Study On Change Point Detection Method Based On Hausdorff Distance In Nonlinear Profile

All the factors that lead to fierce competition come to the competition in quality. So it is important for an industry to take quality monitoring measures. Statistical process control(SPC), as an effective means for quality monitoring, has been studied since 1920 s and now there is a series of well-developed theories and tools to use. Research objects have also widened, from one-variate to multivariate and to profile variate. Profile monitoring is just the process for monitoring quality characteristics well described by profile.Phase I monitoring aims to setting up a statistical in-control model for Phase II by using historical data which has removed out-of-control ones. Due to high dimensionality of profile data, we usually use regression method to reduce the high dimension data into a function model and apply multivariate control chart to monitor parameters. For regressing part, we can use LS, PLS or PCR in linear profile regression. But when a nonlinear profile is considered, especially some complicate ones, it is hard to decide the regression model and R2 can be too small. They all lead to a decrease in monitoring reliability.In this paper, we proposed applying the combination of method based on Hausdorff distance, a measurement for discrete degree between two curves with consideration of curvilinear shape, and binary segmentation to the process of one change point detection in parametric nonlinear profile. In simulation, by comparing assessment indexes with the multivariate T~2 statistic, the method proposed in this paper performs better in the accuracy of location identification in one change point detection of nonlinear profile.Besides, there is no need to figure out regression model parameters when calculating Hausdorff distance between two curves in the proposed method. So it can not only reduces the workload and complexity level but also increases reliability in nonlinear profile control, to some extent.