12.1 Statistical Process Control (SPC) & Shewhart Control Rules

12.1 Statistical Process Control (SPC) & Shewhart Control Rules

Quick Summary: Statistical Process Control (SPC) is an analytical methodology for monitoring, controlling, and improving processes using statistical techniques. Conceived by Walter Shewhart in 1924 and championed globally by W. Edwards Deming, SPC differentiates between common cause variation (inherent random noise) and special cause variation (assignable external disruptions). A process is in a state of statistical control when it exhibits only common cause variation. Green Belts monitor stability via control charts with limits set at $\pm 3\sigma$ and detect instability using pattern rules such as the Western Electric and Nelson rules.

---

The Philosophical Foundations of SPC

In high-volume manufacturing and transactional business environments, variation is inevitable. No two machined shafts, processed invoices, or chemical batches are identical down to the atomic level. In the 1920s at Bell Telephone Laboratories, physicist and statistician Walter A. Shewhart recognized that attempting to eliminate all variation through endless machine adjustments often degraded quality rather than improving it.

Shewhart proved that variation stems from two fundamentally distinct origins:

                         Total Process Variation
                                    │
         ┌──────────────────────────┴──────────────────────────┐
         ▼                                                     ▼
   Common Cause                                          Special Cause
 (Chance / Noise)                                    (Assignable / Signal)
  • Inherent to process design                         • External, unpredictable shock
  • Stable, random, predictable bounds                 • Intermittent, non-random shift
  • Addressed by Management System redesign            • Diagnosed and removed by Operators & Belts
  • Reacting to points = Tampering                     • Ignoring signals = Neglect

Common Cause Variation (Chance Variation)

Common cause variation represents the natural, historical, random background noise inherent to a stable system. It results from hundreds of minor, uncontrollable variables interacting simultaneously—such as ambient humidity fluctuations, normal electrical line voltage drift, minor raw material lot variances, or standard human micro-motion differences.

• Characteristics: It remains statistically stable, follows a repeatable probability distribution over time, and produces outputs within predictable limits.
• Responsibility: Because common cause variation is built into the process technology and workflow architecture, individual machine operators cannot remove it. Reducing common cause variation requires fundamental system redesign, capital investment, or process re-engineering—actions that fall under management responsibility.

Special Cause Variation (Assignable Cause Variation)

Special cause variation arises from specific, identifiable, external events that are not part of the normal process design.

• Characteristics: It is intermittent, non-random, unpredictable, and shifts the process mean, inflates process dispersion, or creates erratic patterns.
• Root Causes: Typical special causes include a broken tool, a batch of unannealed raw stock, a software bug, an untrained substitute operator, a sudden mechanical bearing seizure, or an uncalibrated measurement instrument.
• Responsibility: Operators, technicians, and Six Sigma Green Belts are responsible for rapidly identifying the root cause of a special cause signal, isolating defective output, and implementing immediate countermeasures to restore statistical control.

---

Deming's Two Mistakes and Process Tampering

W. Edwards Deming expanded Shewhart's principles into a comprehensive management philosophy. Deming emphasized that operational personnel routinely commit two disastrous errors when interpreting process data:

1. Mistake 1: Treating a common cause as if it were a special cause (Tampering / Over-adjustment). When an operator observes a part that deviates slightly from the target—yet remains well within the statistical control limits—and tweaks the machine setting, they introduce new variation. This phenomenon, demonstrated mathematically by Deming's famous Funnel Experiment, dramatically increases the process standard deviation and can double or triple process variability.
2. Mistake 2: Treating a special cause as if it were a common cause (Inaction / Complacency). When a true out-of-control signal appears (such as an extreme outlier or an upward drift) and the team dismisses it as "normal production noise," they miss the opportunity to diagnose and eliminate a systemic failure before defective product reaches the customer.

Core Principle: An organization must achieve statistical control before it can assess process capability. A process subject to special causes is unpredictable; any calculated capability index ($C_p$ or $C_{pk}$) on an unstable process is mathematically meaningless.

---

Anatomy of a Shewhart Control Chart

A control chart is a dynamic graphical tool that plots sample statistics (averages, ranges, individual values, or defect rates) in chronological time sequence against mathematically calculated horizontal reference boundaries.

 Value
   ▲
   │                                                            [Special Cause: Outlier]
UCL┼───────────────────────────────────────────────────────────────●────────────────────
   │                                 ●                   ●
   │      Zone A (2σ to 3σ)             ●             ●     ●
+2σ┼- - - - - - - - - - - - - - - - - - - -● - - - - - - - - - - - - - - - - - - - - - -
   │      Zone B (1σ to 2σ)                   ●     ●
+1σ┼- - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - -
   │      Zone C (0 to 1σ)          ●            ●
 CL┼───────────────────────●───────●────────────────────────────────────────────────────
   │      Zone C (0 to 1σ)    ●
-1σ┼- - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - -
   │      Zone B (1σ to 2σ)      ●
-2σ┼- - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - -
   │      Zone A (2σ to 3σ)
LCL┼────────────────────────────────────────────────────────────────────────────────────
   │
   └─────────────────────────────────────────────────────────────────────────────────► Time / Subgroup

A standard control chart features three structural reference lines:

1. Center Line (CL): Represents the historical mathematical average of the process statistic being plotted (e.g., $\bar{\bar{X}}$ for subgroup means, $\bar{X}$ for individual values, $\bar{p}$ for proportions). It reflects central process location.
2. Upper Control Limit (UCL): Positioned exactly three standard errors above the Center Line: $$\text{UCL} = \text{CL} + 3\sigma_{\text{stat}}$$
3. Lower Control Limit (LCL): Positioned exactly three standard errors below the Center Line: $$\text{LCL} = \text{CL} - 3\sigma_{\text{stat}}$$ (Note: For attribute charts or dispersion charts where a statistic cannot be negative, the LCL defaults to zero if the calculation produces a negative number).

Why $\pm 3$-Sigma Limits?

Shewhart chose $3$-sigma limits based on economic and probabilistic balance. Under the assumption of an underlying normal distribution of sample statistics (supported by the Central Limit Theorem):

• The probability of a sample point falling between $\pm 1\sigma$ is $68.26\%$
• The probability of a sample point falling between $\pm 2\sigma$ is $95.44\%$
• The probability of a sample point falling between $\pm 3\sigma$ is $99.73\%$

The probability of a point falling outside the $3$-sigma limits purely due to random chance is:

$$\alpha = 1 - 0.9973 = 0.0027 \quad (0.27\%, \text{ or approximately } 1 \text{ out of every } 370.4 \text{ subgroups})$$

Setting limits at $\pm 3\sigma$ provides an optimal economic balance: it minimizes the risk of false alarms (Type I error, chasing phantom causes) while retaining high statistical power to detect meaningful process shifts (avoiding Type II error).

---

Control Limits vs. Specification Limits: The Critical Boundary

Confusing control limits with specification limits is one of the most persistent errors made by novice quality practitioners and is heavily tested on the CSSC Green Belt examination.

Cardinal Exam Rule: Never draw specification limits on an $\bar{X}$ chart. An $\bar{X}$ chart plots subgroup averages, which exhibit standard error $\sigma_{\bar{x}} = \sigma / \sqrt{n}$, making their spread significantly narrower than the distribution of individual values. Placing individual piece specifications on a chart of averages leads to false confidence, as individual pieces may violate specifications even while subgroup averages sit well within the limits.

---

Control Chart Zones & Shewhart / Western Electric / Nelson Rules

To detect subtle process shifts, drifts, and non-random patterns before single points breach the $3$-sigma control limits, Shewhart's framework subdivides the territory between the Center Line and each control limit into three equal $1\sigma$ zones:

• Zone C (Inner Zone): Between Center Line and $\pm 1\sigma$ ($68.26\%$ of normal data)
• Zone B (Middle Zone): Between $\pm 1\sigma$ and $\pm 2\sigma$ ($27.18\%$ of normal data)
• Zone A (Outer Zone): Between $\pm 2\sigma$ and $\pm 3\sigma$ ($4.28\%$ of normal data)

The Western Electric Rules (1956) and Nelson Rules (1984) codify exact statistical decision criteria based on these zones. When any of these rules are violated, the process is statistically declared out of control, triggering immediate root-cause investigation.

Pattern Diagnostics Visualization:

Rule 1: Point Beyond 3σ          Rule 2: Run (9 on one side)      Rule 3: Trend (6 increasing)
      ● [Out of Control!]                                                    ●
UCL ──┼────────────────────      UCL ───────────────────────      UCL ───────●───────────────
      │                                                                     ●
      │                          CL  ──●──●──●──●──●──●──●──●─    CL  ─────●─────────────────
CL  ──┼──●───────●─────────          (9 consecutive above)                ●
      │    ●   ●                                                         ●
LCL ──┼──────●─────────────      LCL ───────────────────────      LCL ──●────────────────────

Rule 4: Alternation (14 pts)     Rule 7: Stratification (15)      Rule 8: Mixture (8 in A/B, 0 in C)
UCL ───────────────────────      UCL ───────────────────────      UCL ──●───●───────●────────
      ●   ●   ●   ●   ●   ●                                         Zone A/B points
CL  ────●───●───●───●───●──      CL  ──●─●─●─●─●─●─●─●─●─●─●      CL  ───────────────────────
                                     (Hugging Center Line)          [Zone C completely empty]
LCL ───────────────────────      LCL ───────────────────────      LCL ────●───●───●───●──────

---

Step-by-Step Worked Calculation: Establishing Control Chart Zones & Pattern Diagnosis

Scenario: A precision aerospace manufacturing cell turns titanium hydraulic spool valves. An $\bar{X}$ control chart tracks the outer diameter of the spool. Subgroup sampling consists of $n = 5$ parts inspected every hour. Over a baseline historical period of 25 in-control subgroups, the grand mean is $\bar{\bar{X}} = 50.000\text{ mm}$ and the standard error of the mean is calculated as $\sigma_{\bar{x}} = 0.015\text{ mm}$.

Step 1: Calculate the Exact Zone Boundaries

With $\bar{\bar{X}} = 50.000\text{ mm}$ and $1\sigma_{\bar{x}} = 0.015\text{ mm}$:

• Center Line (CL): $\text{CL} = 50.000\text{ mm}$
• Zone C Boundaries ($\text{CL} \pm 1\sigma_{\bar{x}}$):
  • Upper Zone C limit: $50.000 + 0.015 = 50.015\text{ mm}$
  • Lower Zone C limit: $50.000 - 0.015 = 49.985\text{ mm}$
  • Zone C Span: $[49.985\text{ mm}, 50.015\text{ mm}]$
• Zone B Boundaries ($\text{CL} \pm 2\sigma_{\bar{x}}$):
  • Upper Zone B limit: $50.000 + 2(0.015) = 50.030\text{ mm}$
  • Lower Zone B limit: $50.000 - 2(0.015) = 49.970\text{ mm}$
  • Upper Zone B Span: $(50.015\text{ mm}, 50.030\text{ mm}]$
  • Lower Zone B Span: $[49.970\text{ mm}, 49.985\text{ mm})$
• Zone A Boundaries / Control Limits ($\text{CL} \pm 3\sigma_{\bar{x}}$):
  • Upper Control Limit (UCL): $50.000 + 3(0.015) = 50.045\text{ mm}$
  • Lower Control Limit (LCL): $50.000 - 3(0.015) = 49.955\text{ mm}$
  • Upper Zone A Span: $(50.030\text{ mm}, 50.045\text{ mm}]$
  • Lower Zone A Span: $[49.955\text{ mm}, 49.970\text{ mm})$

Step 2: Evaluate a Production Run Sequence

During the subsequent shift, the inspector logs the following sequence of 8 consecutive subgroup averages:

$$\bar{X} = \{50.008, 50.018, 50.025, 50.033, 50.038, 50.041, 50.036, 50.039\}$$

Let us map each observation to its respective zone:

1. Subgroup 1: $50.008\text{ mm} \to$ Zone C (Upper)
2. Subgroup 2: $50.018\text{ mm} \to$ Zone B (Upper)
3. Subgroup 3: $50.025\text{ mm} \to$ Zone B (Upper)
4. Subgroup 4: $50.033\text{ mm} \to$ Zone A (Upper)
5. Subgroup 5: $50.038\text{ mm} \to$ Zone A (Upper)
6. Subgroup 6: $50.041\text{ mm} \to$ Zone A (Upper)
7. Subgroup 7: $50.036\text{ mm} \to$ Zone A (Upper)
8. Subgroup 8: $50.039\text{ mm} \to$ Zone A (Upper)

Step 3: Diagnostic Rule Evaluation

• Rule 1 Check: Are any points $> 50.045\text{ mm}$ (UCL) or $< 49.955\text{ mm}$ (LCL)?
  • Maximum point is $50.041\text{ mm} \le 50.045\text{ mm}$. Rule 1 is not violated.
• Rule 5 Check (2 out of 3 consecutive in Zone A):
  • Looking at Subgroups 4, 5, and 6: Subgroup 4 ($50.033$), Subgroup 5 ($50.038$), and Subgroup 6 ($50.041$) are all in Upper Zone A.
  • This represents 3 out of 3 consecutive points in Zone A, which severely breaches Rule 5 ($2$ out of $3$ in Zone A).
• Rule 6 Check (4 out of 5 consecutive in Zone B or beyond):
  • Looking at Subgroups 2 through 6: all 5 points fall in Zone B or Zone A.
  • This violates Rule 6 ($4$ out of $5$ points in Zone B or beyond).
• Operational Conclusion: Although not a single part has breached the Upper Control Limit ($50.045\text{ mm}$), the process is definitively declared statistically out of control at Subgroup 5 due to Rule 5. The Green Belt must immediately halt production, quarantine parts machined since Subgroup 3, and execute the Out-of-Control Action Plan.

---

Diagnostic Application: Step-by-Step Out-of-Control Investigation

When an automated quality system or quality inspector flags an out-of-control rule breach, the Green Belt executes an Out-of-Control Action Plan (OCAP):

1. Verify Measurement Integrity: Confirm that the flagged point was not caused by a data entry typo, transposition error, or uncalibrated gauge.
2. Isolate Affected Material: Quarantine parts produced between the last known in-control subgroup and the out-of-control point to prevent defective shipment.
3. Classify the Signal:
   • Is it a sudden single-point shock (Rule 1)? Check for sudden power blips, tool breakage, or new material batches.
   • Is it a trend (Rule 3)? Check for gradual degradation, such as cutter wear, coolant temperature rise, or filter clogging.
   • Is it stratification (Rule 7)? Verify whether samples within the subgroup were inadvertently pooled from two separate machines.
4. Identify Assignable Cause: Use Root Cause Analysis tools (5 Whys, Fishbone Diagram) with the process operators.
5. Implement Countermeasure and Document: Eliminate the assignable cause, verify that subsequent points re-establish statistical control, and document the resolution in the process logbook.

---

Critical CSSC Exam Traps

• Trap 1: Plotting Specification Limits on an $\bar{X}$ Chart — Exam questions will present an $\bar{X}$ control chart displaying customer specification lines and ask you to identify the defect. The error is that specification limits govern individual parts ($X$), while $\bar{X}$ charts track subgroup averages. Placing USL/LSL on an $\bar{X}$ chart is an absolute statistical violation.
• Trap 2: Assuming "In Control" Means "Zero Defects" — A process can be in perfect statistical control (predictable, stable common cause variation only) while generating $20\%$ scrap because its natural spread exceeds customer specifications. Control charts monitor stability, not capability.
• Trap 3: Misinterpreting Stratification (Rule 7) as "High Quality" — Candidates often see 15 points clustered tightly around the Center Line and conclude the process has become "hyper-precise." In reality, Rule 7 indicates an error in rational subgrouping—typically pooling samples from two diametrically opposing sources, which artificially inflates the subgroup range $\bar{R}$ and balloons the control limits.
• Trap 4: Tampering vs. Assignable Cause Action — The exam frequently tests whether an operator should adjust a machine when a point is at $+2.2\sigma$ (in Zone A) but no run rule has triggered. Adjusting the machine here is tampering. Operators must only act when a formal Western Electric / Nelson rule is violated.