A Comparative Study on PCIs for Non-normal Data

Process capability indices (PCIs) are applied to assess the ability of a process to manufacture products that meet certain tolerance. The conventional PCIs are based on the assumption of normality, however, process data are often not normally distributed in practice. For non-normal data, there are more than twenty approaches dealing with this problem. Whereas, it is still difficult for us to choose an appropriate method to measure non-normal processes. In this paper, several approaches for non-normal data are comparing by Monte Carlo simulation. Simulation results show that Johnson transformation method and Clements' method perform better than the other methods in accuracy and dispersion by boxplot.