6. Statistical Quality Control and Measurement Systems

Introduction and Context

Quality management depends on evidence. Inspection can identify defective output, but it does not by itself explain why defects occur or whether a process is capable of meeting requirements consistently. Statistical quality control (SQC) addresses this limitation by using data, probability, sampling, and statistical reasoning to distinguish normal process variation from unusual disturbances. Measurement-system analysis complements SQC by asking a prior question: can the organization trust the data on which its decisions are based?

In production management, these issues are inseparable. A process may appear unstable because an instrument is poorly calibrated; a capable process may be judged inadequate because measurements are excessively variable; or an apparently stable process may consistently produce nonconforming goods because its average is incorrectly centred. Effective quality decisions therefore require both process analysis and measurement-system evaluation.

This chapter develops the principal statistical tools used for these decisions. It explains descriptive statistics, probability and sampling, control charts for variables and attributes, process capability and performance, and the evaluation of measurement systems through calibration, bias, stability, linearity, repeatability, reproducibility, and gauge R&R. The emphasis is not merely on calculation but on managerial interpretation and appropriate action.

Learning Objectives

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

1. Explain the role of variation, probability, sampling, and descriptive statistics in quality control.

2. Distinguish common causes from special causes of process variation.

3. Construct and interpret variable and attribute control charts.

4. Select an appropriate chart for a particular type of data.

5. Calculate and interpret process capability and performance indices.

6. Explain calibration, bias, stability, linearity, repeatability, and reproducibility.

7. Evaluate a measurement system using gauge R&R logic.

8. Make appropriate production and improvement decisions from statistical evidence.

Variation, Data, and Statistical Thinking

Every production process exhibits variation. Some variation is inherent in the design, materials, equipment, methods, and environment of the process. This relatively predictable variation is called common-cause variation. Other variation arises from identifiable and unusual events such as tool breakage, incorrect settings, operator error, contaminated material, or instrument malfunction. This is special-cause variation.

The distinction is managerial as well as statistical. When only common causes are present, changing settings in response to each observation may increase variation rather than reduce it. This behaviour is often called tampering. A stable process must be improved by changing the underlying system. In contrast, a special cause requires investigation and corrective action directed at the specific event.

A process is not necessarily satisfactory merely because it is stable. Statistical stability means that the process behaves predictably over time; it does not mean that the predictable output meets customer specifications. Stability and capability are therefore separate questions.

Descriptive Statistics for Quality Data

Descriptive statistics summarize the location and spread of observations. The arithmetic mean is

x̄ = Σxᵢ/n

where xᵢ is an observation and n is the number of observations. The median is the central ordered observation and is less affected by extreme values. The range, R = maximum − minimum, gives a simple measure of within-sample spread. Variance and standard deviation provide more informative measures of dispersion, with the sample standard deviation commonly expressed as

s = √[Σ(xᵢ − x̄)²/(n − 1)].

The mean is useful for identifying process centring, while the standard deviation helps assess consistency. Histograms, box plots, run charts, and scatter diagrams reveal distributional shape, outliers, trends, and relationships that numerical summaries may conceal. A quality analyst should examine both numerical and graphical evidence.

Probability and Sampling

Probability supports decisions under uncertainty. If a process output is approximately normally distributed, the mean and standard deviation can be used to estimate the proportion of output within given limits. The familiar empirical rule suggests that approximately 68%, 95%, and 99.7% of observations fall within one, two, and three standard deviations of the mean, respectively, when the normal model is appropriate.

A sample is used because measuring every unit may be costly, destructive, or impractical. Sampling introduces uncertainty: a sample statistic is an estimate of a population parameter. A sound sampling plan specifies the population, sampling unit, sample size, selection method, frequency, and action rules. Random or systematic sampling is generally preferable to convenient selection because convenience samples may not represent actual process behaviour.

Subgrouping is especially important in control-chart applications. Units in a rational subgroup should be produced under conditions sufficiently similar that within-subgroup variation represents short-term process variation. Differences between subgroups can then reveal shifts over time. Poor subgroup design can hide meaningful variation or create false signals.

Control Charts: Monitoring Stability Over Time

A control chart is a time-ordered display of a process statistic with a centre line and statistically derived upper and lower control limits. Control limits describe expected process behaviour; they are not customer specification limits. Specification limits come from design, engineering, regulation, or customer requirements. Control limits are calculated from process data.

For a normally distributed individual observation, approximate three-sigma limits are μ ± 3σ. In practice, μ and σ are estimated from process data. A point outside the limits is a strong signal of a special cause, but other patterns are also important: a sustained run on one side of the centre line, a trend, cyclic behaviour, or unusual alternation. Interpretation rules should be defined before data are examined to avoid selective or subjective decisions.

A stable chart does not prove that the process is normally distributed or permanently controlled. It indicates that, over the period studied, the observed pattern is consistent with the assumed model. Control charts must therefore be maintained as process conditions change.

Charts for Continuous or Variable Data

Variable data are measured on a continuous scale, such as length, mass, temperature, pressure, or cycle time. The most common chart combination is the X̄–R chart, used when observations are collected in small subgroups.

The X̄ chart monitors subgroup averages and is sensitive to shifts in process location. The R chart monitors within-subgroup variation and is sensitive to changes in short-term dispersion. A typical procedure is:

1. Collect several rational subgroups, each containing a consistent number of observations.

2. Calculate each subgroup mean and range.

3. Estimate the overall average of subgroup means and the average range.

4. Calculate limits using appropriate control-chart constants.

5. Examine the R chart first, because interpretation of the X̄ chart depends on a reliable estimate of within-subgroup variation.

6. Investigate signals before adjusting the process.

For larger subgroups or when standard deviation is preferred, an X̄–S chart is used. For individual measurements, especially when production is slow or testing is expensive, an Individuals–Moving Range (I–MR) chart is appropriate. The individuals chart monitors individual values, while the moving-range chart estimates short-term variation from successive observations.

Charts for Attribute Data

Attribute data classify units or count events. A p chart monitors the fraction or proportion defective when sample sizes may vary. A np chart monitors the number of defective units when the sample size is constant. A c chart monitors the number of defects per inspection unit when the opportunity for defects is constant. A u chart monitors defects per unit when sample sizes or opportunities vary.

The distinction between a defective unit and a defect is essential. A product may contain several defects but still be counted as one defective unit if it fails acceptance criteria. Selecting the wrong chart changes the statistical model and may produce misleading limits.

Attribute charts are particularly useful when measurement is categorical, inspection is rapid, or several defect types are recorded. However, they generally require more observations to detect small process shifts than variable charts. Where economically and technically feasible, reliable variable measurements often provide earlier warning.

From Signals to Decisions

A control chart is a decision-support tool, not an automatic correction mechanism. When a signal appears, the analyst should verify the data, identify the affected period and product, preserve traceability, and investigate plausible causes. Possible evidence includes maintenance records, material batches, operator assignments, environmental conditions, tool life, software settings, and measurement-system status.

If a special cause is confirmed, affected output may require containment or reinspection. The cause should be removed or controlled, and the event documented. Limits should not be recalculated simply to make a point appear in control. Recalculation is justified only after a genuine process change or after a documented special cause has been removed and the resulting data represent the intended process.

Process Capability and Performance

Capability analysis compares the natural variation of a stable process with specification requirements. It should normally be performed only after statistical stability has been demonstrated and the measurement system has been found adequate.

For a two-sided specification with upper specification limit USL and lower specification limit LSL, the potential capability index is

Cp = (USL − LSL)/(6σ).

Cp compares specification width with six standard deviations but does not consider process centring. The capability index accounting for centring is

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

A large Cp with a substantially smaller Cpk indicates that the process has adequate potential spread but is off-centre. If Cp and Cpk are both low, variation is too large, centring is poor, or both conditions exist.

Performance indices Pp and Ppk use overall, long-term standard deviation rather than within-subgroup variation:

Pp = (USL − LSL)/(6s_overall)

Ppk = min[(USL − μ)/(3s_overall), (μ − LSL)/(3s_overall)].

Capability indices estimate potential performance under relatively stable conditions; performance indices describe actual overall performance over the period sampled. Their interpretation depends on distributional assumptions, data quality, stability, specification validity, and adequate sample size. A numerical threshold should not replace engineering judgement or customer requirements.

For a one-sided specification, an appropriate index is used, such as Cpu = (USL − μ)/(3σ) or Cpl = (μ − LSL)/(3σ). Non-normal data require care. Possible approaches include transformation, fitting an appropriate distribution, percentile-based analysis, or using a model suited to the physical process. Blind application of normal capability formulas can produce false confidence.

Measurement Systems: Ensuring Trustworthy Data

A measurement system includes the instrument, software, standards, method, environment, operator, and procedures. Measurement error may be random or systematic. Random error reduces precision; systematic error produces bias. If measurement variation is large relative to process variation or tolerance, control charts and capability indices become unreliable.

Calibration and Related Properties

Calibration compares an instrument with a traceable reference and, where necessary, adjusts or documents its indication. Calibration does not automatically guarantee that the instrument is suitable for a particular application; resolution, range, environmental conditions, and uncertainty must also be considered.

Bias is the difference between the average observed measurement and a accepted reference value. Stability is the consistency of measurement performance over time. Linearity describes whether bias remains consistent across the operating range. A system may be unbiased at one point but biased at another.

Repeatability is variation when the same operator measures the same characteristic on the same part using the same instrument under consistent conditions. Reproducibility is variation attributable to changes such as different operators, appraisers, methods, or laboratories. Together, repeatability and reproducibility form gauge R&R.

Gauge R&R and Measurement-System Analysis

A gauge R&R study normally uses representative parts spanning the process range, multiple operators, and repeated measurements in a randomized sequence. The analysis separates total observed variation into part-to-part variation and measurement variation. A system with excessive gauge variation cannot reliably discriminate among parts or support capability claims.

The percentage contribution, study variation, and percentage of tolerance are different interpretations and should not be confused. A measurement system may be acceptable for broad product screening but inadequate for controlling a tight engineering characteristic. The decision must consider the intended use, cost of improvement, risk of misclassification, and customer or regulatory requirements.

When a measurement system is inadequate, possible responses include improving fixture design, standardizing operating technique, controlling temperature, increasing instrument resolution, automating data capture, retraining operators, revising the measurement method, or selecting a more suitable instrument. Repeating measurements without addressing the source of error does not solve the problem.

Illustrative Case: Capability Confusion in a Machining Process

Background

A manufacturer of precision hydraulic components monitored the diameter of a machined bore. The engineering specification was 50.00 ± 0.05 mm. The production team reported a Cpk of 1.45 and concluded that the process was capable.

Situation and Analysis

A quality engineer reviewed the evidence. The X̄ chart showed a gradual upward trend, while the R chart indicated stable within-subgroup variation. The measurement system had not undergone a gauge R&R study. A study using ten representative parts, three operators, and two repeated measurements showed that measurement variation was substantial relative to the tolerance. Further examination found that the gauge was calibrated at room temperature, whereas measurement occurred near a machining centre that periodically increased local temperature.

The apparent Cpk had been calculated using a short-term estimate of variation and data collected during a period when the process mean was close to the target. It therefore understated long-term variation. In addition, the trending mean suggested that the process was not stable.

Approach and Solution

The organization first controlled the measurement environment and introduced a temperature-equilibration period for samples. Operators received a standardized measurement procedure, and the fixture was redesigned to prevent inconsistent positioning. A new gauge R&R study showed materially lower measurement variation. The team then collected fresh subgroup data, investigated tool-wear effects, and introduced a tool-life replacement rule based on process evidence rather than a fixed calendar interval.

Outcome

After these actions, the process displayed stable mean and range behaviour. Capability analysis was repeated using an appropriate within-subgroup estimate, while overall performance was reported separately for management review. The revised analysis showed that the process was only marginally capable when centred and that tool wear was the principal driver of long-term drift. The production team consequently improved centring and reduced tool-related variation rather than relying on inspection to sort output.

Learning Points

The case illustrates that capability indices are meaningful only when the process is stable and the measurement system is adequate. It also demonstrates why Cpk alone can conceal drift, why Cp and Cpk should be considered together, and why measurement-system improvement is often a prerequisite for process improvement.

Challenges, Risks, and Professional Considerations

Common mistakes include confusing control limits with specification limits, treating every out-of-control point as evidence that the entire process is defective, recalculating limits to suppress signals, ignoring autocorrelation, using non-random samples, and applying normal formulas to markedly non-normal data. Another frequent error is selecting a chart based only on convenience rather than on the data type and sampling design.

Managers should also avoid setting performance targets that encourage concealment, selective sampling, or excessive adjustment. Statistical results may affect product release, worker evaluation, supplier qualification, and customer communication. Data integrity, traceability, confidentiality, and transparent reporting are therefore professional obligations. If a measurement system is known to be inadequate, decisions based on it should be qualified and, where necessary, repeated after corrective action.

Digital quality systems can automate data collection, chart updating, alarm generation, and linkage with maintenance or manufacturing-execution systems. Machine learning may assist anomaly detection, but predictive models require representative historical data, controlled measurement processes, explainable decision rules, and human oversight. Automation should not convert statistical signals into unreviewed production stoppages or unverified product disposition.

Best Practices for Statistical Quality Control

A robust SQC system generally follows these principles:

• Define the quality characteristic, sampling plan, and decision rule before collecting data.
• Verify the measurement system before interpreting process statistics.
• Use rational subgroups and preserve time order.
• Separate stability analysis from capability analysis.
• Investigate signals using process knowledge and traceable evidence.
• Distinguish defects from defective units when selecting attribute charts.
• Report both short-term capability and long-term performance where relevant.
• Use control charts to prevent problems, not merely to document inspection.
• Reassess limits and capability after significant changes in equipment, material, method, or environment.
• Combine statistical evidence with engineering judgement, customer requirements, and risk analysis.

Current Developments

Modern factories increasingly integrate inline sensors, automated vision, statistical alarms, and real-time dashboards. These systems can reduce delay between process change and corrective action. Digital twins and advanced analytics can support prediction of drift, while automated measurement systems can improve repeatability when properly validated. Nevertheless, the foundations remain unchanged: representative data, sound sampling, valid measurement, appropriate statistical models, and disciplined human interpretation.

Chapter Summary

Statistical quality control uses data and probability to understand process behaviour and support decisions. Descriptive statistics summarize location and variation, while sampling provides practical evidence about a larger process. Control charts distinguish predictable common-cause variation from special-cause signals and must be selected according to the data type and sampling structure. Variable charts monitor measured characteristics; attribute charts monitor defective units or defect counts.

Capability indices compare process variation and centring with specifications, but they are meaningful only when the process is stable and the measurement system is trustworthy. Cp measures potential spread, whereas Cpk also reflects centring; Pp and Ppk use overall performance variation. Measurement-system analysis examines calibration, bias, stability, linearity, repeatability, reproducibility, and gauge R&R. Sound quality decisions require the integrated use of these ideas rather than reliance on a single statistic.

Key Terms

Common-cause variation: Predictable variation arising from the normal design and operation of a process.

Special-cause variation: Unusual variation attributable to an identifiable, non-routine source.

Rational subgroup: A sample formed so that within-group variation reflects short-term process variation.

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

Control limit: A statistically derived boundary used to judge process stability.

Specification limit: A customer, engineering, regulatory, or design requirement for a quality characteristic.

Variable data: Quantitative measurements recorded on a continuous scale.

Attribute data: Classified or counted quality information.

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

Defect: An individual nonconformity or departure from a requirement.

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

Cp: A capability index comparing specification width with potential six-sigma process spread.

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

Bias: The difference between average measurement and an accepted reference value.

Stability: Consistency of measurement performance over time.

Linearity: Consistency of measurement bias across the operating range.

Repeatability: Variation when the same operator repeatedly measures the same item with the same system.

Reproducibility: Variation associated with different operators or measurement conditions.

Gauge R&R: Analysis separating repeatability and reproducibility in a measurement system.

Measurement uncertainty: Quantified doubt associated with a measurement result.

Review Questions

Short Answer Questions

1. Distinguish common-cause variation from special-cause variation.

2. Why are control limits different from specification limits?

3. What is the purpose of rational subgrouping?

4. Define repeatability and reproducibility.

5. Why should measurement-system adequacy be checked before capability analysis?

Long Answer Questions

1. Explain the construction, interpretation, and managerial use of an X̄–R chart.

2. Compare p, np, c, and u charts, including the conditions under which each is appropriate.

3. Explain Cp, Cpk, Pp, and Ppk and discuss how their interpretation can differ.

4. Describe the major components of measurement-system analysis and their practical significance.

5. Develop a systematic procedure for responding to an out-of-control signal.

Analytical / Application Questions

1. A process has a high Cp but a much lower Cpk. Diagnose the likely condition of the process and recommend action.

2. A p chart is being used for samples whose sizes vary substantially. Evaluate whether the chart is appropriate and propose a better alternative.

3. A stable control chart shows excellent capability, but a gauge R&R study reports excessive measurement variation. Explain why the capability conclusion is unsafe and identify corrective actions.

22. Multiple Choice Questions with Answers

Question 1

A process is stable on its control chart but produces output outside customer specifications. What is the most appropriate conclusion?

A. The process is statistically predictable but not capable of meeting requirements

B. The control limits should be widened immediately

C. The measurement system must automatically be defective

D. Every observation outside specification is necessarily a special cause

Correct Answer: A. The process is statistically predictable but not capable of meeting requirements

Question 2

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

A. np chart

B. p chart

C. c chart

D. X̄ chart

Correct Answer: B. p chart

Question 3

A process has Cp = 1.60 and Cpk = 0.90. What does this most strongly suggest?

A. The process has excessive random measurement error only

B. The specification limits are identical

C. The process has reasonable potential spread but is poorly centred

D. The process is necessarily unstable

Correct Answer: C. The process has reasonable potential spread but is poorly centred

Question 4

Why is the R chart generally examined before interpreting the X̄ chart in an X̄–R system?

A. The R chart determines customer specifications

B. The X̄ chart cannot display time order

C. The R chart removes all special causes

D. The estimate of within-subgroup variation should be checked first

Correct Answer: D. The estimate of within-subgroup variation should be checked first

Question 5

Which situation represents reproducibility?

A. One operator measures the same part repeatedly with the same gauge

B. Different operators measure parts using the same measurement system

C. The same gauge is calibrated against a standard

D. A gauge gives the same result at two temperatures

Correct Answer: B. Different operators measure parts using the same measurement system

Question 6

A c chart is most appropriate when a quality team counts defects and:

A. the inspection unit and opportunity for defects remain constant

B. the sample size changes for every inspection

C. each unit is classified only as defective or nondefective

D. the measured characteristic is continuous

Correct Answer: A. the inspection unit and opportunity for defects remain constant

Question 7

What is the principal limitation of Cp compared with Cpk?

A. Cp cannot be calculated from specification limits

B. Cp applies only to attribute data

C. Cp does not account for process centring

D. Cp always uses long-term variation

Correct Answer: C. Cp does not account for process centring

Question 8

An instrument agrees with a reference at one point but shows increasing error across its range. Which property is most directly affected?

A. Repeatability

B. Linearity

C. Subgrouping

D. Defective-unit classification

Correct Answer: B. Linearity

Question 9

A point outside a control limit should initially lead the quality team to:

A. recalculate the limits so the point is included

B. adjust the process immediately without investigation

C. verify the data and investigate a possible special cause

D. change the specification limits

Correct Answer: C. verify the data and investigate a possible special cause

Question 10

Why can a capability study based on an inadequate gauge be misleading?

A. Measurement error can distort estimates of process location and variation

B. Gauge studies apply only to attribute charts

C. Capability indices do not use process variation

D. An inadequate gauge always makes Cpk equal to zero

Correct Answer: A. Measurement error can distort estimates of process location and variation