Process Capability Analysis For Non-normal Data And Multivariate Data And Its Application In The Process And Discrete Manufacturing

Based on the process capability analysis, this paper focuses on process capabilityindices for non-normal distribution and multivariate data. For non-normal distributiondata, a Monte Carlo Simulation is conducted on evaluating the accuracy and precisionof seven methods that have been proposed for computing process capability indicesfor non-normal data. The simulation was lead using50sets of data that are generatedfrom Log-normal and Weibull distribution and each set was calculated through thoseseven methods. The results were evaluated by accuracy and precision. Simulationresults show that with the circumstances of sample size of no greater than100,Johnson transformation is superior to Box-Cox transformation and Spmkperforms thebest. The conclusion comes out as a method’s performance depends on its capabilityto capture the characteristic parameters and the tail behavior of the underlyingdistribution. Then the result is used to compute the process capability indices for a setof non-normal data from a real world.For multivariate data, this paper first came up a new method of buildingmultivariate control chart to monitor the process. Inspired by principal componentanalysis which is widely used in reducing the high dimensions to lower ones,comprehensive score is the target to build the multivariate control chart. If themultivariate control chart shows the process is under control, then the multivariateprocess capability indices are followed to evaluate the process ability to meet thecustomers’ satisfaction. There are two methods to compute the indices. One is thegeometric mean of single index; the other based on the comprehensive score.