Wind turbine rolling bearings are crucial components for ensuring the reliability and stability of wind power systems. Their failure can lead to significant economic losses and equipment downtime. Therefore, the accurate diagnosis of bearing faults is of great importance. Although existing deep learning fault diagnosis methods have achieved certain results, they still face limitations such as inadequate feature extraction capabilities, insufficient generalization to complex working conditions, and ineffective multi-scale feature capture. To address these issues, this paper proposes an advanced fault diagnosis method named the two-stream feature fusion convolutional neural network (TSFFResNet-Net). Firstly, the proposed method combines the advantages of one-dimensional convolutional neural networks (1D-ResNet) and two-dimensional convolutional neural networks (2D-ResNet). It transforms one-dimensional vibration signals into two-dimensional images through the empirical wavelet transform (EWT) method. Then, parallel convolutional kernels in 1D-ResNet and 2D-ResNet are used to extract multi-scale features, respectively. Next, the Convolutional Block Attention Module (CBAM) is introduced to enhance the network’s ability to capture key features by focusing on important features in specific channels or spatial areas. After feature fusion, CBAM is introduced again to further enhance the effect of feature fusion, ensuring that the features extracted by different network branches can be effectively integrated, ultimately providing more accurate input features for the classification task of the fully connected layer. The experimental results demonstrate that the proposed method outperforms other traditional methods and advanced convolutional neural network models on different datasets. Compared with convolutional neural network models such as LeNet-5, AlexNet, and ResNet, the proposed method achieves a significantly higher accuracy on the test set, with a stable accuracy of over 99%. Compared with other models, it shows better generalization and stability, effectively improving the overall performance of rolling bearing vibration signal fault diagnosis. The method provides an effective solution for the intelligent fault diagnosis of wind turbine rolling bearings.

Rolling bearings, as critical components of rotating machinery, directly affect the reliability and efficiency of the system. Due to extended operation under high load, harsh environmental conditions, and continuous use, bearings become more susceptible to failure, leading to a higher likelihood of malfunction. To prevent sudden failures, reduce downtime, and optimize maintenance strategies, early and accurate diagnosis of rolling bearing faults is essential. Although existing methods have achieved certain success in processing acoustic and vibration signals, they still face challenges such as insufficient feature fusion, inflexible weight allocation, lack of effective feature selection mechanisms, and low computational efficiency. To address these challenges, we propose a dynamic weighted multimodal fault diagnosis model based on the fusion of acoustic and vibration information. This model aims to enhance feature fusion, dynamically adapt to signal characteristics, optimize feature selection, and reduce computational complexity. The model incorporates an adaptive fusion method based on a multi-branch convolutional structure, enabling unified processing of both acoustic and vibration signals. At the same time, a cross-modal dynamic weighted fusion mechanism is employed, allowing the real-time adjustment of weight distribution based on signal characteristics. By utilizing an attention mechanism for dynamic feature selection and weighting, the robustness of classification is further improved. Additionally, when processing acoustic signals, a depthwise separable convolutional network is used, effectively reducing computational complexity. Experimental results demonstrate that our method significantly outperforms other algorithms in terms of convergence speed and final performance. Additionally, the accuracy curve during training showed minimal fluctuation, reflecting higher robustness. The model achieved over 99% diagnostic accuracy under all signal-to-noise ratio (SNR) conditions, showcasing exceptional robustness and noise resistance in both noisy and high-SNR environments. Furthermore, its superiority across different data scales, especially in small-sample learning and stability, highlights its strong generalization capability.

To address the challenges of acoustic and vibration signals being susceptible to noise interference, coarse modality fusion, and static graph structures in industrial fault diagnosis, we propose a multimodal dynamic graph neural network combining node-level modality attention fusion and learnable soft K-nearest neighbor dynamic graph construction (LSKNN). In our method, each time-window node independently learns attention weights for acoustic and vibration modalities; both the number of neighbors and the distance threshold are trainable parameters for end-to-end adaptive graph structure; cross-modal edges are dynamically generated based on feature similarity. The model is optimized using a combination of cross-entropy loss and supervised contrastive loss. Extensive experiments on a self-collected dataset and the Beijing Jiaotong University bogie testbench public dataset (BJTU-RAO) demonstrate that our method achieves 100% classification accuracy; importantly, in noise robustness evaluations the accuracy remains above 99.8% while baseline methods experience significant performance degradation. Ablation studies further confirm that the three key modules of node-level modality attention, LSKNN dynamic graphs, and contrastive loss are all essential for maintaining high accuracy and noise resilience. This research highlights the indispensable value of systematic noise experiments for validating the robustness and reliability of multimodal fault diagnosis methods.

Amid the accelerating pace of global urbanization, the volume of municipal solid garbage has surged dramatically, thereby demanding more efficient and precise garbage management technologies. In this paper, we introduce a novel garbage classification approach that leverages a dynamic graph neural network based on multimodal feature fusion. Specifically, the proposed method employs an enhanced Residual Network Attention Module (RNAM) network to capture deep semantic features and utilizes CIELAB color (LAB) histograms to extract color distribution characteristics, achieving a complementary integration of multimodal information. An adaptive K-nearest neighbor algorithm is utilized to construct the dynamic graph structure, while the incorporation of a multi-head attention layer within the graph neural network facilitates the efficient aggregation of both local and global features. This design significantly enhances the model’s ability to discriminate among various garbage categories. Experimental evaluations reveal that on our self-curated KRHO dataset, all performance metrics approach 1.00, and the overall classification accuracy reaches an impressive 99.33%, surpassing existing mainstream models. Moreover, on the public TrashNet dataset, the proposed method demonstrates equally outstanding classification performance and robustness, achieving an overall accuracy of 99.49%. Additionally, hyperparameter studies indicate that the model attains optimal performance with a learning rate of 2 × 10⁻⁴, a dropout rate of 0.3, an initial neighbor count of 20, and 8 attention heads.