7.4 One-Sided Tolerances and PPM Defect Calculations

7.4 One-Sided Tolerances and PPM Defect Calculations

Unilateral (One-Sided) Specifications

Not all engineering tolerances have two boundaries. In precision inspection and mechanical testing, quality technicians routinely encounter quality characteristics governed by unilateral (one-sided) specifications, where only an upper limit or only a lower limit is defined.

1. Upper Specification Limit ($USL$) Only

Many physical characteristics have a natural zero physical boundary, and engineering specifications dictate only a maximum allowable value:

• Geometric Dimensioning and Tolerancing (GD&T): Flatness (\le 0.05\text{ mm}), circular runout ($TIR \le 0.025\text{ mm}$), cylindricity, straightness, perpendicularity.
• Surface Finish: Roughness average ($Ra \le 0.8\ \mu\text{m}$, $Rz \le 3.2\ \mu\text{m}$).
• Contamination & Metallurgy: Particulate count in cleanrooms, burr height on stampings, weld porosity percentage, casting flash thickness.

For an upper-limit-only specification, capability is determined exclusively by the upper capability index: $C_{pk} = C_{pu} = \frac{USL - \mu}{3\hat{\sigma}}$

2. Lower Specification Limit ($LSL$) Only

Other physical properties require a minimum threshold for mechanical strength, functional reliability, or endurance, with no upper boundary:

• Mechanical Strength: Tensile break strength ($\ge 450\text{ MPa}$), yield strength ($\ge 310\text{ MPa}$), weld shear failure force ($\ge 8.5\text{ kN}$).
• Material Properties: Core hardness ($\ge 58\text{ HRC}$), case depth ($\ge 0.75\text{ mm}$), coating adhesion peel strength ($\ge 15\text{ N/cm}$).
• Electrical & Fluid Performance: Dielectric breakdown voltage ($\ge 2,500\text{ V}$), hydraulic burst pressure ($\ge 350\text{ bar}$).

For a lower-limit-only specification, capability is determined exclusively by the lower capability index: $C_{pk} = C_{pl} = \frac{\mu - LSL}{3\hat{\sigma}}$

The $C_p$ Dilemma for Unilateral Tolerances

A common point of confusion on the ASQ CQT exam concerns the calculation of $C_p$ for one-sided tolerances:

[!CAUTION] Exam Trap: Standard potential capability ($C_p = \frac{USL - LSL}{6\sigma}$) is UNDEFINED for unilateral specifications because there is no second specification limit! Never substitute zero as the $LSL$ for geometric features like flatness or runout. Zero is a physical boundary, not an engineering tolerance limit. For unilateral tolerances, only $C_{pk}$ ($C_{pu}$ or $C_{pl}$) is technically valid.

---

Converting Capability Indices to Standard Normal $Z$-Scores

A profound mathematical connection links process capability indices to probability theory and the standard normal distribution curve.

The Fundamental $Z$-Score Equation

The standard normal deviate ($Z$-score) represents the number of standard deviations between the process mean and a given specification limit:

• For an upper limit: $Z_U = \frac{USL - \mu}{\sigma}$
• For a lower limit: $Z_L = \frac{\mu - LSL}{\sigma}$

Recall the definitions of $C_{pu}$ and $C_{pl}$: $C_{pu} = \frac{USL - \mu}{3\sigma} = \frac{Z_U}{3} \implies Z_U = 3 \times C_{pu}$ $C_{pl} = \frac{\mu - LSL}{3\sigma} = \frac{Z_L}{3} \implies Z_L = 3 \times C_{pl}$

Therefore, the universal relationship connecting actual capability to the standard normal $Z$-score is: $Z = 3 \times C_{pk}$

This simple, elegant equation allows a quality technician to instantly convert any capability score into a $Z$-score, consult a standard normal distribution table, and determine the exact probability of producing nonconforming product.

---

Calculating Nonconformance in Parts Per Million (PPM)

Once the $Z$-score is established, the expected proportion of nonconforming product ($p$) is calculated from the cumulative standard normal distribution tail area: $p = P(z > Z) = 1 - \Phi(Z)$

To express this defect rate in Parts Per Million (PPM): $\text{PPM} = p \times 1,000,000$

Bilateral vs. Unilateral PPM Calculations

• For Unilateral Specifications (One-Sided): Defective parts can only occur beyond the single specified boundary: $\text{PPM}_{unilateral} = P(z > 3 \times C_{pk}) \times 10^6$
• For Bilateral Centered Specifications (Two-Sided): If a process is centered between $USL$ and $LSL$, defects occur with equal probability in both tails: $\text{PPM}_{bilateral} = 2 \times P(z > 3 \times C_{pk}) \times 10^6$

                          BILATERAL CENTERED DEFECT TAILS
                                   Mean (μ)
                                      │
                                  * * │ * *
                              * *     │     * *
                          * *         │         * *
                      * *             │             * *
                  * *                 │                 * *
           LSL *                      │                      * USL
       ───────▲───────────────────────┼───────────────────────▲───────
      Lower Tail:                    │                    Upper Tail:
      p = P(z < -Z)                   │                    p = P(z > +Z)
      PPM = 1,350 (at Cpk=1.0)        │                    PPM = 1,350 (at Cpk=1.0)
      
      TOTAL BILATERAL DEFECT RATE = 1,350 + 1,350 = 2,700 PPM

---

Comprehensive Reference Table: Capability, $Z$-Scores, and PPM

The following table is an indispensable reference for ASQ CQT exam preparation, summarizing the mathematical equivalence between capability indices, $Z$-scores, and defect rates under a normal distribution:

The Famous Motorola Six Sigma $1.5\sigma$ Shift

Notice the final column in the reference table. In classical Six Sigma methodology developed by Motorola, empirical studies demonstrated that over long operational horizons, manufacturing process averages drift by an average of $1.5\sigma$ due to tool wear, machine warm-up, and material batch changes.

When a Six Sigma process ($C_p = 2.0$, $Z = 6.0$) experiences this maximum anticipated $1.5\sigma$ shift:

• The effective distance to the nearest specification limit drops from $6.0\sigma$ down to: $Z_{effective} = 6.0\sigma - 1.5\sigma = 4.5\sigma$
• Consulting the standard normal table for $Z = 4.50$ reveals a single-tail area of $0.00000340$.
• Multiplying by $10^6$ produces the iconic Six Sigma benchmark: 3.4 Defects Per Million Opportunities (DPMO / PPM)!

---

Capability Analysis for Non-Normal Data

When quality characteristics follow non-normal distributions (such as lognormal cycle times, Weibull fatigue life, or skewed GD&T flatness readings), standard capability calculations break down completely. If a technician blind-calculates $C_{pk} = 1.33$ on heavily skewed data, the actual scrap rate might be 5,000 PPM instead of the expected 64 PPM!

To analyze non-normal data validly, quality technicians apply three primary methodologies:

1. Box-Cox Power Transformation

The Box-Cox transformation is a mathematical algorithm that transforms non-normal continuous data ($x$) into an approximately normal distribution ($y$) by raising the data to a power parameter $\lambda$ (lambda): $y = \begin{cases} \frac{x^\lambda - 1}{\lambda} & \text{if } \lambda \ne 0 \\ \ln(x) & \text{if } \lambda = 0 \end{cases}$

• Requirement: All data values $x$ must be strictly positive ($x > 0$).
• Common $\lambda$ Values:
  • $\lambda = 1.0$: No transformation required (already normal).
  • $\lambda = 0.5$: Square root transformation ($y = \sqrt{x}$).
  • $\lambda = 0.0$: Natural log transformation ($y = \ln(x)$).
  • $\lambda = -1.0$: Reciprocal transformation ($y = 1/x$).
• Execution Protocol: Both the sample data points AND the specification limits ($USL, LSL$) must be transformed using the exact same $\lambda$. Capability indices are then calculated on the transformed scale.

2. Johnson Transformation System

When data contain negative numbers, zero, or severe bi-modal/kurtotic shapes where Box-Cox fails, the Johnson transformation system selects from three distinct distribution families:

• $S_B$ (Bounded): For data bounded by physical limits (e.g., percentages, fractions between 0 and 1).
• $S_L$ (Lognormal): For positively skewed data bounded on the left by zero.
• $S_U$ (Unbounded): For unbounded data with heavy tails.

The algorithm matches sample moments (mean, variance, skewness, kurtosis) to generate an optimal transformation to the standard normal distribution.

3. Non-Parametric Percentile Method (ISO 21747 / DIN 55319)

If data transformation is undesired or distorts engineering interpretation, the technician can use the percentile method, which replaces theoretical $3\sigma$ standard deviation boundaries with empirical sample percentiles:

• $P_{50}$: The 50th percentile (the sample median), replacing the sample mean $\mu$.
• $P_{0.135}$: The 0.135th percentile, representing the equivalent lower $3\sigma$ boundary ($Z = -3.0$).
• $P_{99.865}$: The 99.865th percentile, representing the equivalent upper $3\sigma$ boundary ($Z = +3.0$).

Formulas for Percentile-Based Capability:

$C_p = \frac{USL - LSL}{P_{99.865} - P_{0.135}}$ $C_{pu} = \frac{USL - P_{50}}{P_{99.865} - P_{50}}$ $C_{pl} = \frac{P_{50} - LSL}{P_{50} - P_{0.135}}$ $C_{pk} = \min(C_{pu}, C_{pl})$

This method requires zero distributional assumptions, providing a robust, mathematically sound capability metric for any continuous distribution.

---

Step-by-Step Worked Numerical Calculations

Worked Example 1: One-Sided Tensile Strength Capability and PPM Defect Estimation

Scenario: An aerospace test lab performs destructive tensile pull testing on welded structural brackets. Engineering blueprints require a minimum tensile failure load of $LSL = 15.0\text{ kN}$. There is no upper limit.

A sample of 100 pull tests demonstrates that the process is in statistical control and follows a normal distribution with:

• Sample Mean: $\bar{x} = 17.4\text{ kN}$
• Estimated Standard Deviation: $\hat{\sigma} = 0.60\text{ kN}$

Step 1: Calculate Actual Capability Index ($C_{pk}$)

Since there is only a lower specification limit ($LSL$): $C_{pk} = C_{pl} = \frac{\bar{x} - LSL}{3\hat{\sigma}} = \frac{17.4 - 15.0}{3(0.60)} = \frac{2.4}{1.80} = 1.333 \approx 1.33$

Step 2: Convert Capability to Standard Normal $Z$-Score

$Z_L = 3 \times C_{pk} = 3 \times 1.333 = 4.00$ (Or directly: $Z_L = \frac{\bar{x} - LSL}{\hat{\sigma}} = \frac{17.4 - 15.0}{0.60} = \frac{2.4}{0.60} = 4.00$).

Step 3: Calculate Expected Nonconformance in PPM

Consulting standard normal cumulative distribution tables for $Z = 4.00$: $p = P(z < -4.00) = 0.00003167$ $\text{PPM} = 0.00003167 \times 1,000,000 = 31.67 \approx 31.7 \text{ PPM}$ Conclusion: The welding process meets the standard 4-sigma industrial capability benchmark ($C_{pk} = 1.33$), with an expected failure rate of approximately 32 brackets per million welded.

---

Worked Example 2: One-Sided GD&T Flatness Inspection

Scenario: A technician measures the surface flatness of precision ground aluminum sealing plates using a coordinate measuring machine (CMM). The blueprint specifies Flatness $\le 0.040\text{ mm}$ ($USL = 0.040\text{ mm}$). Zero is the lower physical boundary.

Data analysis reveals $\bar{x} = 0.022\text{ mm}$ and $\hat{\sigma} = 0.004\text{ mm}$. The technician confirms data normality across the operating range.

Step 1: Calculate Capability

$C_{pk} = C_{pu} = \frac{USL - \bar{x}}{3\hat{\sigma}} = \frac{0.040 - 0.022}{3(0.004)} = \frac{0.018}{0.012} = 1.50$

Step 2: Calculate $Z$-Score and PPM

$Z_U = 3 \times 1.50 = 4.50$ From normal distribution tables for $Z = 4.50$: $p = P(z > 4.50) = 0.00000340$ $\text{PPM} = 0.00000340 \times 10^6 = 3.4 \text{ PPM}$ Conclusion: The grinding operation achieves $C_{pk} = 1.50$ with an expected nonconformance rate of only 3.4 plates per million.

---

Quality Technician Inspection Scenarios & Common Exam Traps

Real-World Shop Scenario: Medical Catheter Heat-Seal Peel Adhesion

A quality technician at a medical packaging plant evaluates the sterile barrier peel strength of Tyvek pouches. The requirement is a minimum peel force of $LSL = 1.5\text{ N/cm}$. The machine delivers $\bar{x} = 2.8\text{ N/cm}$ with $\sigma = 0.25\text{ N/cm}$. An operator attempts to report $C_p$ by entering $USL = 4.0\text{ N/cm}$, an arbitrary number invented so the software would output a $C_p$ number. The technician stops the report: fabricating artificial limits distorts capability analysis. Because peel strength has no upper specification, the technician instructs the team to report $C_{pk} = C_{pl} = (2.8 - 1.5) / (3 \times 0.25) = 1.3 / 0.75 = 1.73$, fully validating compliance without fabricated data.

Common Exam Traps for CQT Candidates

• Exam Trap 1: Forgetting that $Z = 3 \times C_{pk}$: A classic exam mistake is assuming $Z = C_{pk}$. If $C_{pk} = 1.33$, $Z$ is not 1.33; $Z = 3 \times 1.33 = 4.00$!
• Exam Trap 2: Doubling Unilateral PPM: For a two-sided centered specification, you must double the single-tail probability ($1,350 \times 2 = 2,700\text{ PPM}$ at $C_{pk} = 1.00$). But for a one-sided specification, do not double it—defects can only occur at the single specified limit ($1,350\text{ PPM}$ at $C_{pk} = 1.00$)!
• Exam Trap 3: Applying Box-Cox to Zero or Negative Numbers: Box-Cox transformations strictly require $x > 0$. If data contain zero (such as zero runout or zero defect counts), you must add a small constant displacement ($x + c$) or use the Johnson transformation.
• Exam Trap 4: Forgetting to Transform Specifications: When using Box-Cox, you cannot compare transformed sample data ($y$) against original untransformed blueprint specifications ($USL$). The specifications must be transformed with the exact same $\lambda$ parameter.