PERFORMANCE OF TWO-SIDED EWMA CEV CONTROL CHARTS WITH MULTIPLE CENSORED DATA

Abstract

Control charts are widely applied in process control. Life testing is time-consuming and expensive; thus, this testing typically has a termination condition that is based on the assumption of a single failure mode. The lifetimes of unfailed units are known as right-censored data, and conditional expected value (CEV) control charts are effective for monitoring right-censored data. However, life testing may produce multiple censored data due to there being multiple failure modes. The occurrence of these failure modes is unpredictable, and thus the censored proportion cannot be known in advance. The exponentially weighted moving average (EWMA) control chart has a high sensitivity for detecting small mean shifts. In this paper, two charting methods for monitoring multiple censored data were developed: a two-sided EWMA CEV control chart and a combination of two single-sided EWMA CEV charts. A multi-objective model was established to optimize the design parameter values for the two types of control charts by using the multi-objective genetic algorithm. The charting methods were compared theoretically and with the practical example of liquid crystal display modules. The results demonstrate that the two-sided EWMA CEV control chart is generally more effective than the combination of two single-sided EWMA CEV control charts for monitoring multiple censored data. Moreover, if a large range and high censoring proportion are used to optimize the model design, increasing the sample size can always improve the performance of the EWMA CEV control chart.