11.3 Process Capability Indices (Cp, Cpk, Cpm, Pp, Ppk) & Sigma Levels

Process Capability Indices ($C_p$, $C_{pk}$, $C_{pm}$, $P_p$, $P_{pk}$) & Sigma Levels

Core Principle: Process capability quantitatively benchmarks the relationship between the natural statistical variation of a process (Voice of the Process) and engineering design tolerances (Voice of the Customer). Capability analysis determines whether an in-control system can manufacture conforming product within acceptable defect limits.

Achieving statistical process control is an indispensable prerequisite for quality engineering, but a process in statistical control is not necessarily a capable process. An operation can exhibit flawless statistical stability while churning out 100% scrap if its natural spread exceeds the permissible engineering tolerances or if its mean is improperly centered. Process capability analysis bridges statistical quality control and systems design engineering. For the PE exam, industrial engineers must distinguish between process potential ($C_p$), centered capability ($C_{pk}$), target-weighted capability ($C_{pm}$), and long-term process performance ($P_p, P_{pk}$).

---

1. Engineering Foundations of Process Capability

Before calculating capability metrics, two foundational engineering criteria must be established:

1. Statistical Control Prerequisite: The process must be verified to be in a state of statistical control via variables control charts ($\bar{X}$-$R$ or $\bar{X}$-$s$). Calculating capability indices on an unstable process subject to special causes yields meaningless, ephemeral numbers that cannot predict future performance.
2. Normality Assumption: Standard capability formulations assume that the quality metric $X$ is normally distributed: $X \sim N(\mu, \sigma^2)$.

Voice of the Process vs. Voice of the Customer

The fundamental tension in quality engineering is between two independent boundaries:

Voice of the Process vs Voice of the Customer

[ LSL ]---------------------[ Nominal Target T ]---------------------[ USL ]  <- Voice of Customer
         |                                                     |
         |<------------------ Specification Width ------------>|
                               (USL - LSL)

         [ -3 Sigma ]---------------[ Mean mu ]---------------[ +3 Sigma ]     <- Voice of Process
               |<-------------- Natural Tolerance Spread ------------>|
                                        (6 Sigma)

• Natural Tolerance Limits (NTL): Defined by the physical physics and variability of the equipment: $\text{NTL} = \mu \pm 3\sigma$. The distance between the upper and lower natural tolerance limits encompasses $99.73\%$ of individual parts produced by a normal process: $\text{Width} = 6\sigma$.
• Engineering Specification Limits: Defined externally by functional design requirements: Lower Specification Limit ($LSL$) and Upper Specification Limit ($USL$). The allowable tolerance width is $\text{Specification Band} = USL - LSL$.

---

2. Process Potential Index ($C_p$)

The Process Potential Index ($C_p$) measures the ratio of the allowable design specification band to the natural $6\sigma$ variation band of the process:

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

Where $\sigma$ represents the within-subgroup process standard deviation estimated from an in-control $\bar{X}$-$R$ or $\bar{X}$-$s$ chart ($\hat{\sigma} = \bar{R}/d_2$ or $\bar{s}/c_4$).

Industrial Benchmarks for $C_p$

---

3. Process Capability Index ($C_{pk}$) and Process Centering

To account for both process spread and process centering, the Process Capability Index ($C_{pk}$) measures the distance from the process mean $\mu$ to the nearest specification limit in units of $3\sigma$.

Mathematical Formulation

First, compute the upper and lower unilateral capability indices:

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

The overall index $C_{pk}$ is the minimum of the two unilateral indices:

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

The Centering Factor ($k$)

The mathematical relationship connecting $C_p$ and $C_{pk}$ is parameterized by the centering factor ($k$), which quantifies the fractional shift of the process mean from the tolerance midpoint $M = \frac{\text{USL} + \text{LSL}}{2}$:

$k = \frac{|M - \mu|}{\frac{\text{USL} - \text{LSL}}{2}} = \frac{2 |M - \mu|}{\text{USL} - \text{LSL}}$

The relationship between the indices is:

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

Interpretation of $C_{pk}$ Values

Relationship Between Process Centering and Capability
├── Case 1: Perfectly Centered (mu = M)
│   ├── k = 0
│   └── C_pk = C_p
├── Case 2: Process Mean Shifted Toward a Spec Limit (LSL < mu < USL)
│   ├── 0 < k < 1
│   └── 0 < C_pk < C_p
├── Case 3: Process Mean Exactly on a Spec Limit (mu = USL or mu = LSL)
│   ├── k = 1
│   └── C_pk = 0  (Exactly 50% of production is nonconforming)
└── Case 4: Process Mean Lies Outside Spec Limits (mu > USL or mu < LSL)
    ├── k > 1
    └── C_pk < 0  (Over 50% of production is nonconforming scrap)

For unilateral (one-sided) engineering specifications—such as minimum burst pressure ($LSL$ only) or maximum surface roughness ($USL$ only)—the bilateral $C_p$ is undefined. Engineers compute solely the relevant unilateral index: $C_{pk} = C_{pl}$ or $C_{pk} = C_{pu}$.

---

4. Taguchi Capability Index ($C_{pm}$)

In classical capability analysis, any part manufactured between $LSL$ and $USL$ is considered equally acceptable (zero defect loss), while any part outside is defective (the step-loss function). Dr. Genichi Taguchi demonstrated that customer dissatisfaction and economic loss do not begin abruptly at the specification limit. Rather, quality loss increases quadratically as the product dimension drifts away from the nominal target value ($T$):

$L(y) = k_L (y - T)^2$

Where $k_L$ is an economic cost constant. To reflect this quadratic loss, Taguchi introduced the Taguchi Capability Index ($C_{pm}$), often termed the process target index:

$C_{pm} = \frac{\text{USL} - \text{LSL}}{6\tau} \qquad \text{where} \quad \tau = \sqrt{E\left[(X - T)^2\right]} = \sqrt{\sigma^2 + (\mu - T)^2}$

Substituting $\tau$ into the equation yields:

$C_{pm} = \frac{\text{USL} - \text{LSL}}{6 \sqrt{\sigma^2 + (\mu - T)^2}} = \frac{C_p}{\sqrt{1 + \left(\frac{\mu - T}{\sigma}\right)^2}}$

Analytical Properties of $C_{pm}$

• When the process mean is centered on target ($\mu = T$), $\tau = \sigma$, and $C_{pm} = C_p$.
• Unlike $C_{pk}$—which evaluates distance only to the nearest boundary—$C_{pm}$ explicitly penalizes any departure from the target $T$. Even if a process has wide tolerances, an off-target mean severely degrades $C_{pm}$.
• When the engineering design specifies an asymmetric tolerance band where target $T$ does not equal midpoint $M$, $C_{pm}$ accurately reflects customer loss, whereas $C_p$ and $C_{pk}$ can give misleading assessments.

---

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

A critical distinction in NCEES examination specifications and AIAG quality standards is the difference between process capability ($C_p, C_{pk}$) and process performance ($P_p, P_{pk}$):

The Performance Diagnostic Gap

Comparing capability indices against performance indices provides a powerful diagnostic tool for industrial engineers:

• If $C_{pk} \approx P_{pk}$: The process is stable over the long term. The variation observed across weeks is identical to the instantaneous variation within subgroups. No significant special causes or thermal drifts exist.
• If $C_{pk} \gg P_{pk}$: A substantial performance gap exists. While the machine tool has excellent intrinsic precision (high short-term $C_{pk}$), the long-term process suffers from between-subgroup variation (shifts, drifts, tool degradation, batch-to-batch material differences, or operator inconsistencies). Engineering effort must focus on external operational controls rather than buying new machinery.

---

6. Six Sigma Quality, Mean Shifts, and Defect Quantification

In the 1980s, Motorola revolutionized industrial quality engineering by establishing the Six Sigma standard. In classical statistics, a $3\sigma$ process produces $99.73\%$ conforming parts, leaving $0.27\%$ defective ($2,700\text{ parts per million}$, PPM). In complex assemblies comprising thousands of components (e.g., cell phones, aerospace avionics, or medical robotics), a $2,700\text{ PPM}$ defect rate per component yields nearly zero first-pass assembly yield.

The $1.5\sigma$ Mean Shift Assumption

Motorola empirically observed that even well-controlled processes do not remain centered on target indefinitely. Over extended operating periods, processes inevitably drift by approximately $1.5\sigma$ due to minor setup errors, raw material shifts, and ambient fluctuations.

Motorola Six Sigma Model with 1.5-Sigma Mean Shift

                       Shifted Mean = Target + 1.5 Sigma
                                   |
                                   v
         LSL                    Target                    USL
          |                        |                       |
          |<----- 7.5 Sigma ------>|<----- 4.5 Sigma ----->|
          |<------------------- 12 Sigma ----------------->|
                                              /=============\
                                             /   Defect Area \
                                            /     3.4 PPM     \
  -----------------------------------------+-------------------+
                                          USL = +6 Sigma from Target
                                              = +4.5 Sigma from Shifted Mean

In a Six Sigma process:

1. The design specification band spans twelve process standard deviations: $\text{USL} - \text{LSL} = 12\sigma$, yielding $C_p = \frac{12\sigma}{6\sigma} = 2.00$.
2. Accounting for the long-term $1.5\sigma$ mean shift, the distance from the shifted mean to the nearest specification limit is:

$\text{Distance} = 6.0\sigma - 1.5\sigma = 4.5\sigma$

1. The resulting long-term capability index is:

$C_{pk} = \frac{4.5\sigma}{3\sigma} = 1.50$

1. The defect rate corresponding to the upper tail of a standard normal distribution at $Z = 4.5$ is:

$P(Z > 4.5) = 1 - \Phi(4.5) = 3.3976 \times 10^{-6} \approx 3.4 \times 10^{-6}$

This derives the famous benchmark: 3.4 Defects Per Million Opportunities (DPMO) / Parts Per Million (PPM).

Master Sigma Level and Defect Conversion Table

---

7. Comprehensive Worked Numerical Problem: Precision CNC Pin Grinding

An industrial engineer at an automotive powertrain facility conducts a comprehensive capability and performance study on a precision CNC centerless grinding cell producing transmission valve spools. The design specifications and baseline parameters are:

• Lower Specification Limit ($LSL$): $24.850\text{ mm}$
• Nominal Engineering Target ($T$): $25.000\text{ mm}$
• Upper Specification Limit ($USL$): $25.150\text{ mm}$
• Tolerance Band: $\text{USL} - \text{LSL} = 25.150 - 24.850 = 0.300\text{ mm}$
• Midpoint ($M$): $(25.150 + 24.850)/2 = 25.000\text{ mm}$ (Symmetric specification, $M = T$)

The Phase I study evaluates $m = 30$ rational subgroups of size $n = 5$ (total parts inspected $N = 150$). Statistical analysis of the in-control $\bar{X}$-$R$ chart yields:

• Grand process mean: $\bar{\bar{X}} = 25.030\text{ mm}$
• Average subgroup range: $\bar{R} = 0.070\text{ mm}$
• Overall sample standard deviation across all 150 parts: $s_{\text{total}} = 0.0380\text{ mm}$

Step 1: Compute Short-Term Within-Subgroup Standard Deviation

From the NCEES factor table for $n = 5$, $d_2 = 2.326$:

$\hat{\sigma}_{\text{within}} = \frac{\bar{R}}{d_2} = \frac{0.070}{2.326} = 0.030095\text{ mm} \approx 0.0301\text{ mm}$

Step 2: Compute Process Potential Index ($C_p$)

$C_p = \frac{\text{USL} - \text{LSL}}{6\hat{\sigma}_{\text{within}}} = \frac{0.300}{6 \times 0.030095} = \frac{0.300}{0.18057} = 1.6614 \approx 1.66$

Step 3: Compute Process Capability Index ($C_{pk}$)

Calculate the unilateral upper and lower capability indices:

$C_{pu} = \frac{\text{USL} - \bar{\bar{X}}}{3\hat{\sigma}_{\text{within}}} = \frac{25.150 - 25.030}{3 \times 0.030095} = \frac{0.120}{0.090284} = 1.3291$ $C_{pl} = \frac{\bar{\bar{X}} - \text{LSL}}{3\hat{\sigma}_{\text{within}}} = \frac{25.030 - 24.850}{3 \times 0.030095} = \frac{0.180}{0.090284} = 1.9937$ $C_{pk} = \min(C_{pu}, C_{pl}) = \min(1.3291, 1.9937) = 1.3291 \approx 1.33$

Verification using the Centering Factor ($k$): $k = \frac{|M - \bar{\bar{X}}|}{\frac{\text{USL} - \text{LSL}}{2}} = \frac{|25.000 - 25.030|}{0.150} = \frac{0.030}{0.150} = 0.200$ $C_{pk} = C_p (1 - k) = 1.6614 \times (1 - 0.200) = 1.6614 \times 0.800 = 1.3291 \approx 1.33$

Step 4: Compute Taguchi Capability Index ($C_{pm}$)

Calculate the total deviation parameter $\tau$ from the nominal target $T = 25.000\text{ mm}$:

$\tau = \sqrt{\hat{\sigma}_{\text{within}}^2 + (\bar{\bar{X}} - T)^2} = \sqrt{(0.030095)^2 + (25.030 - 25.000)^2}$ $\tau = \sqrt{0.0009057 + (0.030)^2} = \sqrt{0.0009057 + 0.0009000} = \sqrt{0.0018057} = 0.042494\text{ mm}$

Compute $C_{pm}$:

$C_{pm} = \frac{\text{USL} - \text{LSL}}{6\tau} = \frac{0.300}{6 \times 0.042494} = \frac{0.300}{0.25496} = 1.1766 \approx 1.18$

Notice that while $C_p = 1.66$ (excellent potential), Taguchi's $C_{pm}$ drops to $1.18$ because the process mean is shifted $0.030\text{ mm}$ off target, which imposes customer loss.

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

Using the overall sample standard deviation $s_{\text{total}} = 0.0380\text{ mm}$:

$P_p = \frac{\text{USL} - \text{LSL}}{6 s_{\text{total}}} = \frac{0.300}{6 \times 0.0380} = \frac{0.300}{0.2280} = 1.3158 \approx 1.32$ $P_{pu} = \frac{\text{USL} - \bar{\bar{X}}}{3 s_{\text{total}}} = \frac{25.150 - 25.030}{3 \times 0.0380} = \frac{0.120}{0.1140} = 1.0526$ $P_{pl} = \frac{\bar{\bar{X}} - \text{LSL}}{3 s_{\text{total}}} = \frac{25.030 - 24.850}{3 \times 0.0380} = \frac{0.180}{0.1140} = 1.5789$ $P_{pk} = \min(1.0526, 1.5789) = 1.0526 \approx 1.05$

Step 6: Engineering Synthesis and Recommendations

1. Short-Term vs Long-Term Gap: The short-term potential is $C_p = 1.66$ (a 5-sigma capability), but long-term performance drops to $P_p = 1.32$ and $P_{pk} = 1.05$. The overall standard deviation ($s_{\text{total}} = 0.0380$) is $26.3\%$ higher than the within-subgroup variation ($\hat{\sigma} = 0.0301$), revealing significant between-subgroup instability (thermal drift on the grinding spindle or coolant temperature swings).
2. Centering Adjustment: Re-centering the grinding head by adjusting the CNC tool offset by $-0.030\text{ mm}$ will immediately align $\bar{\bar{X}} = 25.000\text{ mm}$, raising short-term capability from $C_{pk} = 1.33$ to $C_{pk} = C_p = 1.66$, and long-term performance from $P_{pk} = 1.05$ to $P_{pk} = P_p = 1.32$.

---

8. Common PE Exam Pitfalls for Capability Indices

1. Evaluating Capability on Unstable Processes: If exam problem data shows an out-of-control point on the range or mean chart, any calculation of $C_p$ or $C_{pk}$ is invalid. The correct answer is to eliminate the assignable cause first.
2. Confusing Within-Subgroup $\sigma$ with Overall $s$: Always use $\hat{\sigma} = \bar{R}/d_2$ (or $\bar{s}/c_4$) for $C_p$ and $C_{pk}$. If a question asks for $P_p$ or $P_{pk}$, use the overall sample standard deviation $s_{\text{total}}$. Interchanging these two variance estimators is a frequent trap.
3. Dividing by $3\sigma$ instead of $6\sigma$ for $C_p$: The potential index spans the entire bilateral $6\sigma$ spread ($USL - LSL$). The unilateral indices ($C_{pu}, C_{pl}$) divide by $3\sigma$. Using $3\sigma$ in the denominator of $C_p$ erroneously doubles the result.
4. Assuming $C_p = C_{pk}$ Without Verifying Centering: Examinees often compute $C_p$ and assume it represents actual defect performance. If the mean is shifted, $C_{pk}$ is strictly less than $C_p$. Always check whether $\mu = M$.
5. Misunderstanding Negative $C_{pk}$ Values: If the process mean drifts beyond either specification limit ($\mu > USL$ or $\mu < LSL$), $C_{pk}$ becomes negative. A negative $C_{pk}$ does not mean a mathematical calculation error; it indicates that more than $50\%$ of production is nonconforming scrap.