Rolling bearing health status diagnosis method based on bidirectional feature extraction by fusing CNN and informer

Abstract Conventional single-dimensional feature extraction methods in rolling bearing fault diagnosis often fail to capture the rich multi-scale and hierarchical fault signatures inherent in vibration signals. To address this limitation, this paper proposes a bidirectional feature extraction framework that synergistically combines convolutional neural networks (CNNs) and Informer to enhance fault diagnosis performance. Firstly, CNN is utilized to capture the local detail features of vibration signals, while the Informer model is employed to extract the global temporal dependency features of the signals. Subsequently, a dual attention mechanism (DA) is introduced, and through an adaptive weight fusion strategy, the complementary and enhancing effects of local and global features are achieved, thereby constructing an end-to-end intelligent recognition diagnostic model of CNN-Informer-DA from feature extraction to fault classification. Finally, to evaluate the generalization ability and diagnostic accuracy of the proposed model, comparative experiments are conducted on two authoritative bearing public datasets from Case Western Reserve University (CWRU) and Southeast University (SEU). The results show that the proposed CNN-Informer-DA model achieves fault diagnosis accuracies of 99.11% and 98.44% on the CWRU and SEU datasets, respectively, and demonstrates stronger robustness in noisy environments. This research provides a high-precision and robust technical solution for bearing fault diagnosis and health management in complex industrial scenarios.