7.4 Statistical Process Control (SPC) & Run Charts
Key Takeaways
- Run charts plot process observations chronologically against a center line (median or mean) to evaluate time-series trends, shifts, cycles, and runs before formal control limits are established.
- Statistical Process Control (SPC), pioneered by Walter Shewhart, uses Control Charts with empirical 3-sigma limits (UCL and LCL) to differentiate common cause variation from special cause variation.
- Control limits (UCL/LCL) represent the statistical 'Voice of the Process,' while specification limits (USL/LSL) represent customer requirements ('Voice of the Customer'); specification limits must never be plotted on control charts.
- Standard out-of-control criteria (Western Electric / Nelson Rules) identify non-random special cause patterns, including points beyond 3-sigma limits, runs of 9 consecutive points on one side of the center line, and 6 consecutively increasing or decreasing points.
- Control chart selection is dictated by data type and subgrouping: Variable charts (Xbar-R, Xbar-s, I-MR) for continuous measurements, and Attribute charts (p, np, c, u) for proportions, counts of defectives, and counts of defects.
7.4 Statistical Process Control (SPC) & Run Charts
Understanding, monitoring, and reducing process variation is the defining objective of quality engineering. Pioneered in 1924 by physicist and statistician Walter A. Shewhart at Bell Telephone Laboratories, Statistical Process Control (SPC) provides the scientific methodology to assess process stability, distinguish routine noise from actionable signals, and maintain predictable manufacturing and service operations. On the ASQ CQIA examination, candidates must master time-series Run Charts, Shewhart Control Charts, the taxonomy of variation, the boundary between Control Limits and Specification Limits, the Western Electric / Nelson Rules, and the complete Control Chart Selection Matrix.
1. Time-Series Process Tracking: Run Charts
A Run Chart is a graphical display of process data plotted in chronological sequence over time against a central reference line (most commonly the median of the dataset, though the mean is occasionally used). Run charts do not feature statistical control limits, making them simple, effective preliminary tools for identifying non-random patterns before formal SPC control charts are constructed.
Characteristic
▲
│ • • •
│ • • • • •
────┼───────────────────────•─────────────────────────── [ MEDIAN ]
│ • • • • • • •
│
└───────────────────────────────────────────────────► Time (Order of Production)
Non-Random Patterns on Run Charts
A process behaving with purely random variation fluctuates unpredictably around the median. Four primary non-random patterns indicate special cause disturbances:
- Shift (Run): A sequence of 7 (or 8) consecutive points falling entirely on one side of the median line. This indicates a sudden, sustained change in the process average.
- Trend: A steady sequence of 6 (or 7) consecutive points continuously increasing or continuously decreasing. This signals progressive drift, such as cutting tool wear, chemical bath depletion, or fouling.
- Alternating / Sawtooth Pattern: A sequence of 14 consecutive points alternating strictly up and down. This indicates over-control (tampering), systematic sampling across two alternating operators, or alternating machine fixtures.
- Too Few or Too Many Runs: Evaluating the total number of runs (crossings of the median) against statistical run-test tables to detect clustering or mixture.
2. Fundamentals of SPC & Shewhart Control Charts
A Control Chart elevates a run chart by adding mathematically determined Upper and Lower Control Limits ($UCL$ and $LCL$) positioned at exactly $\pm 3$ standard errors ($3\sigma$) from the central process average ($CL$).
+-------------------------------------------------------------------------+
| ANATOMY OF A SHEWHART CONTROL CHART |
+-------------------------------------------------------------------------+
| |
| UCL ────────────────────────────────────────────────── [+3 Sigma] |
| Zone A (Upper): 2σ to 3σ |
| - - - - - - - - - - - - - - - - - - - - - - - - - [+2 Sigma] |
| Zone B (Upper): 1σ to 2σ |
| - - - - - - - - - - - - - - - - - - - - - - - - - [+1 Sigma] |
| Zone C (Upper): Center to 1σ |
| CL ══════════════════════════════════════════════════ [Center Line] |
| Zone C (Lower): Center to 1σ |
| - - - - - - - - - - - - - - - - - - - - - - - - - [-1 Sigma] |
| Zone B (Lower): 1σ to 2σ |
| - - - - - - - - - - - - - - - - - - - - - - - - - [-2 Sigma] |
| Zone A (Lower): 2σ to 3σ |
| LCL ────────────────────────────────────────────────── [-3 Sigma] |
| |
+-------------------------------------------------------------------------+
The Mathematical Rationale for 3-Sigma Limits
Shewhart selected $3\sigma$ limits because, by the Central Limit Theorem and Chebychev's Inequality, roughly 99.73% of subgroup averages from a normally distributed, stable process will fall naturally within $\pm 3\sigma$ of the mean purely due to chance. A data point falling outside these boundaries has a probability of less than 0.27% ($p < 0.0027$) of occurring by random chance, providing an immediate, statistically sound trigger that an assignable cause has entered the process.
3. Common Cause vs. Special Cause Variation
W. Edwards Deming emphasized that managing quality requires understanding the two fundamentally distinct sources of process variation.
+-------------------------------------------------------------------------+
| VARIATION TAXONOMY & STRATEGY |
+-------------------------------------------------------------------------+
| |
| COMMON CAUSE VARIATION SPECIAL CAUSE VARIATION |
| (Chance / Random / Noise) (Assignable / Non-Random) |
| |
| * Inherent to the design & system * External, sporadic intrusion |
| * Present 100% of the time * Intermittent and unpredictable
| * Affects all units produced * Affects specific lots/shifts |
| * Predictable within 3σ limits * Points fall beyond 3σ limits |
| |
| MANAGEMENT ACTION: FRONTLINE OPERATOR ACTION: |
| * Fundamental process redesign * Halt line / isolate cause |
| * Capital investment, training * Immediate containment fix |
| * DO NOT TAMPER with stable system! * Eliminate root assignable |
| cause |
+-------------------------------------------------------------------------+
The Danger of Tampering (Deming's Funnel Experiment)
Tampering occurs when an operator or engineer adjusts a stable process in response to individual data points fluctuating purely due to common cause variation. Deming proved mathematically (via the Funnel Experiment) that tampering doubles the overall variance of the process, artificially destabilizing a previously predictable system.
4. Control Limits vs. Specification Limits
One of the most heavily tested conceptual distinctions on the ASQ CQIA examination is the difference between Control Limits and Specification Limits.
| Attribute | Control Limits ($UCL$ / $LCL$) | Specification Limits ($USL$ / $LSL$) |
|---|---|---|
| Core Representation | Voice of the Process (VOP) | Voice of the Customer (VOC) |
| Source of Calculation | Calculated mathematically from empirical subgroup data ($\pm 3\sigma_{\bar{X}}$). | Set externally by product design engineers, customer contracts, or regulations. |
| Application Scope | Applied to process summary statistics (subgroup averages $\bar{X}$, ranges $R$, counts). | Applied to individual units or parts ($X_i$). |
| Display on SPC Charts | Plotted directly on Shewhart control charts. | NEVER plotted on standard $\bar{X}$ control charts (subgroup averages cannot be directly compared to individual tolerances). |
| Managerial Goal | Establish Statistical Control (Stability). | Achieve Process Capability ($C_p, C_{pk} \ge 1.33$). |
+-------------------------------------------------------------------------+
| STABILITY VS. CAPABILITY MATRIX |
+-------------------------------------------------------------------------+
| |
| PROCESS IN CONTROL (STABLE) |
| YES NO |
| ┌───────────────────┬───────────────────┐ |
| PROCESS │ IDEAL STATE │ BRINK OF CHAOS │ |
| CAPABLE YES │ Predictable & │ Meets specs now, │ |
| (Meets VOC) │ meets customer │ but unpredictable │ |
| │ specs reliably │ defect surges │ |
| ├───────────────────┼───────────────────┤ |
| NO │ STATISTICAL TRAP │ TOTAL CHAOS │ |
| │ Predictable, but │ Unpredictable & │ |
| │ consistently │ consistently │ |
| │ produces scrap! │ out of spec │ |
| └───────────────────┴───────────────────┘ |
+-------------------------------------------------------------------------+
5. Western Electric & Nelson Out-of-Control Decision Rules
To identify non-random process behavior before a point violates the 3-sigma limits, the Western Electric Company (1956) and Lloyd S. Nelson (1984) codified standardized decision rules based on dividing the control chart into three equal $1\sigma$ zones on either side of the center line:
- Zone C: Center Line to $\pm 1\sigma$
- Zone B: $\pm 1\sigma$ to $\pm 2\sigma$
- Zone A: $\pm 2\sigma$ to $\pm 3\sigma$
The Standard Nelson Out-of-Control Rules
| Rule Number | Visual Condition on Control Chart | Statistical Interpretation & Diagnostic |
|---|---|---|
| Rule 1 | 1 point beyond Zone A (outside the $\pm 3\sigma$ $UCL$ or $LCL$). | Gross special cause disturbance; immediate process shift or severe measurement error ($p = 0.0027$). |
| Rule 2 | 9 (or 8) consecutive points on one side of the Center Line. | Sustained shift in the process average or baseline setup. |
| Rule 3 | 6 consecutive points continuously increasing or decreasing. | Continuous trend or drift (tool wear, chemical depletion, temperature rise). |
| Rule 4 | 14 consecutive points alternating up and down. | Systematic over-control (tampering), fixture swapping, or alternating inspectors. |
| Rule 5 | 2 out of 3 consecutive points in Zone A or beyond (on the same side). | Strong early warning signal of an impending process mean shift ($p = 0.003$). |
| Rule 6 | 4 out of 5 consecutive points in Zone B or beyond (on the same side). | Moderate early warning signal of process drift or centering offset. |
| Rule 7 | 15 consecutive points in Zone C (within $\pm 1\sigma$ of center line on either side). | "Stratification" / Hugging the center line; indicates incorrect control limit calculation or mixed subgroups. |
| Rule 8 | 8 consecutive points on both sides of center line with none in Zone C. | Mixture distribution; sampling from two different machines or streams without proper stratification. |
6. Control Chart Selection Taxonomy & Master Decision Matrix
Selecting the correct control chart depends strictly on two criteria: Data Type (Variable vs. Attribute) and Subgroup Sample Size ($n$).
+-------------------------------------------------------------------------+
| CONTROL CHART SELECTION DECISION TREE |
+-------------------------------------------------------------------------+
| |
| [ DATA TYPE? ] |
| │ |
| ┌───────────────┴───────────────┐ |
| ▼ ▼ |
| [ VARIABLE DATA ] [ ATTRIBUTE DATA ] |
| (Continuous: mm, kg, s) (Discrete: Good/Bad, Count) |
| │ │ |
| [ SUBGROUP SIZE? ] [ WHAT IS COUNTED? ] |
| │ │ │ │ │ |
| ▼ ▼ ▼ ▼ ▼ |
| n = 1 2≤n≤8 n ≥ 9 [ DEFECTIVES ] [ DEFECTS ] |
| │ │ │ (Nonconforming units)(Total flaw counts) |
| │ │ │ │ │ |
| ▼ ▼ ▼ [ SAMPLE SIZE? ] [ SAMPLE SIZE? ] |
| I-MR X̄ - R X̄ - s │ │ │ │ |
| ▼ ▼ ▼ ▼ |
| Variable Constant Constant Variable |
| (p) (np) (c) (u) |
| |
+-------------------------------------------------------------------------+
Comprehensive Master Selection Matrix
| Chart Type | Monitored Parameter | Data Classification | Subgroup Size ($n$) | Center Line ($CL$) | Control Limit Calculations |
|---|---|---|---|---|---|
| $I\text{-}MR$ | Individuals ($I$) & Moving Range ($MR$) | Variable (Continuous) | $n = 1$ (Individual measurements) | $\bar{X}$ and $\bar{MR}$ | $\text{UCL}_I = \bar{X} + 2.66\bar{MR}$<br>$\text{LCL}_I = \bar{X} - 2.66\bar{MR}$ |
| $\bar{X}\text{-}R$ | Subgroup Average ($\bar{X}$) & Range ($R$) | Variable (Continuous) | $2 \le n \le 8$ (Small subgroups) | $\bar{\bar{X}}$ and $\bar{R}$ | $\text{UCL}{\bar{X}} = \bar{\bar{X}} + A_2\bar{R}$<br>$\text{LCL}{\bar{X}} = \bar{\bar{X}} - A_2\bar{R}$<br>$\text{UCL}_R = D_4\bar{R},~\text{LCL}_R = D_3\bar{R}$ |
| $\bar{X}\text{-}s$ | Subgroup Average ($\bar{X}$) & Std Dev ($s$) | Variable (Continuous) | $n \ge 9$ (Large subgroups) | $\bar{\bar{X}}$ and $\bar{s}$ | $\text{UCL}{\bar{X}} = \bar{\bar{X}} + A_3\bar{s}$<br>$\text{LCL}{\bar{X}} = \bar{\bar{X}} - A_3\bar{s}$<br>$\text{UCL}_s = B_4\bar{s},~\text{LCL}_s = B_3\bar{s}$ |
| $p$-Chart | Proportion / Fraction Nonconforming | Attribute (Defective units) | Variable or Constant $n$ | $\bar{p} = \frac{\sum np}{\sum n}$ | $\text{UCL}_p = \bar{p} + 3\sqrt{\frac{\bar{p}(1-\bar{p})}{n}}$<br>$\text{LCL}_p = \bar{p} - 3\sqrt{\frac{\bar{p}(1-\bar{p})}{n}}$ |
| $np$-Chart | Number of Nonconforming Units | Attribute (Defective units) | Constant $n$ strictly required | $n\bar{p}$ | $\text{UCL}{np} = n\bar{p} + 3\sqrt{n\bar{p}(1-\bar{p})}$<br>$\text{LCL}{np} = n\bar{p} - 3\sqrt{n\bar{p}(1-\bar{p})}$ |
| $c$-Chart | Total Count of Nonconformities (Defects) | Attribute (Defect counts) | Constant unit area / size ($n=1$) | $\bar{c} = \frac{\sum c}{k}$ | $\text{UCL}_c = \bar{c} + 3\sqrt{\bar{c}}$<br>$\text{LCL}_c = \bar{c} - 3\sqrt{\bar{c}}$ |
| $u$-Chart | Average Defects per Inspection Unit | Attribute (Defect counts) | Variable or Constant inspection units | $\bar{u} = \frac{\sum c}{\sum n}$ | $\text{UCL}_u = \bar{u} + 3\sqrt{\frac{\bar{u}}{n}}$<br>$\text{LCL}_u = \bar{u} - 3\sqrt{\frac{\bar{u}}{n}}$ |
What is the fundamental conceptual difference between Control Limits (UCL/LCL) and Specification Limits (USL/LSL) in statistical quality control?
A quality inspector in a textile mill inspects fabric rolls of varying lengths (100 meters, 250 meters, and 500 meters) and counts the total number of weave flaws and surface blemishes (defects) per roll. Which statistical control chart should be selected to monitor the defect rate across these variable inspection unit sizes?
According to standard Western Electric and Nelson out-of-control decision rules, which of the following patterns represents a statistically significant special cause signal requiring immediate root-cause investigation?
What is the predictable operational outcome if a machine operator repeatedly adjusts machine calibration settings in response to routine individual measurements that are fluctuating purely due to common cause (chance) variation?