Outlier Detection For Nonlinear Profile Based On SVR

Outliers were once considered as noise data in statistical research.Noise data is not concerned in most cases.With the development of science and technology,more and more scholars have found the importance of outliers.Outliers can be translated into information for reference in a variety of applications.Nowadays,outlier detection has become an important research issue in academic and application fields.With the increasing complexity of data types in modern manufacturing process,the monitoring of non-linear profile data has become a research focus in the field of statistical process control.The problem of outlier recognition for non-linear profile data belongs to Phase ? of process monitoring for non-linear contour data.Phase II of monitoring is to use the data set obtained in Phase ? to establish control charts and monitor the future process data to determine whether the process is in a stable state.At present,the research of nonlinear contour data monitoring mainly focuses on Phase II,most of which assume that the controlled state data are known.However,it is very important for the stability of monitoring state in Phase II whether the anomaly can be identified accurately in Phase ?.Therefore,this thesis focuses on Phase ? of outlier detection for non-linear profile data.In this thesis,based on the normal distribution process and non-normal distribution process,a method based on support vector regression is proposed,which uses data depth and cluster analysis to identify the outlier profile that appear in the process.The simulation test results show that the proposed new method has higher accuracy in identifying nonlinear profile data.The analysis of the vertical density profile of the board shows that the new method can be effectively applied to practical problems.