Fault Detection Based on Hierarchical Mode Analysis with ICA-KICA

For the problem of fault detection in non-Gaussian and nonlinear multimode processes, a fault detection method based on hierarchical mode analysis with independent component analysis (ICA)-kernel independent component analysis (KICA) is presented. Since stable and transition modes coexist in multimode processes and the mode information has uncertainty, hierarchical mode analysis is used for obtaining the hierarchical mode information. Then, the ICA-KICA is presented for modeling the stable and transition modes obtained from the hierarchical mode analysis. Lastly, a mode identification method based on discriminant function is given for fault detection. The penicillin fermentation process is used for the experiments of fault detection and the analyses indicate the performance of the fault detection based on hierarchical mode analysis with ICA- KICA.