Process Capability (Cp / Cpk) Calculator

Enter specification limits, process mean, and standard deviation to compute Cp, Cpu, Cpl and Cpk, with the estimated nonconforming fraction (ppm) from the normal distribution — the core continued-process-verification statistic for ICH Q10.

WHAT THIS CALCULATES

Cp and Cpk — how much of your specification width the process actually uses, and how well centred it is within that width — plus the nonconforming rate those figures imply. Cp asks whether the process could fit; Cpk asks whether it does.

THE METHOD

USL

upper specification limit

LSL

lower specification limit — leave either blank for a one-sided specification

mean

the process mean

σ

the process standard deviation

Cpk

the lower of the two one-sided indices — capability as actually centred

Which σ you enter decides what you are computing. The within-subgroup (short-term) σ gives Cp/Cpk — potential capability. The overall (long-term) σ gives Pp/Ppk — actual performance including drift between batches. The arithmetic is identical; the meaning is not, and reporting one while having calculated the other is the most common error in this analysis. The nonconforming ppm is estimated from the normal distribution.

THE INPUTS, AND WHAT THEY MEAN

Upper / lower specification limit

The registered or established limits — not internal alert levels, and not the range your data happens to span. Leave one blank for a one-sided specification, which is common for impurities and dissolution.

Process mean

The average of the characteristic across the dataset you are assessing. Use enough batches to be meaningful; a handful gives an estimate with wide uncertainty that the index will not show you.

Standard deviation (σ)

Within-subgroup σ for Cp/Cpk, overall σ for Pp/Ppk. Decide which question you are answering before you pick, and label the output accordingly.

HOW TO READ THE OUTPUT

• ›Cp far above Cpk means the process is capable but off-centre — the fix is centring, which is usually cheaper than reducing variation. Cp and Cpk close together means the process is centred and the spread itself is the constraint.
• ›The commonly cited targets are Cpk ≥ 1.33 as capable and ≥ 1.0 as marginal. These are industry convention from SPC practice, not regulatory limits — no GMP regulation specifies a Cpk.
• ›The ppm estimate assumes an in-control, approximately normal process. Skewed data, mixtures of populations, or a process still drifting will make the estimate optimistic, sometimes badly so — check normality before quoting it.
• ›Capability on an out-of-control process is meaningless. Establish stability first; an index computed across a process that shifted mid-dataset describes nothing that exists.

WORKED EXAMPLE

An assay with a 95–105% specification, running at a mean of 101% with a standard deviation of 1.5%.

Specification

LSL 95 · USL 105

Process mean

101

Standard deviation

1.5

Cp of 1.11 says the spread would fit inside the specification. Cpk of 0.89 says it does not, because the mean sits 1 unit above centre and the upper side is the constraint — Cpu 0.89 against Cpl 1.33. Almost all of the estimated 3,862 ppm is failing high. Centring the process would move Cpk close to Cp without reducing variation at all, which is the cheaper intervention and the one the two indices together point at.

REGULATORY BASIS

FDA Process Validation Guidance (2011)

Stage 3 continued process verification expects ongoing assurance that the process remains in a state of control — capability indices are one accepted way to evidence that trend.

ICH Q10 — Pharmaceutical Quality System

Requires monitoring of process performance and product quality, and its use as an input to continual improvement.

ASTM E2281 / AIAG SPC conventions

The source of the Cp/Cpk definitions and the within-subgroup versus overall σ distinction applied here.

LIMITATIONS — READ BEFORE YOU RELY ON THIS

• ›This is an analytical aid, not a validated system. Reproduce the calculation in your own statistical software before it supports a continued-process-verification conclusion — the formulas and the normal-distribution assumption are published above so you can.
• ›It assumes an in-control, approximately normal process, and does not test either assumption. Non-normal data can produce a comfortable Cpk over a process that fails regularly.
• ›It takes a mean and a σ, not your raw data, so it cannot detect trends, shifts, outliers or autocorrelation — all of which matter more than the index.
• ›It does not distinguish Cp/Cpk from Pp/Ppk for you. That distinction lives entirely in which σ you entered, and the output is labelled the same either way.

RELATED ON SPEQ