Understanding the Voice of the Process vs. Customer
Process capability fundamentally compares two voices. The "Voice of the Customer" is defined by the engineering specification limits: the Upper Specification Limit (USL) and Lower Specification Limit (LSL). This represents the acceptable tolerance band.
The "Voice of the Process" is determined by the natural variation of the manufacturing equipment, measured statistically by standard deviation (sigma, denoted as σ). Almost all capability calculations assume that the manufacturing data follows a normal distribution (a bell curve). A capable process is one whose natural bell curve is narrow enough to fit comfortably between the USL and LSL without overlapping the boundaries.
The Difference Between Potential (Cp) and Actual (Cpk)
The Cp index measures the "potential" capability of a process. It answers the question: "If this process were perfectly centered between the specification limits, could it produce acceptable parts?" It compares the total tolerance width against the natural spread of the process. A high Cp means the bell curve is very narrow.
However, Cp ignores where the process is centered. A machine could have a brilliantly narrow bell curve, but if it is running drastically off-center, it will produce 100% scrap. This is why Cpk is the more critical metric. Cpk measures the "actual" capability by evaluating the distance from the process mean to the nearest specification limit. It penalizes the score if the process is not perfectly centered.
How to Calculate Cp and Cpk (Formulas and Example)
The formula for potential capability is Cp = (USL - LSL) / (6 × standard deviation). The formula for actual capability is more strict, requiring two calculations to find the worst-case scenario. Cpk = minimum of [ (USL - Mean) / (3 × standard deviation) ] OR [ (Mean - LSL) / (3 × standard deviation) ].
Let's calculate an example. An automotive shaft has an engineering specification of 10.0 mm ± 2.0 mm. Therefore, the USL is 12.0 and the LSL is 8.0. A sample of parts is measured, showing an average (mean) diameter of 10.5 mm and a standard deviation of 0.4 mm.
First, calculate Cp: (12.0 - 8.0) / (6 × 0.4) = 4.0 / 2.4 = 1.67. This high Cp shows the process has excellent potential. Next, calculate Cpk. The upper distance is (12.0 - 10.5) / (3 × 0.4) = 1.5 / 1.2 = 1.25. The lower distance is (10.5 - 8.0) / (3 × 0.4) = 2.5 / 1.2 = 2.08. We take the minimum of the two. Thus, Cpk = 1.25. Because the process average is shifted toward the upper limit, the actual capability (1.25) is significantly lower than the potential capability (1.67).
Short-Term vs. Long-Term Capability (Pp and Ppk)
While Cp and Cpk measure short-term variation, Pp and Ppk measure long-term performance. Short-term data is usually gathered in controlled bursts (e.g., a 50-piece run during machine validation) without environmental shifts. Long-term data spans weeks or months, encompassing multiple operators, material batches, tool wear, and temperature fluctuations.
The calculations for Pp and Ppk use identical formulas to Cp and Cpk. The mathematical difference lies entirely in how standard deviation is calculated. Cp/Cpk uses "within-subgroup" estimated standard deviation (often utilizing R-bar / d2 constants from control charts), whereas Pp/Ppk uses the overall historical standard deviation of the entire population.
Interpreting Cpk Values and Sigma Levels
Capability indices translate directly into anticipated defect rates, measured in Parts Per Million (PPM). A Cpk of 1.0 indicates the edge of the process curve is touching the specification limit, yielding an expected defect rate of about 2,700 PPM. In modern manufacturing, a Cpk of 1.0 is considered unacceptably risky.
The generally accepted minimum standard across most industries is a Cpk of 1.33, which corresponds to a 4-sigma level and an expected defect rate of 63 PPM. For automotive safety-critical components or aerospace parts, the standard is a Cpk of 1.67 (5-sigma, 0.6 PPM) or even a Cpk of 2.0, which constitutes the famous Six Sigma quality standard yielding just 3.4 defects per million opportunities.
Frequently asked questions
Can a Cpk value be negative?
Yes. A negative Cpk value means that the entire average of the process has shifted outside of the specification limits. It indicates that more than 50% of the parts being manufactured are inherently defective.
Can Cpk be greater than Cp?
No, mathematically it is impossible. Cp represents the maximum potential of the process if it were perfectly centered. If a process is perfectly centered, Cpk will exactly equal Cp. As the process drifts off-center, Cpk drops below Cp.
How many data points do I need to calculate a valid Cpk?
While you can technically calculate the math on a handful of numbers, statistical best practices (such as AIAG PPAP manuals) dictate a minimum of 100 data points collected in sequential subgroups of 3 to 5 to generate a statistically reliable Cpk estimate.
What happens if my data is not normally distributed?
Standard Cp and Cpk formulas assume normal (bell-curve) distribution. If your data is heavily skewed (e.g., flatness or roundness measurements bounded by zero), standard Cpk calculations will yield invalid results. You must use data transformation techniques (like Box-Cox) or non-normal capability models.
Why did my customer ask for Ppk instead of Cpk?
Customers request Ppk when they want to evaluate your long-term, historical production stability over months, rather than relying on a short-term Cpk capability study conducted under artificially perfect conditions.