8.3 Process Capability Basics (Cp & Cpk)

8.3 Process Capability Basics (Cp & Cpk)

In quality engineering and continuous improvement, an operational process can be in statistical control (predictable and free of special causes) yet still produce output that fails to satisfy customer requirements. To determine whether a manufacturing or service process can consistently produce conforming units, quality professionals conduct Process Capability Analysis.

On the ASQ Certified Quality Improvement Associate (CQIA) examination, candidates must understand the relationship between specification limits and process variation, master the calculation and interpretation of capability indices ($C_p$ and $C_{pk}$), recognize essential analytical prerequisites, and differentiate short-term capability from long-term performance ($P_p$ and $P_{pk}$).

---

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

Process capability is fundamentally the mathematical comparison between what the customer demands and what the process naturally delivers.

+-------------------------------------------------------------------------+
|            VOICE OF CUSTOMER (VOC) vs. VOICE OF PROCESS (VOP)           |
+-------------------------------------------------------------------------+
|                                                                         |
|  VOICE OF THE CUSTOMER (VOC)             VOICE OF THE PROCESS (VOP)     |
|  * Specification Limits (USL / LSL)      * Natural Process Spread (6σ)  |
|  * Set externally by engineering,        * Measured internally from     |
|    clients, blueprints, regulations        actual production data       |
|  * Applies to INDIVIDUAL UNITS           * Applies to PROCESS POPULATION|
|  * Independent of process stability      * Dependent on process control |
|  * What the product MUST do              * What the process CAN do      |
|                                                                         |
+-------------------------------------------------------------------------+

                  VOICE OF THE CUSTOMER (Tolerance Width = USL - LSL)
          LSL                                                     USL
           │◄──────────────────────────────────────────────────────►│
           │                                                        │
           │                    Normal Curve                        │
           │                        ▲                               │
           │                       /│\                              │
           │                      / │ \                             │
           │                     /  │  \                            │
           │                   /    │    \                          │
           │               /        │        \                      │
           │           /            │            \                  │
          ─┴─────────/──────────────┴──────────────\────────────────┴─►
                   μ - 3σ           μ            μ + 3σ
                   │◄────────────── 6σ ───────────►│
                         VOICE OF THE PROCESS

Specification Limits vs. Control Limits

Candidates must never confuse Specification Limits with Control Limits:

• Specification Limits (USL and LSL): Engineering tolerances established by product designers, customers, or regulatory bodies. They define whether an individual manufactured part is conforming or non-conforming (scrap/rework).
• Control Limits (UCL and LCL): Statistical boundaries calculated from sample data (typically $\mu \pm 3\sigma_{\bar{x}}$). They reflect the natural boundaries of common cause variation and apply to subgroup averages to monitor process stability over time.

---

2. Prerequisites for Process Capability Analysis

Before calculating or interpreting $C_p$ or $C_{pk}$, three mandatory engineering prerequisites must be satisfied:

+-------------------------------------------------------------------------+
|            MANDATORY PREREQUISITES FOR CAPABILITY ANALYSIS              |
+-------------------------------------------------------------------------+
|                                                                         |
|  1. STATISTICAL STABILITY ──► Process must be in statistical control    |
|                               (free of special causes on control charts)|
|                                                                         |
|  2. NORMAL DISTRIBUTION   ──► Process output data must follow an        |
|                               approximately normal (bell-shaped) curve  |
|                                                                         |
|  3. MEASUREMENT INTEGRITY ──► Measurement System Analysis (MSA / Gage   |
|                               R&R) must confirm acceptable precision    |
|                                                                         |
+-------------------------------------------------------------------------+

1. Statistical Control (Stability): The process must exhibit only common cause (random) variation. If special causes of variation are present (e.g., points beyond control limits, trends, runs), the process distribution is unstable and shifting. Calculating capability indices on an unstable process produces meaningless and misleading numbers.
2. Approximate Normality: Standard $C_p$ and $C_{pk}$ formulas are derived from the mathematical properties of the standard normal distribution (where $\mu \pm 3\sigma$ encompasses $99.73%$ of output). If the process data is skewed or non-normal (e.g., flatness, runout, cycle time), data transformations (Box-Cox) or non-normal percentile methods must be applied.
3. Measurement System Validity: Gage Repeatability & Reproducibility (Gage R&R) must confirm that measurement error accounts for a negligible fraction ($<10%$ preferred, $<30%$ acceptable) of total process variation.

---

3. Potential Process Capability Index ($C_p$)

The Process Capability Index ($C_p$) measures the potential capability of a process. It is the ratio of the allowable customer tolerance width to the natural process spread ($6\sigma$).

Mathematical Formula

$C_p = \frac{\text{Tolerance Width}}{\text{Natural Process Spread}} = \frac{\text{USL} - \text{LSL}}{6\sigma}$

Where:

• $\text{USL} = \text{Upper Specification Limit}$
• $\text{LSL} = \text{Lower Specification Limit}$
• $\sigma = \text{Estimated process standard deviation (short-term within-subgroup variation)}$
• $6\sigma = \text{Natural Process Spread } (\pm 3\sigma \text{ from mean, representing } 99.73% \text{ of output})$

Key Characteristics of $C_p$

• Ignores Process Centering: $C_p$ evaluates strictly whether the process spread is narrow enough to fit within the specification tolerance band. It completely ignores where the process mean ($\mu$) is currently located.
• Best-Case Potential: $C_p$ represents what the capability would be if the process mean were perfectly centered at the nominal specification midpoint: $\mu = \frac{\text{USL} + \text{LSL}}{2}$.

---

4. Actual Process Capability Index ($C_{pk}$)

The Actual Process Capability Index ($C_{pk}$) measures the actual performance of a process by taking into account both the process spread ($6\sigma$) and the process centering (the location of the mean $\mu$ relative to the specification limits).

Mathematical Formulas

$C_{pk}$ evaluates the capability relative to each specification limit independently and selects the minimum (the worst-case side):

$C_{pu} = \frac{\text{USL} - \mu}{3\sigma}$

$C_{pl} = \frac{\mu - \text{LSL}}{3\sigma}$

$C_{pk} = \min(C_{pu}, C_{pl}) = \min\left(\frac{\text{USL} - \mu}{3\sigma}, \frac{\mu - \text{LSL}}{3\sigma}\right)$

Where:

• $C_{pu} = \text{Upper capability index (distance from mean to USL divided by } 3\sigma)$
• $C_{pl} = \text{Lower capability index (distance from mean to LSL divided by } 3\sigma)$
• $\mu = \text{Process mean}$

+-------------------------------------------------------------------------+
|                 COMPARISON OF PROCESS CAPABILITY SCENARIOS              |
+-------------------------------------------------------------------------+
|                                                                         |
|  SCENARIO 1: Centered & Capable       SCENARIO 2: Off-Center (Drifting) |
|  LSL               μ               USL LSL               μ         USL  |
|   │                ▲                │   │                │          ▲│  |
|   │               /│\               │   │                │         /│\  |
|   │              / │ \              │   │                │        / │ \ │
|   │             /  │  \             │   │                │       /  │  \│
|  ─┴───────────/────┴────\───────────┴─ ─┴────────────────┼─────/────┴───\ (Defects
|               │◄── 6σ ──►│                               │     │◄─ 6σ ─►│  Escape!)
|          Cp = 1.67, Cpk = 1.67                      Cp = 1.67, Cpk = 0.67
|                                                                         |
+-------------------------------------------------------------------------+

Fundamental Relationship Between $C_p$ and $C_{pk}$

• Universal Inequality: $C_{pk} \le C_p$ Because $C_{pk}$ penalizes for off-center processes, $C_{pk}$ can never exceed $C_p$.
• Perfect Centering ($C_{pk} = C_p$): When the process mean sits exactly at the specification midpoint ($\mu = \frac{\text{USL} + \text{LSL}}{2}$), $C_{pu} = C_{pl}$, and therefore $C_{pk} = C_p$.
• Off-Center Shifts ($C_{pk} < C_p$): As the process mean drifts toward either the USL or LSL, $C_{pk}$ decreases while $C_p$ remains constant.
• Mean on Specification Limit ($C_{pk} = 0$): When the process mean sits directly on either the USL or LSL ($\mu = \text{USL}$ or $\mu = \text{LSL}$), $C_{pk} = 0$, meaning $50%$ of production is defective.
• Mean Outside Specifications ($C_{pk} < 0$): If the process mean drifts completely outside the tolerance band ($\mu > \text{USL}$ or $\mu < \text{LSL}$), $C_{pk}$ becomes negative.

---

5. Step-by-Step Worked Numerical Examples

Example 1: Perfectly Centered Process

A precision stamping process produces steel pins with specifications $\text{USL} = 55.0\text{ mm}$ and $\text{LSL} = 45.0\text{ mm}$. Statistical analysis of a stable process indicates $\mu = 50.0\text{ mm}$ and standard deviation $\sigma = 1.0\text{ mm}$.

1. Calculate $C_p$: $C_p = \frac{\text{USL} - \text{LSL}}{6\sigma} = \frac{55.0 - 45.0}{6(1.0)} = \frac{10.0}{6.0} \approx 1.67$
2. Calculate $C_{pk}$: $C_{pu} = \frac{55.0 - 50.0}{3(1.0)} = \frac{5.0}{3.0} \approx 1.67$ $C_{pl} = \frac{50.0 - 45.0}{3(1.0)} = \frac{5.0}{3.0} \approx 1.67$ $C_{pk} = \min(1.67, 1.67) = 1.67$
3. Conclusion: Because $\mu$ is centered, $C_{pk} = C_p = 1.67$. The process is highly capable.

Example 2: Off-Center Process (Mean Shift)

Using the same pins ($\text{USL} = 55.0\text{ mm}, \text{LSL} = 45.0\text{ mm}, \sigma = 1.0\text{ mm}$), machine wear causes the process mean to drift upward to $\mu = 53.0\text{ mm}$.

1. Calculate $C_p$: $C_p = \frac{55.0 - 45.0}{6(1.0)} = \frac{10.0}{6.0} \approx 1.67$ (Notice that $C_p$ is unchanged because spread and tolerances are identical).
2. Calculate $C_{pk}$: $C_{pu} = \frac{55.0 - 53.0}{3(1.0)} = \frac{2.0}{3.0} \approx 0.67$ $C_{pl} = \frac{53.0 - 45.0}{3(1.0)} = \frac{8.0}{3.0} \approx 2.67$ $C_{pk} = \min(0.67, 2.67) = 0.67$
3. Conclusion: Even though the potential capability is $1.67$, the actual capability has collapsed to $C_{pk} = 0.67$ due to the mean shift toward USL. The process is generating out-of-spec scrap.

---

6. Capability Benchmarks and Sigma Levels

Industry standards define specific benchmark thresholds for $C_p$ and $C_{pk}$:

+-----------------------------------------------------------------------------------------+
|                        CAPABILITY BENCHMARKS & DEFECT RATES                             |
+-----------------------------------------------------------------------------------------+
| Cpk Value | Sigma Level | Centered Defect Rate (PPM) | Process Capability Status        |
+-----------+-------------+----------------------------+----------------------------------+
| < 1.00    | < 3σ        | > 2,700 ppm (> 0.27%)      | Incapable (Excessive defects)    |
| 1.00      | 3σ          | 2,700 ppm (0.27%)          | Marginally Capable (Bare minimum)|
| 1.33      | 4σ          | 63 ppm                     | Standard Industry Capability     |
| 1.67      | 5σ          | 0.57 ppm                   | Highly Capable (New/Safety parts)|
| >= 2.00   | 6σ          | 0.002 ppm (3.4 ppm w/shift)| World-Class Six Sigma Quality    |
+-----------------------------------------------------------------------------------------+

• $C_{pk} < 1.00$ (Incapable): The natural process spread exceeds specification limits or the process is significantly off-center. Non-conforming units are actively produced.
• $C_{pk} = 1.00$ ($3\sigma$ Minimum): The natural process spread ($6\sigma$) exactly equals the tolerance width. In a centered normal distribution, $99.73%$ of parts conform, leaving $0.27%$ ($2,700 \text{ parts per million}$) defective.
• $C_{pk} = 1.33$ ($4\sigma$ Standard): The universally recognized minimum target for existing manufacturing and transactional service processes ($8\sigma$ tolerance width).
• $C_{pk} = 1.67$ ($5\sigma$ Critical): The standard industry requirement for new product launches, safety-critical aerospace/medical components, or critical-to-quality (CTQ) parameters ($10\sigma$ tolerance width).
• $C_{pk} \ge 2.00$ ($6\sigma$ World Class): Represents Six Sigma quality ($12\sigma$ tolerance width), allowing for a standard $\pm 1.5\sigma$ long-term mean shift while sustaining no more than $3.4 \text{ defects per million opportunities (DPMO)}$.

---

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

A critical distinction on the CQIA exam is the difference between Process Capability ($C_p, C_{pk}$) and Process Performance ($P_p, P_{pk}$):

+-------------------------------------------------------------------------+
|        SHORT-TERM CAPABILITY (Cp / Cpk) vs. LONG-TERM PERFORMANCE (Pp / Ppk) |
+-------------------------------------------------------------------------+
|  Dimension         | Capability (Cp, Cpk)      | Performance (Pp, Ppk)  |
+--------------------+---------------------------+------------------------+
|  Time Horizon      | Short-term snapshot       | Long-term history      |
|  Variation Scope   | Within-subgroup variation | Overall total variation|
|  Standard Dev (σ)  | σ_within (from R-bar/d2)  | s (sample std dev)     |
|  Represents        | Process POTENTIAL         | Process REALITY (VOC)  |
|  Special Causes    | Assumes zero special cause| Captures all drift/wear|
+-------------------------------------------------------------------------+

Mathematical Distinction in Standard Deviation

• $C_p$ and $C_{pk}$ (Short-Term / Within-Subgroup): Estimated from control chart ranges or standard deviations (e.g., $\sigma_{\text{within}} = \frac{\bar{R}}{d_2}$ or $\frac{\bar{s}}{c_4}$). This isolates short-term machine capability by filtering out between-subgroup shifts, batch-to-batch variation, operator changes, and environmental temperature fluctuations.
• $P_p$ and $P_{pk}$ (Long-Term / Overall Total): Calculated using the standard sample standard deviation formula across all raw individual measurements: $s = \sqrt{\frac{\sum_{i=1}^n (X_i - \bar{X})^2}{n - 1}}$ $P_p = \frac{\text{USL} - \text{LSL}}{6s}, \quad P_{pk} = \min\left(\frac{\text{USL} - \bar{X}}{3s}, \frac{\bar{X} - \text{LSL}}{3s}\right)$
• Operational Relationship: Under real-world conditions with tool wear, raw material batch differences, and shift changes over time, overall variation is larger than within-subgroup variation ($s > \sigma_{\text{within}}$). Consequently, $P_{pk} \le C_{pk}$ in almost all production environments.