Can PPAP Be Submitted with a Ppk of Only 0.83? —— Five-Step Method for Sampling and Judging Initial Process Studies

A machining company submitted its PPAP, which was returned the next day with the client's sole reason: "The sampling for the initial process study is invalid." The engineer was indignant: 100 pieces of data, an Excel report, and a Ppk of 1.42—nothing was missing. Upon further inquiry, it was revealed that the 100 pieces were cobbled together from three machines, two shifts, and two batches of materials, including about 20 pieces of machine adjustment and trial cutting parts. The data was genuine, but the conclusion was invalid—this is the most common way PPAP initial process studies fail.

1. PPAP Requires Ppk, Not Cpk

The initial process study focuses on the special characteristics on the drawing, aiming to prove that "this process can consistently produce conforming products" before mass production release. This is different from the Cpk calculated after the process is stabilized by SPC.

Ppk uses the overall standard deviation s, which includes all variations within and between subgroups when all data is considered together; Cpk uses the within-subgroup standard deviation (such as R̄/d2), reflecting only the instantaneous dispersion of the current small segment. When the process has not been stabilized and the variation between subgroups has not been eliminated, the calculated Cpk will overstate the capability. Therefore, the AIAG PPAP manual only recognizes Ppk in the initial study phase.

A simple way to remember: Cpk is the "capability after the process is controlled," while Ppk is the "performance as the data is collected." The client wants the latter.

The two are not mutually exclusive. After the batch is processed, once the control chart stabilizes and the variation between subgroups is eliminated, recalculate the Cpk and compare it with the initial Ppk—narrowing the gap indicates that the process has truly stabilized; if the gap remains, it suggests that the initial variation between subgroups has not been fully resolved, only temporarily suppressed. This comparison is more convincing than a single index.

2. Sample Size and Sampling Window: How to Make It Count

A minimum of 100 pieces is required, and it is recommended to arrange them into at least 25 subgroups, with no fewer than 4 pieces per subgroup. More important than the quantity is the four "samenesses" in sampling: the same equipment, the same shift, the same batch of materials, and the same set of parameters. The production must be continuous, and the pieces must be numbered in the order they are produced; no picking or excluding "bad-looking" points.

Machine adjustment pieces, trial cutting pieces, and first pieces cannot be included in the sample. Data from before and after mid-process tool changes, material batch changes, or long production stops cannot be combined into a single study. For multi-cavity molds, multi-station equipment, and parallel machines, each cavity, station, and machine should be studied separately—mixing data from six mold cavities into 100 pieces averages out the differences between groups, masking the very issues that need improvement.

There are indeed cases where 100 pieces cannot be produced, such as single-piece small batches, bulk materials, or products that can only undergo destructive testing. In such cases, do not fabricate a set of beautiful data to get by. The correct approach is to agree on an alternative plan with the client in advance: using capability data from a similar process, combining key parameter process data with 100% inspection, or clearly stating the alternative monitoring method in the control plan. The premise is the same—clarify the criteria and get the client's approval, rather than leaving it to the auditor to guess.

3. Look at One Ratio First to Identify the Problem Direction

For a certain machined part, the critical dimension is 25.00±0.05mm, and 100 pieces were continuously sampled. The overall standard deviation s was calculated to be 0.0152mm, with the mean biased towards the upper tolerance limit by 0.012mm, resulting in a Ppk of 0.83. The within-subgroup standard deviation s内 for the 25 subgroups was 0.006mm.

The difference between these two numbers points to the problem direction. If s内 is also close to 0.015, it indicates that the dispersion comes from random variations, and the improvement focus should be on tools, parameters, and incoming material consistency. Since s内 is only 40% of the overall standard deviation, the parts within the same subgroup are very similar, but the groups themselves are drifting, indicating issues with the positioning datum or fixture wear, tool change cycles, and thermal deformation affecting the center. Adjusting parameters forcefully will be futile; the focus should be on fixture positioning and tool change strategies.

Focusing solely on whether the Ppk passes can easily misinterpret "group drift" as "too much dispersion," wasting months of effort.

4. Criteria: 1.67, 1.33, and Below 1.33

According to common practice (follow the client's specific requirements if they are clearly defined):

• Ppk ≥ 1.67: Acceptable, proceed directly to the approval process.
• 1.33 ≤ Ppk < 1.67: Typically requires client review, and may necessitate the submission of a corrective action plan first.
• Ppk < 1.33: Generally not accepted, requiring 100% inspection or equivalent containment, along with a clear improvement plan, to be processed according to the client's temporary approval requirements.

The subsequent handling of the company mentioned above is worth learning from: they did not hide the 0.83, but instead attached three items to the report—100% inspection work arrangements and measurement tool capability explanations, improvement plans and timelines for group drift, and a commitment to re-sample after improvements. The client approved the temporary release, and 60 days later, the Ppk was re-measured at 1.47, transitioning to normal monitoring. The cost of hiding data is not just a single return but a loss of the client's trust in the supplier.

The criteria also depend on the nature of the characteristic. The same 1.33, when applied to safety or regulatory parts, typically results in stricter client requirements, even rejecting temporary approval; when applied to general appearance or assembly dimensions, clients often provide an improvement window. Clarifying the category of the characteristic with the client before submission saves a lot of effort compared to repeated explanations afterward.

5. Five-Step Method for Implementation

1. Define Scope: Identify all special characteristics from the PFMEA and control plan, list them in a study checklist, and prepare a separate report for each.
2. Set Timing: Schedule the study during the trial production phase, after process parameters are fixed; data collected before this point is only for process exploration and cannot be used as submission evidence.
3. Collect Data: Continuously produce more than 100 pieces, record them in time order, and include information on equipment, shift, material batch, parameters, and measurement tools. The original data must be traceable.
4. Calculate and Evaluate: Simultaneously calculate Ppk, Cpk, and the within-subgroup and overall standard deviations. First, explain the differences clearly, then draw conclusions based on the criteria.
5. Close the Loop: Archive items that meet the criteria and upgrade them to SPC monitoring for mass production. Items that do not meet the criteria should enter a cycle of containment, improvement, and retesting until the client closes the issue.

6. Three Pitfalls

1. Selecting "Good Data" as Samples: Conveniently deleting a few outliers, this data cannot withstand client scrutiny.
2. Reporting Only One Index Without Original Data and Calculation Criteria: The client cannot verify how s was calculated or how subgroups were divided, making the submission essentially invalid.
3. Using 100% Inspection as a Solution: Containment only buys time; without an improvement plan, the next submission will yield the same result.

The initial process study is not just an Excel report but a defense of whether "this process can work." Valid sampling, clear criteria, and straightforward handling of nonconformities will lead to faster client approval and more stable mass production.

The initial process study is not about submitting 100 pieces of data but providing a chain of evidence that the process "can work."

Knowledge code: 8.3.3

Version: v20260917

Author: QTank QTank is dedicated to providing systematic professional knowledge, methodologies, and practical tools for quality management practitioners, helping companies continuously improve their quality capabilities.