Machine fault detection using bicoherence spectra

This paper describes the detection and identification of an induction motor's asymmetric faults by measuring vibration data and analyzing the nonlinearity of the machine using higher-order spectral (HOS) analysis. Since damaged or abnormal-state machines often generate highly nonlinear signals, it is desirable to use a tool that can detect and analyze nonlinear signatures. The bispectrum has been commonly proposed for such nonlinear analysis. However, we utilize the bicoherence as a novel tool to detect and analyze the machine condition. The principal advantage of the bicoherence is that it is a direct measure of the phase coupling introduced by nonlinearities, but independent of the amplitude of the interacting frequencies. The usefulness and statistical robustness of this method are confirmed in the experiments.