Statistical Process Control and Capability

Introduction

Manufacturing quality is not achieved merely by inspecting finished parts. Inspection detects nonconforming output after resources have already been consumed; statistical process control (SPC) seeks to understand and control the process that produces the output. In machine technology, SPC is applied to dimensions, surface roughness, hardness, cycle time, defect counts, tool wear, and many other characteristics.

SPC combines statistical reasoning with process knowledge. It distinguishes ordinary, inherent variation from unusual variation caused by identifiable events. Capability analysis then asks a different question: assuming the process is stable, can its natural variation meet the engineering specifications?

A process may be stable but incapable, capable but unstable at a particular moment, or neither stable nor capable. These distinctions are central to responsible quality decisions.

Learning Objectives

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

• distinguish common and special causes of variation;
• construct and interpret variable control charts;
• construct and interpret attribute control charts;
• assess process stability and recognize non-random patterns;
• calculate process capability and performance indices;
• state the assumptions and limitations of capability analysis; and
• recommend appropriate corrective and improvement actions.

1. Variation, Processes, and Statistical Control

Every manufacturing process varies. For example, repeated measurements of a turned diameter may differ because of material structure, machine stiffness, temperature, measurement resolution, tool geometry, and operator technique. Variation is not automatically evidence of poor workmanship.

Common and special causes

Common-cause variation is the cumulative effect of many small, relatively stable influences built into the process. Examples include normal machine vibration, routine measurement noise, and small material differences. Reducing common-cause variation usually requires changing the process system: machine design, tooling, maintenance, materials, methods, or environmental control.

Special-cause variation arises from an identifiable and unusual influence. Examples include a broken insert, incorrect offset entry, a loose fixture, a sudden coolant failure, or an improperly calibrated gauge. Special causes should normally be investigated and removed rather than statistically averaged away.

A process is in statistical control when its observed variation is consistent with a stable system of common causes. Statistical control does not mean that every item meets specification; it means that the process behavior is predictable in a statistical sense.

Control limits and specification limits

Control limits are calculated from process data and describe expected process behavior. Specification limits are established by design, function, safety, or customer requirements. The upper and lower control limits are denoted UCL and LCL; the upper and lower specification limits are USL and LSL.

Control limits are not acceptance limits. A point outside a control limit signals unusual process behavior, whereas a point outside a specification limit indicates a nonconforming product. A stable process can have all points within its control limits and still produce many items outside specification.

2. Rational Subgroups and Data Collection

A control chart is only as meaningful as the data-collection plan. A rational subgroup consists of observations collected close enough in time and under similar conditions that short-term variation is represented within the subgroup. Differences between subgroup averages then reveal changes over time.

For machining, five consecutive parts produced under the same setup may form a rational subgroup. If parts from different shifts, machines, or tool conditions are mixed indiscriminately, within-subgroup variation may be inflated and between-subgroup changes may be concealed.

Before collecting data, define:

1. the quality characteristic and its unit;
2. the sampling frequency and subgroup size;
3. the measurement method and gauge resolution;
4. the process conditions to record; and
5. the reaction plan for signals.

The measurement system must be suitable. If gauge variation is large relative to process variation, the chart may show false signals or conceal real ones. A measurement-system analysis, such as a gauge repeatability and reproducibility study, should precede serious capability work.

3. Variable Control Charts

Variable charts use measured numerical values such as diameter, length, mass, temperature, or roughness. The most common paired charts are the X̄–R chart and the X̄–S chart.

3.1 X̄–R charts

For subgroup i containing n observations xᵢ₁, xᵢ₂, …, xᵢₙ, the subgroup mean and range are:

The average mean and average range over k subgroups are:

For subgroup sizes commonly used in practice, control limits are calculated using tabulated constants A₂, D₃, and D₄:

The constants depend on n and assume reasonably independent observations and a stable measurement process. For small subgroup sizes, D₃ is often zero. The R chart is interpreted first because the estimate of short-term dispersion must be trustworthy before interpreting the X̄ chart.

3.2 X̄–S charts

For larger subgroups, the standard deviation sᵢ is generally more informative than the range. The subgroup standard deviation is commonly calculated as:

The average standard deviation is s̄ = (Σsᵢ) ÷ k. Constants c₄, A₃, B₃, and B₄ are used to correct bias and calculate limits:

The X̄–S chart is preferred when subgroup sizes are moderate or large. For individual measurements, an Individuals–Moving Range (I–MR) chart is often used, but it is more sensitive to non-normality, autocorrelation, and estimation assumptions.

Worked Example 1: X̄–R chart

Five subgroups of four shaft diameters, in millimetres, are summarized below.

Thus, x̄̄ = 20.016 mm and R̄ = 0.056 mm. For n = 4, use A₂ = 0.729, D₃ = 0, and D₄ = 2.282.

For the mean chart:

For the range chart:

All subgroup means and ranges lie within their limits. With no additional non-random pattern, the process is provisionally stable. This conclusion concerns statistical stability only; specification limits are still required to assess capability.

4. Attribute Control Charts

Attribute data classify observations or count occurrences. They are appropriate when measurement is impractical or when the requirement is naturally categorical, such as pass/fail, defective/nondefective, or number of surface defects.

p chart: fraction nonconforming

If nᵢ units are inspected in subgroup i and dᵢ are nonconforming, the fraction nonconforming is:

The pooled fraction is p̄ = (Σdᵢ) ÷ (Σnᵢ). For a constant subgroup size n:

Limits below zero are set to zero. The usual model is binomial: each unit is classified independently with a common probability of nonconformance.

np chart: number nonconforming

For constant n:

The np chart is easier to interpret in counts but cannot normally accommodate varying subgroup sizes.

c and u charts: defect counts

A c chart counts defects when the inspection unit or opportunity is constant. If c̄ is the average number of defects:

A u chart counts defects per unit when the number of units or opportunities varies:

For varying nᵢ, the limits are:

These charts assume a Poisson-type defect-counting process. Overdispersion, clustering, or multiple defect mechanisms may make that assumption unsuitable.

5. Interpreting Control Charts

A point outside a three-sigma limit is a strong signal of a special cause, but control-chart interpretation should not rely exclusively on that rule. Common supplementary rules include:

• two of three successive points beyond the same two-sigma zone;
• four of five successive points beyond the same one-sigma zone;
• eight successive points on one side of the center line;
• six successive points steadily increasing or decreasing; and
• an obvious cycle, stratification, or repeating pattern.

The exact rule set should be defined before routine monitoring. Excessive rules increase false alarms; too few rules delay detection. A signal is an invitation to investigate, not automatic proof of a particular cause.

A practical response is to preserve the evidence, identify affected time periods and lots, check machine and measurement conditions, contain suspect product, and document the confirmed cause. Do not adjust the process after every ordinary fluctuation. Such over-adjustment, called tampering, can increase variation.

6. Process Capability and Performance

Capability indices compare process spread and centering with specification requirements. They are meaningful only after the process is stable and the measurement system is adequate.

Let μ be the process mean, σ the process standard deviation, USL the upper specification limit, and LSL the lower specification limit. The potential capability index is:

C_p measures the ratio of specification width to six standard deviations. It ignores centering. The centered capability index is:

The two one-sided indices reveal which specification boundary is more critical. A process centered between symmetric limits has approximately C_pk = C_p. If it is off-center, C_pk < C_p.

Capability calculations generally assume a stable process, a meaningful estimate of short-term σ, independent observations, and an appropriate distributional model. The familiar “six sigma” interpretation is most direct for a stable, approximately normal distribution. Non-normal data require transformation, a suitable distributional method, or percentile-based analysis.

Performance indices

Performance indices use overall, long-term standard deviation s, often including shifts and other sources of variation:

The distinction is conceptual: Cp and Cpk commonly use within-subgroup or short-term variation, while Pp and Ppk use overall observed variation. Organizations must define their estimation conventions clearly because software labels and defaults differ.

Worked Example 2: capability

Suppose a stable shaft-diameter process has USL = 20.10 mm, LSL = 19.90 mm, mean μ = 20.04 mm, and within-process σ = 0.025 mm.

The upper and lower indices are:

Therefore:

The process spread could fit within the specification width if centered, as indicated by Cp = 1.33, but the actual mean is too close to the USL. Centering improvement is the immediate priority. Reducing variation is also valuable, but it will not by itself correct the mean shift.

Capability indices are not universal pass/fail laws. A customer may require Cpk ≥ 1.33, while a safety-critical process may require a higher value and documented controls. The requirement must be connected to risk, product function, and applicable standards.

7. Illustrative Case: CNC Turning Process

Background

A fictional precision-machining company produces a 20 mm bearing-seat diameter on a CNC lathe. The specification is 20.00 ± 0.05 mm. Five consecutive parts are sampled every hour, and an X̄–R chart is maintained.

Problem

After a tool insert change, the quality engineer observes seven successive subgroup means above the center line. No point exceeds the upper control limit, but the pattern suggests a sustained shift.

Analysis

The range chart remains stable, indicating that short-term dispersion has not materially changed. The mean-chart pattern suggests a special cause affecting location rather than spread. Investigation finds that the replacement insert was loaded with a nominal wear offset but the offset compensation procedure was not followed. The gauge calibration is verified, and the affected production interval is identified.

Approach and solution

The team contains parts produced after the suspected change, corrects the offset, and confirms the correction with a short run. The control chart is not automatically recalculated using the shifted data. Instead, the special-cause points are documented, and a new baseline is established only after the process is demonstrably stable. The work instruction is revised to require independent verification of tool offsets.

Outcome

The subgroup means return to the previous stable region, while the range chart remains acceptable. A later capability study gives Cp = 1.45 and Cpk = 1.38, indicating adequate potential and reasonably good centering under the selected estimation method.

Learning points

The case illustrates that control-chart evidence can identify a process shift before widespread nonconformance occurs. It also shows why the range chart and mean chart must be interpreted together, why special causes should be investigated rather than averaged into the baseline, and why capability should be calculated only after stability has been established.

8. Limitations, Governance, and Contemporary Practice

SPC is not a substitute for engineering judgment. Autocorrelated data can make ordinary control limits misleading because successive observations are not independent. Tool wear may produce a predictable trend rather than random variation. In such cases, regression, time-series methods, or tool-life models may supplement a conventional chart.

Non-normal distributions, changing production rates, mixed streams from several machines, and rare-event data require special care. Combining data from unlike machines can produce an apparently stable average that represents no actual machine. Subgrouping should reflect the physical process, not merely the convenience of data collection.

Digital manufacturing systems now collect measurements automatically from coordinate-measuring machines, in-process probes, and machine controllers. Automated alerts can improve reaction speed, but algorithms must be validated. Artificial intelligence may detect complex patterns or predict tool wear, yet it can also generate false alarms, obscure causal reasoning, or reproduce biased historical decisions. Human review, traceable data, cybersecurity, access control, and versioned decision rules remain necessary.

Quality records should identify the data source, measurement method, chart constants, exclusion rules, analyst, and corrective action. Removing an outlier is acceptable only when an assignable cause is demonstrated and documented; deletion merely to improve capability is a serious governance failure. SPC results should support product safety and truthful communication rather than be manipulated to satisfy an arbitrary index target.

Best practice is to:

1. validate the measurement system;
2. define rational subgroups;
3. establish limits from stable baseline data;
4. investigate signals using process knowledge;
5. separate control-limit decisions from specification decisions;
6. verify stability before capability analysis;
7. report the estimation method and assumptions; and
8. use charts as part of a broader improvement system.

Chapter Summary

Statistical process control monitors process behavior over time and separates common-cause variation from special-cause variation. Variable charts, including X̄–R and X̄–S charts, monitor measured characteristics; p, np, c, and u charts monitor classified units or defect counts. Rational subgrouping, suitable measurement systems, and appropriate distributional assumptions are essential.

Control limits describe expected process behavior, whereas specification limits describe product requirements. A stable process may still be incapable. Capability indices Cp and Cpk compare specification width and centering with short-term variation; Pp and Ppk use overall variation. Capability interpretation requires stability, reliable measurements, and explicit assumptions. Responsible SPC combines statistical signals with engineering investigation, documented reaction plans, and ethical data governance.

Key Terms

Common cause: Routine variation inherent in a stable process.

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

Statistical control: A condition in which process variation is predictable from common causes.

Control limit: A statistically calculated boundary used to detect unusual process behavior.

Specification limit: An engineering or customer requirement defining acceptable output.

Rational subgroup: A set of observations collected under closely similar conditions.

X̄ chart: A chart monitoring subgroup means.

R chart: A chart monitoring subgroup ranges.

S chart: A chart monitoring subgroup standard deviations.

p chart: A chart for the fraction nonconforming.

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

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

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

Capability: The ability of a stable process to meet specifications.

Cp: Potential capability based on specification width and short-term spread.

Cpk: Capability accounting for process centering.

Pp: Performance index using overall process spread.

Ppk: Performance index using overall spread and centering.

Tampering: Unnecessary adjustment of a stable process in response to ordinary variation.

Measurement-system analysis: Evaluation of measurement error and consistency.

Review Questions

Short Answer Questions

1. What is the difference between common-cause and special-cause variation?
2. Why must control limits be distinguished from specification limits?
3. What is the purpose of a rational subgroup?
4. When is a p chart preferred to an np chart?
5. Why should process stability be assessed before capability?

Long Answer Questions

1. Explain the construction and interpretation of an X̄–R chart, including the roles of the mean chart and range chart.
2. Compare p, np, c, and u charts, stating the data conditions appropriate for each.
3. Explain the conceptual difference between Cp and Cpk and describe how a process can have a satisfactory Cp but an unsatisfactory Cpk.
4. Discuss the assumptions underlying normal-theory capability indices and explain how non-normality may affect conclusions.
5. Develop a practical reaction plan for a special-cause signal in a CNC machining process.

Analytical / Application Questions

1. A stable process has USL = 50.10 mm, LSL = 49.90 mm, μ = 50.06 mm, and σ = 0.020 mm. Calculate Cp and Cpk, identify the limiting specification side, and recommend an improvement priority.
2. A p chart shows eight successive subgroup fractions below the center line, while the subgroup sizes are constant and the range of the related measurement chart is stable. What process interpretations are plausible, and what investigation should be performed?
3. A plant combines measurements from three machines into one capability study and obtains Ppk = 0.92. Explain why this result may be misleading and propose a more appropriate analysis plan.

Multiple Choice Questions with Answers

Question 1

Which statement best distinguishes a control limit from a specification limit?

A. A control limit is set by product design, while a specification limit is calculated from samples.

B. A control limit detects unusual process behavior, while a specification limit defines product requirements.

C. A control limit applies only to attribute data, while a specification limit applies only to variable data.

D. They are two names for the same statistical boundary.

Correct Answer: B. A control limit detects unusual process behavior, while a specification limit defines product requirements.

Question 2

For an X̄–R chart, which chart should generally be examined first?

A. The X̄ chart, because dispersion is irrelevant.

B. The specification chart, because specifications determine control limits.

C. The R chart, because the estimate of short-term variation must be reliable.

D. The p chart, because all processes are binomial.

Correct Answer: C. The R chart, because the estimate of short-term variation must be reliable.

Question 3

A process has USL = 10.10, LSL = 9.90, and σ = 0.025. What is Cp?

A. 0.67

B. 1.33

C. 2.67

D. 4.00

Correct Answer: B. 1.33

Question 4

If Cp = 1.50 but Cpk = 0.80, the most likely conclusion is that the process is:

A. Well centered and highly variable.

B. Centered exactly at the upper specification limit.

C. Potentially narrow enough but substantially off-center.

D. Unable to have any specification limits.

Correct Answer: C. Potentially narrow enough but substantially off-center.

Question 5

Which chart is designed for the fraction of nonconforming units?

A. p chart

B. c chart

C. X̄ chart

D. R chart

Correct Answer: A. p chart

Question 6

A u chart is most appropriate when:

A. The number of defects per unit is monitored and inspection opportunity may vary.

B. The subgroup mean of a measured dimension is monitored.

C. The number of nonconforming units is counted with constant sample size only.

D. The process standard deviation is known to be zero.

Correct Answer: A. The number of defects per unit is monitored and inspection opportunity may vary.

Question 7

What does a sequence of points steadily increasing on a control chart commonly suggest?

A. A possible trend or systematic process change.

B. Proof that the process is perfectly random.

C. That specification limits should be removed.

D. That all measurements are identical.

Correct Answer: A. A possible trend or systematic process change.

Question 8

Which condition is essential before a conventional capability study?

A. The process must have no specification limits.

B. The process should be statistically stable and measured reliably.

C. Every individual value must equal the target exactly.

D. The sample size must always be one.

Correct Answer: B. The process should be statistically stable and measured reliably.

Question 9

Which index uses overall, long-term standard deviation rather than commonly used within-subgroup variation?

A. Cp

B. Cpk

C. Pp

D. A₂

Correct Answer: C. Pp

Question 10

Unnecessary adjustment of a stable process in response to ordinary random fluctuation is called:

A. Stratification

B. Calibration

C. Tampering

D. Rational subgrouping

Correct Answer: C. Tampering