Bearing fault diagnosis method based on enhanced VMD and adaptive-optimized SDAE.

Motor rolling bearing is a fundamental component of industrial production, and its vibration signal extraction and fault diagnosis are challenging because of the effect of operating characteristics and external noise. This research initially proposes an adaptive variational mode decomposition approach based on dung beetle optimization algorithm to decompose and extract signals. At the same time, a composite optimization indicator function based on Tanimoto coefficient, permutation entropy and kurtosis are presented as the fitness function of decomposition to increase the flexibility and robustness of the technique. Next it combines with composite multiscale permutation entropy to finish feature extraction and create feature vectors. Finally, an enhanced inertia weights and Cauchy chaotic mutation-Sine Cosine Algorithm is utilized to optimize the hyperparameters of the stacked denoising auto-encoders network and construct a fault diagnosis model. The CWRU open bearing dataset is used to comprehensively evaluate the performance of the method, and the experimental results will be compared to show that the method proposed in this paper can effectively extract signal features in the situation of strong noise, while ensuring a high prediction accuracy, and has stronger adaptability and noise resistance compared with other methods.