Machining process capability Cpk: what it measures and how to run a valid study
Cpk compares the spread of a machined feature to its tolerance band, then subtracts points for being off target. This page shows how to collect data that survives an audit, which benchmark to use, and how the number should change your inspection plan. Written for engineers and buyers who receive capability reports, not just produce them.

In this article
- 1
- 2
- 3
- 4
- 5
- 6
- 7
- 8
Key takeaways
What machining process capability Cpk actually measures
Cpk answers one question: how many standard deviations fit between the process mean and the nearest specification limit. The formula is Cpk = min[(USL − mean), (mean − LSL)] / (3σ). USL and LSL are the upper and lower limits from the drawing, and σ is the process standard deviation. A Cpk of 1.33 means the nearer limit sits four standard deviations from the mean. It is a ratio, so it has no units.
The number is only as good as the σ under it. If σ comes from twenty consecutive parts turned on one bar, you measured repeatability, not capability. Cpk is meant to describe what the process will do over a full tool life, several bars, and more than one operator. Collect data that way and the index starts to mean something.
On the shop floor, Cpk is a planning tool. It tells you how much room the process has before a feature hits a limit, and therefore how often to measure. It does not tell you the process is in control. That is what the control chart is for.
For a feature turned to Ø25.000 +0.021/0 mm with σ = 0.003 mm, the spread alone gives Cp = 1.17. If the mean sits at 25.018 mm, Cpk drops to about 0.33. The process is tight and badly aimed. That is the case Cpk is built to catch.
Cp versus Cpk: why centering decides the result
Cp uses the same denominator and ignores where the mean sits: Cp = (USL − LSL) / (6σ). A process can post Cp 2.00 and Cpk 0.80. That combination means the operator has plenty of spread margin but is running near one limit, usually because a tool offset was set once and never checked against the nominal.
The fix for a low Cpk with a high Cp is not a slower program or a new machine. It is a tool offset change. Move the mean back to the middle of the band and Cpk climbs toward Cp immediately. We see this most on bores and shaft diameters where the operator dials in the first part and then walks away.
When Cp and Cpk are both low, the spread is the problem. Look at rigidity, tool runout, coolant stability, and fixture repeatability. Adding a spring pass or a finish pass at 0.1–0.2 mm depth of cut often narrows the spread more than a parameter tweak.
Always put both numbers on the report. A supplier who quotes Cpk alone is telling you half of what you need to know about the process.
Benchmark values and what they mean in production
The common benchmarks are 1.00, 1.33, and 1.67. At 1.00, roughly 0.27% of parts are expected to fall outside the limits, which is marginal for anything shipping in volume. At 1.33, expected fallout is about 63 ppm, and that is the usual minimum for production. At 1.67, it drops to around 0.6 ppm, the level expected on safety-critical or high-value features.
Automotive and aerospace customers often write 1.33 minimum into the purchase order and 1.67 on critical characteristics. Medical work frequently lands in the same range. Check the drawing and the quality agreement before you quote a number back to a customer.
These figures assume a normal distribution and a stable process. Real machining data is often not normal. A turned diameter may show a slight skew from tool wear, and a reamed hole may be truncated because the reamer cannot cut below its own size. When the distribution is not normal, discuss the assumption with the customer rather than reporting a percentage that will not hold.
Bored features held to ±0.005 mm sit near the edge of what a standard CNC process can deliver reliably. On those, we plan the process around 1.67 instead of 1.33, because the margin between a good part and a scrap part is small.
Sampling and data requirements for a valid study
A credible study needs 25 to 30 subgroups of 3 to 5 parts each. Collect them across the real production sequence, not from one uninterrupted run. Samples should span at least one full tool life and one shift change. If the job runs across two machines, keep the data separate.
Only consecutive parts from one bar measure short-term repeatability. That data excludes lot-to-lot variation, operator changes, tool changes, and thermal drift from a cold start. It will inflate Cpk, sometimes by a full point, and the number will not survive the first month of production.
Measurement system matters as much as the parts. A gauge with resolution of 0.01 mm cannot resolve a σ of 0.003 mm. Run a gauge R&R before the study. If the gauge contributes more than 10% of the tolerance, fix the measurement first, or the Cpk figure describes the gauge, not the process.
Record the conditions next to the numbers: machine, program revision, tool grade, coolant, bar lot, and ambient temperature. Without that, nobody can reproduce the study when a feature starts trending.
Why machining processes drift: wear, heat, and fixturing
Carbide inserts in steel can wear 0.02 to 0.05 mm on diameter over a 30 to 60 minute tool life. That walks a bore steadily toward its upper limit. A process that looked centered at 8 a.m. is off center by 10 a.m. Cpk calculated from the morning parts only will overstate what the afternoon produces.
Spindle and ballscrew thermal growth can move a feature 0.005 to 0.015 mm in the first hour of a shift. Bar stock diameter tolerance of ±0.05 mm changes the depth of cut and therefore the cutting force. Coolant temperature swings shift the whole machine frame. None of this shows up in a twenty-part study.
Fixture clamp variation is the quiet one. A soft jaw that was bored on the machine is repeatable. A jaw swapped in from a shelf is not. If Cpk drops after a fixture change, check the jaw setup before you touch the program.
Cpk and Ppk answer different questions. Cpk uses within-subgroup σ and reflects short-term capability after special causes are removed. Ppk uses the overall standard deviation of all points and includes drift between subgroups. A turning cell with Cpk 1.67 and Ppk 1.10 is stable inside each subgroup but drifting across the shift. That is a tool-life or thermal problem, not a precision problem.
Step by step: running a Cpk study on a machined feature
- 11. Fix the drawing revisionConfirm the tolerance, datum scheme, and any modifier from the current drawing revision before any data is collected. A study run to an obsolete tolerance is wasted work.
- 22. Qualify the gaugeRun a gauge R&R on the feature. Resolution should be at least one tenth of the tolerance band. If the gauge contributes more than 10% of tolerance, change the gauge or the method first.
- 33. Define the subgroupsPlan 25–30 subgroups of 3–5 parts. Spread them over a full tool life and a shift change. Log machine, program revision, tool grade, and bar lot beside each subgroup.
- 44. Machine under production conditionsRun the actual program, fixture, and coolant. No hand feeding, no trial speeds, no operator hovering with an offset change mid-run. Note every offset adjustment with a timestamp.
- 55. Measure in random orderMeasure parts in an order that does not match the machining sequence. This breaks the tendency to read the gauge with an expectation of what the number should be.
- 66. Plot before you calculatePut the data on an X-bar and R chart first. Remove points only with a documented, physical cause such as a power interruption or a wrong bar. Never remove a point because it is inconvenient.
- 77. Calculate Cp, Cpk, and PpkUse within-subgroup σ for Cp and Cpk, and overall σ for Ppk. Report all three with the sample size and the period covered. Round to two decimals.
- 88. Decide and actCompare against the customer requirement. If Cpk is under 1.33, tighten the process, adjust the offset, or open a tolerance discussion with design before full release.
What to do with the Cpk result
Benchmarks assume a stable, roughly normal process.
| Cpk range | Process state | Inspection plan | Action |
|---|---|---|---|
| Above 1.67 | Centered, tight spread | Skip-lot or periodic sampling | Keep the control chart running |
| 1.33 to 1.67 | Capable, some drift | SPC with defined reaction plan | Watch tool life and offsets |
| 1.00 to 1.33 | Marginal | 100% or high-frequency check | Tighten process before volume |
| Below 1.00 | Not capable | Sort every lot | Fix spread or centering first |
| Cp high, Cpk low | Tight but off target | Same as Cpk band | Adjust the tool offset |
Common questions on machining process capability Cpk
How many parts do we need for a valid Cpk study?
Plan for 25 to 30 subgroups of 3 to 5 parts, so roughly 75 to 150 measurements. Fewer than 25 subgroups gives a σ estimate that moves too much between studies.
Spread the subgroups across a full tool life and a shift change. A study built from one uninterrupted run measures repeatability, not capability.
Can Cpk be calculated from a pre-production trial?
A trial on a hand-fed lathe or a prototype setup does not represent production. The fixture, the operator, and the cycle are all different.
Use trial data to set expectations, not to report capability. Report Cpk only from parts made under the production process.
What if the data is not normally distributed?
Machined features often skew from tool wear or truncate at a limit the tool cannot cross. The ppm figures behind 1.33 and 1.67 assume normality.
Discuss the distribution with the customer and consider a non-normal capability method. Reporting a percentage that assumes normality when the data clearly is not normal will not hold up in an audit.
Why does Cpk drop after a few weeks in production?
Usually tool wear, thermal growth at shift start, or a fixture change. Compare Cpk from the original study with current Ppk. A wide gap points to drift between subgroups.
Check the tool change interval and the offset log first. Most drops trace to a worn insert or an offset set once and never revisited.
Should we report Cpk or Ppk to the customer?
Report both. Cpk describes short-term capability with special causes removed. Ppk describes what the process actually delivered over the study period.
A customer who only sees Cpk may be surprised by the first shipment. A customer who sees both understands the process.
Does a Cpk of 1.33 mean no parts will be out of tolerance?
No. At 1.33, roughly 63 ppm are still expected outside the limits. That is small, but not zero.
Inspection planning should reflect the Cpk band, not assume the process is perfect. Features above 1.67 can move to periodic sampling; features below 1.33 need action before volume.
Send us the drawing and the tolerance
We review the feature, the tolerance, and the inspection plan, then quote with a capability target in writing. Quotation and free DFM analysis within 12 hours.
12-hour quote100% inspectionTolerances to ±0.005 mm