Multimode Processes Monitoring Using Global–Local MIC-PCA-SVDD

Abstract

A multimode processes monitoring method using global–local MIC-PCA-SVDD is presented. Our method contains the procedures of mode division stage, offline modelling stage and online monitoring stage. At mode division stage, mode division using spectral clustering and multimode processes continuous characteristic is developed. It can divide multimode processes into multiple modes without priori multimode information. At offline modelling stage, considering multimode, global similarity and local non-similarity characteristics, global–local MIC-PCA-SVDD constructs multiple local models and a global model for monitoring. Our method considers dissimilarity between different modes and similarity in multimode processes. At online monitoring stage, different radiuses and distances between testing samples and the centre of the spheres using SVDD models are obtained for multimode processes monitoring. The advantages of SVDD in dealing with non-Gaussian and nonlinear data are used in our method. SVDD has no distribution assumption in which multimode processes data can be mapped to the high-dimensional feature space to construct multiple hyperspheres for global and local monitoring. The experiments of the penicillin fermentation processes are used to validate the feasibility and availability.