Integrating Taguchi Method and Durbin-Watson Statistic for Independent Component Selection: With Application to Etch Process Fault Detection  
Author Chun-Chin Hsu

 

Co-Author(s)

 

Abstract Independent Component Analysis (ICA) based monitoring method had been proved to be more efficient than Principal Component Analysis (PCA) when the process inhered non-Gaussianity. For PCA, the dominate components can be extracted from previous largest variance via ranked covariance matrix. However, the covariance matrix of the Independent Components (ICs) exhibits an identity matrix which means there provides no information for choosing dominant ICs via covariance matrix. Recently, the 2 L norm and Durbin-Watson (DW) statistic are proposed to select the dominant ICs. However, both selection methods have the common disadvantage of the need of users’ subjective judgment. Therefore, this study endeavors to develop an objective method for opting the dominate ICs for ICA algorithm. The proposed method first utilizes Orthogonal Array (OA) for experimenting possible ICs combination. Afterward, the DW statistic based Signal-to-Noise (SN) ratio is developed for determining the significant ICs. The proposed method will be applied for an etch process fault detection. Results demonstrated the proposed approach can use lesser ICs to achieve satisfactory monitoring results when compare to traditional methods.

 

Keywords Independent Component Analysis, Durbin-Watson statistic, Etch process, Orthogonal Array, Signal-to-Noise ratio
   
    Article #:  1914
 
Proceedings of the 19th ISSAT International Conference on Reliability and Quality in Design
August 5-7, 2013 - Honolulu, Hawaii, U.S.A.