Bearing Fault Diagnosis Method Based on Optimized VMD and Deep Timing Fusion

In the current field of rolling bearing fault diagnosis, two critical challenges persist. First, it is difficult to effectively extract fault features from nonlinear and non-stationary vibration signals. Second, precise diagnosis remains a challenge, especially when distinguishing between different fault types and capturing incipient faults with weak characteristic information. To address these issues, this paper proposes a novel fault diagnosis method based on adaptively optimized Variational Mode Decomposition (VMD) and Deep Temporal Fusion. First, the method improves the traditional Sparrow Search Algorithm (SSA). It enhances SSA’s global optimization capability through strategies like chaotic population initialization and adaptive perturbation. This improved SSA enables efficient global optimization of VMD’s key parameters. Leveraging these optimized parameters, the method decomposes modal signal components with different center frequencies from the vibration signal. This process effectively separates fault-related components from noise and interference, laying the groundwork for subsequent feature analysis. Second, the paper innovatively constructs a deep temporal fusion model. This model integrates Transformer and Bidirectional Gated Recurrent Unit (BiGRU), aiming to realize high-precision fault mode identification. The adaptive optimization of VMD’s key parameters, which is powered by the improved SSA, enhances both the accuracy and robustness of vibration signal decomposition under various fault conditions. Concurrently, the deep temporal fusion model can effectively capture long-range global dependencies in vibration signals and extract local dynamic temporal features. These two capabilities not only significantly improve diagnostic accuracy compared to traditional methods but also overcome a key limitation of single-model approaches: their inability to balance global and local feature focus. Experimental results across multiple public datasets demonstrate that the proposed method exhibits higher parameter optimization efficiency and more accurate fault identification capability. Thus, it provides valuable references and technical guidance for future research in the field of rolling bearing fault diagnosis.