A quality control (QC) chart is a visual tool used to monitor whether a process or product characteristic is stable and operating within expected limits. QC charts show sampled measurements over time and flag signals that indicate special-cause (nonrandom) variation so teams can investigate and correct problems before defects escalate.
Key takeaways
– QC charts separate common (random) variation from special (assignable) causes of variation.
– Choose the chart type based on the data type: variable data (measurements) use X̅, R or S charts; attribute data (counts or proportions) use p, np, c, or u charts. Multivariate charts (e.g., Hotelling’s T²) monitor several correlated variables at once.
– Control limits are statistical—typically set at ±3 sigma—and differ from specification/tolerance limits set by product requirements.
– A chart’s value lies as much in the interpretation rules (patterns and trends) and response plan as in plotting the numbers.
Understanding quality control charts
Purpose
– Detect changes in process behavior quickly so you can find and remove assignable causes.
– Verify that process improvements hold over time.
– Provide a visual, data-driven basis for decision-making about production/process adjustments.
Basic elements of a QC chart
– X axis: time or sample sequence.
– Y axis: measured characteristic (e.g., dimension, density, defect count).
– Center line (CL): typically the process mean (for X̅ charts) or average statistic (e.g., average proportion for a p-chart).
– Upper control limit (UCL) and lower control limit (LCL): statistical bounds that reflect expected random variation. Points outside these limits indicate likely special-cause variation.
Types of QC charts (high level)
– Variable charts (continuous measurements)
• X̅ and R charts: monitor subgroup means (X̅) and subgroup ranges (R) when subgroups are small (n ≤ ~10).
• X̅ and S charts: use subgroup standard deviations (S) when estimating variability from larger subgroups.
• Individuals (XmR) charts: for single observations (n = 1).
– Attribute charts (count/proportion)
• p-chart: proportion defective in a sample.
• np-chart: number defective in a sample (constant sample size).
• c-chart: count of defects per unit (constant inspection area).
• u-chart: defects per unit (variable inspection area or sample size).
– Multivariate charts
• Hotelling’s T², multivariate control charts: monitor several correlated measurements simultaneously.
Important: control limits vs. specifications
– Control limits are derived from process data and reflect variability; they help detect changes in the process.
– Specification/tolerance limits are customer or design requirements. A process may be “in control” but still produce items outside specifications (and vice versa).
How to build and use a quality control chart — practical steps
Step 1 — Define objective and metric
– Choose the quality characteristic to monitor (e.g., diameter, weight, density, proportion defective).
– Decide whether the metric is variable (continuous) or attribute (count/proportion).
Step 2 — Design your sampling plan
– Select subgroup size (n) and sampling frequency. For variable data, typical subgroup sizes are 2–10; for individuals charts, n = 1.
– Samples should be representative and taken consistently (same place, shift, machine, etc.) to make interpretation meaningful.
Step 3 — Collect baseline data
– Collect enough initial samples to estimate the process center and variability. Common practice: 20–25 subgroups (or more) to estimate control limits reliably.
Step 4 — Calculate summary statistics
– For X̅–R charts (example)
• For each subgroup calculate the subgroup mean (X̅i) and range (Ri).
• Grand mean: X̅̿ = average of subgroup means.
• Average range: R̅ = average of subgroup ranges.
– For p-charts, compute the average proportion defective p̅ across samples, etc.
Step 5 — Compute control limits
– X̅ chart with R:
• CL = X̅̿
• UCL = X̅̿ + A2 × R̅
• LCL = X̅̿ − A2 × R̅
• (A2 is a constant that depends on subgroup size n; use standard tables.)
– R chart:
• CL = R̅
• UCL = D4 × R̅
• LCL = D3 × R̅
• (D3 and D4 depend on n; use standard tables.)
– For p-charts:
• CL = p̅
• UCL = p̅ + 3 × sqrt[p̅(1 − p̅)/n]
• LCL = p̅ − 3 × sqrt[p̅(1 − p̅)/n] (often floored at 0)
Note: For X̅ charts paired with S charts, use S̅ and constants B3/B4 or A3 (see statistical references and tables). When in doubt, refer to a trusted table or statistical software.
Step 6 — Plot and monitor
– Plot subgroup statistics and control limits over time.
– Apply run rules to detect nonrandom behavior (beyond ±3 sigma or pattern rules—see Step 7).
Step 7 — Interpret using rules and signals
– Common signals indicating special-cause variation:
• Any point outside UCL or LCL.
• Two out of three consecutive points beyond ±2 sigma on the same side of CL.
• A run of 7–8 points all on one side of CL.
• Trends (7–8 points steadily increasing or decreasing).
• Cycles or systematic patterns.
– Standard rule sets include the Western Electric rules and Nelson rules; apply a consistent set and document actions.
Step 8 — Investigate and act
– When you detect an out-of-control signal:
• Stop making arbitrary adjustments. Use the chart to decide whether action is warranted.
• Investigate potential assignable causes (tools, materials, operators, environment, settings).
• Take corrective action and document root cause analysis.
• If changes are made, re-estimate control limits using the new stable data (but retain a record of the change).
Step 9 — Maintain and review
– Periodically re-evaluate the sampling plan and control limits.
– Use the chart for trend analysis and continuous improvement.
Worked example — Bob’s widget press (X̅–R chart)
Scenario: Bob samples 4 widgets at five times (subgroup size n = 4, number of subgroups k = 5). He measures a buoyancy index (higher = more airy).
Sample subgroup means and ranges:
– Subgroup 1: X̅1 = 3.15, R1 = 0.30
– Subgroup 2: X̅2 = 3.10, R2 = 0.20
– Subgroup 3: X̅3 = 3.30, R3 = 0.20
– Subgroup 4: X̅4 = 3.00, R4 = 0.20
– Subgroup 5: X̅5 = 3.20, R5 = 0.20
Compute grand mean and average range:
– X̅̿ = (3.15 + 3.10 + 3.30 + 3.00 + 3.20) / 5 = 3.15
– R̅ = (0.30 + 0.20 + 0.20 + 0.20 + 0.20) / 5 = 0.22
Using statistical constant A2 for n = 4 (A2 ≈ 0.729):
– UCL_x̅ = 3.15 + 0.729 × 0.22 ≈ 3.31
– LCL_x̅ = 3.15 − 0.729 × 0.22 ≈ 2.99
For the R chart, using D3 = 0 and D4 ≈ 2.282 (for n = 4):
– UCL_R = 2.282 × 0.22 ≈ 0.50
– LCL_R = 0 × 0.22 = 0
Interpretation:
– All subgroup means (3.00–3.30) lie between 2.99 and 3.31 → no immediate out-of-control points.
– All ranges (0.2–0.3) are below UCL_R → variability appears stable.
– Continue monitoring and apply run rules to detect pending trends.
Common pitfalls and practical tips
– Don’t confuse control limits with specifications. A process “in control” can still produce items outside specifications.
– Use consistent sampling—mixing samples from different shifts or machines can mask problems.
– Avoid over-reacting to single points unless they violate control rules or are clearly caused by assignable events.
– Build an action plan ahead of time: decide who investigates, how to document root causes, and when to recalculate control limits after a proven change.
– Use automation/statistical software for real-time monitoring when possible—software also reduces calculation errors.
– For multivariate situations, don’t rely solely on multiple univariate charts if variables are correlated; use multivariate methods (e.g., Hotelling’s T²).
When to use attribute vs variable charts
– Use variable charts when you can measure a continuous characteristic (diameter, weight, density). They are usually more sensitive (fewer samples needed) than attribute charts.
– Use attribute charts for pass/fail, defect counts, or when measurements are impractical.
Further reading and references
– Investopedia — “Quality Control Chart”:
– NIST/ITL Engineering Statistics Handbook — Control Charts:
– ASQ — Control Chart resource:
– Montgomery, D.C., Introduction to Statistical Quality Control (standard textbook for formulas, constants, and rules).
Editor’s note: The following topics are reserved for upcoming updates and will be expanded with detailed examples and datasets.