Bearing Fault Diagnosis Using PSO-VMD and a Hybrid Transformer-CNN-BiGRU Model
Hualin Dai, Daoxuan Yang, Liying Zhang, Guorui Liu
Abstract
Reliable bearing fault diagnosis is essential for the steady running of mechanical systems. However, existing diagnostic models still face significant limitations in feature extraction, primarily due to the non-stationary and nonlinear characteristics of vibration signals, which lead to a decline in diagnostic performance. To address this issue, this paper proposes a novel diagnostic framework that combines Particle Swarm Optimization-based Variational Mode Decomposition (PSO-VMD) for feature extraction with a deeply integrated Transformer-Convolutional Neural Network-Bidirectional Gated Recurrent Unit (TCB) model for fault classification. Bearing fault diagnosis is crucial for the stable operation of mechanical equipment, yet existing models often suffer from limited feature extraction and low detection accuracy. To address this, PSO-VMD is employed to extract informative, band-limited features from vibration signals, yielding a highly correlated feature set; a composite model TCB, combining a Transformer, a CNN, and a bidirectional GRU (BiGRU), is then used for fault classification. To prevent window-level leakage, the dataset is split before windowing and normalization, and all baselines are aligned under identical preprocessing and training settings. On the CWRU benchmark, the model attains 98.9% accuracy, 98.8% precision, 99.4% recall, 99.1% F1, and macro-F1 = 0.9766 over five runs. The approach offers a favorable accuracy –latency trade-off and yields interpretable, band-limited modes, supporting reproducible deployment in practice.