Dynamic evolution-driven domain adaptation with wavelet dual-path structure for cross-domain fault diagnosis

Domain adaptation (DA) techniques have shown effectiveness in cross-domain fault diagnosis; however, their performance is constrained by the closed-set assumption, which is often unrealistic in complex industrial environments. The inherent uncertainty and complexity of mechanical systems in industrial big data often lead to the emergence of novel and unknown fault types. To address this challenge, a practical open-set DA approach was proposed, named open-set classifier-guided dynamic joint domain adaptation (OC-DJDA). Built upon the domain adversarial neural network, OC-DJDA incorporates two key strategies, as suggested by its name. An open-set classifier (OC) is introduced to detect unknown samples and delineate clear boundaries between known and unknown classes without requiring predefined thresholds. Meanwhile, target domain-specific features are isolated to form a shared domain, thereby mitigating the influence of unknown categories during feature alignment. A dynamic joint distribution strategy is employed to adaptively align both marginal and conditional feature distributions across domains through a dynamic weighting mechanism. Experimental results demonstrate the robustness of OC-DJDA and highlight its potential as a reliable solution for open-set fault diagnosis in complex industrial scenarios. Notably, OC-DJDA surpasses several state-of-the-art methods, achieving classification accuracies of 94.79% and 91.70% in open transfer tasks within the same machine and across different machines, respectively.

The scarcity of labeled fault data in rotating machinery diagnosis poses significant challenges for model reliability, particularly when encountering novel operational scenarios or cross-domain applications. While unsupervised domain adaptation (UDA) has emerged as a promising solution, conventional UDA methods often exhibit static architectures that fail to dynamically adjust to target-domain distribution shifts, thereby limiting diagnostic accuracy and practical utility. To address these limitations, we propose an adaptive dynamic cooperative multisource domain adaptation network (ADCMDAN) for rotating machinery fault diagnosis. Firstly, a multi-perspective dynamic cooperative feature extractor is constructed to enable the model to dynamically adjust its architecture, thereby enhancing its ability to extract domain-invariant features across different domains. Secondly, a dynamic cooperative training mechanism is designed to adaptively adjust the training strategy that conducts targeted training for the model’s bi-module in different training phases, so as to strengthen the domain adaptation performance. Finally, a dual-object dynamic optimization loss is introduced to treat distinctively different source domains, thereby improving the utilization of existing diagnostic knowledge. Extensive experimental evaluations are conducted based on two different cases to rigorously assess the diagnostic performance of ADCMDAN. The results conclusively demonstrate that the proposed method achieves exceptional fault identification accuracy even in the absence of labeled target-domain data.

In machinery fault diagnosis, labeled data are always difficult or even impossible to obtain. Transfer learning can leverage related fault diagnosis knowledge from fully labeled source domain to enhance the fault diagnosis performance in sparsely labeled or unlabeled target domain, which has been widely used for cross domain fault diagnosis. However, existing methods focus on either marginal distribution adaptation (MDA) or conditional distribution adaptation (CDA). In practice, marginal and conditional distributions discrepancies both have significant but different influences on the domain divergence. In this paper, a dynamic distribution adaptation based transfer network (DDATN) is proposed for cross domain bearing fault diagnosis. DDATN utilizes the proposed instance-weighted dynamic maximum mean discrepancy (IDMMD) for dynamic distribution adaptation (DDA), which can dynamically estimate the influences of marginal and conditional distribution and adapt target domain with source domain. The experimental evaluation on cross domain bearing fault diagnosis demonstrates that DDATN can outperformance the state-of-the-art cross domain fault diagnosis methods.

Electroencephalogram (EEG) has been widely used in brain computer interface (BCI) due to its convenience and reliability. The EEG-based BCI applications are majorly limited by the time-consuming calibration procedure for discriminative feature representation and classification. Existing EEG classification methods either heavily depend on the handcrafted features or require adequate annotated samples at each session for calibration. To address these issues, we propose a novel dynamic joint domain adaptation network based on adversarial learning strategy to learn domain-invariant feature representation, and thus improve EEG classification performance in the target domain by leveraging useful information from the source session. Specifically, we explore the global discriminator to align the marginal distribution across domains, and the local discriminator to reduce the conditional distribution discrepancy between sub-domains via conditioning on deep representation as well as the predicted labels from the classifier. In addition, we further investigate a dynamic adversarial factor to adaptively estimate the relative importance of alignment between the marginal and conditional distributions. To evaluate the efficacy of our method, extensive experiments are conducted on two public EEG datasets, namely, Datasets IIa and IIb of BCI Competition IV. The experimental results demonstrate that the proposed method achieves superior performance compared with the state-of-the-art methods.

Semi-supervised multi-adversarial domain-adaptive fault diagnosis for hydraulic pumps from pressure simulation data to experimental data

A robust domain distribution alignment discriminative network driven by physical samples for rotor-bearing fault diagnosis

Domain adaptation is a research hotspot in the field of bearing fault diagnosis; however, extremely scarce bearing fault sample information limits the development of domain adaptation techniques. Thus, improving the generalization ability of fault diagnosis models is an urgent problem to be solved. Thus, this study proposed a dynamic feature separation domain generalization method for bearing fault diagnosis. The proposed method facilitated the dynamic adjustment of the influence of internal and mutual invariant features on the domain generalization process by learning the relationship between internal invariant features within the domain and mutual invariant features between the domains. This ensured the maximum utilization of invariant information in the features and obtainment of a domain-invariant bearing fault diagnostic model. Consequently, the fault diagnosis capability of the model was improved and the shortcomings of the poor generalization capability of existing models were alleviated. Finally, the effectiveness of the method was verified by comparing several bearing-fault datasets.

Domain adaptation is widely-used in deep neural networks to address the problem of data distribution shift. Most of the deep CNN models use the Maximum Mean Discrepancy(MMD) to measure the distribution difference between the source and task domains, which have achieved great success on transfer learning tasks. However, these conventional domain adaptation methods have limited use in dynamic knowledge adaptation due to the constant transfer coefficient for all adaptation layers. In this paper, we propose an efficient dynamic domain adaptation method on deep CNN models, which transfers knowledge dynamically according to different layers and training extent. Specifically, we first present a deep understanding of how transferable each layer in various deep CNN models is, including the different VGG and AlexNet architectures. Then, a dynamic transfer coefficient is proposed based on the deep understanding. By doing this, our paper gives guidance on how to choose transferring layers and adaptation coefficients aptly instead of using empirical constant transfer parameters in conventional methods. Extensive experiments conducted on standard benchmark datasets demonstrate that the proposed method achieves the state-of-the-art results by using dynamic domain adaptation parameters compared with conventional methods.

A complex process fault diagnosis method based on manifold distribution adaptation

Transfer learning can realize cross-domain fault diagnosis of rotating machinery, where the model trained on many labeled samples collected in one working condition can be transferred to insufficient samples collected in the target working condition. Currently, the data features cannot be completely extracted by existing methods when the data distribution gap of the samples collected in different working conditions is quite large. In order to fully extract the data features of rotating machinery to achieve cross-domain fault diagnosis, this paper investigated a cross-domain fault diagnosis model of rotating machinery based on graph feature extraction. The proposed method can realize unsupervised fault diagnosis on rotating machinery running under different working conditions by extracting the numerical and structural features of source and target domains. First of all, data features with large data distribution gaps need to be fully extracted, so a convolutional network based on a deformable convolutional network (De-conv) is designed to extract the features with large differences in data distribution under various working conditions. Secondly, features are extracted based on a convolutional neural network for data values in existing domain adaptation (DA) methods while the structure features of machine monitoring data are ignored. Therefore, a composite spectral-based graph convolutional network is designed to extract structural features of data. Thirdly, fully extracted features are input into a universal DA network to achieve cross-domain fault diagnosis of unknown faults in rotating machinery under changing working conditions. Finally, a benchmarking data set and a data set collected from a practical experimental platform are used to verify the effectiveness of the proposed model, and the results show that it is more suitable for cross-domain fault diagnosis of rotating machinery than other comparison models.

The computational grid provides a promising platform for the efficient execution of parallel coarse grain tasks computing over very large sample space. Scheduling such applications is challenging for the heterogeneity, autonomy, and dynamic adaptability of grid resources. Assuming resource owners have a preemptive priority, we propose an adaptive algorithm of jobs scheduling based on time balancing strategy, which solves the parallel computing tasks by using the idle resources in computational grid. A mathematical model is developed to predict performance, which also considers systems with heterogeneous machine utilization and heterogeneous service distribution. The model separates the influence of machine utilization, job service rate and parallel task allocation on the completion time. According to the time balancing policy, a task is partitioned into several subtasks and scheduled, and the costs of communication are reduced. The expected value of parallel task completion time is predicted with performance model. To get better parallel computing performance, an optimal subset of heterogeneous resources with the shortest parallel executing time of tasks can be selected with the efficient algorithm. Remapping strategy is applied during scheduling, which is more suitable for the dynamic adaptability and domain autonomy in the grid.

Cross-domain fault diagnosis using deep learning plays a critical role in ensuring the reliability and safety of mechanical systems. However, real-world industrial scenarios often involve unknown fault classes, which introduce significant challenges beyond environmental differences between training and testing phases. These unknown fault classes, which do not appear in the training data, create a cross-domain open set fault diagnosis problem where the target domain includes both known and unknown fault types with distinct distribution characteristics. Traditional domain adaptation methods that align source and target domains often overlook the spatial distribution of each class in the feature space, leading to potential negative transfer and misclassification of unknown faults. To address these challenges, this paper proposes a k-nearest neighbors based adaptive thresholding (KNNAT) method, which dynamically adjusts classification thresholds based on the spatial distribution of each class in the feature space. This approach effectively isolates unknown faults, reducing their impact on domain adaptation and improving the reliability of the diagnostic process. Extensive experiments on the publicly available CWRU bearing and PHM09 datasets demonstrate that the proposed KNNAT method outperforms other state-of-the-art methods, achieving higher accuracy and robustness in identifying known faults while successfully isolating unknown faults. These results highlight the potential of using the KNNAT method to enhance the reliability of mechanical systems in cross-domain fault diagnosis applications.

In response to the problems of inaccurate feature alignment, loss of source domain information, imbalanced sample distribution, and biased class decision boundaries in traditional unsupervised domain adaptation methods, this paper proposes a class decision boundary-based dynamic expansion network unsupervised domain adaptation method called CDE-Net. Specifically, our method dynamically expands the autoencoder-based network structure, which can preserve source domain feature information while gradually adapting to the target domain data distribution and learning useful feature information from the target domain. Meanwhile, by minimizing clustering loss and conditional entropy loss, CDE-Net can explore the intrinsic structure of the data and push class decision boundaries away from dense data areas. We experimentally verify our method on three medical image datasets, chest X-rays, intracranial hemorrhage, and mammography, and achieve an average AUC improvement of 25.8% or more compared to non-transfer methods. In addition, we compare our method with previous unsupervised domain adaptation methods, and the experimental results show that our method achieves better classification accuracy and generalization performance.

With the deepening advancement of smart manufacturing, regular fault diagnosis plays a crucial role in optimizing productivity and maintaining equipment safety and stability. Joint domain adaptation (JDA) techniques for fault diagnosis have garnered significant attention; however, their ability to characterize nonlinear, high-dimensional features remains limited. Additionally, low pseudo-label confidence levels result in insufficient distribution alignment, leading to domain drift that compromises cross-domain fault diagnosis performance. Based on this, this paper proposes a multi-kernel higher-order moment matching joint domain adaptive network (MKHoJDA) for cross-domain mechanical fault diagnosis. Firstly, a multi-scale dilatable residual convolution (MDRC) module is proposed to extract transferable features from both the source and target domains more accurately. Secondly, to address the limitation of traditional JDA in statistically measuring higher-order features, a multi-kernel higher-order moment matching (MKHoMM) statistical metric strategy is designed. Multiple kernel functions are combined to create a more expressive composite kernel, enhancing the power of feature mappings, reducing distribution discrepancies between domains. Additionally, to address the dependency of conditional distribution alignment (CDA) on pseudo-label accuracy, a prototype-aware pseudo-label feedback (PPF) mechanism was developed. This mechanism evaluates sample confidence using Jensen-Shannon (JS) divergence, selects high-confidence target domain samples for pseudo-label generation, and dynamically updates prototype centers. This approach effectively mitigates the negative impact of pseudo-labels while improving CDA performance. Finally, multiple cross-domain fault diagnosis experiments were conducted on the JNU and PU datasets. The average accuracy significantly surpassed that of other advanced comparative models, demonstrating that the proposed algorithm effectively mitigates the impact of data distribution differences and exhibits exceptional cross-domain fault diagnosis performance.

The diagnosis of ceramic bearings based on the fusion of heterogeneous data from multiple sources can reflect fault information from multiple dimensions, representing an advanced method of fault diagnosis. Nevertheless, the extant cross-domain fault diagnosis methods for ceramic bearings based on heterogeneous data fusion are constrained by the dimensionality of the data, the network’s inadequate capacity to comprehend the interrelationships between the data, and the inconsistency of the fused fault features, which impede cross-domain fault diagnosis for ceramic bearings. In this paper, we propose a meta-learning network based on the fusion of heterogeneous data from vibration, sound, and infrared sources for cross-domain fault diagnosis of ceramic bearings under variable operating conditions. First, the non-parametric encoding of Gram Angle Field is employed to align and fuse the one-dimensional and two-dimensional heterogeneous data dimensions, thereby addressing the issue of dimensional disparity inherent in heterogeneous data. Second, a multi-source feature reconstruction module is designed to enhance the network’s ability to comprehend heterogeneous data features. To improve fault feature consistency, a multi-channel feature consistency module is developed to facilitate more consistent and generalizable feature representation. The results of the experiments demonstrate that the method attains an average diagnosis accuracy of 98.91% across six distinct cross-domain scenarios and exhibits robust performance in noise resistance experiments. The proposed method combines meta-learning with multimodal data fusion to address the limitations of single-modal sensing of rotating machinery under multi-physical field coupling conditions. It provides some practical application value for multimodal data monitoring of rotating systems such as aero-engines, high-speed machine tool spindles, and wind power gearboxes.