Fault classification and detection using an improved statistical analysis method

Abstract

In this paper, a new statistical method for fault classification and fault detection based on independent component analysis (ICA) and Fisher discriminant analysis (FDA) is proposed. In this method, The ICA method is used to extract feature from original data space. Then, FDA method is performed in ICA feature space for fault classification and detection. Based on such a mixing method, the performance of fault classification and detection is improved. The proposed method is applied to Iris classification and Tennessee Eastman process (TEP). The results show proposed method has more superior performance of fault classification and fault detection.