What is the Process Capability Index (Cp, Cpk)?
Cp and Cpk are core metrics in Statistical Process Control (SPC) and quality engineering. They compare the natural spread of a manufacturing process (six standard deviations of its output) against the engineering specification window defined by a lower specification limit (LSL) and an upper specification limit (USL). This is a universal statistical tool with no country-specific rules.
Cp describes the potential capability assuming the process is perfectly centered. Cpk describes the actual capability by also penalizing any offset of the process mean from the center of the spec window.
How to use this calculator
Paste or type your measured data values into the Data field, separated by commas, spaces, tabs or new lines. Enter the lower and upper specification limits in the same unit as your data. The calculator parses the list, computes the mean, variance, standard deviation, 3 sigma and 6 sigma, and then derives Cp and Cpk.
The formula explained
For data x₁..xₙ the mean is \(\mu = \frac{\sum x_i}{n}\). The population variance is \(\frac{\sum (x_i - \mu)^2}{n}\) and \(\sigma\) is its square root. Then:
$$C_p = \frac{\text{USL} - \text{LSL}}{6\,\sigma} \qquad C_{pk} = \min\!\left( \frac{\text{USL} - \mu}{3\,\sigma},\; \frac{\mu - \text{LSL}}{3\,\sigma} \right)$$Note this tool uses the population standard deviation (divide by n) so \(\sigma\) feeds the capability formulas consistently.
Guidance: Cp = Cpk = 1 means the +/-3 sigma spread exactly equals the specification width. A commonly recommended target is 1.33 (+/-4 sigma). Higher is better; values below 1.0 indicate the process is not capable. A 0 result for Cp/Cpk here means sigma = 0 (no variation) or an invalid specification range.
Worked example
Data = 45, 46, 44, 47, 43, 48, 45, 46, 44, 47 (n = 10), LSL = 40, USL = 50. Sum = 455, so \(\mu = 45.5\). The sum of squared deviations is 22.5, giving variance = 2.25 and \(\sigma = 1.5\). Then \(3\sigma = 4.5\), \(6\sigma = 9.0\).
$$C_p = \frac{10}{9} \approx 1.1111$$$$C_{pk} = \min\!\left[ \frac{50 - 45.5}{4.5},\; \frac{45.5 - 40}{4.5} \right] = \min[1.0,\; 1.2222] = 1.0$$Key Terms and Variables
| Term / symbol | Meaning |
|---|---|
| LSL — lower specification limit | The smallest acceptable value of the characteristic, in the measurement unit (mm, g, MPa, °C, …). Set by engineering on the drawing or specification. |
| USL — upper specification limit | The largest acceptable value, same unit as the measurement. |
| Tolerance width | \(\text{USL} - \text{LSL}\), the total width of the acceptable window, in measurement units. Sometimes written \(2T\) around a nominal target. |
| \(\mu\) — process mean | Average of the measured values, \(\mu = \frac{1}{n}\sum x_i\); same unit as the data. Estimated from the sample as \(\bar{x}\). |
| \(\sigma\) — standard deviation | Spread of the characteristic, same unit as the data. This tool uses the population form \(\sigma = \sqrt{\frac{1}{n}\sum (x_i - \mu)^2}\) (divide by \(n\)); the sample form divides by \(n-1\). |
| \(3\sigma\) and \(6\sigma\) spread | \(3\sigma\) is the half-width of the natural process spread (one side of the mean); \(6\sigma\), from \(\mu - 3\sigma\) to \(\mu + 3\sigma\), is the full natural spread, containing about 99.73% of a normal distribution. Capability compares this spread against the tolerance width. |
| \(C_p\) — potential capability | \(C_p = (\text{USL} - \text{LSL})/(6\sigma)\). Dimensionless ratio of tolerance width to natural spread; ignores where the mean sits, so it is the best capability achievable if the process were perfectly centered. |
| \(C_{pu}\), \(C_{pl}\) — one-sided indices | \(C_{pu} = (\text{USL} - \mu)/(3\sigma)\) measures room to the upper limit; \(C_{pl} = (\mu - \text{LSL})/(3\sigma)\) measures room to the lower limit. Either can be reported alone for a one-sided specification. |
| \(C_{pk}\) — actual capability | \(C_{pk} = \min(C_{pu}, C_{pl})\) — capability at the nearer limit, so it reflects both spread and centering. Always \(C_{pk} \le C_p\); negative if \(\mu\) falls outside a limit. |
| \(k\) — centering offset | Dimensionless fraction of the half-tolerance by which the mean misses the midpoint: \(k = \lvert m - \mu \rvert / \big((\text{USL} - \text{LSL})/2\big)\) with midpoint \(m\). Then \(C_{pk} = (1-k)\,C_p\); \(k = 0\) means perfectly centered. |
| \(P_p\), \(P_{pk}\) — performance indices | Same formulas as \(C_p\) and \(C_{pk}\) but computed with the overall (long-term) standard deviation of all the data, including between-subgroup drift. \(C_p\)/\(C_{pk}\) in the strict SPC sense use a within-subgroup estimate such as \(\bar{R}/d_2\). The gap between the two pairs is a measure of process instability. |
| Statistical control | A state in which the process shows only common-cause (random) variation — a stable mean and stable spread over time — so its future output is predictable. Confirmed with a control chart (e.g. X̄–R or individuals–moving range), not with a capability index. |
| DPMO / PPM | Defects per million opportunities / parts per million nonconforming: the estimated fraction outside specification multiplied by 1,000,000. For a normal, centered process it is derived from the normal tail areas, so it is a model-based estimate rather than a count. |
| Specification limits vs control limits | Specification limits (LSL, USL) are requirements set by engineering, the customer or a standard — they come from the design intent and never from the data. Control limits (LCL, UCL) are computed from the process data at ±3 standard errors of the plotted statistic and describe what the process actually does. They answer different questions and must never be mixed or plotted interchangeably: a point outside a control limit signals instability; a value outside a specification limit is a nonconforming part. |
FAQ
What is a good Cpk value? A value of 1.33 (corresponding to \(\pm 4\sigma\)) is a common target. Values below 1.0 indicate the process cannot reliably meet specification.
Why are Cp and Cpk different? Cp ignores centering, while Cpk drops as the process mean drifts away from the midpoint of the spec window. Cpk is always less than or equal to Cp.
What if all values are identical? Then \(\sigma = 0\) and the capability is mathematically infinite (zero variation), so this tool returns 0 as a guard against dividing by zero.