Research On Nonlinear Profile Outlier Detection Method

Statistical process control(SPC)has been the most popular tool in the field of Quality Management.These charts are used to monitor the stability of process and then decrease the variability of product quality.Through process monitoring,analysis of the production process of the operating state and identify the abnormal point and then to ensure that the process is in a stable state and then the quality of products and services to be guaranteed.With the continuous development of science and technology,the monitoring object rarely takes a simple one-dimensional form.In the case of quality control,the quality of the process or product can be depicted as a Profile or Functional Data because the data presents a very complex state.Profile is a relationship between a response variable and one or more explanatory variables.Profile monitoring is divided into two phases:the first stage is based on the collected sample data to determine whether the sample points out of control.The second stage is based on the first phase of the process has been controlled then monitor real-time sample points and detect real-time variation.In this thesis,the problem of profile outlier detection is in the first stage of profile monitoring.Detection of small part of outliers in the process in the first stage is very important.Outliers indicate the abnormal state in the system.These outliers are needed to be removed,but sometimes outliers also show some useful information,even a major discovery.Outliers detection has more applications in the field of network security,visual monitoring,remote sensing technology,medical diagnostics and other fields.In the field of quality control,profile data is usually expressed in non-linear state,so a more complex model is needed for analysis.In this thesis,a new profile detection method for non-normal variation is proposed by using wavelet analysis,data depth,clustering analysis and other data analysis methods.In this thesis,we use the simulation technology to compare the proposed method and the ?~2 control chart method.The results show that the proposed method shows better outliers recognition performance and identify outliers with higher stability and accuracy.Finally,the method is applied to a real data set.The results show that the new method can effectively identify the abnormal data.