Control Chart

Anatomy of a Control Chart

A control chart (SPC chart) plots a process metric over time together with statistical limits. It separates random variation (process "in control") from systematic deviations (special causes) — and is the standard Control-phase tool to keep an improved process in check long-term.

Center line (CL): The mean of observed values or subgroup means. It represents the current process level.

UCL and LCL (±3σ): Upper and lower control limits, typically at 3 standard deviations from the mean. Deviations within are treated as random; points outside indicate a special cause.

Zones A, B, C: The region between center line and limits is divided into three zones of one standard deviation each. Western Electric and Nelson rules use these zones to detect non-random patterns.

Special cause: An influence outside normal process variation — e.g. a tool break, a new operator, a material change. Must be identified and removed before the process continues.

Chart types: I-MR for individuals, X̄-R and X̄-S for subgroups, p and np for proportions (defective/not), c and u for counts (defects per unit). Choice depends on data type and subgroups.

Control charts prevent processes from drifting unnoticed. They are not quality verdicts but early-warning systems: they show something is happening before the spec is violated.

Approach

• Pick chart type by data type and subgroup structure.
• Collect enough baseline data — typically 20–25 subgroups or 100+ individuals.
• Compute center line and limits from the baseline (Phase I).
• Check for outliers and patterns — exclude special causes from Phase I and recompute limits.
• Start ongoing monitoring (Phase II) — each new point is checked against fixed limits.
• On a rule violation, find, document, and eliminate the cause.
• Recompute limits only when the process was demonstrably and intentionally changed.

Rules for Special Causes

• 1 point outside the 3σ limits — clearly out of control.
• 9 points in a row on the same side of the center line — level shift.
• 6 points rising or falling in a row — trend.
• 14 points alternating — systematic zig-zag.
• 2 of 3 points in zone A (>2σ) — near-limit excursion more often than expected.
• 4 of 5 points in zone B or beyond (>1σ) — process spreading wider than normal.

The rules are hints, not verdicts. Every violation should trigger a cause search, not an automatic alarm. Too many active rules create false alarms.

Pitfalls

Spec limits instead of control limits: USL/LSL come from the customer, UCL/LCL from process data. Mixing them (drawing spec limits on the chart and treating them as control limits) is among the most common SPC mistakes.

Wrong chart type: Applying an I-MR chart to proportion data or a p-chart to individual measurements yields nonsense limits. Data type and subgroup structure must fit the chart type.

Too few baseline data: With 5 subgroups the limits are unstable and keep shifting. At least 20–25 subgroups are needed for a reliable baseline.

Too many rules enabled: Every extra rule increases the false-alarm rate. Enabling all Nelson rules gives a "hit" every few points on average — and the team stops taking the chart seriously.

Recomputing limits after every change: Pulling limits along with every violation hides the drift. Only adjust limits when there is a documented reason (new process, new equipment).

Chart without reaction: A control chart without a defined reaction procedure is decoration. Per violation, it must be clear: who looks, who decides, who documents.

Available in the following cycles

• DMAIC: Control
• DMADV: Verify