Research on Fault Diagnosis of Motor Bearings Based on SBOA-VMD-TCN

Motor bearings are critical components in rotating machinery, and their failures are among the primary causes of motor breakdowns, often leading to unplanned downtime and significant economic losses. Therefore, accurate and early fault diagnosis for motor bearings is of paramount importance for predictive maintenance. The core task of data-driven fault diagnosis is to identify characteristic fault signatures (e.g., specific frequency components) from measured signals, such as vibration or stator current. However, this remains a challenging problem due to the low signal-to-noise ratio in practical industrial environments, the weak nature of incipient fault features, and the complexity of signal patterns under varying operational conditions. To address these challenges, this paper proposed a hybrid intelligent fault diagnosis model that combined Variational Mode Decomposition (VMD), correlation-based Intrinsic Mode Function (IMF) filtering, and a Time Convolutional Network (TCN). Unlike pure end-to-end deep learning approaches, the proposed method utilized VMD and physically-informed heuristic features (nine time-domain statistical features) to construct a more discriminative input representation, thereby significantly enhancing the performance and interpretability of the TCN classifier. By leveraging the global search capabilities of the Secretary Bird Optimization Algorithm (SBOA) to achieve higher accuracy while maintaining computational efficiency, and by enhancing the efficient decomposition capabilities of VMD, the proposed SBOA-VMD-TCN model effectively captured both global and local features, enabled robust temporal modeling capabilities, and significantly improved diagnostic performance. The experimental results showed that on two different datasets, the model reduced the error by 39% and improved the accuracy by 5% on the Case Western Reserve University (CWRU) dataset, while also reducing the total processing time by 31%. On the Southeast University (SEU) dataset, it reduced the error by 37%, improved the accuracy by 7%, and reduced the total processing time by 52%.