7.2 Potential Capability (Cp) and Actual Capability (Cpk)

7.2 Potential Capability (Cp) and Actual Capability (Cpk)

The Dual Dimensions of Process Capability

When assessing a manufacturing process, quality technicians must evaluate two entirely independent characteristics:

1. Process Spread (Dispersion): How wide is the distribution of individual part dimensions? Does the equipment possess the inherent precision to fit inside the blueprint tolerance band?
2. Process Centering (Location): Where is the process distribution positioned relative to the target nominal or specification midpoint? Is the process running centered, or has it drifted toward the upper or lower specification limit?

To separate and evaluate these two dimensions, the quality profession relies on two foundational indices: $C_p$ (Process Potential) and $C_{pk}$ (Actual Process Capability).

---

Potential Capability Index ($C_p$): Measuring Process Spread

The potential capability index ($C_p$) measures the maximum theoretical capability the process could achieve if its distribution were centered perfectly between the engineering specification limits. It represents the ratio of the allowable tolerance width to the inherent $6\sigma$ natural process spread.

Mathematical Formula for $C_p$

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

Where:

• $USL$: Upper Specification Limit
• $LSL$: Lower Specification Limit
• $USL - LSL$: Total allowable engineering tolerance width (Voice of Customer)
• $6\hat{\sigma}_{within}$: Total inherent natural process spread (Voice of Process)
• $\hat{\sigma}_{within}$: Within-subgroup standard deviation, representing short-term common-cause variation

Estimating Within-Subgroup Standard Deviation ($\hat{\sigma}_{within}$)

In standard capability studies, $\sigma$ is not calculated using the sample standard deviation formula across all pooled parts. Instead, it is estimated directly from the rational subgroup statistics of a stable control chart:

• From an $\bar{X}$ and $R$ Chart: $$\hat{\sigma}_{within} = \frac{\bar{R}}{d_2}$$ (where $\bar{R}$ is the average subgroup range and $d_2$ is the Shewhart bias correction factor for subgroup size $n$).
• From an $\bar{X}$ and $s$ Chart: $$\hat{\sigma}_{within} = \frac{\bar{s}}{c_4}$$ (where $\bar{s}$ is the average subgroup standard deviation and $c_4$ is the bias correction factor for subgroup size $n$).

Standard Unbiasing Constants Table ($d_2$ and $c_4$)

Why $C_p$ is Blind to Process Centering

Notice that the formula for $C_p$ contains only $USL$, $LSL$, and $\hat{\sigma}_{within}$. The process average ($\mu$ or $\bar{\bar{X}}$) does not appear in the formula!

Consequently, a process can have a stellar $C_p$ of 2.50, but if the machine tool is set incorrectly and running 5 millimeters above the upper specification limit, 100% of production will be scrap! $C_p$ tells you only whether the process spread could fit inside the specifications, never whether it is actually inside them.

Industrial Capability Benchmarks for $C_p$

---

Actual Capability Index ($C_{pk}$): Measuring Centering and Spread

To assess how well the process is actually performing relative to specifications, technicians calculate the actual capability index ($C_{pk}$). The subscript "k" originates from the Japanese word katayori, meaning "bias" or "offset."

$C_{pk}$ calculates capability to the upper specification limit ($C_{pu}$) and lower specification limit ($C_{pl}$) independently, and selects the minimum of the two.

Mathematical Formulas for $C_{pk}$

$$C_{pu} = \frac{USL - \mu}{3\hat{\sigma}_{within}}$$ $$C_{pl} = \frac{\mu - LSL}{3\hat{\sigma}_{within}}$$ $$C_{pk} = \min(C_{pu}, C_{pl})$$

Where $\mu$ is the estimated process mean (the grand average $\bar{\bar{X}}$ from the control chart).

Why $C_{pk}$ Takes the Minimum

The divisor in $C_{pu}$ and $C_{pl}$ is $3\sigma$ (representing the distance from the mean to one natural process limit), rather than $6\sigma$.

By selecting $\min(C_{pu}, C_{pl})$, $C_{pk}$ reflects the performance at the worst-case specification boundary—the specification limit that is closest to the process average. Defective parts will always emerge first at the specification limit closest to the mean!

                  LSL                   Mean (μ)          USL
                   |                      |                |
                   |                      |                |
         Process   |          * * * *     |                |
       Distribution:      * * * * * * * * |                |
                   |    * * * * * * * * * * * *            |
                   +----------------------+----------------+
                   |<---- Cpl (3.0σ) ---->|<-- Cpu (1.0σ)->|
                   
                   Worst-case boundary is USL: Cpk = min(Cpl, Cpu) = Cpu = 1.0

Interpreting Extreme $C_{pk}$ Values

• $C_{pk} = C_p$: The process mean is located exactly at the specification midpoint. Maximum possible capability is achieved.
• $0 < C_{pk} < 1.00$: The process is producing nonconforming product because the natural limit exceeds the nearest specification limit.
• $C_{pk} = 0$: The process mean is located directly on one of the specification limits ($\mu = USL$ or $\mu = LSL$). Exactly 50% of production is nonconforming.
• $C_{pk} < 0$ (Negative $C_{pk}$): The process mean has drifted completely outside the engineering specification limits (e.g., $\mu > USL$). More than 50% of production is nonconforming scrap.

---

The Mathematical Relationship: $C_p$, $C_{pk}$, and the $k$-Factor

The mathematical bridge connecting $C_p$ and $C_{pk}$ is governed by the process centering factor ($k$).

Defining the Centering Factor ($k$)

Let $m$ represent the engineering specification midpoint, and $d$ represent the half-specification width:

$$m = \frac{USL + LSL}{2}$$ $$d = \frac{USL - LSL}{2}$$

The centering factor $k$ measures the ratio of the mean's offset from midpoint to the available half-tolerance:

$$k = \frac{|\mu - m|}{d} = \frac{|\mu - m|}{\frac{USL - LSL}{2}}$$

Properties of $k$

• $k = 0$: The process is perfectly centered on target ($\mu = m$).
• $0 < k < 1$: The process mean is inside the specification band but off-center.
• $k = 1$: The process mean sits exactly on a specification limit ($|\mu - m| = d$).
• $k > 1$: The process mean lies outside the specification limits.

The Fundamental Identity

The relationship between potential capability, actual capability, and centering is:

$$C_{pk} = C_p(1 - k)$$

From this equation, two immutable mathematical truths emerge:

1. $C_{pk} \le C_p$ always: Actual capability can never exceed potential capability ($C_{pk}$ cannot be greater than $C_p$).
2. $C_{pk} = C_p$ if and only if $k = 0$: The only way $C_{pk}$ can equal $C_p$ is for the process to be centered with zero offset.

---

Diagnostic Matrix: Comparing $C_p$ and $C_{pk}$ to Guide Action

A quality technician uses the comparison between $C_p$ and $C_{pk}$ as a powerful shop-floor diagnostic tool to decide whether to adjust machine settings or call for tooling overhaul.

---

Step-by-Step Worked Numerical Calculations

Worked Example 1: Full Capability Calculation from Control Chart Data

Scenario: A technician at an aerospace machine shop monitors a titanium fastener diameter on an automated CNC lathe. Blueprints specify a diameter of $12.000 \pm 0.050\text{ mm}$ ($USL = 12.050\text{ mm}, LSL = 11.950\text{ mm}$).

A capability study of 25 rational subgroups of $n = 5$ parts confirms the process is in statistical control. The control chart yields:

• Grand Average: $\bar{\bar{X}} = 12.015\text{ mm}$
• Average Range: $\bar{R} = 0.028\text{ mm}$
• For subgroup size $n = 5$, the unbiasing factor is $d_2 = 2.326$.

Step 1: Estimate Within-Subgroup Standard Deviation ($\hat{\sigma}_{within}$)

$$\hat{\sigma}_{within} = \frac{\bar{R}}{d_2} = \frac{0.028}{2.326} = 0.0120378\text{ mm}$$

Step 2: Calculate Total Specification Width and Process Spread

$$\text{Tolerance Width} = USL - LSL = 12.050 - 11.950 = 0.100\text{ mm}$$ $$6\hat{\sigma}_{within} = 6 \times 0.0120378 = 0.072227\text{ mm}$$ $$3\hat{\sigma}_{within} = 3 \times 0.0120378 = 0.036113\text{ mm}$$

Step 3: Calculate Potential Capability ($C_p$)

$$C_p = \frac{USL - LSL}{6\hat{\sigma}_{within}} = \frac{0.100}{0.072227} = 1.3845 \approx 1.38$$

Interpretation: Since $C_p = 1.38 > 1.33$, the process possesses adequate potential capability.

Step 4: Calculate Upper and Lower Capability ($C_{pu}$ and $C_{pl}$)

$$C_{pu} = \frac{USL - \bar{\bar{X}}}{3\hat{\sigma}_{within}} = \frac{12.050 - 12.015}{0.036113} = \frac{0.035}{0.036113} = 0.9692 \approx 0.97$$ $$C_{pl} = \frac{\bar{\bar{X}} - LSL}{3\hat{\sigma}_{within}} = \frac{12.015 - 11.950}{0.036113} = \frac{0.065}{0.036113} = 1.7999 \approx 1.80$$

Step 5: Determine Actual Capability ($C_{pk}$)

$$C_{pk} = \min(C_{pu}, C_{pl}) = \min(0.97, 1.80) = 0.97$$

Interpretation: Even though potential capability is acceptable ($C_p = 1.38$), actual capability fails customer standards ($C_{pk} = 0.97 < 1.33$). The process is running closer to the upper specification limit, creating nonconformance risk at the upper boundary.

Step 6: Verify Using the Centering Factor ($k$)

$$m = \frac{12.050 + 11.950}{2} = 12.000\text{ mm}$$ $$d = \frac{0.100}{2} = 0.050\text{ mm}$$ $$k = \frac{|12.015 - 12.000|}{0.050} = \frac{0.015}{0.050} = 0.300 \quad (30\% \text{ offset})$$ $$C_{pk} = C_p(1 - k) = 1.3845 \times (1 - 0.300) = 1.3845 \times 0.700 = 0.9692 \approx 0.97$$

Exact verification confirms the calculation.

---

Worked Example 2: Determining Required Machine Adjustment

From Worked Example 1, the technician must recommend a corrective action.

• Because $C_p = 1.38$ is already capable, the machine has plenty of precision.
• To achieve $C_{pk} = C_p = 1.38$, the technician needs $k = 0$.
• Action: Shift the machine tool offset down by exactly $0.015\text{ mm}$ (from $\bar{\bar{X}} = 12.015\text{ mm}$ to target $m = 12.000\text{ mm}$).
• Once re-centered, $C_{pk}$ will immediately jump from $0.97$ to $1.38$ without spending a single dollar on tooling or machine overhauls!

---

Quality Technician Inspection Scenarios & Common Exam Traps

Real-World Shop Scenario: CNC Swiss-Lathe Medical Bone Screw

A technician inspecting titanium bone screw thread pitch diameters calculates $C_p = 2.10$ and $C_{pk} = 1.15$. The operator argues that the machine is performing at a "world-class Six Sigma level" because $C_p > 2.00$. The technician must correct the operator: while the machine's inherent dispersion is exceptional ($C_p = 2.10$), the actual process is currently unacceptable ($C_{pk} = 1.15 < 1.33$) due to thermal expansion in the lathe headstock causing the mean to run hot toward the upper spec. A simple coolant temperature adjustment will re-center the mean and unlock the machine's true potential.

Common Exam Traps for CQT Candidates

• Exam Trap 1: Calculating $C_{pk}$ with Pooled Sample Standard Deviation: Never use $s$ (from all pooled data) to calculate $C_p$ or $C_{pk}$! Standard capability formulas strictly require within-subgroup variation ($\bar{R}/d_2$ or $\bar{s}/c_4$). Using $s$ calculates process performance ($P_p, P_{pk}$), not capability.
• Exam Trap 2: Selecting the Maximum Instead of Minimum for $C_{pk}$: Always choose $\min(C_{pu}, C_{pl})$. An exam distracter will invariably offer the maximum of the two or their simple average.
• Exam Trap 3: Believing $C_{pk}$ Can Exceed $C_p$: If you calculate $C_{pk} = 1.45$ and $C_p = 1.20$, STOP immediately. Mathematically, $C_{pk} \le C_p$. You have swapped formulas or made an arithmetic mistake.
• Exam Trap 4: Forgetting the Factor of 3 vs. 6: Remember that $C_p$ divides the full tolerance by $6\sigma$, while $C_{pu}$ and $C_{pl}$ divide the single-sided distance by $3\sigma$.