Process Capability for a Non-Normal Quality Characteristics Data

Process capability indices (PCI) are widely used in manufacturing industry today. These statistical measures (Cp, Cpk) provide a quantitative measure of the process performance for decision makers. They are based on normality assumptions and provide a better estimation of process parameters if the process data is normally distributed. Unfortunately, this assumption is often violated in practice. In most cases, the distribution of a process characteristic data is non-normal. Application of conventional methods for calculation of process capability indices based on normal assumption will therefore give erroneous results that could lead to wrong decisions. In non-normal data situations, estimation of accurate PCI is critical for process improvement purposes. This paper explores application of a novel method (Pei-Hsi Liu and Feng-Long Chen, 2006) based on Burr distribution for PCI calculations when the process data is not normally distributed and compares simulation results with the commonly used Clements´ method. Finally, an example illustrating application of this method with real world data is presented