A Change Point Identifying Method Based On Data Depth

In industrial production process, The importance of statistical process control(SPC) has been widely recognized in quality management, control models in SPC is important tools for monitoring whether a manufacturing process is statistically in control or not. We can divide SPC in two stage: Phase I, analysis the historical data, identify correct baseline data which meet the quality requirements; Phase II, Parameter estimation, establish control charts and continue real-time monitoring, the control charts will alarm when the process data is out of control, then we should analysis the reason and take appropriate measures to adjust.In the stage of Phase I, it is very important to identify change point accurately. But most of the traditional method is based on the assumption of normal distribution. Our new method proposes an idea of change point identification based on the proximity of the data center. The propose method transfer individual observations sequences into points in multi-dimensional space, and the variable is constructed based on data depth, then a change point location rule is developed. The proposed method is tested on several simulation experiments from a literate, and compared with the traditional method( LRT, AHC), it is proved even better to identify the baseline period without the normal distribution assumption.Finally, this paper gives a summary about the method’s advantages and limitations draw some useful conclusions and make prospects for future development.