KPCA algorithm based on improved wavelet kernel function

Conventional PCA method for fault detection and diagnosis has a high rate of false positives and false negatives due to most industrial processes with nonlinear and non-Gaussian characteristics. For a class of the problem of fault detection and diagnosis associated with nonlinear and non-Gaussian, an improved wavelet kernel KPCA method is proposed, It adopts kernel principal component analysis methods based on traditional wavelet kernel function, and uses gauss comparability rule of multidimensional variable to optimize kernel parameters. the proposed method is applied to detect and diagnose the faults in three-tank process, Simulation results show effectiveness and higher correct recognition rate of this method.