Research on robust kernel independent component analysis based on Kurtosis in fault detection

The paper is proposed the novel non-linear process monitoring method based on robust kernel independent component analysis component analysis (Robust KICA). Combined with the kernel principle component analysis (KPCA) and the robust independent component based on Kurtosis, the basic theorem of RobustKICA are applied to realize nonlinear mapping, and made the data linearly structured in feature space by using KPCA after wavelet packet de-noising of the original data. RobustICA optimized by iterative technique is employed to separate independent components which are driven by a process in the KPCA-transformed space. The statistics (I², I²_e and SPE) are established, and two combined monitoring indices for RobustKICA are also adopted for fault detection. The simulation results of TE model are clearly demonstrated the effectiveness and advantages of the proposed method.