Learning based on kernel-PCA for abnormal event detection using filtering EWMA-ED.

Abstract

Multivariate statistical approaches have been widely applied to monitoring complex process, however incipient and small−magnitude faults may not be properly detected with the above techniques. In this paper, a learning approach based on kernel-PCA with filtering EWMA-ED is proposed to improve the detection of these types of faults. The proposal was tested on the Tennessee Eastman (TE) process where it is observed a significant decrease in the missing alarms, whereas the latency times are reduced.