QPRC 2016

A Strategy for Best Distribution Selection for Process Capability Indices

Karl Pazdernik

North Carolina State University

Process capability indices (PCIs) are heavily relied upon in industry to evaluate manufacturing process health and, consequently, profitability. Implicit in the calculation of these metrics is an assumption of some underlying distribution or the use of non-parametric density estimation. This paper explores common goodness-of-fit tests and model selection techniques with the goal of determining the best distribution choice, particularly as it pertains to Cp and Cpk calculation. A simulation study on the effects of distribution choice for these PCIs is provided and an algorithm is proposed to guide practitioners towards the best estimation of both Cp and Cpk. This algorithm could be easily automated in any standard statistical software.