Process Capability Calculator

The Process Capability Calculator compares a process against its specification limits and reports how much of the tolerance the process actually uses. It returns the capability indices Cp and Cpk, the performance indices Pp and Ppk, the estimated defect rate in parts per million above and below each limit, the yield, and the sigma level.

You give it a process mean and two standard deviations, one within-subgroup and one overall, together with the upper and lower specification limits. The specification can be two sided, upper only or lower only, and the indices that are meaningful for your choice are the ones reported.

Two cautions carry the whole result. Every defect rate here assumes the characteristic is normally distributed, and for a skewed or bounded characteristic the predicted parts per million can be badly wrong even when the indices look healthy. And capability indices are meaningless for a process that is not in statistical control, which this tool does not test. Use a control chart first.

Purpose and model boundary

This calculator estimates normal-model process capability and performance for two-sided, upper-only, or lower-only specifications. It supports either directly entered mean/within/overall standard deviations or a grid of individual measurements. It reports applicable Cp/Cpk and Pp/Ppk components, estimated tail PPM and yield, centring for a two-sided specification, and shifted/unshifted sigma-level equivalents.

Capability indices describe a distribution relative to specification limits. They do not demonstrate statistical control, normality, measurement-system adequacy, or product conformance by themselves.

Inputs and units

Choose Enter mean and standard deviation to provide process mean mu, within-subgroup sigma sigma_w, and overall sigma sigma_o. Choose Compute from the sample grid to enter up to 30 individual measurements; at least five are required. Enter measurements contiguously from the top and leave the unused tail rows wholly blank. In grid mode the workbook calculates the sample mean and STDEV.S, then uses that one sample standard deviation as both sigma_w and sigma_o.

Choose Two sided, Upper limit only, or Lower limit only. USL and LSL use the same unit as the measurements. Both limits are required and must be ordered only for the two-sided mode; the irrelevant limit is hidden for a one-sided mode.

Governing relationships

For a two-sided specification,

• Cp = (USL - LSL) / (6*sigma_w);
• Cpu = (USL - mu) / (3*sigma_w);
• Cpl = (mu - LSL) / (3*sigma_w);
• Cpk = min(Cpu, Cpl);
• Pp = (USL - LSL) / (6*sigma_o);
• Ppu = (USL - mu) / (3*sigma_o);
• Ppl = (mu - LSL) / (3*sigma_o);
• Ppk = min(Ppu, Ppl).

In an upper-only mode the minimum index is the applicable upper component; in a lower-only mode it is the lower component. Inapplicable two-sided outputs are blank.

With z_upper = (USL-mu)/sigma_o and z_lower = (mu-LSL)/sigma_o, the workbook evaluates applicable standard-normal tails with the Zelen-Severo rational approximation, converts their sum to PPM, and calculates yield = 1 - upper_tail - lower_tail. For the named sigma outputs, the current workbook uses sigma_short = 3*Cpk and sigma_long = sigma_short - 1.5. The 1.5 difference is a convention rather than an observed change in performance.

For a two-sided specification, the centring ratio is

k = abs((USL + LSL)/2 - mu) / ((USL - LSL)/2).

Calculation sequence

1. Map the data-source and specification-type selections to workbook modes.
2. In grid mode, count numeric observations, calculate their arithmetic mean and sample standard deviation; otherwise read the three manual statistics.
3. Select the applicable limit branches and calculate within-sigma capability and overall-sigma performance indices.
4. Evaluate the applicable normal tails, PPM, yield, sigma equivalents, and the two-sided centring ratio.
5. Populate the distribution chart with density bars and applicable LSL/USL reference guides, then apply status precedence.

Outputs and interpretation

Cp/Pp express potential spread relative to a two-sided tolerance and assume centring. Cpk/Ppk use the nearer applicable limit and therefore include off-centring. The C indices use within-subgroup variation; the P indices use overall variation. In grid mode these pairs are equal by construction because one unstructured sample standard deviation feeds both. PPM and yield are fitted-normal estimates, not observed defect counts.

Validation and status logic

The workbook evaluates status in this order:

The zero-sigma test precedes the grid-count message; therefore an empty or constant grid can return the zero-standard-deviation message first. The single measurement column does not declare per-cell blank optionality: unused tail rows are accepted by the grid row rule, while an all-blank submitted table is refused because it would erase the model rather than request its defaults. Input visibility and selector values are also enforced before calculation.

Assumptions and limitations

Tail probabilities assume a stable, approximately normal process and trustworthy specification limits. The tool does not transform non-normal data or estimate confidence intervals. Grid mode contains individuals rather than rational subgroups and deliberately assigns the same STDEV.S to within and overall sigma; it cannot separate short- and long-term variation, so Cp equals Pp and Cpk equals Ppk there. Five values is only the workbook's computing minimum, not a claim of adequate evidence. One-sided results intentionally omit the opposite-side and width-based indices.

Restrictions and non-computing states

The workbook does not transform non-normal data, estimate confidence intervals, or separate short- and long-term variation in grid mode. Fewer than five usable values, zero variation, or invalid specification-limit combinations produce a non-computing state identified by the status message. A partial measurement grid may contain a contiguous block of values followed by blank capacity rows; omitting the grid requests the shipped defaults, while submitting an entirely blank grid is refused.

Errors and warnings

A rejected entry means a selector, value, or grid entry did not satisfy the published input rules. Workbook NOT VALID means the active mode cannot produce meaningful capability statistics. Workbook CHECK preserves computed values while flagging low capability or materially different variation estimates. A network or calculation-service failure is a service error, not a capability finding.

The workbook reproduces no table, chart or text from ANSI, ASQ, AIAG, ISO or any other standard. The normal distribution values come from a published closed-form approximation, cited below.

• Zelen, Marvin, and Norman C. Severo. Probability Functions, Chapter 26 in Abramowitz, Milton, and Irene A. Stegun, editors, Handbook of Mathematical Functions with Formulas, Graphs, and Mathematical Tables. National Bureau of Standards Applied Mathematics Series 55, 1964. Formula 26.2.17. https://personal.math.ubc.ca/~cbm/aands/page_932.htm (the rational approximation used for the standard normal cumulative distribution, and therefore for every parts-per-million figure on this page)
• Wikipedia. Process capability index. https://en.wikipedia.org/wiki/Process_capability_index
• Wikipedia. Normal distribution, section "Cumulative distribution function". https://en.wikipedia.org/wiki/Normal_distribution

The 1.5 sigma shift used for the short-term sigma level is an industry convention rather than a measured property, and it is an editable constant in the workbook. ANSI, ASQ, AIAG and ISO are trademarks of their respective owners, and none endorses this workbook.

Additional source notes migrated from Methodology

The workbook uses the standard Cp/Cpk/Pp/Ppk relationships, the Zelen-Severo normal-CDF approximation published in Abramowitz and Stegun 26.2.17, and an explicitly identified 1.5-sigma reporting convention. The dedicated References section contains method provenance and licensing details.