This thesis is concerned with designs of multivariate cumulative sum(MCUSUM) and exponentially weighted moving average(MEWMA) control charts for skewed populations. Most standard MCUSUM and MEWMA control charts assume that quality characteristics follow multivariate normal distribution. In practice, however, distributions of measurements from chemical and, semiconductor processes, etc are often skewed. For such a skewed population, standard control charts can not be directly applied.
The proposed control charts are based on the `weighted standard deviations' obtained by decomposing the standard deviation into upper and lower deviations. The method modifies the charting statistics of standard MCUSUM and MEWMA charts in accordance with the degree and direction of skewness. False alarm rates of the proposed charts are compared with those of standard charts under various skewness and correlation structures when the distribution is skewed.
