Local and Global Unsupervised Kernel Extreme Learning Machine and Its Application in Nonlinear Process Fault Detection

Abstract

Unsupervised extreme learning machine (UELM) only concentrates on the local structure analysis of the observation data and thus may perform unsatisfactorily for process monitoring. In this paper, local and global unsupervised kernel extreme learning machine and support vector data description (LGUKELM-SVDD) is proposed for effective fault detection. LGUKELM model incorporates the global structure analysis into the standard UELM model. Meanwhile, the kernel trick is adopted to avoid the problem of explicitly selecting the number of the hidden layer nodes. After the nonlinear data features are extracted from LGUKELM model, SVDD is applied to build a monitoring index to detect fault. The simulation results on the continuous stirred tank reactor system show that the proposed method can detect process fault more effectively than conventional methods.