Fault diagnosis with between mode similarity analysis reconstruction for multimode processes

Abstract

In this paper, a new approach on between mode similarity analysis (BMSA) reconstruction is proposed. Compared with the traditional method, similarity between different modes, higher-order and independent statistics are considered to separate independent subspaces, which contain increased subspace, decreased subspace, unchanged subspace and residual subspace. And further, the fault amplitude and fault directions are accurately extracted from the independent components to realize fault reconstruction, thus fault feature is highlighted, and the diagnosis performance is improved. Fault diagnosis indices are developed based on BMSA reconstruction for various fault alarms. The proposed method is applied to penicillin fermentation process, and is compared to traditional multiple modeling method. Experiment results show that the proposed method can more accurately diagnose fault than traditional multiple modeling method.