Why Process Capability Matters in Manufacturing
Producing a single prototype part that passes dimensional inspection during First Article Inspection (FAI) does not guarantee that your manufacturing process can produce 10,000 parts without generating defective scrap. Process Capability Analysis statistically evaluates whether an inherent manufacturing process (CNC milling, turning, stamping, or molding) is capable of consistently holding tolerances over high-volume production runs.
Major manufacturing standards — such as AS9100 Rev D, ISO 9001, and AIAG PPAP — require suppliers to prove process capability on all Key Characteristics (KC) and Critical to Quality (CTQ) dimensions.
The Core Indices: Cp, Cpk, Pp, and Ppk
1. Cp (Process Capability Potential)
Cp measures the ratio of the total allowable customer tolerance width (USL - LSL) to the inherent width of the process ($6\sigma$). It reflects what the process could achieve if it were perfectly centered on the nominal target, ignoring process shift.
2. Cpk (Process Capability Index — Short Term)
Cpk accounts for both process spread and process centering relative to the Upper Specification Limit (USL) and Lower Specification Limit (LSL). It uses the "within-subgroup" standard deviation ($\sigma_{within}$, estimated via $\bar{R} / d_2$ or $\bar{s} / c_4$) from rational subgroups of parts produced close together in time.
Cpl = (μ - LSL) / (3 × σ_within)
Cpk = Min(Cpu, Cpl)
3. Pp and Ppk (Process Performance Indices — Long Term)
While Cpk evaluates what the process is capable of doing under short-term stable conditions, Ppk (Process Performance) measures actual historical performance across long-term production. It uses the total overall sample standard deviation ($s_{overall}$), capturing variations caused by machine warm-up, tool wear, raw material lot variations, and shift-to-shift operator changes.
Key Differences: Cpk vs Ppk
| Characteristic | Cpk (Capability Index) | Ppk (Performance Index) |
|---|---|---|
| Time Horizon | Short-term (e.g., initial production run or validation batch) | Long-term (full sustained production over weeks/months) |
| Standard Deviation | $\sigma_{within}$ (estimated from sub-groups: $\bar{R}/d_2$) | $s_{overall}$ (calculated from the root-sum-square of all data) |
| Stability Requirement | Requires the process to be statistically in control with no special causes | Accounts for both common and special-cause real-world variation |
| Common Benchmark | Cpk ≥ 1.33 (Capable), Cpk ≥ 1.67 (Critical/Aerospace) | Ppk ≥ 1.33 for ongoing production validation |
| Primary Question Answered | Can our machine hold this tolerance when dialed in? | Did our production process actually hold this tolerance over time? |
Interpreting Capability Values: Defect Rates (PPM)
The table below illustrates the relationship between capability values, process sigma level, and expected defect rate in Parts Per Million (PPM), assuming a centered normal distribution:
| Cpk / Ppk Value | Sigma Level | Expected Defect Rate (PPM) | Quality Assessment |
|---|---|---|---|
| < 1.00 | < 3σ | > 2,700 PPM | Unacceptable — High risk of shipping non-conforming scrap. 100% sorting required. |
| 1.00 | 3σ | 2,700 PPM | Marginal — Process spread exactly equals tolerance limits. Any shift creates scrap. |
| 1.33 | 4σ | 63 PPM | Industry Standard — Minimum requirement for general automotive and industrial parts. |
| 1.67 | 5σ | 0.57 PPM | Aerospace Standard — Mandated for Flight Safety Parts and Key Characteristics. |
| 2.00 | 6σ | 0.002 PPM (3.4 PPM with 1.5σ shift) | World-Class Six Sigma — Virtually defect-free manufacturing process. |
How to Conduct a Capability Study on a CNC Machine
- Verify Measurement System (Gage R&R): Before collecting parts, perform a Gage Repeatability and Reproducibility study. Your measurement system variation (%GRR) must be under 10% of total tolerance to yield valid capability statistics.
- Select Rational Subgroups: Sample consecutive parts (e.g., 5 parts every 2 hours over several shifts). Minimum standard sample size is 25 subgroups of 5 parts ($N = 125$ pieces total).
- Verify Normality: Test measured data for a normal Gaussian distribution using Anderson-Darling or Shapiro-Wilk tests. Non-normal distributions (common in true position, flatness, and runout) require Box-Cox power transformations or Weibull analysis.
- Calculate Control Limits: Plot data on an $\bar{X}$-$R$ control chart to confirm the process is in statistical control before calculating Cpk.
Prepare Inspection Characteristics for SPC in Minutes
BalloonPlot AI automatically flags critical dimensions, tolerance limits, and engineering notes on your drawings, exporting standardized inspection tables ready for statistical process capability analysis.