Scale Space Guided Feature Mode Decomposition Method for Extracting Fault Feature of Rolling Bearings

Decomposition methods are capable of separating fault information from the heavy background noises. However, the performance of decomposition methods is always affected by vital parameters. Feature mode decomposition (FMD) is widely used in fault diagnosis due to its adaptive decomposition capability; nevertheless, the unreasonable parameters setting in the FMD method leads to a wrong decomposition result. This article introduces a scale space guided FMD fault diagnosis method, designed to eliminate the need for manual parameter tuning. By leveraging scale parameters for multiscale smoothing, the scale space method effectively suppresses noise in the signal and extracts critical fault characteristic frequencies using Gaussian kernel convolution. In addition, boundaries are identified through local minima, enabling comprehensive feature capture from fine details to global structures within the signal. Simulation and experimental validation results demonstrate that the method outperforms up-to-date decomposition methods in terms of noise suppression and fault feature extraction. Adaptive FMD provides a more robust solution for health monitoring of mechanical equipment in complex environments.