Quality and Inspection

Process Capability and Cpk: A Practical Guide for Manufacturing Teams

Cpk measures how well a process fits within its specification limits relative to variation. Here is how to calculate it, what the thresholds mean, and how to act on the number.

Cpk equals the minimum of (USL - mean) / (3 x sigma) and (mean - LSL) / (3 x sigma). USL is upper spec limit, LSL is lower spec limit, mean is the process average, and sigma is process standard deviation. A Cpk of 1.0 means the process average is 3 sigma from the nearest spec limit and yields about 2700 defects per million, or 0.27%. A Cpk of 1.33 cuts that to about 64 ppm, while 1.67 is roughly 0.6 ppm. This is why many manufacturers require at least 1.33 before releasing a process to volume production.

Cp equals (USL - LSL) / (6 x sigma) and measures capability assuming the process is centered, while Cpk also accounts for centering. A process with Cp of 1.5 but Cpk of 0.8 has enough potential spread but is sitting too close to one spec limit. Sigma should come from measured data using rational subgroups, not from the tolerance itself. A common minimum study is 25 to 30 subgroups of 4 to 5 parts each, which gives about 125 to 150 data points. Measurement system quality also matters, because poor gauge repeatability can make sigma look worse than the process really is.

The most common Cpk error is using the specification tolerance to estimate sigma. That guarantees a neat answer but tells you nothing about actual process variation. Another mistake is trying to calculate Cpk from a tiny sample, such as 10 parts from one setup. Plants also confuse Cpk with Ppk. Cpk uses within-subgroup variation and reflects short-term capability, while Ppk uses overall variation and is usually lower because it includes drift between subgroups.

Use Cpk to make release and improvement decisions. If a new machine setting raises yield from 94% to 97%, a Cpk study tells you whether the improvement is statistically credible and how much spec margin you really have. If Cpk is below 1.33, you typically need to reduce variation, recenter the mean, or both before promising the process can hold production. That directs action toward tool wear, fixture repeatability, temperature control, or incoming material variation instead of chasing symptoms. Cpk turns a pile of measurements into a go or no-go decision.

When Cpk is low, attack dominant variation sources before asking engineering for a wider tolerance. Tightening sigma through better process control is more durable than widening the spec window. Related metrics such as Ppk, scrap rate, and first-pass yield help show whether the issue is short-term centering, long-term drift, or basic instability. Recheck Cpk after tool changes, material changes, or fixture revisions because each can shift the process mean. Capability is not a one-time certificate, it is an ongoing check on whether the process still deserves the print tolerance.

Published 2026-05-28.