OPTIMAL STATISTICAL DESIGNS OF A MULTIVARIATE CUSUM CHART BASED ON ARL AND MRL

摘要

Optimal statistical designs of the multivariate CUSUM (MCUSUM) chart for multivari-ate individual observations based on ARL and MRL are proposed. Statistical design procedures refer to choices of the reference value, k and the control limit, H to ensure that the MCUSUM chart's performance meets certain statistical criteria. The primary criterion is the average run length (ARL) which is the most commonly used measure of a control chart's performance, while the median run length (MRL) which is the 50th percentage point of the run length distribution is suggested to be used as a potential alternative to the ARL or as a secondary criterion in the evaluation of the performance of the MCUSUM control chart. Although the MRL is used in the optimal design of the univariate EWMA and univariate CUSUM charts but the design of an optimal multivariate CUSUM chart based on MRL is not yet given in any literature. This paper also suggests a systematic approach of designing an optimal MCUSUM chart based on the average run length (ARL). The MRL profiles are considered as supplements to the ARL profiles for the control scheme. Examples of optimal designs of the MCUSUM chart based on both MRL and ARL are also presented. Tables are provided to determine the optimal chart parameters for the design of the chart based on both ARL and MRL.