Abstract: A novel dynamic time-frequency process monitoring method based on multivariate spectral decomposition is proposed.Based on the existing spectral independent component analysis(spectral ICA) and non-negative constrained decomposition,a moving time window is introduced,and multiple dominant spectral components are extracted within the short-time window.Various trend illustrations are presented,and significant-index plots to indicate the degrees of contribution of process variables are given.The proposed approach can effectively monitor the changing trends of the dominant frequency and energy as well as the process variable contributions,and is feasible for non-stationary process monitoring,fault detection and localization.Simulation results demonstrate the feasibility of the proposed method.