Statistics kernel principal component analysis for nonlinear process fault detection

Traditional kernel principal component analysis (KPCA) considers the mean and variance-covariance of the data in the kernel space and can't make use of higher-order statistics to get more useful information from observed data. In this paper, a new nonlinear fault detection method called statistics kernel principal component analysis (SKPCA) is developed. First, change the original data space into a statistics space based on statistics pattern analysis framework; then use KPCA in the statistics space to extract some dominant principal components. SKPCA provides more meaningful knowledge by involving the higher-order statistics in the statistics space compared with KPCA. The effectiveness of the proposed monitoring approach are illustrated through a numerical example and the complicated Tennessee Eastman (TE) benchmark process.