Nonlinear process monitoring using wavelet kernel principal component analysis

Conventional principal component analysis (PCA) performs poorly in nonlinear process monitoring due to it only can capture the linear structure of the process variables. To overcome this nonlinear problem, a new nonlinear process monitoring technique based on kernel principal component analysis (KPCA) is proposed. In order to enhance the ability of capturing the nonlinear feature for KPCA, a marr wavelet kernel is constructed and proved. Then the proposed method is applied to the fault detection in the Tennessee Eastman process (TEP). The simulation results showed superior process monitoring performance compared with PCA.