Underdetermined blind separation of bearing faults in hyperplane space with variational mode decomposition

Abstract In the health monitoring of rotating machinery, there often coexists multiple fault sources. Thus a multi-source compound fault signal will be excited and collected by sensors. Moreover, due to the practical limitations, the number of sensors is usually less than that of the source signals, which makes it an underdetermined blind source separation (BSS) problem to identify the fault signals. Because the observed signals are usually not sparse enough in low-dimension transform domains, it is not ideal to solve the underdetermined BSS problem in the low-dimension space with traditional methods. In this study, we solve the problem by exploring more effective features of the signal in the hyperplane space with variational mode decomposition. We construct the hyperplane from a variational mode decomposition of the compound fault signal, and develop a mixed matrix estimation model based on that in the hyperplane space. Simulation and experiments show that the proposed methodology can effectively separate the compound fault signals of rotating machinery and detect the fault features.