Monitoring and Diagnosis of Nonlinear Profile Variations Using Wavelets and Support Vector Machine
| Autor: | Wei-Shan Peng, 彭薇珊 |
|---|---|
| Druh dokumentu: | 學位論文 ; thesis |
| Popis: | 103 In conventional statistical process control (SPC) applications, it is commonly assumed that quality characteristics can be well represented by a single measurement from a univariate distribution or multiple measurements from a multivariate distribution. However, in some practical applications, there is a demand in monitoring multiple measurements constituting a line or curve that is referred to as a profile or function. This research aims at monitoring and identifying the changes in a nonlinear profile. For smooth nonlinear profiles, we propose the use of non-parametric regression method to construct a reference (baseline) profile. Some relevant statistics obtained from the statistical analysis together with the distance-based metrics are used to construct a feature vector for a support vector machine-based monitoring procedure. For nonlinear profiles that have complex, irregular, non-smooth behavior, we propose using wavelet transform to smooth the profile data. A novel approach taken here is to construct a data pattern by examining the differences of wavelet coefficients between each new profile and the in-control profile. A set of features obtained from the data pattern are used to construct a monitoring and diagnosis procedure. By mapping the changed locations and the type of data patterns in differences of wavelet coefficients into profile segments, the proposed method can characterize various types of variation patterns. The advantages of the proposed method are illustrated using examples taken from the literature. |
| Databáze: | Networked Digital Library of Theses & Dissertations |
| Externí odkaz: |
|