Multiple Change-points Detection Of Nonlinear Profiles Via Cluster Method

As one of the most important quality control tool,statistic process control has become more and more popular among industry factories.However,the supervisory object is no longer univariate as the technology improves.The data which shall be monitored are increasing rapidly in this new situation in quality control,so here comes a new quality characteristic that can be explained by more than one variate using a function.We call this kind of function data set profile,and the way to monitor it is called profile monitor.With the ability of process large data,profile monitor can be applied to fields like health,climate,economic,etc.According to the difference of profiles' change pattern,we can separate it into two parts: local change and global change.The former one indicates that the change on the out of control profile is its segment's shift and the latter one shows the hole profile's shape's change.Recent researches are concentrated in single change point which means once the change happens,the profiles after the moment are all changed in the same way.But this hypothesis is not always true in the realistic production as the change's number,form and position are always unknown.Thus,the single change point method's application is limited.According to the facts above,we propose a new method combine wavelet transform and cluster method to solve this problem.In order to find out how our method works,we use Matlab to make a series of simulations to compare our method with binary segmentation method on the multiple change points situations.The result show that our method has a better performance on sensitivity,reliability and stability.