Optimal IMF selection and unknown fault feature extraction for rolling bearings with different defect modes

Abstract

Rolling bearings are widely used in the rotating machinery. The fault types and fault feature frequencies are usually unknown when rolling bearings fail in the engineering applications, which brings difficulties to accurately recognize the health status of rolling bearings. Realizing the extraction of unknown fault features is of great significance for rolling bearing fault diagnosis in real engineering scenarios. Based on the ensemble empirical mode decomposition (EEMD) and stochastic resonance (SR), an adaptive fault diagnosis method for unknown rolling bearing faults is proposed by designing an effective frequency cut-off criteria and constructing two novel evaluation indicators. For a target signal containing complex frequency components, a frequency cut-off criterion is applied to remove redundant intrinsic mode function (IMF) components and improve the accuracy of unknown fault diagnosis in the process of using EEMD to decompose target signal. For the remaining effective IMFs, the spectral amplification factor (SAF) index is built to estimate the energy of each IMF for selecting the optimal IMF. The adaptive SR with the piecewise mean value (PMV) index is performed to extract the unknown fault features from the optimal IMF. The proposed frequency cut-off criteria obviously reduce the interference of other unknown frequencies. Compared with traditional kurtosis, SAF has good noise immunity and stability. The effectiveness of the proposed method is successfully verified by three rolling bearing experiments with different defect modes. Moreover, the vibration waveform characterizing unknown fault feature is reproduced to some extent based on the adaptive SR response. The proposed method accurately realizes unknown fault diagnosis and provides a reference for condition monitoring of rolling bearings in engineering practice.