Results
Your results appear here. No data? Load the example to see how it works.
Paste your measurements and enter the spec limits. You get Cp, Cpk, Pp, Ppk, expected PPM and a normality check, with what each number means in plain English and the formulas behind it.
Your results appear here. No data? Load the example to see how it works.
A screenshot shows a result. Your customer wants the whole study, traceable and signed off, and you want to know what to do about it.
Standard definitions, nothing proprietary. Open any topic for the formula and what it means for your process.
On every change, automated tests compare this calculator with the NIST worked example and reject-rate table, Montgomery Examples 6.1 and 8.1–8.2, and the Mensuro statistics engine (built on SciPy and statsmodels). It must agree with the engine to at least 9 significant digits. See the tests.
Cp and Cpk use within (short-term) sigma: the variation from one part to the next, without slow drift or shifts.
σwithin = MR̄ / 1.128 (one at a time)
σwithin = R̄ / d2 (in groups; d2 depends on group size)
Cp = (USL − LSL) / 6σwithin
Cpk = min(USL − x̄, x̄ − LSL) / 3σwithin
MR̄ is the average difference between consecutive measurements; R̄ is the average range within each group.
Pp and Ppk use the standard deviation of every measurement, so they include drift and shifts over the whole study.
s = √[Σ(x − x̄)² / (n − 1)]
Pp = (USL − LSL) / 6s
Ppk = min(USL − x̄, x̄ − LSL) / 3s
Same formula, different sigma. Ppk is usually the one asked for in an initial PPAP study; Cpk shows what the process can do short-term. If Ppk is much lower than Cpk, the process is drifting or shifting over time.
Cp compares the tolerance width with the process spread and ignores where the process sits. Cpk measures from the mean to the nearer limit. A high Cp with a low Cpk means the process is capable but off-center, often an easy fix.
The verdict uses PPAP-style criteria on Ppk by default:
That's the default, Launch (PPAP). A short launch study usually looks better than months of production, so 1.67 up front leaves room for tool wear, material lots and shift changes. Once a process is approved, many customers accept 1.33 or more for ongoing production: choose Ongoing production. If your customer sets its own minimum, choose Customer minimum and enter it.
Expected PPM is the share of a normal distribution beyond each limit, in parts per million, shown for both sigmas:
PPM = [Φ((LSL − x̄) / σ) + 1 − Φ((USL − x̄) / σ)] × 1,000,000
Observed PPM counts the measurements actually outside the limits.
With 8 or more measurements, an Anderson-Darling test checks whether a normal curve fits, using the D'Agostino-Stephens small-sample adjustment. If p is below 0.05, the data may not be normal, and the indices and expected PPM should be read with caution.
Choose one at a time if each measurement is a separate part taken in sequence. Choose in small groups if you pulled several parts together at each check, for example 5 parts every hour, and enter the measurements group by group. This only changes the short-term sigma behind Cp and Cpk; Pp and Ppk are the same either way.
Leave the other limit blank. You get the one-sided index (Ppu or Ppl, Cpu or Cpl); Cp and Pp need both limits and aren't shown.
The process should be stable (in statistical control), and the measurements in the order they were taken, because the short-term sigma depends on order. Use 30 or more measurements for a dependable estimate; with fewer, the result carries a warning.
No. The calculation runs in your browser. The measurements and limits you enter are never sent to Mensuro or anyone else, and this page doesn't save them.