To save time, cost and labor, there are many studies that have been conducted about the detection of faults in industrial processes. Most of the previous studies used only Independent Component Analysis (ICA) or Principal Component Analysis (PCA) for detection, but they can not form close enough boundaries to reject outliers. This paper proposes an ICA-based approach to detect outliers in a process by forming close boundaries. The basic idea of the proposed method is to apply ICA to convert original data into independent components, and then apply Durbin Watson (DW) criterion to select important independent components. Hereafter, Support Vector Data Description (SVDD) has been applied for outlier detection by forming much tighter boundaries. The efficiency of proposed ICA-based approach is investigated via a simulated multivariate process example. To demonstrate the identification capability of the proposed ICA-based approach, the traditional Hotelling’s T2 chart is constructed for the simulated data set.
Mu-Chen Chen, National Chiao Tung University, Taiwan
Chun-Chin Hsu, Chaoyang University of Technology, Taiwan
Bharat Malhotra, Indian Institute of Technology, India
Manoj Kumar Tiwari, Indian Institute of Technology, India
Stream: New Realities through Artificial Intelligence
This paper is part of the ACTIS2014 Conference Proceedings (View)
View / Download the full paper in a new tab/window