11.4 Process Capability & Performance Analysis

11.4 Process Capability & Performance Analysis

Specification Limits vs. Control Limits: VOC vs. VOP

In the entire discipline of quality engineering and inspection metrology, no conceptual distinction is more critical—or more frequently tested on ASQ certification examinations—than the difference between Specification Limits and Control Limits.

• Specification Limits (The Voice of the Customer - VOC):
  • Defined by product design engineers, customer contracts, or industry standards (such as ASME Y14.5).
  • Denoted as $USL$ (Upper Specification Limit) and $LSL$ (Lower Specification Limit).
  • Physical Application: Specification limits apply strictly to individual physical parts ($X$). They define the boundary between functionally acceptable product and nonconforming scrap or rework.
  • Authority: Quality inspectors and manufacturing personnel cannot alter specification limits. Modifying a specification requires a formal Engineering Change Notice (ECN) approved by design engineering.
• Control Limits (The Voice of the Process - VOP):
  • Calculated mathematically from sample inspection data collected over time.
  • Denoted as $UCL$ (Upper Control Limit) and $LCL$ (Lower Control Limit).
  • Statistical Application: Control limits apply strictly to sample statistics (such as subgroup averages $\bar{X}$, ranges $R$, standard deviations $s$, or attribute fractions $p$).
  • Purpose: Control limits define the boundaries of natural, common cause variation. They indicate whether the manufacturing process is behaving consistently over time.

+-----------------------------------------------------------------------------+
|                   SPECIFICATION LIMITS VS. CONTROL LIMITS                   |
+-----------------------------------------------------------------------------+
|  USL = 1.005 in  ==================================== [Voice of Customer]  |
|                                                                             |
|  UCL = 1.003 in  - - - - - - - - - - - - - - - - - -  [Voice of Process]   |
|                          *   *       *                                      |
|  X̄̄  = 1.000 in  ---------*-------*-------*-------    [Process Center Line] |
|                                  *                                          |
|  LCL = 0.997 in  - - - - - - - - - - - - - - - - - -  [Voice of Process]   |
|                                                                             |
|  LSL = 0.995 in  ==================================== [Voice of Customer]  |
+-----------------------------------------------------------------------------+

The Deadly Fallacy: Plotting Spec Limits on $\bar{X}$ Charts

A catastrophic error committed by untrained personnel is plotting engineering specification limits directly onto an $\bar{X}$ control chart. Subgroup averages have a standard deviation that is dramatically smaller than individual piece dimensions ($\sigma_{\bar{X}} = \sigma / \sqrt{n}$). For a subgroup of $n = 5$, the spread of averages is $\sqrt{5} \approx 2.24$ times narrower than the spread of individual parts! Plotting spec limits on an $\bar{X}$ chart gives operators a completely false sense of security: subgroup averages may fall comfortably inside the spec limits while individual parts are being scrapped.

---

The Mandatory Prerequisite: Statistical Control

A universal axiom of statistical quality control states:

Process capability analysis can ONLY be legitimately performed on a process that is in a state of statistical control.

• The Mathematical Reality: Process capability indices ($C_p, C_{pk}$) are predictive metrics. They estimate what percentage of future production will satisfy customer specifications. However, if a control chart reveals out-of-control conditions (points beyond limits, runs, trends, or special causes), the process mean and variance are unstable and drifting unpredictably.
• The Rule: Calculating $C_p$ or $C_{pk}$ on an unstable process yields a meaningless statistical snapshot. Before conducting capability studies, inspectors must first eliminate all special causes and verify statistical control using variable control charts (such as $\bar{X}$ and $R$).

---

Short-Term Potential Capability: $C_p$

The Process Capability Index ($C_p$) measures the potential capability of a process. It compares the total allowable tolerance band defined on the blueprint to the inherent process spread ($6\hat{\sigma}$), under the theoretical assumption that the process is centered perfectly between the specification limits.

Mathematical Formula for $C_p$

$C_p = \frac{\text{USL} - \text{LSL}}{6\hat{\sigma}}$

Where $\hat{\sigma}$ represents the within-subgroup standard deviation estimated from an in-control range or standard deviation chart: $\hat{\sigma} = \frac{\bar{R}}{d_2} \qquad \text{or} \qquad \hat{\sigma} = \frac{\bar{s}}{c_4}$

Interpretation and Blind Spots of $C_p$

• If $C_p > 1.0$: The allowable tolerance spread is wider than the natural 6-sigma process spread. The process has the potential to produce 100% conforming parts.
• If $C_p = 1.0$: The allowable tolerance width exactly equals the 6-sigma process spread ($100%$ of tolerance consumed).
• If $C_p < 1.0$: The process spread is wider than the tolerance band; even if perfectly centered, the process will generate nonconforming scrap.
• The Fatal Blind Spot of $C_p$: $C_p$ is completely location-blind. The formula considers only the distance between $USL$ and $LSL$; it does not include the process mean $\bar{\bar{X}}$. A machining operation could have an outstanding $C_p = 2.50$, but if the tool offset is shifted so that the mean is located above the $USL$, 100% of the parts produced will be scrap! Therefore, $C_p$ alone is never sufficient to evaluate capability.

---

Short-Term Actual Capability: $C_{pk}$

The Process Capability Index ($C_{pk}$) measures the actual performance of the process by taking into account both the process spread ($3\hat{\sigma}$) and the physical location of the process mean relative to the specification limits.

Mathematical Formulas for $C_{pk}$

$C_{pk}$ evaluates the upper capability ($C_{pu}$) and lower capability ($C_{pl}$) independently, and defines actual capability as the worst-case (minimum) value: $C_{pu} = \frac{\text{USL} - \bar{\bar{X}}}{3\hat{\sigma}}$ $C_{pl} = \frac{\bar{\bar{X}} - \text{LSL}}{3\hat{\sigma}}$ $C_{pk} = \min(C_{pu}, \; C_{pl})$

One-Sided Specifications (Single Limit Characteristics)

When an engineering drawing specifies a one-sided tolerance, only one capability ratio can be computed:

• Upper Spec Only (e.g., maximum surface roughness $Ra$, maximum GD&T true position, runout, flatness): $C_{pk} = C_{pu} = \frac{\text{USL} - \bar{\bar{X}}}{3\hat{\sigma}}$.
• Lower Spec Only (e.g., minimum tensile strength, minimum plating thickness, minimum hardness): $C_{pk} = C_{pl} = \frac{\bar{\bar{X}} - \text{LSL}}{3\hat{\sigma}}$.

The Mathematical Relationship Between $C_p$ and $C_{pk}$

1. $C_{pk} \le C_p$ Always: Because $C_{pk}$ penalizes for any departure from nominal center, $C_{pk}$ can never exceed $C_p$.
2. $C_{pk} = C_p$ (Perfect Centering): $C_{pk}$ is mathematically equal to $C_p$ if and only if the process mean $\bar{\bar{X}}$ is centered perfectly at the nominal midpoint of the tolerance band: $\bar{\bar{X}} = (USL + LSL) / 2$.
3. $C_{pk} = 0$ (Boundary Condition): If the process mean falls exactly on either the $USL$ or the $LSL$, the numerator of the corresponding index becomes zero, yielding $C_{pk} = 0$. Under this condition, exactly 50% of the production output is nonconforming.
4. $C_{pk} < 0$ (Negative Capability): If the process mean drifts completely outside the specification limits, $C_{pk}$ becomes negative! A negative $C_{pk}$ indicates that more than 50% of production is defective.

---

Long-Term Process Performance: $P_p$ and $P_{pk}$

In automotive (AIAG PPAP), aerospace, and medical device sectors, quality standards demand a clear metrological separation between short-term potential capability ($C_p, C_{pk}$) and long-term process performance ($P_p, P_{pk}$).

The Mathematical Difference: $\hat{\sigma}$ vs. $s_{\text{overall}}$

The sole mathematical difference between capability ($C_p, C_{pk}$) and performance ($P_p, P_{pk}$) lies in how the standard deviation is calculated:

• $C_p$ and $C_{pk}$ (Short-Term): Use the within-subgroup standard deviation estimated from range or sample standard deviation: $\hat{\sigma} = \bar{R} / d_2$ or $\bar{s} / c_4$. This metric filters out between-subgroup shifts, representing what the process could achieve under ideal, short-term conditions.
• $P_p$ and $P_{pk}$ (Long-Term): Use the sample overall standard deviation ($s_{\text{overall}}$) calculated by pooling every individual measurement taken across all batches, shifts, and setups: $s_{\text{overall}} = \sqrt{\frac{\sum_{i=1}^{N} (X_i - \bar{X})^2}{N - 1}}$ $P_p = \frac{\text{USL} - \text{LSL}}{6s_{\text{overall}}}, \qquad P_{pk} = \min\left(\frac{\text{USL} - \bar{X}}{3s_{\text{overall}}}, \; \frac{\bar{X} - \text{LSL}}{3s_{\text{overall}}}\right)$

Metrological Diagnostic Value

Comparing $C_{pk}$ to $P_{pk}$ provides profound diagnostic insight for inspectors:

• If $C_{pk} \approx P_{pk}$: The process exhibits minimal shift over time; the long-term spread is identical to short-term spread.
• If $C_{pk}$ is significantly higher than $P_{pk}$ (e.g., $C_{pk} = 1.65$, but $P_{pk} = 1.05$): The machine has outstanding inherent precision, but the process is suffering from between-subgroup shifts—such as thermal drift across shifts, operator setup differences, or batch-to-batch raw material hardness variation.

---

Industry Capability Benchmarks

Over decades of manufacturing evolution, global industry consortia have standardized minimum capability benchmarks:

The Six Sigma $1.5\sigma$ Shift Rule: A Six Sigma process maintains $C_p = 2.0$ ($12\sigma$ total width). Dr. Mikel Harry and Motorola proved that over long-term production, processes naturally experience an average mean shift of up to $\pm 1.5\sigma$. Even with this $1.5\sigma$ drift, a Six Sigma process maintains $C_{pk} = 1.50$, resulting in a maximum long-term defect rate of no more than $3.4\text{ Parts Per Million (PPM)}$.

---

Comprehensive Worked Capability Calculation Table

To prepare quality inspectors for practical shop evaluations and certification exams, consider this end-to-end machining case study:

The Shop Problem

An aerospace CNC lathe turns titanium hydraulic spool valve outer diameters. The blueprint specifies: $\varnothing 0.7500 \pm 0.0015\text{ in}$.

• $\text{USL} = 0.7515\text{ in}$
• $\text{LSL} = 0.7485\text{ in}$
• $\text{Total Tolerance Band} = 0.7515 - 0.7485 = 0.0030\text{ in}$

An inspector measures 25 rational subgroups of $n = 5$ parts each from an in-control process:

• Grand Average: $\bar{\bar{X}} = 0.7506\text{ in}$
• Average Range: $\bar{R} = 0.0007\text{ in}$
• Standard factor for $n = 5$: $d_2 = 2.326$

+-----------------------------------------------------------------------------+
|                   STEP-BY-STEP CAPABILITY CALCULATION WORKED                |
+-----------------------------------------------------------------------------+
| Step 1: Estimate Inherent Within-Subgroup Standard Deviation (σ̂):           |
|   σ̂ = R̄ / d2 = 0.0007 in / 2.326 = 0.000301 in                              |
|                                                                             |
| Step 2: Calculate Short-Term Potential Capability (Cp):                     |
|   Cp = (USL - LSL) / (6σ̂)                                                   |
|   Cp = (0.7515 - 0.7485) / (6 * 0.000301) = 0.0030 / 0.001806 = 1.66       |
|                                                                             |
| Step 3: Calculate Upper Capability (Cpu):                                   |
|   Cpu = (USL - X̄̄) / (3σ̂)                                                    |
|   Cpu = (0.7515 - 0.7506) / (3 * 0.000301) = 0.0009 / 0.000903 = 0.997 ≈ 1.00|
|                                                                             |
| Step 4: Calculate Lower Capability (Cpl):                                   |
|   Cpl = (X̄̄ - LSL) / (3σ̂)                                                    |
|   Cpl = (0.7506 - 0.7485) / (3 * 0.000301) = 0.0021 / 0.000903 = 2.325 ≈ 2.33|
|                                                                             |
| Step 5: Determine Actual Capability (Cpk):                                  |
|   Cpk = min(Cpu, Cpl) = min(1.00, 2.33) = 1.00                             |
+-----------------------------------------------------------------------------+

Metrological Audit Interpretation and Inspector Action

1. Diagnostic Evaluation: The potential capability is exceptional ($C_p = 1.66$, exceeding the aerospace benchmark of $1.67$). The machine has ample precision to satisfy the blueprint.
2. The Problem: The actual capability is marginal ($C_{pk} = 1.00$). The process mean is shifted high towards the $USL$ ($0.7506\text{ in}$ vs. nominal $0.7500\text{ in}$).
3. Inspector Remediation: The inspector instructs the machinist to adjust the lathe X-axis tool wear offset by $-0.0006\text{ in}$. This simple offset change shifts the process mean back to nominal ($0.7500\text{ in}$), immediately elevating $C_{pk}$ from $1.00$ to $1.66$ without spending a single dollar on tooling or machine maintenance.

---

Real Shop Inspection Scenarios & Common Exam Traps

• Real Shop Scenario — The Rejection of a "1.80 Cp" Process Submission: A tier-1 automotive supplier submits a PPAP package showing $C_p = 1.82$ for a critical steering knuckle bore diameter. However, the customer quality engineer rejects the submission. The supplier's technician is furious, citing that $1.82$ exceeds the $1.67$ benchmark. Metrological Audit: The quality engineer points to $C_{pk}$, which is $0.91$. The boring tool was set off-center, causing parts to graze the lower tolerance limit. Furthermore, the overall performance index $P_{pk}$ was only $0.78$ due to temperature shifts between day and night shifts. The supplier proved the machine was potentially capable, but failed to demonstrate actual capability and long-term control.

• Exam Trap: Confusing $\hat{\sigma} = \bar{R}/d_2$ with Sample Standard Deviation $s$: Exam questions often give both $\bar{R}$, $d_2$, and the sample standard deviation $s$. To calculate $C_p$ or $C_{pk}$, you must use $\hat{\sigma} = \bar{R}/d_2$. If you use $s$, you have mistakenly calculated $P_p$ or $P_{pk}$.

• Exam Trap: Mathematical Relationship Claiming $C_{pk} > C_p$: Never select an answer choice stating $C_{pk}$ is greater than $C_p$. $C_{pk}$ is always less than or equal to $C_p$ ($C_{pk} \le C_p$). Any calculation where $C_{pk} > C_p$ represents an arithmetic error.

• Exam Trap: The Negative $C_{pk}$ Sign: A classic exam question asks: "What condition results in a negative Cpk value?" Candidates often guess that variance was zero or sample size was too small. The correct answer: The process mean falls outside the specification limits (either $\bar{X} > USL$ or $\bar{X} < LSL$).