Fault diagnosis of multimode processes based on similarities

In this paper, a new large-scale process monitoring method based on knowledge mining is proposed. The contributions are as follows: 1) between-mode independent similarities are explored to reveal the between-mode relationship for multimode model development and online monitoring; 2) comprehensive subspace decomposition is performed in each mode regarding their relative similarities and influences on process monitoring; 3) each mode is separated into three different systematic subspaces and one residual subspace (RS) based on the between-mode similarities; and 4) different variations are modeled, respectively, for online monitoring to identify mode affiliation and detect the fault status. The proposed method is applied to fault detection of Tennessee Eastman process (TE Process). The monitoring results show the effectiveness of the proposed method, compared to the conventional monitoring method.