Statistical Quality Control

Learning Objectives

  • Understand the difference between common and assignable cause variation.
  • Interpret control charts for variables and attributes.
  • Calculate and evaluate process capability ratios.
  • Understand the fundamentals of acceptance sampling and Six Sigma.

Statistical Quality Control (SQC) is the application of statistical methods to monitor and control a process to ensure it operates at its full potential. Control charts for variables (Xˉ\bar{X}, RR, SS) and attributes (pp, npnp, cc, uu), process capability ratios, and natural tolerance limits are essential tools. In civil engineering, this means ensuring that construction materials (like concrete, steel, or asphalt) consistently meet design specifications.

The Nature of Variation

The Nature of Variation

Understanding why no two manufactured items are exactly identical. Even in the most tightly controlled manufacturing environment, variation is inevitable. SQC aims to distinguish between two fundamentally different types of variation.

Common Cause (Chance) Variation

Inherent, natural variation in the process due to countless small, uncontrollable factors (e.g., slight fluctuations in ambient temperature, minor inconsistencies in raw materials).

  • A process experiencing only common cause variation is said to be In Statistical Control.
  • Its behavior is predictable over time under the selected process model; stability does not by itself imply a normal distribution.

Assignable Cause (Special) Variation

Variation caused by specific, identifiable factors that are not part of the normal process (e.g., a broken machine part, a new untrained operator, a defective batch of raw material).

  • A process experiencing assignable cause variation is Out of Control.
  • Its behavior is less predictable. Investigate a credible signal and take corrective or containment action when the evidence and risk justify it.

Control Charts

Control Charts Overview

The primary tool of Statistical Process Control (SPC). A control chart is a time-series plot of a quality characteristic, featuring a Center Line (CL) and two Control Limits: the Upper Control Limit (UCL) and the Lower Control Limit (LCL).

Anatomy of a Control Chart

  • Center Line (CL): The historical average or target value of the process.
  • Control Limits (UCL, LCL): Common Shewhart limits are set about three standard errors of the charted statistic from its center line; the formulas depend on the chart and sampling plan. For a normal model, about 99.7% of individual observations fall within ±3σ\pm 3\sigma, but a point beyond a control limit is a signal to investigate—not proof of a specific assignable cause.

1. Control Charts for Variables

Variables Charts

Used for continuous, measurable data (e.g., concrete compressive strength, steel rebar diameter). Variables charts are almost always used in pairs: one to monitor the process average, and one to monitor the process variability.

Select a Variable Control Chart

For a continuous measured characteristic, choose a chart pair based on the sampling plan and number of observations at each time point.

Use an individuals/moving-range pair when there is one observation per time point. With rational subgroups, pair an X-bar chart with a suitable R or S chart; the subgroup size and adopted chart method determine the spread statistic and limits.

Select a Variable Control ChartFor a continuous measured characteristic, choose a chart pair based on the sampling plan and number of observations at each time point.. Use an individuals/moving-range pair when there is one observation per time point. With rational subgroups, pair an X-bar chart with a suitable R or S chart; the subgroup size and adopted chart method determine the spread statistic and limits.. Continuous measured characteristic → One observation per time point?; One observation per time point? — Yes → Use individuals and moving-range charts; One observation per time point? — No → Pair X-bar with a suitable R or S chart for the subgroup plan

Continuous measured characteristic → One observation per time point?; One observation per time point? — Yes → Use individuals and moving-range charts; One observation per time point? — No → Pair X-bar with a suitable R or S chart for the subgroup plan

  • Continuous measured characteristic: terminator
  • One observation per time point?: decision
  • Use individuals and moving-range charts: terminator
  • Pair X-bar with a suitable R or S chart for the subgroup plan: terminator

The Xˉ\bar{X} (X-bar) Chart

Monitors the average value of the process over time. Samples of size nn are taken periodically, and their means (xˉ\bar{x}) are plotted. It detects shifts in the process center.

Center Line for X-bar Chart

The grand mean of the sample means.

CL=xˉˉCL = \bar{\bar{x}}

Variables

SymbolDescriptionUnit
CLCLCenter Line-
xˉˉ\bar{\bar{x}}The grand mean (average of sample means)-

The RR Chart and SS Chart

Monitors the variability (spread) of the process over time.

  • RR Chart (Range): Plots each subgroup's range and is straightforward for smaller subgroups.
  • SS Chart (Standard Deviation): Plots each subgroup's standard deviation and is often preferred for larger subgroups. Use subgroup-size guidance and limits from the selected chart method.

Order of Interpretation

For a paired Xˉ\bar{X}-RR or Xˉ\bar{X}-SS method, review the spread chart before interpreting mean-chart signals. If the spread chart signals instability, investigate it and consider its effect on the paired mean-chart limits; do not treat the result as automatic proof of a cause or as a reason to adjust without investigation.

2. Control Charts for Attributes

Attribute Charts

Used for discrete, countable data (e.g., number of defective bricks, pass/fail inspections). Attribute charts monitor characteristics that cannot be easily measured on a continuous scale.

Fraction Defective Charts (pp and npnp)

Used when inspecting items that are simply classified as "conforming" or "non-conforming" (defective).

  • pp Chart: Plots the proportion of defective items in a sample. Essential when the sample size nn varies from batch to batch.
  • npnp Chart: Plots the number of defective items. Used only when the sample size nn is constant.

Defects per Unit Charts (cc and uu)

Used when an item can have multiple defects, but the item itself is not necessarily "defective" or rejected outright (e.g., the number of surface blemishes on a precast concrete panel).

  • cc Chart: Plots the total number of defects in a constant area of opportunity (e.g., one defined fixed-length pipe segment). The usual model is Poisson.
  • uu Chart: Plots the average number of defects per unit. Used when the area of opportunity varies (e.g., inspecting pipes of different lengths).
Select an Attribute Control Chart

For ordered categorical inspection data, match the chart to whether the sample counts nonconforming units or nonconformities and to the inspection opportunity.

Use an np chart for the number of nonconforming units when subgroup size is fixed, and a p chart for the fraction when subgroup size varies. Use a c chart for nonconformity counts with constant inspection opportunity and a u chart when that opportunity varies. Confirm the distributional assumptions and use suitable limits for the actual data.

Select an Attribute Control ChartFor ordered categorical inspection data, match the chart to whether the sample counts nonconforming units or nonconformities and to the inspection opportunity.. Use an np chart for the number of nonconforming units when subgroup size is fixed, and a p chart for the fraction when subgroup size varies. Use a c chart for nonconformity counts with constant inspection opportunity and a u chart when that opportunity varies. Confirm the distributional assumptions and use suitable limits for the actual data.. Collect ordered attribute data → What does each sample count?; What does each sample count? — Units → Is the number of inspected units fixed per subgroup?; Is the number of inspected units fixed per subgroup? — Fixed size → Use an np chart for number nonconforming; Is the number of inspected units fixed per subgroup? — Variable size → Use a p chart for fraction nonconforming; What does each sample count? — Defects → Is inspection opportunity constant?; Is inspection opportunity constant? — Constant → Use a c chart for counts per constant opportunity; Is inspection opportunity constant? — Variable → Use a u chart when opportunity varies

Collect ordered attribute data → What does each sample count?; What does each sample count? — Units → Is the number of inspected units fixed per subgroup?; Is the number of inspected units fixed per subgroup? — Fixed size → Use an np chart for number nonconforming; Is the number of inspected units fixed per subgroup? — Variable size → Use a p chart for fraction nonconforming; What does each sample count? — Defects → Is inspection opportunity constant?; Is inspection opportunity constant? — Constant → Use a c chart for counts per constant opportunity; Is inspection opportunity constant? — Variable → Use a u chart when opportunity varies

  • Collect ordered attribute data: terminator
  • What does each sample count?: decision
  • Is the number of inspected units fixed per subgroup?: decision
  • Use an np chart for number nonconforming: terminator
  • Use a p chart for fraction nonconforming: terminator
  • Is inspection opportunity constant?: decision
  • Use a c chart for counts per constant opportunity: terminator
  • Use a u chart when opportunity varies: terminator

Process Capability

Process Capability Overview

Is an "in-control" process actually good enough? A process can be perfectly "in statistical control" (only common cause variation), but still produce terrible products if its natural variation is wider than the customer's specifications.

Natural Tolerance Limits vs. Specification Limits

  • Natural Tolerance Limits (NTL): For a stable, approximately normal process, μ±3σ\mu \pm 3\sigma is a useful description of its central process spread—not an absolute bound on every possible observation. This represents the "Voice of the Process."
  • Specification Limits (USL, LSL): The engineering tolerances required by the design or the client. This represents the "Voice of the Customer."

Process Capability Ratios (CpC_p and CpkC_{pk})

Metrics that quantify how well the process fits within the specification limits.

  • CpC_p (Capability Potential): Measures the ratio of the specification width to the process width. It ignores whether the process is centered.

If Cp<1C_p < 1, the process spread is wider than the specification width. This does not make every unit nonconforming; the expected nonconformance also depends on centering and the distribution. A target such as Cp≥1.33C_p \ge 1.33 is a convention in some contexts, not a universal acceptance rule.

  • CpkC_{pk} (Capability Index): Takes centering into account. It is the minimum of capability relative to the upper and lower specs.

For a stable process with an appropriate distribution model, Cpk<1C_{pk} < 1 indicates that the nearest specification is less than three modeled standard deviations from the process mean; interpret the expected defect rate using the model and context.

Capability Potential

Measures the ratio of the specification width to the process width.

Cp=USL−LSL6σC_p = \frac{USL - LSL}{6\sigma}

Variables

SymbolDescriptionUnit
CpC_pCapability potential-
USLUSLUpper Specification Limit-
LSLLSLLower Specification Limit-
σ\sigmaProcess standard deviation-

Capability Index

Measures process capability taking centering into account.

Cpk=min⁡(USL−μ3σ,μ−LSL3σ)C_{pk} = \min \left( \frac{USL - \mu}{3\sigma}, \frac{\mu - LSL}{3\sigma} \right)

Variables

SymbolDescriptionUnit
CpkC_{pk}Capability index-
USLUSLUpper Specification Limit-
μ\muProcess mean-
σ\sigmaProcess standard deviation-
LSLLSLLower Specification Limit-
Stability and capability answer separate questions

A stable process can still miss specification requirements, and an apparently capable snapshot may be misleading if the process is unstable. Establish a suitable stable baseline and distribution model before interpreting capability indices.

Control limits describe time-ordered process stability while specification limits describe engineering requirements; stability is assessed before capability.
Assess Stability Before Process Capability

Apply stated signal rules to time-ordered observations, establish whether the process is stable, and assess capability only for a suitable continuous-data model.

A signal prompts investigation, not automatic adjustment. Correct only verified causes and establish a justified baseline before resuming monitoring. Control limits describe process behavior; specification limits are engineering requirements. Capability indices require a stable process and suitable distribution/variation model.

Assess Stability Before Process CapabilityApply stated signal rules to time-ordered observations, establish whether the process is stable, and assess capability only for a suitable continuous-data model.. A signal prompts investigation, not automatic adjustment. Correct only verified causes and establish a justified baseline before resuming monitoring. Control limits describe process behavior; specification limits are engineering requirements. Capability indices require a stable process and suitable distribution/variation model.. Start with a chart suited to the data → Plot ordered data with declared signal rules; Plot ordered data with declared signal rules → Is there evidence of special-cause variation?; Is there evidence of special-cause variation? — Yes → Investigate; correct verified causes and establish a new baseline; Is there evidence of special-cause variation? — No → Is the process stable over the study period?; Is the process stable over the study period? — No → Continue monitoring; capability is premature; Is the process stable over the study period? — Yes → Is a continuous characteristic judged against specifications?; Is a continuous characteristic judged against specifications? — No → Maintain the suitable chart; do not calculate Cp or Cpk; Is a continuous characteristic judged against specifications? — Yes → Is the distribution or variation model suitable?; Is the distribution or variation model suitable? — Yes → Calculate and report capability with assumptions; Is the distribution or variation model suitable? — No → Use a distribution-appropriate capability method

Start with a chart suited to the data → Plot ordered data with declared signal rules; Plot ordered data with declared signal rules → Is there evidence of special-cause variation?; Is there evidence of special-cause variation? — Yes → Investigate; correct verified causes and establish a new baseline; Is there evidence of special-cause variation? — No → Is the process stable over the study period?; Is the process stable over the study period? — No → Continue monitoring; capability is premature; Is the process stable over the study period? — Yes → Is a continuous characteristic judged against specifications?; Is a continuous characteristic judged against specifications? — No → Maintain the suitable chart; do not calculate Cp or Cpk; Is a continuous characteristic judged against specifications? — Yes → Is the distribution or variation model suitable?; Is the distribution or variation model suitable? — Yes → Calculate and report capability with assumptions; Is the distribution or variation model suitable? — No → Use a distribution-appropriate capability method

  • Start with a chart suited to the data: terminator
  • Plot ordered data with declared signal rules: process
  • Is there evidence of special-cause variation?: decision
  • Investigate; correct verified causes and establish a new baseline: terminator
  • Is the process stable over the study period?: decision
  • Continue monitoring; capability is premature: terminator
  • Is a continuous characteristic judged against specifications?: decision
  • Maintain the suitable chart; do not calculate Cp or Cpk: terminator
  • Is the distribution or variation model suitable?: decision
  • Calculate and report capability with assumptions: terminator
  • Use a distribution-appropriate capability method: terminator

Acceptance Sampling and Six Sigma

Additional Concepts

Other fundamental concepts in statistical quality control.

Acceptance Sampling

A quality-control method that uses a prescribed sample and decision rule to determine a production lot's disposition.

Single-Sample Attribute Plan

Proceed only when the approved plan covers the defined lot and defect criterion. If the sampling criteria or n,cn,c plan are not authorized, hold and obtain an applicable plan before sampling. Under a single-sample attributes plan, draw nn units at random, count dd nonconforming units, and accept the lot when d≤cd \le c; otherwise reject it or follow its authorized disposition. The sample decision does not estimate long-term process quality or replace process control. Double, multiple, sequential, and variable plans use their own stated rules.

Disposition a Lot with a Single-Sample Attribute Plan

Use an approved sample size and acceptance number to decide lot disposition from a random attribute sample.

This chart covers a single-sample attributes plan: n is the sample size, c is the acceptance number, and the lot is accepted when the sample count d of nonconforming units is at most c. Proceed only after confirming the plan fits the defined lot and defect, and its criteria and n,c are authorized. If no approved plan exists, specify relevant quality points and producer/consumer risks (for example AQL and LTPD), meet the required operating-characteristic properties, and obtain authorization; otherwise hold. Double, multiple, sequential, and variable plans use their own authorized decision rules. Lot acceptance sampling does not estimate long-term process quality or replace process control.

Disposition a Lot with a Single-Sample Attribute PlanUse an approved sample size and acceptance number to decide lot disposition from a random attribute sample.. This chart covers a single-sample attributes plan: n is the sample size, c is the acceptance number, and the lot is accepted when the sample count d of nonconforming units is at most c. Proceed only after confirming the plan fits the defined lot and defect, and its criteria and n,c are authorized. If no approved plan exists, specify relevant quality points and producer/consumer risks (for example AQL and LTPD), meet the required operating-characteristic properties, and obtain authorization; otherwise hold. Double, multiple, sequential, and variable plans use their own authorized decision rules. Lot acceptance sampling does not estimate long-term process quality or replace process control.. Define lot and nonconforming-unit criteria → Approved single-sample attribute plan?; Approved single-sample attribute plan? — Yes → Check the approved plan against lot and defect; Approved single-sample attribute plan? — No → Set quality/risk points; propose n and c; Check the approved plan against lot and defect → Plan fits the lot and is authorized?; Set quality/risk points; propose n and c → Plan fits the lot and is authorized?; Plan fits the lot and is authorized? — Yes → Lock approved sample size n and acceptance number c; Plan fits the lot and is authorized? — No → Hold; obtain an applicable authorized plan; Lock approved sample size n and acceptance number c → Draw n units at random from the lot; Draw n units at random from the lot → Inspect the sample; count d nonconforming units; Inspect the sample; count d nonconforming units → Is d ≤ c?; Is d ≤ c? — Yes → Accept the lot under this plan; Is d ≤ c? — No → Reject the lot; follow authorized disposition

Define lot and nonconforming-unit criteria → Approved single-sample attribute plan?; Approved single-sample attribute plan? — Yes → Check the approved plan against lot and defect; Approved single-sample attribute plan? — No → Set quality/risk points; propose n and c; Check the approved plan against lot and defect → Plan fits the lot and is authorized?; Set quality/risk points; propose n and c → Plan fits the lot and is authorized?; Plan fits the lot and is authorized? — Yes → Lock approved sample size n and acceptance number c; Plan fits the lot and is authorized? — No → Hold; obtain an applicable authorized plan; Lock approved sample size n and acceptance number c → Draw n units at random from the lot; Draw n units at random from the lot → Inspect the sample; count d nonconforming units; Inspect the sample; count d nonconforming units → Is d ≤ c?; Is d ≤ c? — Yes → Accept the lot under this plan; Is d ≤ c? — No → Reject the lot; follow authorized disposition

  • Define lot and nonconforming-unit criteria: terminator
  • Approved single-sample attribute plan?: decision
  • Check the approved plan against lot and defect: process
  • Set quality/risk points; propose n and c: process
  • Plan fits the lot and is authorized?: decision
  • Hold; obtain an applicable authorized plan: terminator
  • Lock approved sample size n and acceptance number c: process
  • Draw n units at random from the lot: process
  • Inspect the sample; count d nonconforming units: process
  • Is d ≤ c?: decision
  • Accept the lot under this plan: terminator
  • Reject the lot; follow authorized disposition: terminator

Six Sigma Principles

A set of process-improvement methods. Under the widely quoted convention, a centered normal process with a short-term six-sigma distance to its nearest specification and an assumed 1.5-standard-deviation long-term shift leaves a 4.5-sigma distance, about Cpk=1.5C_{pk}=1.5; its one-sided tail is about 3.4 defects per million opportunities. This is a comparison convention, not a universal performance promise, and the result depends on how defect opportunities are defined. Methods such as DMAIC organize improvement work.

Interactive Simulation

Interact with the simulation below to explore statistical quality control charts.

Engineering Data Analysis

Control Chart (Xˉ\bar{X}) Simulator

Process Parameters

Target Mean (μ\mu)100
Std Dev (σ\sigma)5
Subgroup Size (nn)5
Mean Shift (Assignable Cause)+0

Applies a step shift to the mean starting at subgroup 11.

UCL0.00
CL (Xˉˉ\bar{\bar{X}})0.00
LCL0.00

Interactive Simulation

Explore the difference between process control and process capability using the visualizer below to adjust specification limits, process mean, and process variation.

Engineering Data Analysis • Topic 13

Process Capability Index (Cₚ & Cₚₖ) Visualizer

Process Mean (μ\mu)100.0
Std Dev (σ\sigma)3.50
Lower Limit (LSL)90
Upper Limit (USL)110
Cp Index0.95
Cpk Index0.95
EvaluationNot Capable (Produces Defects)
Loading chart...

• Cₚ (Process Capability): Measures potential capability if the process was perfectly centered. Formula: Cp=USL−LSL6σC_p = \frac{\text{USL} - \text{LSL}}{6\sigma}

• Cₚₖ (Actual Capability): Accounts for the centering of the mean relative to limits. Formula: Cpk=min⁡(USL−μ3σ,μ−LSL3σ)C_{pk} = \min\left(\frac{\text{USL} - \mu}{3\sigma}, \frac{\mu - \text{LSL}}{3\sigma}\right)

Key Takeaways
  • Common vs. Assignable Cause: Distinguishing natural noise and specific problems.
  • Xˉ\bar{X} and R/SR/S Charts: Review the spread chart when interpreting its paired mean chart; a signal prompts investigation rather than proving a cause.
  • Attribute Charts: Monitor discrete data; use pp for varying sample sizes (proportions), cc for Poisson counts.
  • Control Limits: Common Shewhart limits are set about three model-based standard errors of the charted statistic from its center line; the chart and sampling plan determine the formula, and control limits are not engineering specifications.
  • Process Capability (Cp,CpkC_p, C_{pk}): Interpret indices only after establishing a suitable stable process and variation model; they compare modeled process spread and centering with specification limits.
  • Acceptance Sampling: Use an authorized sample plan for lot disposition; a single-sample attributes plan accepts only when d≤cd \le c and does not replace process monitoring.
  • Six Sigma convention: The often-quoted 3.4 defects per million opportunities depends on a normal-model calculation and an assumed 1.5-standard-deviation long-term shift; it is not a universal guarantee.

References