Abstract: Although the band delineation method has made great strides in locating the fault resonance bands, there is still the problem of inaccurate resonance band location. To address the problem that it is difficult to locate the resonance bands generated by bearing faults, an adaptive banding method is proposed to extract the resonance bands generated by bearing faults. The core of the method is to use the power spectral density (PSD) of the signal to iteratively convolve with a gaussian kernel function. The convolved PSD will become smoother, and then the local minima after convolution will be used as the boundary basis for adaptive spectrum partitioning, followed by the use of the squared envelope spectral harmonic noise ratio (SESHIR) to lay out the spectral plane, and the maximum envelope spectral harmonic noise ratio to determine the resonant band, and carry out envelope demodulation analysis to extract fault features. Through simulation and experimental data analysis, it is demonstrated that the method can extract resonance bands and identify fault information in bearings, and has better performance than fast kurtogram (FK)，Autogram and variational mode decomposition (VMD).

Key words: bearing fault diagnosis, adaptive spectrum division, power spectral density, squared envelope spectrum harmonic-to-interference ratio