Fault detection based on global-local PCA-SVDD for multimode processes

Abstract

Fault detection is of importance for industrial processes with complex characteristics including multimode. In this paper, a fault detection method based on global-local PCA-SVDD is proposed for multimode industrial processes. Firstly, mode division based on spectral clustering is presented, which divides multimode processes into multiple modes without priori multimode information. Then, considering the multimode characteristic, global similarity and local non-similarity, the global-local PCA-SVDD models are built, which decomposes fault detection into a global model and multiple local models. Finally, different statistics and confidence limits are used for different models. The experiment results of the penicillin fermentation processes illustrate the feasibility and effectiveness of the proposed method for multimode industrial processes.