause variation from special-cause variation and to assess whether a process is predictable. Variable charts such as X-bar–R, X-bar–S, and I-MR charts monitor measured characteristics, while p, np, c, and u charts monitor attribute data. Correct chart selection depends on the data type, sample structure, and inspection opportunity.

Control limits are statistically derived and must not be confused with specification limits. Interpretation requires attention to points beyond limits as well as runs, trends, cycles, and changes in dispersion. A reaction plan converts a signal into disciplined containment, investigation, correction, and verification.

Process capability is meaningful only when the measurement system is suitable and the process is stable. Cp describes potential spread capability, while Cpk incorporates centring; Pp and Ppk use overall variation and are useful for longer-term performance. Digital systems can strengthen SPC through automated data collection and alerts, but governance, data integrity, human judgment, and ethical practice remain essential.

Key Terms

Common cause: Inherent, routine variation produced by the normal process system.

Special cause: An identifiable, unusual source of process variation.

Statistical process control: The use of statistical methods to monitor and improve process behaviour.

Control chart: A time-ordered chart with a centre line and statistically calculated control limits.

Control limit: A boundary indicating expected process variation under specified conditions.

Specification limit: A customer, design, regulatory, or engineering requirement for acceptable output.

Rational subgroup: A set of observations collected under sufficiently similar conditions for meaningful analysis.

Stability: A condition in which process behaviour is predictable and affected only by common causes.

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

Cp: A capability index comparing specification width with six times within-process standard deviation.

Cpk: A capability index that accounts for both process spread and centring.

Pp: A long-term performance index based on overall standard deviation.

Ppk: A long-term performance index accounting for centring and overall variation.

Nonconforming unit: A unit that fails to meet one or more specified requirements.

Defect: A failure to meet a characteristic requirement; one unit may contain multiple defects.

p chart: A chart for the fraction or proportion of nonconforming units.

np chart: A chart for the number of nonconforming units at constant sample size.

c chart: A chart for defect counts with a constant inspection opportunity.

u chart: A chart for defects per unit when inspection opportunity varies.

Reaction plan: A documented response to a control-chart signal or process abnormality.

Review Questions

Short Answer Questions

1. What is the difference between common-cause and special-cause variation?

2. Why must control limits not be confused with specification limits?

3. When is an X-bar–R chart preferred to an individuals–moving range chart?

4. What distinction exists between a defect and a nonconforming unit?

5. Why should capability analysis generally follow stability analysis?

Long Answer Questions

1. Explain the construction and interpretation of X-bar–R charts, including the role of the R chart.

2. Compare p, np, c, and u charts and identify the data conditions appropriate for each.

3. Explain Cp and Cpk, including how centring affects their interpretation.

4. Discuss the elements of an effective SPC reaction plan.

5. Examine the role of measurement-system quality, rational subgrouping, and data integrity in capability studies.

Analytical / Application Questions

1. A process has a stable mean but a Cpk much lower than Cp. Diagnose the likely condition and recommend managerial action.

2. A u chart for defects per unit shows a rising trend only on one shift. Design an investigation and reaction plan that avoids blaming operators without evidence.

3. A production manager wants to recalculate control limits after every point outside the limits. Evaluate this practice and propose a statistically and operationally sound alternative.

22. Multiple Choice Questions with Answers

Question 1

A process has all observations within the control limits, but eight consecutive points occur above the centre line. What is the most appropriate interpretation?

A. The process is necessarily incapable

B. The pattern may indicate a process shift or special cause

C. The specification limits should be widened

D. The data must be discarded

Correct Answer: B. The pattern may indicate a process shift or special cause

Question 2

Which chart is most appropriate for the proportion of nonconforming units when sample sizes vary?

A. c chart

B. np chart

C. p chart

D. X-bar–R chart

Correct Answer: C. p chart

Question 3

What does Cpk measure that Cp does not?

A. The number of defects per unit

B. The effect of process centring relative to specifications

C. The sample size used for inspection

D. The frequency of special causes

Correct Answer: B. The effect of process centring relative to specifications

Question 4

A process has Cp = 1.80 and Cpk = 0.95. Which conclusion is most justified?

A. The process has excessive spread but is perfectly centred

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

C. The process is unstable by definition

D. The specification limits are identical

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

Question 5

Which situation best supports use of an individuals–moving range chart?

A. Five consecutive parts are measured every hour

B. A constant sample of 100 units is classified as conforming or nonconforming

C. One destructive test result is available at a time

D. Defects are counted on panels of identical area

Correct Answer: C. One destructive test result is available at a time

Question 6

Why should an unstable process generally not be subjected to capability analysis?

A. Capability indices cannot be calculated for measured data

B. The estimated distribution may combine different process conditions and be unpredictable

C. Control charts and capability indices use the same limits

D. Stable processes cannot have specification limits

Correct Answer: B. The estimated distribution may combine different process conditions and be unpredictable

Question 7

A production unit counts scratches on products whose inspected surface area varies. Which chart is generally most appropriate?

A. u chart

B. np chart

C. X-bar chart

D. R chart

Correct Answer: A. u chart

Question 8

Which action is an example of tampering?

A. Investigating a confirmed tool breakage

B. Segregating output after a verified special-cause signal

C. Adjusting a stable process after every ordinary fluctuation

D. Checking the measurement system before capability analysis

Correct Answer: C. Adjusting a stable process after every ordinary fluctuation

Question 9

A large difference between Cpk and Ppk may suggest that:

A. Long-term variation is greater than within-subgroup variation

B. The process has no specification limits

C. The measurement system is automatically perfect

D. The process mean is exactly centred

Correct Answer: A. Long-term variation is greater than within-subgroup variation

Question 10

Which feature is essential in a useful reaction plan?

A. A requirement to alter the process after every point

B. A defined signal, responsible person, containment action, investigation, and verification

C. Removal of all observations outside specifications

D. Automatic widening of control limits

Correct Answer: B. A defined signal, responsible person, containment action, investigation, and verification