Cp / Cpk calculator
Paste individual measurements, enter your spec limits, and get process capability indices the way Minitab computes them. Nothing is uploaded.
| Sample size (n) | 30 | Mean | 10.0080 |
|---|---|---|---|
| σ within (MR̄/d2) | 0.0348 | σ overall (sample SD) | 0.0262 |
| Cp | 1.435 | Pp | 1.910 |
| Cpk | 1.358 | Ppk | 1.808 |
| CPL / CPU | 1.511 / 1.358 | PPL / PPU | 2.012 / 1.808 |
| Expected PPM (within) | 25.9 | Expected PPM (overall) | 0.0 |
Need a capability histogram, subgrouped data, Box-Cox, or a control chart first? Open SpcLite →
How Cp and Cpk are calculated
Cp compares the width of the specification to the spread of the process. Cpk also accounts for where the process is centered, so it is never larger than Cp.
Cp = (USL − LSL) / (6σ)CPU = (USL − μ) / (3σ)andCPL = (μ − LSL) / (3σ)Cpk = min(CPU, CPL)— with a one-sided spec, Cpk is the index for that side
Cpk vs Ppk: within vs overall σ
Cp and Cpk use the within standard deviation — for individual measurements, the average moving range divided by d2 = 1.128. They describe the potential of a process if it stayed stable. Pp and Ppk use the overall sample standard deviation, so they include drift and shifts between subgroups. A Ppk much lower than Cpk is a sign the process is not in statistical control.
What is a good Cpk?
| Cpk | Expected defects (centered, both sides) | Typical interpretation |
|---|---|---|
| < 1.00 | > 2,700 PPM | Not capable — the process produces out-of-spec parts |
| 1.00 | ≈ 2,700 PPM | Barely capable; any shift creates defects |
| 1.33 | ≈ 63 PPM | Common minimum for ongoing production (AIAG / PPAP) |
| 1.67 | ≈ 0.6 PPM | Common requirement for new or safety-critical characteristics |
| 2.00 | ≈ 0.002 PPM | Six Sigma level (before the 1.5σ shift convention) |
Before you trust the number
- The process should be stable: check an I-MR or Xbar-R control chart before reporting capability.
- The data should be roughly normal; skewed data needs a Box-Cox transform or a non-normal model.
- The gage should be adequate — run a Gage R&R so measurement error does not inflate σ.
- Use at least 30, ideally 100+, measurements; Cpk estimates from small samples are wide.