Intelligent fault diagnosis based on multi-source information fusion and attention-enhanced networks

Deep learning has been widely applied in the field of intelligent fault diagnosis, achieving remarkable progress in feature extraction and classification performance. However, most existing methods still face challenges in simultaneously capturing the temporal information and global features during the bearing operation process, which leads to insufficient acquisition of fault-related information. Moreover, under complex and harsh working environments, single-source fault diagnosis methods often struggle to stably extract fault features. To address these issues, this paper proposes an intelligent fault diagnosis method based on multi-source information fusion, aiming to enhance stability through the extraction and integration of rich feature representations. Specifically, vibration and current signals are transformed from raw time-domain data into time-frequency representations using continuous wavelet transform. At the image level, a grayscale-weighted fusion strategy is employed to effectively integrate multi-source information. In terms of model design, a diagnostic framework combining convolutional neural networks with attention mechanisms is constructed, enabling effective capture of both temporal information and global feature dependencies of bearing faults. Experimental results on a publicly available bearing fault dataset demonstrate that the proposed method consistently outperforms existing single-source and multi-source diagnosis models across various evaluation metrics, achieving higher fault recognition accuracy.