278 / 2025-06-15 22:32:43

Multi-Source Data Driven Parallel Branch VIT Intelligent Diagnosis Method for Rolling Bearing Faults

Intelligent fault diagnosis,Multi-branch ViT,Mixed attention mechanism,Gamma correction

Special Sessions > SSL 05 Condition monitoring, fault diagnosis and intelligent maintenance of mechanical equipment

Traditional feature fusion fails to characterize true bearing faults under structural complexity and multi-factor influences, this paper proposes a multi-source fusion intelligent fault diagnosis method based on parallel Vision Transformer (ViT) and hybrid attention mechanisms. The method performs time-frequency feature extraction on multi-source vibration data collected from multiple vibration sensors, obtaining feature image data that undergoes Gamma correction preprocessing. A diagnostic model architecture is constructed comprising multiple parallel ViT networks, incorporating a hybrid attention mechanism in the fusion layer to dynamically adjust the contribution weights of features from each sensor. The integration of multiple parallel ViT network branches with an attention-based feature fusion module effectively captures interdependencies within multi-sensor data, thereby enhancing system performance. The results show that the proposed method can effectively achieve high-precision image classification and rolling bearing fault recognition.