Machine Condition Monitoring Utilizing a Novel Bispectral Change Detection

Since a damaged or abnormal-state machine often generates highly nonlinear signals, it is desirable to use a tool that can effectively detect and analyze nonlinear signatures. The bicoherence has been proposed for such nonlinear analysis since it is a measure of the phase coupling between interacting frequency components. However, the bicoherence has some difficulties in machine condition monitoring due to the challenge of distinguishing between the intrinsic nonlinear signature of a healthy machine and the nonlinear signature of a faulted machine. To address this issue, we propose a novel method exploiting the bispectral change detection (BCD) to detect and analyze the machine faults. The principal advantages of the proposed BCD method are that it can suppress the intrinsic nonlinear signature of healthy machines and emphasize the fault-induced nonlinearities. Therefore, the proposed BCD method can discriminate between fault-induced nonlinearities and intrinsic nonlinearities, and thus can be a strong and sensitive diagnostic for machine faults. The usefulness and statistical robustness of this method are demonstrated via some experimental results.