The customer is requesting proof of process capability, the audit is in three weeks, and the question is: How do you go from the measured values to a Cpk that holds up? Many teams immediately start crunching numbers and overlook the real hurdle in the process. A Cpk is only as valid as the process from which it is derived, and it is precisely this validity that must be established before the first metric even makes sense.
The process capability study is more than just a calculation. It is a sequence of verifications that build upon one another: first, measurement capability; then stability; then distribution; and only at the very end, capability. If a step is skipped, it invalidates all subsequent steps. An auditor immediately recognizes a metric derived in this way and rejects it, regardless of how high the value may be.
This article guides you through all six steps of the process capability study, lists the three prerequisites and how to verify them, explains the required sample size, and shows why stability always comes before capability. By the end, you’ll know how to properly conduct a process capability study according to IATF 16949 and present a Cpk that will stand up to an audit.
THE MOST IMPORTANT POINTS AT A GLANCE
|
IN SHORTThe Process Capability Study (PFU) demonstrates that a manufacturing process is stable and capable over time. It begins by verifying the capability of the measuring equipment, then assesses stability using a control chart, and only then calculates the capability indices Cp, Cpk, Pp, and Ppk. The standard scope consists of at least 25 subgroups. An unstable process does not yield a reliable Cpk value. |
A process capability study, or PCS for short, is a procedure that verifies whether a process is capable of producing parts within the required tolerance over an extended period of time. Unlike a machine capability study, it considers the entire process along with all real-world influences: material, operator, tool changes, environment, and time.
The result is a set of capability indices. Cpk and Ppk are the relevant values because they account for the mean. The PFU is regulated in the automotive industry by IATF 16949 Section 8.5.1 and the AIAG-VDA SPC Guide.
The difference from machine capability is fundamental, yet it is often blurred. The MFU asks whether the machine is capable. The PFU asks whether the entire process is capable over time. A process may include a capable machine and still be incapable if the material, the operator, or the environment introduce too much variation. It is precisely these real-world influences that the PFU makes visible.
Before the first key figure is calculated, three conditions must be met. If these are skipped, the result is meaningless, no matter how accurately the calculation was performed.
| Requirement | Why it’s necessary | Verification |
|---|---|---|
| Capable measuring equipment | Measurement variation must not dominate the result | MSA, gage RR less than 10 percent |
| Stable process | Capability is meaningful only when the process is under control | Control chart without out-of-control points |
| Normally distributed data | The Cpk formula assumes a normal distribution | Distribution test, e.g., Anderson-Darling |
Of the three prerequisites, measurement equipment capability is most often overlooked. Yet it is the foundation: If the measurement equipment itself exhibits excessive variation, its variation becomes inextricably intertwined with the process variation. A gage RR above 30 percent renders any statement about capability meaningless, because it then becomes unclear whether the variation is due to the process or the measuring instrument. Verifying measurement capability is therefore deliberately given top priority.
The three prerequisites form a hierarchy. The measuring instrument comes first, because its variation is incorporated into every subsequent measurement. Second is stability, because without it, no performance indicator has predictive value. Third is the distribution, because it determines the choice of calculation method. Those who adhere to this hierarchy address the prerequisites in the most economically sensible order.
The PFU follows a set sequence. Each step builds on the previous one.
| Step | Action | Result |
|---|---|---|
| 1 | Verify measuring equipment capability | Gage RR passed |
| 2 | Collect subgroup data over time | Raw data |
| 3 | Maintain a control chart | Stability Confirmed |
| 4 | Check distribution | Normal distribution confirmed |
| 5 | Calculate Cp, Cpk, Pp, Ppk | Capability indices |
| 6 | Evaluate and document against requirements | Audit-ready documentation |
The fixed sequence of the six steps is not a mere formality, but a chain of dependencies. Each step provides the prerequisite for the next. Without capable measuring equipment, the raw data is unusable; without stability, the distribution test is meaningless; without a confirmed distribution, the Cpk calculation is incorrect. Skipping a step invalidates all subsequent steps.
In practice, it is worthwhile to use the six steps as a checklist and to mark off each step individually. This not only creates order but also provides evidence during an audit that the prerequisites were indeed checked in the correct order. Such a documented sequence distinguishes robust evidence from a collection of numbers compiled after the fact.
The sample size determines the reliability of the key metrics. Too few data points result in an unreliable estimate of the standard deviation and, consequently, an unreliable Cpk.
The AIAG-VDA SPC Guide recommends at least 25 subgroups. With a typical subgroup size of four to five parts, this results in 100 to 125 individual data points. It is crucial that the subgroups are distributed over a representative period of time so that the actual process drift is captured.
| Size | Subgroups | Individual values | Suitability |
|---|---|---|---|
| Minimum | 25 | 100 to 125 | Standard verification |
| Recommended | 30 and more | 150 and more | Critical features |
| Too small | less than 20 | less than 80 | Not durable |
The reason for the distribution over time is to capture actual drift. If all 125 values were collected within one hour, shift changes, batch changes, and tool wear would go unnoticed. The sample must include the factors that affect production in series operation; otherwise, the Cpk describes an ideal state that never actually occurs.
A common compromise in practice concerns the data collection period. If the sample is collected too quickly, the actual drift is not captured; if it is collected too slowly, the validation is delayed. A proven middle ground is to collect data over several shifts and at least one batch change, so that the most important sources of variation are included without unnecessarily prolonging the detection process.
The analysis begins with the control chart, not with the capability calculation. The short-term variation is estimated from the mean range of the subgroups, and the long-term variation is estimated from the standard deviation, which includes all individual values. Only then are the key metrics calculated.
The control chart is not a tedious intermediate step. It is the proof that the Cpk actually means something. Without proven stability, the Cpk merely describes a random state.
Amadeus Lederle, Chief Technology Executive, CSP Intelligence GmbH
A common mistake in the analysis lies in the choice of dispersion measure. If the total variation is used for the Cpk, it will be too low. If the short-term variation is used for the Ppk, it will be too high. Both metrics require their respective correct sources of variation, and a proper control chart analysis provides both separately.
A process may appear capable but still be out of control. If systematic factors cause the mean to fluctuate, the calculated Cpk may happen to be high by chance, even though the process is not predictable. Such a Cpk is merely a snapshot with no predictive value.
That is why the order is stability, then capability. Only when the control chart shows that only random variation is present does the Cpk describe a consistent characteristic. This is why auditors ask to see the control chart first and only then inquire about the Cpk.
PRACTICAL NOTEA high Cpk from an unstable process is more dangerous than a low one from a stable process. The high value creates a false sense of security that does not exist. Always check the control chart first for out-of-control points, trends, and runs before interpreting the Cpk. |
Stability is verified using defined test criteria on the control chart. In addition to points outside the control limits, prolonged one-sided runs, sustained trends, and unusual patterns are also considered signs of a lack of process control. Only when none of these criteria are met is the process considered stable and the Cpk considered meaningful.
One point that is often overlooked is the handling of outliers. A single point outside the control limits must not simply be deleted to establish stability. It requires a root cause analysis. Only when a clear, eliminated special cause has been verified may the data point be excluded from the capability calculation, and even this must be documented. Otherwise, the Cpk is being artificially inflated in an unacceptable manner.
A one-time PFU serves as proof; continuous PFU acts as an early warning system. The CSP Manufacturing OS automatically maintains control charts, checks stability in real time, and continuously calculates Cpk and Ppk. If a process falls outside the capability limits, the IPM module triggers an alarm before scrap is produced. Every inspection remains traceable back to the component via the serial number and is archived in an audit-proof manner.
The shift from a one-time to a continuous PFU is also the shift from reaction to prevention. A one-time inspection confirms a current state. A continuous inspection detects changes as they occur. Only then does the capability verification become a tool that prevents scrap rather than merely documenting it.
A process capability study (PFU) demonstrates whether a process is stable over time and capable of producing a characteristic within the tolerance. It examines the entire process and is evaluated using Cp, Cpk, Pp, and Ppk.
The AIAG-VDA SPC Guide recommends at least 25 subgroups. With a subgroup size of four to five, this results in 100 to 125 individual data points, distributed over a representative time period.
The MFU evaluates only the machine under ideal conditions using Cm and Cmk. The PFU evaluates the entire process over time using Cp, Cpk, Pp, and Ppk and takes all real-world influences into account.
Capability indices require a controlled process. In an unstable process, the Cpk value merely describes a random state and has no predictive value. That is why proof of stability comes first.
Three prerequisites: a verified, capable measuring instrument; a stable process as indicated by the control chart; and normally distributed data. All three must be met before the capability indices can be calculated.
In the automotive industry, process capability studies are governed by IATF 16949 Section 8.5.1 and the AIAG-VDA SPC Guide. For distribution-related methods, ISO 22514 also applies.
For non-normally distributed data, the classic Cpk formula yields incorrect values. In such cases, a distribution transformation or a distribution-free method in accordance with ISO 22514 is used.