11.2 Control Charts for Variables

11.2 Control Charts for Variables

Shewhart's Philosophy: Common Cause vs. Special Cause Variation

In 1924, Dr. Walter A. Shewhart of Bell Telephone Laboratories revolutionized modern manufacturing by introducing the Statistical Process Control (SPC) chart. Prior to Shewhart, quality control consisted almost entirely of terminal inspection—sorting finished parts into conforming and scrap bins. Shewhart recognized that variation is an inescapable physical reality in every manufacturing process, but that variation stems from two fundamentally distinct sources:

1. Common Cause Variation (Chance Causes / Inherent Noise):
   • Common cause variation is the natural, background variability inherent in a stable system. It arises from the combined interaction of dozens of minor, unavoidable variables: slight micro-structural variations in raw steel bars, minor ambient temperature fluctuations in the shop, standard electrical line noise driving machine servomotors, or normal bearing runout.
   • System Behavior: Common cause variation is statistically stable, predictable, and follows a consistent probability distribution over time.
   • Management Action: Individual machine operators cannot eliminate common cause variation through machine tweaks. Reducing common cause variation requires systemic management intervention: purchasing higher-precision machine tools, installing climate-controlled HVAC systems, or sourcing tighter-tolerance raw stock.
2. Special Cause Variation (Assignable Causes / Specific Disturbances):
   • Special cause variation arises from external, specific, identifiable events that are not part of the standard process design. Examples include a chipped carbide cutting insert, an improper machine offset entered by a substitute operator, a batch of bar stock with incorrect hardness, a malfunctioning coolant pump, or a loose fixture clamp.
   • System Behavior: Special cause variation is sporadic, non-random, unstable, and unpredictable.
   • Shop Action: Quality inspectors and operators must detect special causes immediately, identify the root cause, take immediate containment action, and permanently eliminate the assignable cause to restore the process to a state of statistical control.

+-----------------------------------------------------------------------------+
|                   COMMON CAUSE VS. SPECIAL CAUSE VARIATION                  |
+-----------------------------------------------------------------------------+
|  DIMENSION                                                                  |
|     ^                                                                       |
|     |           * (Special Cause: Chipped Tool Insert)                      |
| UCL + - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - +
|     |      *         *       *                                              |
|     |  *       *   *       *   *   *       *   *                            |
|  CL +----*-------*-------*-------*-------*-------*--------------------------+
|     |        *               *       *   *                                  |
| LCL + - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - +
|     |                                                                       |
|     +------------------------------------------------------------> TIME     |
|        [ Inherent Common Cause Noise ]        [ Special Cause Disturbance ] |
+-----------------------------------------------------------------------------+

The Cardinal Danger: Process Tampering (Deming's Rule)

Dr. W. Edwards Deming frequently emphasized the destructive operational error known as process tampering. Tampering occurs when an operator treats common cause variation as if it were a special cause. If a machinist measures a single component that is slightly high (yet well within the 3-sigma statistical control limits) and immediately adjusts the tool offset downward, the operator has introduced artificial variation. This constant over-correction destabilizes a naturally stable process, actually doubling the overall variance and creating out-of-tolerance parts.

---

Principles of Rational Subgrouping

A control chart is only as reliable as the sampling strategy used to construct it. The concept of Rational Subgrouping—conceived by Shewhart—dictates how inspection data must be collected over time.

The Golden Rule of Rational Subgrouping

Subgroups must be chosen so that the opportunity for variation within each subgroup is minimized (reflecting only inherent common causes), while the opportunity for variation between subgroups is maximized (allowing assignable special causes to emerge).

1. Minimizing Within-Subgroup Variation:
   • To capture pure short-term common cause noise, the parts comprising a single subgroup ($n = 4$ or $5$) should be manufactured consecutively over the shortest practical time interval, under virtually identical operating conditions: the same operator, same machine spindle, same tool, same pallet, and same batch of raw material.
   • Under this protocol, the variation observed inside the subgroup ($R$ or $s$) is uncontaminated by long-term shifts.
2. Maximizing Between-Subgroup Variation:
   • Subgroups are collected periodically over time (e.g., 5 consecutive parts sampled every hour, or 5 parts sampled at the beginning, middle, and end of each shift).
   • This temporal spacing allows external disturbances—such as progressive tool wear, thermal growth of the machine frame, operator shift changes, and raw material heat changes—to manifest as shifts between the subgroup averages ($\bar{X}$).

+-----------------------------------------------------------------------------+
|                        RATIONAL SUBGROUPING MATRIX                          |
+-----------------------------------------------------------------------------+
|  SUBGROUP 1 (8:00 AM)  : [Part 1, Part 2, Part 3, Part 4, Part 5]           |
|                          -> Produced in 2 minutes; measures WITHIN spread.  |
|                                                                             |
|  SUBGROUP 2 (9:00 AM)  : [Part 1, Part 2, Part 3, Part 4, Part 5]           |
|                          -> Measures shift BETWEEN 8:00 AM and 9:00 AM.     |
|                                                                             |
|  CRITICAL RULE         : NEVER combine parts from Spindle 1 and Spindle 2   |
|                          into the same subgroup! Mixing streams artificially|
|                          inflates within-subgroup range, ruining limits.    |
+-----------------------------------------------------------------------------+

---

The $\bar{X}$ and $R$ Control Chart (Subgroups $n = 4$ to $6$)

The $\bar{X}$ (average) and $R$ (range) chart combination is the industry workhorse for monitoring variable quality characteristics when subgroup sizes are small ($n = 4$ to $6$, with $n = 5$ being standard).

• $\bar{X}$ Chart: Tracks the central tendency (process location) over time.
• $R$ Chart: Tracks process variability (dispersion) over time.

Mathematical Formulas for Center Lines and 3-Sigma Limits

Given $k$ rational subgroups, each of sample size $n$:

1. Calculate Subgroup Averages and Ranges: $\bar{X}_i = \frac{\sum X}{n}, \qquad R_i = X_{\text{max}} - X_{\text{min}}$
2. Calculate Grand Average and Average Range: $\text{Center Line for } \bar{X}: \quad \bar{\bar{X}} = \frac{\sum_{i=1}^{k} \bar{X}_i}{k}$ $\text{Center Line for } R: \quad \bar{R} = \frac{\sum_{i=1}^{k} R_i}{k}$
3. Calculate 3-Sigma Control Limits for the $\bar{X}$ Chart: $UCL_{\bar{X}} = \bar{\bar{X}} + A_2 \bar{R}$ $LCL_{\bar{X}} = \bar{\bar{X}} - A_2 \bar{R}$
4. Calculate 3-Sigma Control Limits for the $R$ Chart: $UCL_R = D_4 \bar{R}$ $LCL_R = D_3 \bar{R}$

Standard Shewhart Factors Table for Variables Charts

Crucial Inspection Detail: For subgroup sizes of $n \le 6$, the factor $D_3$ is equal to zero. Consequently, the Lower Control Limit for the Range chart is zero ($LCL_R = 0$). An inspector cannot have a range less than zero.

The Cardinal Operational Rule: Interpret the Range Chart FIRST!

A fundamental protocol tested on the ASQ CQI examination is the sequential priority of chart analysis:

Quality inspectors must ALWAYS inspect and evaluate the $R$ chart before analyzing the $\bar{X}$ chart.

• The Statistical Rationale: The control limits on the $\bar{X}$ chart ($UCL_{\bar{X}}$ and $LCL_{\bar{X}}$) are mathematically derived from $\bar{R}$ via the formula $\bar{\bar{X}} \pm A_2 \bar{R}$. If the $R$ chart exhibits an out-of-control condition (such as a point above $UCL_R$), it proves that the process dispersion is unstable. Because the spread is erratic, the calculated $\bar{R}$ is invalid, which renders the control limits on the $\bar{X}$ chart completely meaningless.
• Action Sequence: If the $R$ chart is out of control, stop immediately. Investigate the cause of excessive dispersion (e.g., loose tooling, severe material non-uniformity). Only when the $R$ chart demonstrates statistical stability can the inspector legitimately interpret the $\bar{X}$ chart for process centering and shifts.

---

The $\bar{X}$ and $s$ Control Chart (Subgroups $n \ge 7$)

While the range $R$ is simple to compute, it suffers from a major mathematical limitation: it uses only two numbers ($X_{\text{max}}$ and $X_{\text{min}}$) and discards all intermediate data. For larger subgroup sizes ($n \ge 7$, and especially $n \ge 10$), the sample standard deviation ($s$) becomes a far more efficient and reliable estimator of process dispersion.

Mathematical Formulas for $\bar{X}$ and $s$ Charts

1. Calculate Subgroup Standard Deviations ($s_i$): $s_i = \sqrt{\frac{\sum (X - \bar{X}_i)^2}{n - 1}}$
2. Center Lines: $CL_{\bar{X}} = \bar{\bar{X}} = \frac{\sum \bar{X}_i}{k}, \qquad CL_s = \bar{s} = \frac{\sum s_i}{k}$
3. Control Limits for $\bar{X}$ Chart (using $s$): $UCL_{\bar{X}} = \bar{\bar{X}} + A_3 \bar{s}$ $LCL_{\bar{X}} = \bar{\bar{X}} - A_3 \bar{s}$
4. Control Limits for $s$ Chart: $UCL_s = B_4 \bar{s}$ $LCL_s = B_3 \bar{s}$

Just as with range charts, for $n \le 5$, the factor $B_3 = 0$, meaning $LCL_s = 0$. For $n \ge 6$, $B_3$ becomes positive, establishing a non-zero lower boundary for process standard deviation.

---

Individual and Moving Range ($I-MR$ / $X-MR$) Charts

In many modern manufacturing and inspection environments, gathering subgroups of $n = 4$ or $5$ parts is technically impossible or economically prohibitive. In such cases, the Individual and Moving Range ($I-MR$) chart (also called the $X-MR$ chart) is applied to single measurements ($n = 1$).

Practical Inspection Applications for $I-MR$ Charts

1. Destructive Testing: Tensile pull testing of aerospace fasteners, burst pressure testing of hydraulic hoses, or metallurgical grain size verification, where destroying 5 consecutive parts every hour would bankrupt production.
2. Slow, Low-Volume Production: Large CNC machining centers manufacturing high-value turbine casings where only 1 or 2 parts are completed per shift.
3. Continuous Chemical and Fluid Processes: Measuring the pH of a passivating bath, viscosity of coating paint, or chemical concentration of an anodizing tank, where consecutive samples drawn seconds apart are virtually identical and subgrouping provides no added value.

Mathematical Construction of the $I-MR$ Chart

Consecutive individual measurements ($X_1, X_2, X_3, \dots$) are plotted on the Individuals chart. The moving range ($MR$) is calculated as the absolute difference between successive pairs of observations: $MR_i = |X_i - X_{i-1}|$ Because each moving range is computed from $n = 2$ consecutive points, the statistical factors for $n = 2$ apply ($d_2 = 1.128, D_4 = 3.267, D_3 = 0$, and $3/d_2 = 3/1.128 \approx 2.66$):

1. Center Lines: $CL_X = \bar{X} = \frac{\sum X}{k}, \qquad CL_{MR} = \bar{MR} = \frac{\sum MR}{k - 1}$
2. Control Limits for Individuals ($X$) Chart: $UCL_X = \bar{X} + 2.66 \bar{MR}$ $LCL_X = \bar{X} - 2.66 \bar{MR}$
3. Control Limits for Moving Range ($MR$) Chart: $UCL_{MR} = 3.267 \bar{MR}$ $LCL_{MR} = 0$

Metrological Caution with $I-MR$ Charts: Because consecutive moving ranges share data points ($MR_2 = |X_2 - X_1|$ and $MR_3 = |X_3 - X_2|$), moving ranges are mathematically correlated. A single extreme outlier will inflate two adjacent moving ranges. Inspectors must also recognize that the Individuals chart is sensitive to departures from normality; if individual measurements are skewed, standard 3-sigma limits may yield false alarms.

---

Establishing Trial vs. Revised Control Limits

Control charts are not static drawing callouts; they are dynamic statistical tools established through a structured multi-phase workflow:

+-----------------------------------------------------------------------------+
|                   CONTROL LIMIT ESTABLISHMENT WORKFLOW                      |
+-----------------------------------------------------------------------------+
|  PHASE 1: BASELINE DATA COLLECTION                                          |
|  - Collect 20 to 25 rational subgroups (minimum 100 total observations).    |
|  - Ensure process operates under standard production conditions.            |
+-----------------------------------------------------------------------------+
|  PHASE 2: CALCULATE TRIAL CONTROL LIMITS                                    |
|  - Compute X̄̄ and R̄.                                                        |
|  - Calculate trial UCL and LCL for both R and X̄ charts.                     |
+-----------------------------------------------------------------------------+
|  PHASE 3: EVALUATE STATISTICAL STABILITY                                    |
|  - Plot baseline points against trial limits.                               |
|  - Audit for points exceeding limits or exhibiting non-random patterns.     |
+-----------------------------------------------------------------------------+
|  PHASE 4: ASSIGNABLE CAUSE INVESTIGATION                                    |
|  - Found assignable special cause? Discard subgroup; recalculate limits.    |
|  - Common cause noise / no cause found? DO NOT DISCARD; process is unstable.|
+-----------------------------------------------------------------------------+
|  PHASE 5: LOCK REVISED CONTROL LIMITS                                       |
|  - Lock limits and extend forward for active real-time monitoring.          |
+-----------------------------------------------------------------------------+

The Revision Protocol: When Can Points Be Discarded?

A major compliance issue on ASQ quality audits is the unauthorized deletion of out-of-control data points. Quality inspectors must adhere to strict rules:

• Legitimate Discard: An out-of-control subgroup may be excluded from the recalculation of limits only if a rigorous root cause analysis confirms a documented, assignable special cause (e.g., maintenance records confirm a cracked boring bar insert, or calibration logs prove a digital micrometer had a loose anvil). Once the assignable cause is confirmed and documented, that subgroup is removed, and revised control limits are recalculated using the remaining subgroups.
• Illegitimate Deletion: If an inspector observes an out-of-control point but cannot identify an assignable cause, the point cannot be removed. Deleting unexplained points is data falsification that produces artificially narrow, invalid control limits.

---

Real Shop Inspection Scenarios & Common Exam Traps

• Real Shop Scenario — The False Out-of-Control Alarm from Tool Wear: A CNC operator machining bronze bushings plots an $\bar{X}$ and $R$ chart ($n = 5$). Over an 8-hour shift, the $R$ chart remains perfectly flat and stable near $\bar{R} = 0.0006\text{ in}$. However, the $\bar{X}$ chart displays seven consecutive points drifting steadily upward, with Point 8 breaching the $UCL_{\bar{X}}$. The operator assumes the machine is broken. Metrological Diagnosis: The inspector recognizes that continuous cutting causes predictable abrasive tool wear on the boring tool. As the insert wears down, the bore diameter gradually decreases (or outside diameter increases). This steady drift is a special cause trend. The solution is not overhauling the machine, but establishing a scheduled tool-offset compensation interval (e.g., adjusting the tool wear offset $0.0004\text{ in}$ every 50 parts) to neutralize the slope before limits are breached.

• Exam Trap: Specification Limits vs. Control Limits on $\bar{X}$ Charts: The most common trap in all quality testing is confusing Specification Limits with Control Limits:

  • Specification Limits ($USL / LSL$) represent the Voice of the Customer and apply strictly to individual parts ($X$).
  • Control Limits ($UCL / LCL$) represent the Voice of the Process and apply to sample statistics (such as subgroup averages $\bar{X}$).
  • NEVER plot specification limits on an $\bar{X}$ chart! Because subgroup averages exhibit a narrower standard deviation ($\sigma / \sqrt{n}$) than individual parts, plotting blueprint tolerances on an $\bar{X}$ chart gives a completely false impression of conformance.

• Exam Trap: Lower Control Limit on Range Charts for Small Samples: Examination questions often present a scenario with $n = 4$ or $n = 5$ and ask candidates to calculate the Lower Control Limit for the Range chart ($LCL_R$). A distractor choice will offer a negative number or state that $LCL_R$ cannot be calculated. For any subgroup size $n \le 6$, $D_3 = 0$, which mathematically mandates that $LCL_R = 0$.