Approximate entropy as a nonlinear feature parameter for fault diagnosis in rotating machinery

Feature extraction is always a crucial step for fault diagnosis in rotating machinery. When faults occur, rotating machinery always manifests nonlinear dynamic behavior. It is necessary to extract the nonlinear features hidden in the vibration signal for more accurate diagnosis. Approximate entropy (ApEn) is the nonlinear parameter identification method for measuring the irregularity of the stochastic signal or the stochastic process. In this paper, ApEn is used as a nonlinear feature parameter to measure the irregularity of the vibration signals for fault diagnosis in rotating machinery. Four typical faults are considered, which are imbalance, misalignment, shaft rubbing and oil whirl. To improve the distinguishability of the ApEn values of the different faults, the empirical mode decomposition (EMD) method is used to remove the basic frequency component from the signals of the various faults. The experimental study results demonstrate that EMD can separate the basic frequency component from the original signals satisfactorily. After removing the basic frequency component, the distinguishability of the ApEn values of the residual signals is improved greatly. The proposed strategy for the ApEn calculation of the various faults is proved effective. In addition, the simulation study is presented to investigate some characteristics of ApEn, which will benefit better application of ApEn in the field of fault diagnosis.