Search

Recent intelligent fault diagnosis of rotating machinery focuses on extracting inherent features using the powerful deep learning. Although the deep learning methods are efficient, they ignore maximizing functions of the available working condition and health condition prior knowledge. Hence, a novel diagnosis method referred to as deep sparse topology network (DSTN) is proposed to extract highly regular and fault-sensitive features. To maximize the health condition label function in fault-sensitive feature extraction, a regular and sparse feature topology is developed through a parameterized sparse label matrix. Along with topology optimization, working condition labels are used utilizing the high-order Kullback–Leibler divergence, to enhance feature robustness to working condition variation and noise. Finally, with these highly regular and fault-sensitive features, the proposed DSTN directly points out the predicted health condition labels. To the best of our knowledge, the proposed network is the first attempt to directly diagnose faults through highly regular fault-sensitive features. The mechanism analysis and experimental investigations validate the efficiency and robustness of DSTN. [ABSTRACT FROM AUTHOR]

Highlights • Tuned noise can boost network performance in a certain degree. • Conditional distribution discrepancy can degenerate common fault diagnosis methods. • Joint distribution alignment via discriminative features is also feasible. Abstract Vibration signals are closely linked with health conditions of rotating machines and widely used in fault diagnosis. Unfortunately, traditional vibration signal-based fault diagnosis methods are under a universal assumption that the target vibration signals for application and the available vibration signals for model training are collected from the same distribution, which is always impractical in real-world scenarios due to working condition variation. For robust fault diagnosis under variable working conditions, although some transfer learning-based methods are proposed, they mostly aim at aligning only the marginal distribution discrepancy of datasets, which is validated not sufficient in some cases. Hence, we propose a new transfer learning method called improved joint distribution adaptation (IJDA) to align both the marginal and conditional distributions of datasets more comprehensively. Meanwhile, built on it, a working condition-robust fault diagnosis method is developed, which utilizes vibration signals and is mainly composed of three parts. Firstly, a new data augmentation method is developed to generate more useful samples for imbalanced vibration signals, which innovatively uses noise to boost network performance. Secondly, sparse filtering (SF) is employed to reduce the input dimension of IJDA. Finally, IJDA is utilized to firstly extract both sharing and principal features and then diagnose the features. Experiments on vibration signal datasets of roller bearings and a gearbox and comparisons with other methods verify its effectiveness and applicability. [ABSTRACT FROM AUTHOR]

The rolling bearing is an important part of the train’s running gear, and its operating state determines the safety during the running of the train. Therefore, it is important to monitor and diagnose the health status of rolling bearings. A convolutional neural network is widely used in the field of fault diagnosis because it does not require feature extraction. Considering that the size of the network model is large and the requirements for monitoring equipment are high. This study proposes a novel bearing fault diagnosis method based on lightweight network ShuffleNet V2 with batch normalization and L2 regularization. In the experiment, the one-dimensional time-domain signal is converted into a two-dimensional Time-Frequency Graph (TFG) using a short-time Fourier transform, though the principle of graphics to enhance the TFG dataset. The model mainly consists of two units, one for extracting features and one for spatial down-sampling. The building units are repeatedly stacked to construct the whole model. By comparing the proposed method with the origin ShuffleNet V2, machine learning model and state-of-the-art fault diagnosis model, the generalization of the proposed method for bearing fault diagnosis is verified.

• A new cross-level fusion neural network (CLFNN) is proposed by introducing a cross-level fusion method for FSL, which compares the similarity of feature embedding from different feature space perspectives and calculates the distribution differences of different data sets based on second-order statistical information. It can also obtain the abstract information of macroscopic data distribution and convolutional kernel scale and adaptively fuses the obtained similarity information. Through these methods, the data within a class are more tightly distributed in a smaller feature space, and more discriminative task-specific features are generated. This also allows CLFNN to capture comprehensive similarities and refine features at different levels of vibration signals in parallel. Effectively alleviates the domain shift problem in the data sparsity scenario. • A new CVWCF is proposed as an example of the domain shift problem in a data sparsity scenario. The CVWCF couples a variety of complex tasks that cannot be handled by traditional intelligent fault diagnosis methods at the same time, which is closer to the actual industrial needs. The proposed method will be compared in this more complex scenario to verify the superiority. • Ablation experiments are implemented on the dataset to find the contribution of different levels and the effectiveness of cross-level fusion methods. The proposed CLFNN belongs to the FSL method, which alleviates the problem of relying on the target domain label information and improves the adaptability to complex coupling conditions. The results show that it has excellent fault diagnosis ability of CVWCF. Rotating machinery fault diagnosis based on deep learning has been successfully applied in modern industrial equipment. However, many existing types of research suffer from two significant deficiencies. First, most deep neural networks are based on a single or same kind of similarity measurement method, which cannot fully exploit the data to extract different levels of feature information. Second, most intelligent fault diagnosis methods can only partially solve the data sparsity and domain shift problems caused by small samples, noise, variable working conditions or compound faults. The model's performance will degenerate rapidly when the above problems occur simultaneously. To address this problem, this paper develops a cross-level fusion neural network method that extracts abundant information on features by calculating spatial-level, channel-level, and second-order statistical information and adaptively fusing the three levels to obtain the final relationship score. First, the signal is input into the embedding module through a Fast Fourier Transform to obtain the feature embedding of the one-dimensional sequence signal. Then, the cross-level metrics learning module calculates the similarity of query sets and support sets at different levels. Finally, the similarities of different levels are fused through the adaptive fusion module to output the final relationship score. The bearing fault diagnosis experiments in the compound variable condition scenario show that the proposed method improves at least 78.53% compared to the traditional deep learning method, at least 3.22% and at most 35.52% compared to multiple few-shot learning methods. In addition, the ablation test analyzes the contribution of different level measurement methods to the model, and the maximum difference between them will reach 32.49%. In summary, the cross-level fusion method can effectively alleviate the data sparsity and domain shift problems. [ABSTRACT FROM AUTHOR]

The rolling bearing is an important part of the train's running gear, and its operating state determines the safety during the running of the train. Therefore, it is important to monitor and diagnose the health status of rolling bearings. A convolutional neural network is widely used in the field of fault diagnosis because it does not require feature extraction. Considering that the size of the network model is large and the requirements for monitoring equipment are high. This study proposes a novel bearing fault diagnosis method based on lightweight network ShuffleNet V2 with batch normalization and L2 regularization. In the experiment, the one-dimensional time-domain signal is converted into a two-dimensional Time-Frequency Graph (TFG) using a short-time Fourier transform, though the principle of graphics to enhance the TFG dataset. The model mainly consists of two units, one for extracting features and one for spatial down-sampling. The building units are repeatedly stacked to construct the whole model. By comparing the proposed method with the origin ShuffleNet V2, machine learning model and state-of-the-art fault diagnosis model, the generalization of the proposed method for bearing fault diagnosis is verified. [ABSTRACT FROM AUTHOR]