12.4 Process Capability & Performance Indices: Cp, Cpk, Pp, Ppk

12.4 Process Capability & Performance Indices: Cp, Cpk, Pp, Ppk

Quick Summary: Process capability analysis quantifies the relationship between the Voice of the Customer (VOC) (specification limits) and the Voice of the Process (VOP) (natural statistical spread). While $C_p$ measures process potential (assuming perfect centering), $C_{pk}$ measures actual capability by penalizing for mean shifts. Long-term performance is captured by $P_p$ and $P_{pk}$ using total overall variation. In the Control Phase, sustaining high capability requires a living Control Plan, an Out-of-Control Action Plan (OCAP), mistake-proofing, and formal project handover to the process owner.

---

Mandatory Prerequisites for Capability Analysis

Before calculating capability metrics ($C_p, C_{pk}$), two strict statistical conditions must be satisfied:

                  Process Capability Prerequisites
                                 │
         ┌───────────────────────┴───────────────────────┐
         ▼                                               ▼
  1. Statistical Control                          2. Normality of Data
  • No special causes on control charts          • Data follows normal distribution
  • Predictable mean and dispersion               • Skewed data invalidates 6σ spread
  • Unstable process = Meaningless Cpk           • If non-normal: Transform (Box-Cox) or
                                                   use non-parametric Weibull methods

1. The Process Must Be in Statistical Control: This is the non-negotiable golden rule. If special causes are present, the process distribution is shifting erratically over time. Any capability index computed on an unstable process represents merely a fleeting historical snapshot with zero predictive validity for future output.
2. The Output Must Be Normally Distributed: The standard capability formulas assume that $6\sigma$ represents $99.73%$ of production. If data is heavily skewed or multimodal, standard $C_p / C_{pk}$ equations produce wildly inaccurate defect forecasts. Non-normal data must be normalized via Box-Cox or Johnson transformations, or analyzed using non-parametric percentile methods.

---

Process Potential ($C_p$) vs. Actual Capability ($C_{pk}$)

1. Process Potential ($C_p$)

$C_p$ evaluates the potential capability of the process, answering the question: "Could the process fit comfortably within customer tolerances if it were perfectly centered?"

$C_p = \frac{\text{Voice of the Customer (Tolerance)}}{\text{Voice of the Process (Natural Spread)}} = \frac{\text{USL} - \text{LSL}}{6\sigma_{\text{st}}}$

• Formula Components: $\text{USL} - \text{LSL}$ is the total customer allowable tolerance width. $6\sigma_{\text{st}}$ is the natural process spread, where $\sigma_{\text{st}}$ is the short-term within-subgroup standard deviation (estimated via $\bar{R} / d_2$ or $\bar{s} / c_4$).
• Limitation: $C_p$ completely ignores the location of the process mean! A process with a $C_p = 2.5$ could be operating entirely outside the Upper Specification Limit, producing $100%$ defective scrap, yet its $C_p$ would remain high because its spread is narrow.

2. Actual Process Capability ($C_{pk}$)

$C_{pk}$ measures the actual capability of the process by accounting for both process spread and process centering relative to specifications. It evaluates the upper and lower capability halves independently and selects the worst-case scenario:

$\text{CPU} = \frac{\text{USL} - \mu}{3\sigma_{\text{st}}} \qquad \text{CPL} = \frac{\mu - \text{LSL}}{3\sigma_{\text{st}}}$ $C_{pk} = \min(\text{CPU}, \text{CPL}) = \min\left(\frac{\text{USL} - \mu}{3\sigma_{\text{st}}}, \frac{\mu - \text{LSL}}{3\sigma_{\text{st}}}\right)$

Capability Visualization: Centered vs. Shifted

     LSL                        Nominal                        USL
      │                            │                            │
Case 1: Perfectly Centered         │
      │                 ┌──────────┴──────────┐                 │
      │               ┌─┘                     └─┐               │
      │             ┌─┘                         └─┐             │
      │─────────────┴─────────────────────────────┴─────────────│
                     ◄─────── 6σ Spread ──────►
                     [Cp = 1.67, Cpk = 1.67 (Cp = Cpk)]

Case 2: Shifted Toward USL         │
      │                                   ┌──────────┴──────────┐
      │                                 ┌─┘                     └─┐
      │                               ┌─┘                         └─┐ Defect Tail!
      │───────────────────────────────┴─────────────────────────────┴───►
                                       ◄─────── 6σ Spread ──────►
                     [Cp = 1.67, Cpk = 0.83 (Cp > Cpk)]

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

1. $C_p \ge C_{pk}$ always: For bilateral (two-sided) specifications, $C_{pk}$ can never exceed $C_p$.
2. $C_p = C_{pk}$: Occurs if and only if the process mean is perfectly centered midway between the specification limits ($\mu = \frac{\text{USL} + \text{LSL}}{2}$).
3. $C_p > C_{pk}$: Indicates that the process is off-center. The magnitude of difference ($C_p - C_{pk}$) directly quantifies the capability lost due to decentering:
   $k = \frac{|\text{Target} - \mu|}{\frac{\text{USL} - \text{LSL}}{2}}, \qquad C_{pk} = C_p(1 - k)$
4. $C_{pk} \le 0$: Indicates that the process mean has drifted outside one of the specification limits, meaning more than $50%$ of production is defective.

---

Capability Benchmarks and Sigma Levels

In industrial and transactional Six Sigma deployment, standardized benchmark thresholds define whether a process is acceptable for commercial release:

---

Short-Term ($C_p, C_{pk}$) vs. Long-Term ($P_p, P_{pk}$)

A common point of confusion on certification exams is the operational distinction between Process Capability ($C_p, C_{pk}$) and Process Performance ($P_p, P_{pk}$).

Diagnostic Insight: If $C_{pk} \approx P_{pk}$, the process is stable and free from long-term drift. If $C_{pk} \gg P_{pk}$ (e.g., $C_{pk} = 1.60$ but $P_{pk} = 1.05$), the process has high inherent capability but is suffering from substantial unmanaged between-subgroup shifts (such as thermal cycling, shifting supplier batches, or poor machine maintenance).

---

Step-by-Step Worked Calculation Example: Complete Evaluation of $C_p, C_{pk}, P_p, P_{pk}$

Scenario: A precision CNC machining center turns stainless steel pins with customer diameter specifications of $10.00 \pm 0.15 \text{ mm}$ (Nominal $= 10.00 \text{ mm}$, $\text{USL} = 10.15 \text{ mm}$, $\text{LSL} = 9.85 \text{ mm}$).

A Green Belt collects $k = 25$ rational subgroups of size $n = 5$ over three shifts. Statistical control charts confirm the process is stable. The baseline data yields:

• Process grand mean: $\bar{\bar{X}} = 10.03 \text{ mm}$
• Average subgroup range: $\bar{R} = 0.0465 \text{ mm}$
• Factor for $n = 5$: $d_2 = 2.326$

Over an extended 3-month production run, all $N = 125$ individual parts are pooled, yielding:

• Overall sample mean: $\bar{X} = 10.035\text{ mm}$
• Overall sample standard deviation: $s_{\text{overall}} = 0.035\text{ mm}$

Step 1: Estimate Within-Subgroup Standard Deviation ($\sigma_{\text{st}}$)

$\hat{\sigma}_{\text{st}} = \frac{\bar{R}}{d_2} = \frac{0.0465}{2.326} = 0.020 \text{ mm}$

Step 2: Compute Process Potential ($C_p$)

$C_p = \frac{\text{USL} - \text{LSL}}{6\hat{\sigma}_{\text{st}}} = \frac{10.15 - 9.85}{6(0.020)} = \frac{0.30}{0.120} = 2.50$ Interpretation: With a $C_p$ of $2.50$, the process spread ($0.12 \text{ mm}$) occupies only $40%$ of the customer tolerance band ($0.30 \text{ mm}$). The process has world-class potential.

Step 3: Compute Upper and Lower Capability Indices ($CPU$ and $CPL$)

$\text{CPU} = \frac{\text{USL} - \bar{\bar{X}}}{3\hat{\sigma}_{\text{st}}} = \frac{10.15 - 10.03}{3(0.020)} = \frac{0.12}{0.060} = 2.00$ $\text{CPL} = \frac{\bar{\bar{X}} - \text{LSL}}{3\hat{\sigma}_{\text{st}}} = \frac{10.03 - 9.85}{3(0.020)} = \frac{0.18}{0.060} = 3.00$

Step 4: Determine Actual Capability ($C_{pk}$)

$C_{pk} = \min(\text{CPU}, \text{CPL}) = \min(2.00, 3.00) = 2.00$ Analysis: Because the process mean is shifted slightly toward the upper specification limit ($10.03 > 10.00$), $C_{pk}$ ($2.00$) is lower than $C_p$ ($2.50$). However, because $C_{pk} = 2.00$, the process comfortably achieves World-Class Six Sigma capability.

Step 5: Compute Long-Term Process Performance ($P_p$ and $P_{pk}$)

Using the overall standard deviation $s_{\text{overall}} = 0.035\text{ mm}$: $P_p = \frac{\text{USL} - \text{LSL}}{6 s_{\text{overall}}} = \frac{10.15 - 9.85}{6(0.035)} = \frac{0.30}{0.210} \approx 1.43$ $\text{PPU} = \frac{\text{USL} - \bar{X}}{3 s_{\text{overall}}} = \frac{10.15 - 10.035}{3(0.035)} = \frac{0.115}{0.105} \approx 1.10$ $\text{PPL} = \frac{\bar{X} - \text{LSL}}{3 s_{\text{overall}}} = \frac{10.035 - 9.85}{3(0.035)} = \frac{0.185}{0.105} \approx 1.76$ $P_{pk} = \min(\text{PPU}, \text{PPL}) = \min(1.10, 1.76) = 1.10$

Step 6: Comparative Diagnosis ($C_{pk}$ vs. $P_{pk}$)

Comparing short-term capability to long-term performance reveals a massive gap: $C_{pk} = 2.00 \quad \text{vs.} \quad P_{pk} = 1.10$ This substantial difference ($2.00 - 1.10 = 0.90$) indicates that while the CNC lathe is inherently capable of Six Sigma performance over short batches, it suffers from severe unmanaged between-subgroup variation over months—such as spindle thermal expansion, progressive cutting tool wear, and raw material bar stock hardness variations. The Green Belt must introduce automated tool wear compensation and stricter environmental temperature controls to bridge this gap.

---

The Control Plan: Sustaining Process Improvements

The ultimate goal of the DMAIC methodology is not simply to achieve improvement, but to sustain it indefinitely. The Control Plan is the formal operational document that binds process requirements, measurement methods, sampling frequencies, and reaction mechanisms into a standardized operational system.

Core Architecture of a Six Sigma Control Plan

A comprehensive Control Plan contains eight mandatory columns:

The Out-of-Control Action Plan (OCAP)

The Out-of-Control Action Plan (OCAP) is the emergency response protocol embedded directly within the Control Plan. An OCAP eliminates guesswork on the shop floor by prescribing a step-by-step diagnostic tree:

1. Contain: Immediately stop the machine, flag the station, and quarantine all parts produced since the last verified in-control check.
2. Diagnose: Execute the standardized diagnostic flowchart (check tool wear, inspect raw material lot code, verify gauge calibration).
3. Correct: Implement the verified corrective action (replace worn insert, adjust coolant flow).
4. Verify & Document: Produce and measure test pieces to confirm that the process has returned to statistical control before releasing the station for full production, and record the event in the shift log.

Project Closure, Auditing, and Ownership Handover

The DMAIC project concludes with a formal transition of authority from the Six Sigma Project Team to the operational Process Owner:

• Standard Operating Procedures (SOPs): Updated with improved operating parameters and visual work instructions.
• Process Owner Sign-off: The functional manager signs the project charter, formally accepting operational responsibility for executing the Control Plan.
• Post-Implementation Audits: The Green Belt and financial controller schedule mandatory audits at 30, 60, and 90 days post-launch to verify that financial savings are real, process capability ($C_{pk}$) remains sustained, and control charts are actively maintained.
• Documentation Archiving: All project documentation, baseline data, hypothesis tests, and final reports are archived in the organization's knowledge management repository for organizational learning.

---

Critical CSSC Exam Traps

• Trap 1: Calculating $C_{pk}$ on an Unstable Process — An exam question will provide subgroup data exhibiting a strong out-of-control trend and ask you to compute $C_{pk}$. The correct answer is that capability cannot be calculated because the process lacks statistical stability.
• Trap 2: Assuming $C_p$ Can Be Smaller than $C_{pk}$ — Mathematically, $C_{pk} = C_p(1 - k)$. Since $k \ge 0$, $C_p$ is always greater than or equal to $C_{pk}$ for two-sided tolerances. If an exam option states $C_p = 1.10$ and $C_{pk} = 1.45$, it is mathematically impossible.
• Trap 3: Confusing Within-Subgroup $\sigma$ with Overall $s$ — When calculating $C_p$ and $C_{pk}$, you must divide by $\sigma_{\text{st}} = \bar{R} / d_2$. If you divide by the overall sample standard deviation $s$, you have mistakenly calculated $P_p$ and $P_{pk}$.
• Trap 4: Conflating $C_{pk} = 1.00$ with Six Sigma Quality — A $C_{pk}$ of $1.00$ corresponds to a $3\sigma$ process (2,700 defects per million opportunities). True Six Sigma quality requires a $C_{pk} \ge 2.00$ (or $C_{pk} \ge 1.50$ accounting for long-term drift).