The presence of autocorrelation has long been recognized as a natural phenomenon in manufacturing industries. Observations from these environments do not conform to the normality and independence assumptions, which are the basic assumptions of traditional control charts. This paper investigates the impact of autocorrelated data on the the average run length (ARL) of the multivariate Shewhart-type chart (X2 chart), MCUSUM chart, and MEWMA chart. The performances for detecting process changes among the X2 chart, residual chart, UCL-adjusted MCUSUM chart, and UCL-adjusted MEWMA chart when applied to autocorrelated processes are evaluated. In the study, the processes are assumed to be VAR(1) models. The performance measure is ARL. The results indicate that when the process is positively autocorrelated, the multivariate Shewhart chart appears to be unaffected from the presence of autocorrelation. From the evaluation of the ARL, the MCUSUM and MEWMA charts with adjusted control limits are more sensitive in detecting small shifts than the multivariate Shewhart chart. However, the Shewhart-type chart is significantly more robust than the MEWMA and MCUSUM chart.
DOI 인용 스타일