Fault Detection using Common and Specific Variable Decomposition for Nonlinear Multimode Process

Due to the change of operating conditions including external environments and production schemes, there are usually multiple modes in most industrial processes. Moreover, different linear and nonlinear correlations simultaneously coexist in different modes, which brings difficulty to fault detection. Aiming at this problem, the paper proposes a fault detection method using common and specific variable decomposition for nonlinear multimode process. The proposed method takes into account the linear and nonlinear correlations in different modes and utilizes the common and specific modeling ideas for enhancing the performance of fault detection in multimode process. First, the multimode process data are divided into the common variable subspace and the specific variable subspace based on sliding window low-order and high-order statistics. Then, the nonlinearity coefficient is used to evaluate the linear and nonlinear correlations for each variable subspace. Subsequently, the common and specific PCA-KPCA-SPCA-based multimode process modeling is developed for fault detection. The fault detection results of the Tennessee Eastman (TE) process demonstrate the feasibility and effectiveness of the proposed method.