The multisource unsupervised domain adaptation (MUDA) scenario poses a significant challenge in the field of intelligent fault diagnosis (IFD), where the goal is to transfer the knowledge learned from multiple labeled source domains to an unlabeled target domain. Existing IFD-oriented MUDA approaches frequently fail to recognize the distinct importance of each source domain relative to specific target samples, or lack flexibility in integrating diagnostic insights from multiple sources. In response, a novel MUDA approach is proposed for IFD, termed point-to-set metric-gated mixture of experts (PSMMoEs). This method leverages a mixture-of-experts (MoEs) framework to automatically integrate the complementary information from multiple source domains. It develops a deep point-to-set distance (PSD) metric learning technique within the MoE's gating mechanism, effectively fusing domain-specific features by assessing the similarity between individual target samples and each source domain. The method ensures balanced training across progressive stages, harmonizing multitask learning with joint training for the MoE framework. Furthermore, a multilayer maximum mean discrepancy (MMD) measurement is employed for domain alignment, ensuring feature alignment across different domains at multiple levels. In order to assess the efficacy of the proposed method, it is compared with several leading domain adaptation methods on publicly available and laboratory-based rotating machinery fault datasets. The experimental results demonstrate superior classification and adaptation capabilities of the proposed fault diagnosis method.

In recent years, domain adaptation has emerged as a promising method for overcoming limitations in labeled data availability and shifting data distributions in fault diagnosis. However, current unsupervised domain adaptation methods primarily emphasize global alignment between source and target domain distributions, ignoring the impact of hard-to-transfer classes and individual samples on transfer performance. To address this, a dynamic confusion-aware correlation network (DCACN) is proposed to achieve significant transfer gains by focusing on aspects with high confusion of the transfer process. For class-level confusion, class confusion is incorporated into the target domain class correlation matrix through a reweighting mechanism based on the source domain confusion matrix. The minimum class confusion (MCC) theory is leveraged to mitigate inter-class confusion, specifically for high-confusion classes. For sample-level confusion, a dynamic Focal Loss is proposed by integrating a transfer confusion factor derived from the confusion-aware correlation matrix, enhancing the ability to focus on highly confusing samples. Experimental evaluations on public and real-world datasets demonstrate the effectiveness of the proposed method, showing significantly improved accuracy in cross-domain fault classification compared to multiple state-of-the-art methods.

Recently, intelligent fault diagnosis has become the focus of research. The powerful feature extraction capability enables it to show better results for large amounts of data. Because of this, the data requirements of fault diagnosis methods consume a lot of manpower and material resources, especially the data annotation process. Sufficient labeled data cannot be met in real industrial conditions. We propose a novel framework for contrastive learning and semi-supervised learning, where the pseudo labels from the SemiSL model assist in the clustering of the contrast space by additional guidance, thereby improving the feature extraction effect of contrastive learning. Meanwhile, we design a new class-aware loss that can promote the compactness of learning similar representations and the separability of different samples. Through experimental verification on classic bearing datasets, the reliability of the method proposed has been sianificantly improved.

Fault diagnosis in rotating machinery is crucial for maintaining operational reliability and safety. However, developing models that reliably detect faults across diverse operational domains from limited training data remains a challenge. Current domain generalization methods often struggle to adequately represent the variability and complexity of unseen data patterns, leading to suboptimal diagnostic accuracy. This paper introduces a novel data generation approach, utilizing diffusion models to significantly bolster the generalization ability of fault diagnosis systems. By imposing innovative technique of novel style synthesis, our conditional diffusion model generates realistic and diverse fault signals that expand beyond the source domain's data distribution. A hybrid training strategy, leveraging both original and generated data, substantially improves the performance of the fault classifier, equipping it with enhanced capabilities to handle unseen data distributions. Rigorous testing on two fault datasets for rotating machinery demonstrates the superior diagnostic accuracy of our method against previous leading domain generalization methods, confirming its exceptional cross-domain generalization potential.

Motor fault diagnosis is imperative to enhance the reliability and security of industrial systems. However, since motors are often operated under nonstationary conditions, the high complexity of vibration signals raises notable difficulties for fault diagnosis. Therefore, considering the special physical characteristics of motor signals under nonstationary conditions, in this article, we propose a multiscale kernel based residual convolutional neural network (CNN) for motor fault diagnosis. Our contributions mainly fall into two aspects. First, we notice that each motor fault category has various patterns in vibration signals due to the changing operational conditions of the motor. To capture these patterns, a multiscale kernel algorithm is applied in the CNN architecture. Second, since the motor vibration signals are made up of many different components from different transfer paths, they are very complex and variable. To enable the architecture to extract fault features from deep and hierarchical representation spaces, sufficient depth of the network is needed, which will lead to the degradation problem. In the proposed method, residual learning is embedded into the multiscale kernel CNN to avoid performance degradation and build a deeper network. To validate the effectiveness of the proposed networks, a normal motor and five motors with different failures are tested. The results and comparisons with state-of-the-art methods highlight the superiority of the proposed method.

Fault diagnosis and remaining useful life (RUL) prediction are always two major issues in modern industrial systems, which are usually regarded as two separated tasks to make the problem easier but ignore the fact that there are certain information of these two tasks that can be shared to improve the performance. Therefore, to capture common features between different relative problems, a joint-loss convolutional neural network (JL-CNN) architecture is proposed in this paper, which can implement bearing fault recognition and RUL prediction in parallel by sharing the parameters and partial networks, meanwhile keeping the output layers of different tasks. The JL-CNN is constructed based on a CNN, which is a widely used deep learning method because of its powerful feature extraction ability. During optimization phase, a JL function is designed to enable the proposed approach to learn the diagnosis-prognosis features and improve generalization while reducing the overfitting risk and computation cost. Moreover, because the information behind the signals of different problems has been shared and exploited deeper, the generalization and the accuracy of results can also be improved. Finally, the effectiveness of the JL-CNN method is validated by run-to-failure dataset. Compared with support vector regression and traditional CNN, the mean-square-error of the proposed method decreases 82.7% and 24.9%, respectively. Therefore, results and comparisons show that the proposed method can be applied for the intercrossed applications between fault diagnosis and RUL prediction.

Identifying impulsive features from massive amounts of dynamic signals for wind turbine systems is like finding needles in haystacks, leading to a major challenge for the Shannon-sampling-theorem-based fault detection techniques. Therefore, this paper describes and analyzes a novel impulsive feature identification technique based on compressed sensing model and convex optimization techniques. One important point of this work is to establish the prior information that periodic impulsive component is frequency compressible. On the other hand, based on a small set of nonadaptive linear measurements, a convex optimization algorithm generated from a popular algorithmic framework (alternating direction of multiplier method) is developed to recover the impulsive features. Consequently, the main highlight of this work is to enable the high-accuracy recovery of impulsive feature signals from far few measurements than the Shannon sampling theory requires. Extensive numerical studies are implemented to quantitatively evaluate the performance of the proposed technique, and its feasibility and superiority are verified simultaneously. More importantly, the validity and applicability of our impulsive feature detection technique is comprehensively investigated and confirmed based on a practical engineering dataset from a wind turbine gearbox in a wind farm.

In most current intelligent diagnosis methods, fault classifiers of electric machine are built based on complex handcrafted features extractor from raw signals, which depend on prior knowledge and is difficult to implement intelligentization authentically. In addition, the increasingly complicated industrial structures and data make handcrafted features extractors less suited. Convolutional neural network (CNN) provides an efficient method to act on raw signals directly by weight sharing and local connections without feature extractors. However, effective as CNN works on image recognition, it does not work well in industrial applications due to the differences between image and industrial signals. Inspired by the idea of CNN, we develop a novel diagnosis framework based on the characteristics of industrial vibration signals, which is called dislocated time series CNN (DTS-CNN). The DTS-CNN architecture is composed of dislocate layer, convolutional layer, sub-sampling layer and fully connected layer. By adding a dislocate layer, this model can extract the relationship between signals with different intervals in periodic mechanical signals, thereby overcome the weaknesses of traditional CNNs and is more applicable for modern electric machines, especially under nonstationary conditions. Experiments under constant and nonstationary conditions are performed on a machine fault simulator to validate the proposed framework. The results and comparison with respect to the state of the art in the field is illustrated in detail, which highlights the superiority of the proposed method in industrial applications.

It is always a primary challenge in fault diagnosis of a wind turbine generator to extract fault character information under strong noise and nonstationary condition. As a novel signal processing method, sparse representation shows excellent performance in time-frequency analysis and feature extraction. However, its result is directly influenced by dictionary, whose atoms should be as similar with signal's inner structure as possible. Due to the variability of operation environment and physical structure in industrial systems, the patterns of impulse signals are changing over time, which makes creating a proper dictionary even harder. To solve the problem, a novel data-driven fault diagnosis method based on sparse representation and shift-invariant dictionary learning is proposed. The impulse signals at different locations with the same characteristic can be represented by only one atom through shift operation. Then, the shift-invariant dictionary is generated by taking all the possible shifts of a few short atoms and, consequently, is more applicable to represent long signals that in the same pattern appear periodically. Based on the learnt shift-invariant dictionary, the coefficients obtained can be sparser, with the extracted impulse signal being closer to the real signal. Finally, the time-frequency representation of the impulse component is obtained with consideration of both the Wigner-Ville distribution of every atom and the corresponding sparse coefficient. The excellent performance of different fault diagnoses in a fault simulator and a wind turbine proves the effectiveness and robustness of the proposed method. Meanwhile, the comparison with the state-of-the-art method is illustrated, which highlights the superiority of the proposed method.