Skip to main content
    QHSEtech
    Updated
    All glossary terms
    Content date: · No individual check logged· AI use

    Process Capability

    Also known as: Process capability index, Cp and Cpk, Capability study

    Definition

    How the natural spread of a stable process compares with its specification limits, usually stated as indices such as Cp and Cpk worked out from measured data.

    Editorial definition: The QHSE Standard, updated 3 October 2026 · https://qhsetech.com/glossary/process-capability

    Scope: An index means little unless the process is in statistical control, the data are close to normal and the sample is large. Thresholds such as 1.33 come from customers or contracts, not from the definition, and the texts of ISO 22514 and the AIAG & VDA SPC Manual were not read.

    Example for your evaluation: Follow one study from the chart that showed stability, through the normality check and sample size, to the index, its confidence interval and the customer's threshold.

    Sources checked: . Examples are editorial illustrations.

    What the NIST/SEMATECH e-Handbook says about Cp, Cpk and the conditions behind them

    Section 6.1.6 of the NIST/SEMATECH e-Handbook of Statistical Methods defines process capability as a comparison between the output of an in-control process and its specification limits: the distance between the upper and lower limits set against the spread of the process, taken as six standard deviations. Cp is that ratio alone. Cpk takes the smaller of the two distances from the mean to a limit, each divided by three standard deviations, so it drops as the process drifts off centre and can never exceed Cp. Cpm also penalises distance from a target value, and Cpu and Cpl serve a one-sided specification. In the handbook's worked example, with specification limits 8 and 20, mean 16 and standard deviation 2, Cp is 1.0 but Cpk only about 0.67.

    Its glossary defines the Cpk index for a stable process, measured against common-cause variation, and section 6.1.6 speaks of an in-control process. The indices assume normally distributed data; for data that are not, the remedies it names are transforming them, with Box-Cox as the popular choice, or using non-parametric versions built from percentiles. Estimates generally need around 50 independent values to be valid. The confidence-limit approximations need at least 25, capability studies in general need 100 or more, and the section warns that the indices are themselves random. The glossary adds that it takes at least 33 distinct repetitions to bring a 95% interval for Cp down to about a quarter of the estimate either side, and that roughly the same holds for Cpk.

    The handbook's table turns Cp into expected rejects for a centred, normal process: 0.27% at a Cp of 1.00, 64 parts per million at 1.33, 0.6 ppm at 1.66 and 2 parts per billion at 2.00. In the glossary, a Cpk above 1.0 makes a process capable, while 1.33, 1.5 and 2.0 appear as stricter criteria that some promote, mostly reserved for the label manufacturable. The worked example calls its own process poor and names less spread, better centring or both as the cure. The handbook pages read here do not define Pp or Ppk.

    ISO's catalogue lists ISO 22514-2:2026, edition 3, published in February 2026. Its summary says it sets out statistics for estimating capability or performance particularly when a process does not stay in statistical control, sorts results into eight distribution types and covers continuous characteristics only. AIAG's page for the first edition of the AIAG & VDA SPC Manual lists machine capability, preliminary process performance and ongoing process capability studies. Both are sold and neither was read. For US medical devices, 21 CFR 820.250, in the eCFR text in force on 1 January 2026, required makers, where appropriate, to keep procedures for choosing valid statistical techniques to establish, control and check the acceptability of process capability and product characteristics. The current part 820, the Quality Management System Regulation, has no such section and incorporates ISO 13485:2016 by reference. This is general information, not regulatory advice.

    What we could not verify on 22 October 2026: a public primary source that defines Pp and Ppk; how ISO 22514-2:2026 and the AIAG & VDA manual compute their indices or set any threshold; and whether, and where, ISO 13485:2016 covers what section 820.250 asked for.

    Sources opened :

    What a process capability study record holds in software

    • The characteristic, its specification limits and any target, with the drawing or specification revision they came from.
    • The control chart that showed stability over the study period, with any signals and their causes.
    • The data: number of values, subgroup size and how they were sampled, with the gauge and its calibration status.
    • The normality check made and, where the data were transformed or a non-parametric index used, which one.
    • Each index reported, such as Cp, Cpk, Pp, Ppk or Cpm, with the standard deviation behind it and the formula or software version.
    • The confidence interval, or at least the sample size, behind every estimate.
    • The acceptance threshold, where it comes from (customer, contract or internal) and the decision taken.

    Capability versus control, and capability versus performance indices

    Control asks whether a process is predictable; capability asks whether that predictable spread fits the specification. A process can be in control and not capable, and an index computed from an unstable process predicts nothing. The handbook's estimators use the sample standard deviation s of the data, so a figure called Cpk in one report may be worked out the way another report works out Ppk: ask which standard deviation was used. A gauge capability study, in the glossary's metrology terms, looks at the measuring instrument, not the process.

    A hypothetical example: a clip with enough spread margin but off centre

    The firm, the clip and every figure here are fictional. A moulder makes a plastic clip whose latch width has a specification of 4.90 to 5.10 mm, and the customer's contract asks for a Cpk of at least 1.33 on that feature. The quality engineer takes 25 subgroups of 5 clips, or 125 values, while the chart shows no signals, and a normal probability plot shows no clear departure from normality. The mean is 5.03 mm and the standard deviation 0.020 mm. Cp is 0.20 ÷ 0.12, or 1.67, but Cpk is the smaller of 0.07 ÷ 0.06 and 0.13 ÷ 0.06, or 1.17. With the handbook's normal approximation, the 95% interval for that Cpk runs from about 1.01 to 1.32.

    The spread is fine and the centring is not. The toolmaker adjusts the cavity, and a fresh run of 125 values, again from a stable chart, gives a mean of 5.005 mm and a standard deviation of 0.021 mm. Cp is now 1.59 and Cpk 1.51, with an interval of about 1.31 to 1.70. The report gives the estimate, the interval, the 125 values and the standard deviation used, so the customer can see that the lower end sits just under 1.33 and apply its own rule.

    Common misunderstandings about process capability

    • An index is not a pass mark for an unstable process. The handbook defines the indices for a process in statistical control.
    • A high Cp does not mean the parts conform. Cp ignores where the mean sits, while Cpk takes it into account.
    • 1.33 is not a universal requirement. The handbook's glossary calls a process capable above 1.0; the threshold that binds you is the one your customer or contract sets.
    • Thirty values do not give a precise index. They fall short of the handbook's sample sizes for valid estimates and for capability studies.
    • The reject figures assume a centred, normal process. An off-centre process rejects more than its Cp suggests.

    How process capability relates to other glossary terms

    Capability is judged on a stable process against a stated requirement. These terms cover the process, the measurements and the requirement.

    Records and evidence that usually exist for process capability

    • Capability study reports with the data, indices, intervals and decision.
    • The control charts covering each study period.
    • Gauge calibration records, and any gauge study, for the measurements used.
    • The specification and drawing revisions the limits came from.
    • Customer or contract requirements naming the index and its threshold.
    • Repeat studies after a change of process, tool or material.

    What a software tool needs to capture for process capability

    • Indices computed from stored raw values, not typed in by hand.
    • A label on each index for the standard deviation behind it, within-subgroup or overall.
    • A stability check, such as the chart for the study period, attached to each study.
    • A normality check and a record of any transformation applied.
    • A confidence interval, or at least the sample size, shown beside every index.
    • Thresholds held per characteristic and customer, with their source.
    • Export of data and results so that a customer can recompute them.

    Catalog profiles documenting SPC or process-capability analysis

    The list uses the same rule as the statistical process control software guide, so its entries match the SPC page. It does not show that a product labels the standard deviation behind each index, reports confidence intervals or checks normality, which should be tried with your own data set.

    Same rule for every product: tagged “Statistical Process Control (SPC)”, “SPC”, “Manufacturing Quality”, “Quality Inspection” or “Quality Management”, with a sourced SPC or process-capability capability in its record, or one the vendor stated to us, labelled as vendor-stated. Documented and conditional capabilities qualify; check the linked scope before treating a module as included.

    Tekmon pays for a sponsored placement (above); its place in this list follows the same rule as every entry. How lists are ordered

    All 4 matching profiles are shown. Documentation review, not hands-on testing or a claim of compliance.

    Related software

    Frequently asked questions

    What is the difference between Cp and Cpk?

    In the NIST handbook, Cp sets the width of the specification against six standard deviations of the process and ignores where the mean sits. Cpk uses the distance from the mean to the nearer limit over three standard deviations, so it is never larger than Cp and equals it only when the process is centred. The handbook's own example shows a Cp of 1.0 with a Cpk of about 0.67.

    What Cpk value is good enough?

    No single figure appears in the sources read. NIST's glossary draws the line for capable at a Cpk over 1.0, and adds that some push 1.33, 1.5 or 2.0, values it mostly keeps for a manufacturable process. Customers, contracts and sector manuals set their own thresholds, and an estimate from a small sample can sit well away from the true value.

    Are Pp and Ppk the same as Cp and Cpk?

    Not quite, and the handbook pages read define Cp, Cpk and Cpm but not Pp or Ppk. ISO 22514 and the AIAG & VDA SPC Manual cover capability and performance studies, but both are paid and were not read. A common convention computes Pp and Ppk from the overall standard deviation of all values, so ask which definition your customer applies.

    More quality terms

    Related guides