Process Capability

Cp = (USL − LSL) ÷ 6σ and Cpk = min(USL − x̄, x̄ − LSL) ÷ 3σ, where σ is the within-subgroup spread: the mean subgroup range divided by d2, or the mean moving range divided by 1.128 for individual readings. Pp and Ppk use the same formulas with the overall sample standard deviation, so they include shifts between subgroups. A one-sided limit gives Cpu or Cpl alone. Cpm adds the distance from the target. This is the AIAG / ISO 22514 convention; d2 and c4 are computed from their definitions and checked against the NIST e-Handbook factors.

Expected parts per million outside the limits come from a normal model on each sigma; the observed PPM is simply the share of your readings already outside. The histogram shows the readings against the limits so a shift or a second peak is visible at a glance. An Anderson–Darling test (with estimated mean and sigma) flags data that do not look normal — Cpk and PPM then describe a model your process does not follow, and a transformation or a different distribution is the right next step, not a bigger sample.

Not a certified SPC package and not a PPAP submission: the indices are only as good as the sampling behind them, fewer than about 30 readings give unstable values (the page says so), and no control-chart stability check is performed here — capability assumes a stable process. Measurements stay in your browser; the CSV is built locally; the same four anonymous usage counts as the rest of the site apply.