Fault Diagnosis of Complex Electromechanical Systems Based on Nonlinear Spectrum Characteristic and Support Vector Machine

A novel approach of frequency domain fault diagnosis of complex electromechanical systems is proposed in this paper. In order to avoid the large calculation amount of identifying general frequency respond function (GFRF), the mapping relationship between input frequency spectrum and output frequency spectrum of nonlinear system is firstly obtained, and it is approximated by least squares support vector machine(LSSVM). Then, the LSSVM reference model is constructed to produce the reference output value. The error function between real output frequency spectra and reference output frequency spectra is used to detect and diagnose fault. The proposed approach only need obtain reference model by learning that avoids direct solving generalized frequency response function. Finally, a fault diagnosis simulation experiment of the mine hoist completed, experimental results verify the effectiveness of the proposed approach.