Process Capability Index Calculator

Process capability indices compare the natural spread of a process with the allowed engineering specification interval. Cp measures potential capability by dividing the distance between the upper and lower specification limits by six standard deviations. A normally distributed, stable process places about 99.73% of observations within three standard deviations of its mean, so six standard deviations represent its conventional natural width. A larger Cp means the tolerance band is wide relative to process variation. Cp ignores where the process mean sits inside the limits. Cpk adds centering by comparing the mean with each specification boundary. Cpu measures the upper side as USL minus the mean divided by three standard deviations. Cpl measures the lower side as the mean minus LSL divided by three standard deviations. Cpk is the smaller of Cpu and Cpl because capability is constrained by the nearest boundary. A centered process has equal side indices and Cpk equals Cp. A shifted process has Cpk below Cp. Organizations often use 1.00 as the point where the estimated natural process width just fits the specification width, 1.33 as a common minimum for a capable routine process, and 1.67 or higher for more critical applications. These are conventions rather than universal laws. Customer risk, measurement error, process maturity, distribution shape, and the cost of defects should determine the acceptance criterion. Capability analysis is meaningful only after statistical stability has been established. A process affected by special causes does not have a dependable mean or standard deviation, so its future performance cannot be summarized reliably by one index. Review control charts first and use a within-process estimate of variation that matches the sampling plan. Specifications must come from engineering or customer requirements, not from the observed data. Cp and Cpk also rely on an approximately normal continuous distribution when interpreted as defect performance. Skewed, bounded, multimodal, or autocorrelated data may require transformation, a nonnormal capability model, or direct percentile analysis. Confirm the measurement system is adequate and use enough rationally subgrouped observations. The calculator supplies transparent arithmetic, but a complete study should report the sampling period, stability evidence, distribution assessment, confidence intervals, and operational context.