Cp and Cpk are qualification indicators used to compare the dimensional distribution in a stable production process to technical tolerances. Cp looks at the width of the distribution, while Cpk looks at the width as well as the position of the average relative to the tolerance limits. These values help to understand dimensional consistency; they do not alone indicate overall product quality or service life.
This approach can be used for critical diameter, connection size, or machining length in hydraulic cylinder production. However, meaningful results begin with the correct measurement system and appropriate data selection. Combining different models or different tolerances in a single pool can produce misleading numbers.
Technical checklist
| Indicator or condition | The subject under examination | Contribution to the decision |
|---|---|---|
| Cp | Distribution width and tolerance | Variability assessment |
| Cpk | Distance to the nearest border | Centering effect |
| Stability | Behavior over time | The representativeness of the analysis |
| Measurement system | Reliability of the result | Separating the actual change from the measurement effect. |
First, assess whether the process is stable.
Capability analysis requires an understanding of the statistical behavior of the process. Tool changes, setup shifts, or different material groups can affect the measurement distribution. Summarizing all data with a single mean and standard deviation without explaining these variations can obscure the truth.
It is important to know the date, machine, and product version to which the measurements relate. Data selection should not only include good results. Maintaining the measurement sequence makes it easier to see changes over time. The qualification value should not be used in place of the process monitoring record.
The main difference between Cp and Cpk
Cp compares the width between the upper and lower tolerance limits to six standard deviations. Cpk, on the other hand, considers the distance of the mean from the nearest tolerance limit. Even if the distribution is narrow, Cpk may decrease if the mean approaches a limit.
This distinction affects production decisions. If the distribution is too wide, variability needs to be reduced. If the distribution is narrow enough but off-center, process adjustment or the cause of the off-center is investigated. It is more useful to understand what the problem is rather than wanting a single “good value”.
An example of a hypothetical measure.
To illustrate the method, let's assume the lower limit of a measurement is 49.90 millimeters and the upper limit is 50.10 millimeters. With an average of 50.00 millimeters and a standard deviation of 0.02 millimeters, the tolerance width is 0.20 millimeters. Dividing this width by six standard deviations (0.12 millimeters) gives approximately Cp (Cp) = 1.67.
If the mean of the same distribution shifts to 50.06 millimeters, Cp remains unchanged; however, the distance from the upper limit decreases to 0.04 millimeters. Since three standard deviations are 0.06 millimeters, Cpk in this example is approximately 0.67. The calculation shows the effect of the center shift. The values are not actual Hidromas production data or the acceptance threshold for any product.
The measurement system comes before the result.
The resolution and calibration of the measuring instrument, operator influence, and the temperature distribution of the part can all have an effect. A definitive judgment about the process should not be made without separating the actual part change from the change reported by the measuring system. Measurement uncertainty is part of the technical decision.
Especially with tight tolerances, how the same part is measured by different people or devices becomes important. Method definition, contact points, and measurement conditions must be standardized. Collecting more data does not automatically correct systematic measurement error.
What should be asked in the supplier report?
Ask which measurement, tolerance version, and production period the report pertains to. Sample selection, data volume, and the calculation approach used should be explained. Results should not be converted into defect rate estimates without evaluating the suitability of assumptions such as normal distribution.
The acceptance threshold is determined according to customer and product requirements. The limit used in a project is not automatically transferred to different dimensions or applications. The report also does not replace the material, surface, or assembly suitability of the part; it only covers the analyzed feature.
How does the outcome lead to a decision to improve?
When centroid shift is observed, the relationship between setting and time is investigated; for wide distributions, process variables are examined. After correction, new data should be collected using the same method. The conditions for before and after comparison must match.
In the Hidromas product evaluation, you can define your dimensional consistency expectations along with critical drawing dimensions. Clarifying which characteristic and production scope the required evidence represents will make the technical quality discussion more concrete.
Distinguish between the tolerance limit and the control limit.
Technical tolerances define the design requirements of a product. Statistical control limits, on the other hand, are used to evaluate the observed behavior of process data. These are not created for the same purpose and are not interchangeable. A measurement remaining within the control limits does not necessarily mean it meets the technical tolerance. Similarly, a measurement within the tolerance does not, by itself, prove that the process has not drifted over time.
In reports where these two boundaries are shown on the same graph, the explanation is particularly important. The purchasing team must know which line represents customer requirements and which represents process behavior. Otherwise, natural change and technical nonconformity can be confused. Correct interpretation of the results is as important as the calculation of the numbers; a well-prepared graph presented with incorrect headings can reduce the quality of decision-making.
Evaluate the measurements while maintaining the chronological order.
If production data is sorted only from smallest to largest, time-dependent behaviors such as tool wear or adjustment changes may become invisible. Preserving measurements with production sequence, shift, and related process information facilitates research. Even in a dataset where the mean remains the same, the behavior of different periods can vary. Therefore, the record-keeping structure behind a single summary value should be examined.
For example, let's assume that measurements are collected at different locations before and after a tool change. A wider distribution can be obtained when the two periods are calculated together. This result may include not only the overall variability of the machine but also the effect of the adjustment application. During the research, it should be determined which groups are technically meaningful to evaluate together. Dividing the data should not be done solely for the purpose of obtaining better results.
Consider the sample size along with the confidence required for the results.
A qualification value calculated from a small number of measurements may be insufficient to represent the long-term behavior of the process. The amount of data, the sampling method, and the production conditions examined should be clearly stated in the report. A large amount of data alone is also insufficient; measurements collected only from the same short time interval may not cover different shift or material conditions.
The desired confidence level and analysis method should be determined by the quality team. A customer's acceptance threshold should not be presented as a mandatory standard for another project. Furthermore, an appropriate analysis approach may be required for unilateral tolerances, abnormal distributions, or other special circumstances. Unquestioningly applying conventional formulas to all data can produce results that appear correct but lack technical significance.
Monitor the improvement using the same measurement system.
If the measurement method is changed after the process setting is modified, it may become unclear which effect caused the observed improvement. Before and after data should be kept as comparable as possible in terms of equipment, method, and sampling conditions. If a change is necessary, its effect should be explained. The aim is not to demonstrate a higher Cpk number, but to ensure more consistent production of the critical measure.
The quality expectation prepared for Hidromas should specify the measurements to be analyzed, the relevant drawing version, and the evaluation period. The results should be read in conjunction with the measurement report and other product controls. This framework transforms statistical indicators from marketing claims into concrete tools for evaluating production consistency. When purchasing and quality teams use the same definitions, decisions made based on the report become clearer.
Frequently Asked Questions
Does high Cp always mean a suitable process?
No. It may have shifted to the average tolerance limit. Stability, Cpk, and the measurement system should also be examined. An odd number does not encompass the entire evaluation.
Does CPK indicate product lifespan?
It doesn't show it. It compares the tolerance with the distribution of a specific measurement. Different data such as load, design and usage are needed for product life.
Can the measurements of different products be combined in a single analysis?
They should not be combined without common statistical and technical justification. Different nominal values, tolerances, or processes may render the result meaningless.
Technical resources
1. NIST process capability and capability indices
2. NIST measurement uncertainty assessment






