High-speed train engine rolling bearings play a crucial role in maintaining engine health and minimizing operational losses during train operation. To solve the problems of low accuracy of the diagnostic model and unstable model due to the influence of noise during fault detection, a rolling bearing fault diagnosis model based on cross-attention fusion of WDCNN and BILSTM is proposed. The first layer of the wide convolutional kernel deep convolutional neural network (WDCNN) is used to extract the local features of the signal and suppress the high-frequency noise. A Bidirectional Long Short-Term Memory Network (BILSTM) is used to obtain global time series features of the signal. Cross-attention combines the WDCNN layer and the BILSTM layer so that the model can recognize more comprehensive feature information of the signal. Meanwhile, to improve the accuracy, Variable Modal Decomposition (VMD) is used to decompose the signals and filter and reconstruct the signals using envelope entropy and kurtosis, which enables the pre-processing of the signals so that the data input to the neural network contains richer feature information. The feasibility of the model is tested and experimentally validated using publicly available datasets. The experimental results show that the accuracy of the model proposed in this paper is significantly improved compared to the traditional WDCNN, BILSTM, and WDCNN-BILSTM models.

Aiming at the problem that the vibration signals of rolling bearings in high-speed rail traction motors are often affected by noise when they are in a fault state, which makes it very difficult to extract the fault features during fault diagnosis and causes obstruction in fault classification. The article proposes a rolling bearing fault diagnosis based on optimized variational mode decomposition (VMD) combined with signal features and an improved convolutional neural network (CNN). The golden jackal optimization (GJO) algorithm is employed to optimize the key parameters of the VMD, enabling effective signal decomposition. The decomposed signals are then filtered and reconstructed using criteria based on kurtosis and interrelationship measures. The time-domain features of the reconstructed signals are computed, and the feature vectors are constructed, which are used as inputs to the deep learning network; the CNN combined with the support vector machine (SVM) network model is used for the extraction of the features and the classification of the faults. The experimental results show that the method can effectively extract fault features in noise-covered signals, and the accuracy is also significantly improved compared with traditional methods.

Traditional bearing fault diagnosis methods struggle to effectively extract distinctive, domain-invariable characterizations from one-dimensional vibration signals of high-speed train (HST) bearings under variable load conditions. A deep migration fault diagnosis method based on the combination of a domain-adversarial network and signal reconstruction unit (CRU) is proposed for this purpose. The feature extraction module, which includes a one-dimensional convolutional (Cov1d) layer, a normalization layer, a ReLU activation function, and a max-pooling layer, is integrated with the CRU to form a feature extractor capable of learning key fault-related features. Additionally, the fault identification module and domain discrimination module utilize a combination of fully connected layers and dropout to reduce model parameters and mitigate the risk of overfitting. It is experimentally validated on two sets of bearing datasets, and the results show that the performance of the proposed method is better than other diagnostic methods under cross-load conditions, and it can be used as an effective cross-load bearing fault diagnosis method.

Fault diagnosis of rolling bearings is crucial for ensuring the stable operation of mechanical equipment and production safety in industrial environments. However, due to the nonlinearity and non-stationarity of collected vibration signals, single-modal methods struggle to capture fault features fully. This paper proposes a rolling bearing fault diagnosis method based on multi-modal information fusion. The method first employs the Hippopotamus Optimization Algorithm (HO) to optimize the number of modes in Variational Mode Decomposition (VMD) to achieve optimal modal decomposition performance. It combines Convolutional Neural Networks (CNN) and Gated Recurrent Units (GRU) to extract temporal features from one-dimensional time-series signals. Meanwhile, the Markovian Transition Field (MTF) is used to transform one-dimensional signals into two-dimensional images for spatial feature mining. Through visualization techniques, the effectiveness of generated images from different parameter combinations is compared to determine the optimal parameter configuration. A multi-modal network (GSTCN) is constructed by integrating Swin-Transformer and the Convolutional Block Attention Module (CBAM), where the attention module is utilized to enhance fault features. Finally, the fault features extracted from different modalities are deeply fused and fed into a fully connected layer to complete fault classification. Experimental results show that the GSTCN model achieves an average diagnostic accuracy of 99.5% across three datasets, significantly outperforming existing comparison methods. This demonstrates that the proposed model has high diagnostic precision and good generalization ability, providing an efficient and reliable solution for rolling bearing fault diagnosis.

For mechanical equipment to operate normally, rolling bearings—which are crucial parts of rotating machinery—need to have their faults diagnosed. This work introduces a bearing defect diagnosis technique that incorporates three-channel feature fusion and is based on enhanced Residual Networks and Bidirectional long- and short-term memory networks (ResNet-BiLSTM) model. The technique can effectively establish spatial-temporal relationships and better capture complex features in data by combining the powerful spatial feature extraction capability of ResNet and the bidirectional temporal modeling capability of BiLSTM. Specifically, the one-dimensional vibration signals are first transformed into two-dimensional images using the Continuous Wavelet Transform (CWT) and Markov Transition Field (MTF). The upgraded ResNet-BiLSTM network is then used to extract and combine the original one-dimensional vibration signal along with features from the two types of two-dimensional images. Finally, experimental validation is performed on two bearing datasets. The results show that compared with other state-of-the-art models, the computing cost is greatly reduced, with params and flops at 15.4 MB and 715.24 MB, respectively, and the running time of a single batch becomes 5.19 s. The fault diagnosis accuracy reaches 99.53% and 99.28% for the two datasets, respectively, successfully realizing the classification task.

In recent years, deep learning technology has been widely applied to bearing fault diagnosis, achieving significant progress. However, noise often masks fault characteristics, making feature extraction difficult and reducing diagnostic accuracy, which in turn hinders practical application. Therefore, this paper proposes a fault diagnosis model based on a gating mechanism and multi-scale feature fusion (multi-scale distributed convolution (MSDC)-Dense DGated convolutional neural network (CNN)). This model enhances noise resistance through a two-stage feature learning architecture: The first stage employs an MSDC module, which incorporates three parallel group convolutions with distinct dilation rates. This simultaneously filters noise while capturing multi-scale temporal features. Group CNN reduces parameter redundancy and strengthens local feature extraction. The second stage introduces a Dense DGated CNN block. This block dynamically suppresses noise interference through a product operation between feature branches and dilation-gated branches. A 1–1 convolution layer is added after convolutions with high dilation rates to mitigate the grid effect. Dense connections are employed to strengthen feature correlations. Additionally, the model incorporates a squeeze and excitation module that utilizes channel attention to amplify critical fault features while suppressing noise-dominated redundant information adaptively. The proposed model was validated on the CWRU and self-test datasets under various noise conditions. Experimental results demonstrate that the model achieves a diagnostic accuracy of 89.70% even in high-noise environments with a signal-to-noise ratio of −10 dB. Characterized by minimal parameters and strong anti-interference capabilities, it is well-suited for fault detection tasks in industrial time-series signals that are heavily influenced by substantial noise interference.