Improved PCA-SVDD based monitoring method for nonlinear process

Conventional principal component analysis (PCA) is limited to Gaussian process data due to its monitoring statistics. This paper introduces an improved PCA based method for nonlinear process monitoring using support vector data description (SVDD) by constructing two new monitoring statistics. Different from the traditional PCA method, monitoring statistics based on SVDD model have no Gaussian assumption. Thus the new monitoring statistics have no restriction to the distribution of process data, which is effective for nonlinear process monitoring. A corresponding fault diagnosis method is also proposed. To demonstrate the efficiency, detailed comparisons between the new approach and conventional methods are presented. The monitoring performance of the proposed method is examined through a numerical example and the Tennessee Eastman (TE) benchmark process.