A novel method for detecting processes with multi-state modes

Aiming at the muiltimode non-Gaussian process with within-mode nonlinearity, a fuzzy clustering multiple-model based inferential detecting method was proposed in this article. A clone-differential evolution-harmony search algorithm (CloneDE-HS) is used to search the best clustering centers of the process data. Then the operating data were classified as different modes. After that, maximum variance unfolding (MVU) were used to reduce the dimensions of each submodel variables. Furthermore the monitoring indices were constructed to detect the process fault. The model based support vector data description (SVDD) was built to detect the process. Finally, the proposed method was applied to detect an ethylene cracking furnace to demonstrate its efficiency.