分野別セミナー

Statistical process control (SPC) charts are extensive used in industrial manufacturing, disease screening, climate monitoring and many other fields of application. Traditional SPC control charts are studied under the assumption that the process observations are independent and identically distributed. Furthermore, it is typically assumed that the observations obtained under the in-control process follow a parametric distribution. In practice, it is difficult to satisfy these assumptions.

Firstly, we propose a nonparametric exponentially weighted moving average (EWMA) control chart for monitoring scale parameter, which combines the Ansari-Bradley test and the framework of change point detection. We construct the control chart that achieves the same purpose with a very small number of historical observations.

Secondly, we propose a Phase II nonparametric control chart based on EWMA and variable sampling interval for jointly monitoring location and scale parameters. We combine the control chart with variable sampling interval, which not only significantly improves the monitor efficiency but also has a lower sampling cost. Moreover, the proposed control chart is capable of simultaneously detecting changes in location and scale parameters, which has a wider application areas.

Finally, we present a new self-starting EWMA chart for monitoring multivariate serially correlated processes. The multivariate chart is based on data decorrelation, combined with Gaussian kernel density estimation (KDE). In contrast to recent nonparametric control charts that categorize or rank the observed data, our new method only uses the observed data to fit the underlying distribution. The control chart is also self-starting and all relevant IC quantities can be recursively updated.