Abstract:

A fault detection method based on kernel state space independent component analysis (KSSICA) was proposed in this paper considering the nonlinear and dynamic characteristics of industrial processes. Kernel canonical variate analysis (KCVA) was adopted to project the nonlinear and dynamic process data into the kernel state space, and the state data which were uncorrelated were obtained. Based on the state data’s time structure matrix which is the weighted sum of the state data’s different timedelayed covariance matrices, an ICA statistical model was constructed to extract the independent component feature data from the state data, and the monitoring statistics were built to detect process faults. The fault detection results on the Tennessee Eastman benchmark process demonstrate that the proposed KSSICAbased fault detection method can detect the process faults more agilely and obtain a higher fault detection rate than the conventional fault detection method based on dynamic kernel principal component analysis (DKPCA).

Key words:  , fault detection； nonlinearity； dynamic characteristic； kernel canonical variate analysis； independent component analysis； fault detection rate