Cp, Cpk, Pp and Ppk in CNC Machining: How to Read Them Correctly

Cp, Cpk, Pp and Ppk all compare observed variation with engineering specification limits, but they are not interchangeable. Cp and Cpk usually use an agreed estimate of within-subgroup or short-term variation. Pp and Ppk use the overall standard deviation across the study period. Cp and Pp compare spread with tolerance width, while Cpk and Ppk also penalize a mean that is close to one specification limit. A high value does not by itself prove process stability, normality, a capable measurement system or that every part passed. Always identify the characteristic, sampling plan, subgroup definition, period and calculation method before interpreting the number.

USL and LSL come from the drawing or customer requirement. Control limits are calculated from process behavior over time. A small lot can remain inside tolerance while showing special-cause instability; a stable process can still be incapable against a very narrow tolerance. Do not use specification limits as control-chart limits, and do not assume an in-control chart means every part conforms. Capability analysis is normally meaningful only after stability is evaluated. Machine, fixture, program, tool, material, operator, shift and measurement changes should be retained in the data structure instead of pooled blindly.

Overall standard deviation includes variation across shifts, tool changes, material lots, restarts and environmental changes. Ppk can therefore be lower than Cpk when long-term drift or subgroup differences exist. The gap is a diagnostic signal, not proof that one index is wrong. Short subgroups may hide long-term risk, while a start-up period with deliberate changes may not represent later stable production. The RFQ or control plan should define the study period, subgroup, sampling frequency and reported index. A blanket Cpk requirement for every drawing dimension is rarely efficient; focus on identified critical characteristics.

Capability estimates come from samples and have uncertainty. NIST cautions that sample size must be large enough and confidence intervals matter. There is no universal count because risk, subgroup logic, tool cycle and customer rules differ. The measurement system needs suitable resolution, calibration and geometry. Gauge R&R, bias, linearity or measurement uncertainty may need review. Preserve sufficient digits and the time, machine, tool, operator and measurement context. Premature rounding or combining different measurement methods can distort the distribution.

Turned diameters, milled sizes, hole positions and ground thicknesses should be subgrouped according to how variation is created. Consecutive parts between tool adjustments may form a rational subgroup, but the actual process must define it. When size drifts with tool life, one pooled index can hide the cycle. Review the time sequence, histogram, special signals, tool changes, offsets and measurement results together. First-off inspection verifies setup and dimensions; a few first parts rarely demonstrate long-term capability. Capability indices also do not replace first article, final inspection, cosmetic checks or material records.

Instead of writing only Cpk ≥ 1.33, identify the characteristic, limits and units; required index; sigma method; subgroup; sample size; frequency; study period; and treatment of special-cause data. State MSA, control chart, raw-data, precision, report and reaction requirements. A threshold such as 1.33 or 1.67 is a contract or quality-policy choice, not a universal law. If a pilot run cannot support a capability claim, an agreed temporary control such as 100% inspection of a critical feature may be appropriate. VietTech reviews measurement time and data requirements during quotation because statistical documentation affects inspection workload.

• Characteristic, balloon, USL/LSL and unit
• Cp/Cpk or Pp/Ppk and sigma method
• Subgroup, frequency and study window
• Sample size and treatment of start-up/tool change
• MSA, control chart and raw-data requirements
• Acceptance threshold and reaction plan

This article is based on the NIST/SEMATECH Engineering Statistics Handbook sections on process capability and graphical assessment against specification limits. Contract work may also invoke ISO 22514, AIAG SPC or sector-specific customer rules; use the licensed edition named by the customer. This article sets no mandatory Cpk threshold, assumes no universal normal distribution and does not replace statistical expertise. VietTech confirms any capability plan only after reviewing the drawing, volume, measurement method, customer rules and demonstrated process behavior.