Deep neural network (DNN), with the capacity for feature inference and nonlinear mapping, has demonstrated its effectiveness in end-to-end fault diagnosis. However, the intermediate learning process of the DNN architecture is invisible, making it an uninterpretable black-box model. In this paper, a stacked residual multi-attention network (SRMANet) is proposed as a means of feature extraction of vibration signals, and visualizing the model training process, designing Squeeze-excitation residual (SE-Res) blocks to obtain additive features with minimal redundancy and sparsity. This study recommends the use of the attention fusion unit to ensure the interpretability of the model and ultimately to obtain representative features. By feeding the output gradient of the attention layer back to the original signal, the key feature components in the time domain signal can be effectively captured. Finally, the interpretability, identification accuracy and adaptability of the model under different operating conditions are verified on 12 different fault tasks in the planetary gearbox.

Mechanical condition monitoring data in real engineering are often severely unbalanced, which can lead to a decrease in the stability and accuracy of intelligent diagnosis methods. In this paper, a fault diagnosis method based on the SMOTE + Tomek Link and dual-channel feature fusion is proposed to improve the performance of the sample imbalance fault diagnosis method, taking the piston pump of a turnout rutting machine as the research object. Combining the data undersampling method and the oversampling method to redistribute the collected normal data and fault data makes the diagnostic model have better diagnostic performance in the case of insufficient fault samples. And, in order to fully utilize the global features and local features, a global–local feature complementary module (GLFC) is proposed. Firstly, the generated data similar to the original data are constructed using the SMOTE + Tomek Link method; secondly, the generated data are input into a GLFC module and BiGRU at the same time, the GLFC module extracts the spatial global features and local features of the original vibration data, and BiGRU extracts the temporal information features of the original vibration data, and fuses the extracted feature information, and inputs the fused features into the attention layer; finally, a GLFC module is proposed by the SMOTE + Tomek Link method to make full use of the global features and local features. The extracted feature information is fused, and the fused features are input to the attention layer; finally, the fault classification is completed by the softmax classifier. In this paper, the accuracy and robustness of the proposed model are demonstrated through experiments.

Rotating machinery is widely used in modern industrial systems, and its health status can directly impact the operation of the entire system. Timely and accurate diagnosis of rotating machinery faults is crucial for ensuring production safety, reducing economic losses, and improving efficiency. Traditional deep learning methods can only extract features from the vertices of the input data, thereby overlooking the information contained in the relationships between vertices. This paper proposes a Legendre graph convolutional network (LGCN) integrated with a self-attention graph pooling method, which is applied to fault diagnosis of rotating machinery. The SA-LGCN model converts vibration signals from Euclidean space into graph signals in non-Euclidean space, employing a fast local spectral filter based on Legendre polynomials and a self-attention graph pooling method, significantly improving the model’s stability and computational efficiency. By applying the proposed method to 10 different planetary gearbox fault tasks, we verify that it offers significant advantages in fault diagnosis accuracy and load adaptability under various working conditions.

A fault diagnosis method based on meta-learning and an improved multi-channel convolutional neural network (MAML-MCCNN-ISENet) was proposed to solve the problems of insufficient feature extraction and low fault type identification accuracy of vibration signals at small sample sizes. The signal is first preprocessed using adaptive chirp mode decomposition (ACMD) methods. A multi-channel input structure is then employed to process the multidimensional signal information after preprocessing. The improved squeeze and excitation networks (ISENets) have been enhanced to concurrently enhance the network’s adaptive perception of the significance of each channel feature. On this basis, a meta-learning strategy is introduced, the learning process of model initialization parameters is improved, the network is optimized by a multi-task learning mechanism, and the initial parameters of the diagnosis model are adaptively adjusted, so that the model can quickly adapt to new fault diagnosis tasks on limited datasets. Then, the overfitting problem under small sample conditions is alleviated, and the accuracy and robustness of fault identification are improved. Finally, the performance of the model is verified on the experimental data of the fault diagnosis of the laboratory plunger pump and the vibration dataset of the centrifugal pump of the Saint Longoval Institute of Engineering and Technology. The results show that the diagnostic accuracy of the proposed method for various diagnostic tasks can reach more than 90% on small samples.

In practical engineering, the asymmetric problem of the domain label space is inevitable owing to the prior fault information of the target domain being difficult to completely obtain. This implies that the target domain may include unseen fault classes or lack certain fault classes found in the source domain. To maintain diagnostic performance and knowledge generalization across different speeds, cross-domain intelligent fault diagnosis (IFD) models are widely researched. However, the rigid requirement for consistent domain label spaces hinders the IFD model from identifying private fault patterns in the target domain. In practical engineering, the asymmetric domain label space problem is inevitable, as the target domain’s fault prior information is difficult to completely obtain. This means that the target domain may have unseen fault classes or lack some source domain fault classes. To address these challenges, we propose an asymmetric cross-domain IFD method with label position matching and boundary sparse learning (ASY-WLB). It reduces the IFD model’s dependence on domain label space symmetry during transient speed variation. To integrate signal prior knowledge for transferable feature representation, angular resampling is used to lessen the time-varying speed fluctuations’ impact on the IFD model. We design a label-positioning information compensation mechanism and weighted contrastive domain discrepancy, accurately matching unseen class label information and constraining the diagnosis model’s decision boundary from a data conditional distribution perspective. Finally, extensive experiments on two time-varying speed datasets demonstrate our method’s superiority.

As a powerful tool for sparse signal representation, dictionary learning has gained traction in data-driven fault diagnosis. However, noise interference exists in actual industrial environments; noise as a common feature across all classes inevitably covers discriminative features, thereby impairing the diagnostic performance of methods reliant on dictionary learning. To tackle this challenge, this paper proposes a multi-sensor dictionary fusion-driven sparse representation classification method. The initial step involves leveraging a low-rank shared dictionary learning technique to extract discriminative dictionaries from multi-sensor data. A dictionary fusion metric is introduced based on dictionary reconstructability, dictionary discriminativity, and sparse code discriminability to fuse dictionaries from different sensors. Second, a sparse code discriminative enhancement method is proposed, which suppresses sparse code in non-correct projection regions by setting them to zero, thereby reducing the contribution of sparse code from incorrect projection areas. A classification framework grounded in sparse representation is then formulated. Validation experiments on two distinct planetary gearbox datasets (a laboratory planetary gearbox dataset and the southeast university planetary gearbox dataset) reveal that the multi-sensor dictionary fusion-driven sparse representation classification method attains 99.95% and 99.94% classification accuracy. Comparative analyses further substantiate its exceptional diagnostic precision against contemporary advanced techniques.