Dynamic kaiser feature mode decomposition: a robust method for motor bearing fault diagnosis under noisy conditions

Abstract Motor bearing signals exhibit significant nonlinear and non-stationary behaviors under environmental noise, speed fluctuation, and multi-source interference, resulting in weak fault feature. The defects of traditional feature mode decomposition, such as strong noise sensitivity and parameter dependence on empirical selection. To address these challenges, a novel diagnostic approach termed dynamic kaiser feature mode decomposition (DKFMD) is proposed. Firstly, to enhance the noise suppression capability and spectral resolution, the Kaiser window is introduced to construct the kaiser feature mode decomposition that can adaptively decompose complex and high-dimensional signals. Secondly, the enhanced sparrow search algorithm (ESSA) is designed for the autonomous optimization of the key parameters in the decomposition (the number of modes and the length of the filters) so as to solve the problem of the subjectivity in the parameter selection. Subsequently, a sensitive mode selection mechanism based on the dispersion entropy kurtosis ratio (DEK) is established to facilitate the identification of fault-relevant modes and signal reconstruction. This mechanism quantitatively evaluates the complexity of mode components and the differences in their impact characteristics, enabling the effective selection of sensitive modes. Finally, the weak fault features under noise interference are extracted by combining with the envelope spectrum. Simulation and experimental results show that the proposed DKFMD can still maintain strong robustness under noise scenarios, and the performance is better than existing methods, providing a systematic adaptive diagnosis scheme for motor bearing faults.