5.2 Control Limits vs. Specification Limits

5.2 Control Limits vs. Specification Limits

Among quality practitioners, few concepts create more dangerous confusion on the production floor than the distinction between control limits and specification limits. When operators or supervisors see a subgroup average plotted near a control limit, they frequently ask: 'Are we making bad parts?' Conversely, when an out-of-control point triggers an alarm, technicians are often pressured with: 'Why shut down the line when all the parts are well inside blueprint tolerance?'

To pass the ASQ CQT exam and lead effective shop-floor quality initiatives, you must understand that control limits and specification limits serve completely different purposes, are derived from completely different sources, govern completely different statistical entities, and must never be interchanged or combined.

---

1. Voice of the Process (VOP) vs. Voice of the Customer (VOC)

The fundamental distinction between control limits and specification limits centers on who is speaking:

   VOICE OF THE PROCESS (VOP)                 VOICE OF THE CUSTOMER (VOC)
  ============================               =============================
  • Represented by: Control Limits           • Represented by: Spec Limits
    (UCL, LCL)                                 (USL, LSL)
  • Source: Actual Process Data              • Source: Engineering Blueprints
  • Evaluates: Statistical Stability         • Evaluates: Part Conformance
  • Answers: 'What is the process doing?'    • Answers: 'What do we need the
                                                        process to do?'
  • Entity: Subgroup Statistics (X-bar, R)   • Entity: Individual Parts (X)

The Voice of the Process (VOP)

Control limits express the Voice of the Process (VOP). They describe the empirical, statistical reality of what the machine and tooling are physically capable of delivering right now under standard operating conditions. The machine does not know, nor does it care, what the customer's engineering drawing states. Control limits tell the technician: 'This is the natural, expected boundary of variation when only common causes are present.'

The Voice of the Customer (VOC)

Specification limits express the Voice of the Customer (VOC). They represent engineering design requirements, functional clearances, mating tolerances, and contractual commitments necessary for the final product to perform safely and reliably in the field. The customer does not know, nor do they care, how old the lathe is, what brand of cutting fluid is used, or what the machine's natural standard deviation is. Specification limits tell the technician: 'This is the tolerance envelope that individual parts must satisfy to be fit for use.'

---

2. Mathematical Definition and Calculation of Control Limits

Control limits are computed strictly from empirical inspection data collected during a period of stable production. They are established at exactly three standard errors ($3\sigma$) on either side of the process centerline.

For an $\bar{X}$ chart monitoring subgroup averages of sample size $n$: $\mu_{\bar{X}} = \mu$ $\sigma_{\bar{X}} = \frac{\sigma}{\sqrt{n}}$ $UCL_{\bar{X}} = \mu + 3\sigma_{\bar{X}} = \mu + \frac{3\sigma}{\sqrt{n}}$ $LCL_{\bar{X}} = \mu - 3\sigma_{\bar{X}} = \mu - \frac{3\sigma}{\sqrt{n}}$

Why 3-Sigma Limits? (Shewhart's Economic Rationale)

Why did Walter Shewhart select $3\sigma$ rather than $2\sigma$ or $4\sigma$?

• If control limits were set at $\pm 2\sigma$, the probability of a point falling outside the limits due to chance alone would be approximately 4.55% (an Alpha risk of $\alpha = 0.0455$). On a shop floor pulling 20 subgroups per shift, operators would spend hours chasing false alarms and tampering with stable machines almost every single shift.
• If control limits were set at $\pm 4\sigma$, the Alpha risk drops to 0.006%, but the limits become so wide that the chart becomes sluggish and insensitive. Real process shifts of $1.0\sigma$ to $2.0\sigma$ would go completely undetected for dozens of subgroups (high Beta risk $\beta$), allowing thousands of nonconforming parts to be produced.
• At $\pm 3\sigma$, the probability of a false alarm under a normal distribution is exactly 0.27% ($p = 0.0027$, or roughly 1 in 370 subgroups). Over a century of industrial experience has proven that $3\sigma$ provides the optimal economic compromise between the cost of false alarms and the cost of missed detections.

---

3. Definition of Specification Limits ($USL, LSL$)

Specification limits are fixed engineering boundaries established by product designers, standards organizations (e.g., ASME, ISO, SAE), or customer contracts. They consist of:

• Upper Specification Limit (USL): The maximum permissible dimensional, electrical, or physical value for an acceptable part.
• Lower Specification Limit (LSL): The minimum permissible dimensional, electrical, or physical value for an acceptable part.

Natural Tolerance Limits (NTL)

In contrast to specification limits, the physical spread of individual parts produced by a stable process is defined by its Natural Tolerance Limits (NTL), also called the process capability spread: $UNTL = \mu + 3\sigma$ $LNTL = \mu - 3\sigma$ $\text{Total Natural Spread} = 6\sigma$

While natural tolerance limits describe the spread of individual pieces ($X$), control limits describe the spread of subgroup averages ($\bar{X}$). Because subgroup averages vary far less than individual pieces, control limits are much narrower than natural tolerance limits by a factor of $\sqrt{n}$: $\text{Control Limit Width} = \frac{6\sigma}{\sqrt{n}}$

---

4. The Cardinal Rule of SPC: Why Spec Limits Must NEVER Appear on an $\bar{X}$ Chart

In quality assurance, one rule is absolute: Specification limits must NEVER be drawn on an $\bar{X}$ control chart.

Drawing $USL$ and $LSL$ on an $\bar{X}$ chart is an egregious statistical error for three fundamental reasons:

A. Comparing Averages to Individual Limits (Apples to Oranges)

Specification limits apply strictly to individual measurements ($X_i$). An $\bar{X}$ control chart plots subgroup averages ($\bar{X}$). Because the standard error of the mean is smaller than individual piece variation by $\sqrt{n}$ (for $n=4$, $\sigma_{\bar{X}} = 0.5\sigma$; for $n=9$, $\sigma_{\bar{X}} = 0.33\sigma$), placing individual blueprint tolerances on a chart of averages creates an invalid, mathematically distorted comparison.

B. The Illusion of False Security (Hiding Scrap in the Averages)

When specification limits are plotted on an $\bar{X}$ chart, operators falsely assume that as long as the plotted point is within the spec lines, all parts produced are conforming. This is dangerously false!

Consider this real-world machining example:

• Engineering Specification: $10.000 \pm 0.050\text{ mm}$ ($LSL = 9.950\text{ mm}, USL = 10.050\text{ mm}$).
• Subgroup Size: $n = 4$.
• A technician measures 4 consecutive parts: $[9.920, 10.080, 10.010, 9.990]$.
• The subgroup average is: $\bar{X} = \frac{9.920 + 10.080 + 10.010 + 9.990}{4} = 10.000\text{ mm}$

If $USL$ and $LSL$ were drawn on the chart, the operator would see a plotted point at exactly $10.000\text{ mm}$—dead center on nominal! The operator would smile, believing quality is perfect. Yet in reality, 50% of the parts in that subgroup are scrap! Part 1 ($9.920\text{ mm}$) is below $LSL$, and Part 2 ($10.080\text{ mm}$) is above $USL$. The averaging process mathematically masked the severe process dispersion.

   Individual Parts (X):      [9.920]                      [10.080]
                              (SCRAP)                      (SCRAP)
                                 |                            |
   Spec Limits:         LSL = 9.950                      USL = 10.050
                                 |                            |
   Subgroup Mean (X-bar):        +----------- 10.000 ---------+
                                          (LOOKS PERFECT!)

C. Inducing False Alarms

If an inexperienced technician mistakenly compresses the specification limits by dividing by $\sqrt{n}$ to draw 'modified spec limits' on an $\bar{X}$ chart, operators will treat points crossing the modified spec as part rejections, shutting down machines and scrapping conforming lots.

[!WARNING] ASQ Exam Rule: Where Do Spec Limits Belong? Specification limits belong exclusively on histograms, run charts of individual readings, or gage inspection check sheets. They NEVER belong on Shewhart variables control charts ($\bar{X}$, $R$, or $s$).

---

5. Donald J. Wheeler's Four Process States

Statistical process control and engineering capability are completely independent dimensions. A process can be in statistical control or out of control; simultaneously, it can be capable of meeting specifications or incapable. Dr. Donald J. Wheeler formalized this relationship into The Four Process States:

                                  PROCESS STABILITY
                             (Control Limits / VOP)

                        IN STATISTICAL CONTROL      OUT OF CONTROL
                     +--------------------------+--------------------------+
          CAPABLE    |         STATE 1:         |         STATE 3:         |
                     |    IN CONTROL & CAPABLE  |   OUT OF CONTROL BUT     |
                     |     (The Ideal State)    |         CAPABLE          |
  PROCESS            |                          |   (The Brink of Chaos)   |
  CAPABILITY         +--------------------------+--------------------------+
  (Spec Limits       |         STATE 2:         |         STATE 4:         |
   / VOC)  INCAPABLE |   IN CONTROL BUT         |   OUT OF CONTROL &       |
                     |         INCAPABLE        |        INCAPABLE         |
                     |   (The Threshold State)  |    (The State of Chaos)  |
                     +--------------------------+--------------------------+

State 1: In Control and Capable (The Ideal State)

• Operational Condition: The process is statistically stable (no special causes, all points within control limits) and highly capable ($C_p \ge 1.33, C_{pk} \ge 1.33$). The natural tolerance spread ($6\sigma$) fits comfortably inside the specification envelope ($USL - LSL$) with room to spare.
• Defect Rate: Virtually zero nonconformances.
• Technician Action: Maintain standard SPC monitoring. Practice continuous improvement (Kaizen) to reduce common cause variation further.

State 2: In Control but Incapable (The Threshold State / Predictably Defective)

• Operational Condition: The process is in statistical control (stable and predictable), but the natural process spread exceeds the specification width ($6\sigma > USL - LSL$), or the process is significantly off-center ($C_{pk} < 1.0$).
• Defect Rate: Generates scrap or rework at a constant, predictable, unrelenting rate (e.g., exactly 3.2% nonconforming every shift).
• Technician Action: Do not blame operators! Because the process is in statistical control, all defects stem from common causes built into the system. Line operators cannot eliminate this scrap. Management and engineering must intervene to overhaul tooling, purchase tighter-tolerance stock, or invest in higher-precision machinery. In the interim, 100% sorting inspection must be instituted to protect the customer.

State 3: Out of Control but Capable (The Brink of Chaos)

• Operational Condition: The process exhibits special cause instability (points outside control limits, runs, or trends), but customer tolerances are so wide ($USL - LSL \gg 6\sigma$) that all parts currently produced happen to meet blueprint specifications.
• Defect Rate: Low or zero defects currently escaping, but catastrophe is imminent.
• Technician Action: Eliminate complacency! Supervisors often argue: 'Why investigate an out-of-control point when all parts passed the go/no-go gage?' The technician must explain that the process is in the 'Brink of Chaos.' A special cause is actively perturbing the system. Without immediate root cause analysis and elimination of the assignable cause, the next process shift could push the distribution across the specification limits, generating massive scrap without warning.

State 4: Out of Control and Incapable (The State of Chaos)

• Operational Condition: The process is mathematically unstable (special causes present) and incapable of meeting specifications ($C_{pk} \ll 1.0$). Points are flying outside control limits, and individual parts are failing blueprint tolerances.
• Defect Rate: High, erratic, and completely unpredictable defect spikes.
• Technician Action: Emergency containment! Stop the production line immediately. Red-tag and quarantine all parts manufactured since the last confirmed in-control subgroup. Initiate immediate root-cause troubleshooting to eliminate special causes and restore basic stability before addressing capability.

Comparison Reference: The Four Process States

---

6. Consequences of Confusing Process Control with Product Acceptance

Confusing control limits with specification limits leads to two catastrophic shop-floor failure modes:

Failure Mode A: Releasing an Unstable Process Because Parts Pass Inspection

A machining line exhibits a 7-point upward trend on the $\bar{X}$ chart, but the quality supervisor refuses to stop the machine because dial indicator readings on individual pieces are still inside blueprint tolerance. Within three hours, the thermal expansion of the spindle pushes the process mean completely over the Upper Specification Limit, resulting in 450 scrapped titanium shafts worth $90,000.

Failure Mode B: Rejecting Material Solely Based on Control Limit Violations

An incoming receiving inspection technician plots hardness readings from a raw material heat lot on a supplier tracking control chart. One subgroup average falls slightly below the Lower Control Limit. The technician immediately issues a Supplier Corrective Action Request (SCAR) and rejects the entire $50,000 shipment, even though every individual bar in the heat lot tests well within the ASTM material specification. The technician committed a fundamental error: Control limits govern process monitoring; they are never valid criteria for material disposition or product rejection. Product acceptance is governed exclusively by specification limits.