Time–Frequency Mode Adaptive Decomposition Based on the Maximum Kurtosis for Extracting Fault Component of Bearings

Separating different modes from a vibration is a critical step for health status monitoring of bearings by signal surveillance. The challenge results from the modulation of signals, which often introduces modulation sidebands in spectra, obscuring genuine components and leading to erroneous decompositions. To resolve this problem, an algorithm named time-frequency mode adaptive decomposition based on the maximum kurtosis (TFMAD-MK) is proposed on the foundation of frame decomposition. The method highlights the critical role of appropriate window length in mitigating modulation sidebands within the short-time Fourier transform (FT) spectrum. This method undergoes comparative analysis with popular adaptive decomposition techniques, such as the empirical Fourier decomposition (EFD), the empirical wavelet transform (EWT), and the ensemble empirical mode decomposition (EEMD), using a sample signal for evaluation. Further validation is conducted through two sets of experimental signals, demonstrating that the algorithm has the capability to effectively isolate fault components from signals.