Unsupervised domain adaptation (UDA) has recently gained attention in fault diagnosis due to its ability to address domain shift problems arising from changes in working conditions. However, when faced with the continual domain shift problem inherent in real-world industries with dynamic working conditions, UDA often suffers from catastrophic forgetting. To address this challenge, we propose a novel replay-free continual UDA framework, CoUDA, for fault diagnosis under dynamic working conditions. In CoUDA, prototype contrastive learning is employed in source domain pre-training in order to improve the model generalization ability in preparation for the adaptation to the subsequent target domains. Then, source discriminator constraint is employed to ensure that the acquired source domain knowledge serves as an anchor, and source feature knowledge distillation is applied to prevent catastrophic forgetting without replay in sequential target domain adaptation. In addition, for better domain adaptation, local domain alignment and information entropy minimization are utilized to achieve fine-grained domain alignment. Experimental results demonstrate the superiority of the proposed CoUDA in achieving robust fault diagnosis under dynamic working conditions.

Multi-source domain adaptation (MSDA) has demonstrated superior performance in intelligent fault diagnosis (IFD) compared to single-source domain adaptation (SSDA), as it can provide more comprehensive and diverse information from multiple fully-labeled source domains. However, in many real industrial scenarios, acquiring multiple fully-labeled source domains is challenging because labeling all the source domains is as expensive and laborious as labeling the target domain. Given this concern, a cross-supervised multisource prototypical network (CSMPN) is proposed for multi-source few-shot fault diagnosis. Specifically, a domain-shared and a domain-individual branch are constructed to realize shared domain alignment across all the source and target domains and individual domain alignment of source-target domain pairs, respectively. Within two branches, domain alignment is realized by the designed prototypical contrastive learning (PCL) module. In the PCL module, we propose a prototype calibration strategy to address the issue of biased prototype estimation owing to outlier samples. In addition, a two-stage pseudo-labeled sample selection mechanism is proposed to enhance the feature representation ability of two branches. At the end of the two branches, we design a cross-supervised learning (CSL) module to realize mutual and collaborative learning between the two branches, which can further improve the diagnosis performance on the target domain. Experiments on two different bearing datasets are implemented to verify the superiority of the proposed method compared with the comparison methods. Our code is available at https://github.com/YNWA-Zhang/CSMPN .

Machine learning models have been widely successful in the field of intelligent fault diagnosis. Most of the existing machine learning models are deployed in static environments and rely on precollected datasets for offline training, which makes it impossible to update the models further once they are established. However, in the open and dynamic environment in reality, there is always incoming data in the form of streams, including new categories of data that are constantly generated over time. In addition, the operating conditions of mechanical equipment are time-varying, which results in continuous stream data that are nonindependently and homogeneously distributed. In industrial applications, the diagnosis problem of nonindependent and identically distributed continuous streaming data is referred to as the cross-domain class incremental diagnosis problem. To address the cross-domain class incremental problem, a novel cross-domain class incremental broad network (CDCIBN) is proposed. Specifically, to solve the nonindependent identically distributed problem, a novel domain-adaptation learning loss function is first designed, which enables the conventional broad network to handle the category increment task well. Then, a cross-domain class incremental learning mechanism is designed, which learns new categories while retaining the knowledge of old categories well enough without replaying old category data. The effectiveness of the proposed method is evaluated through multiple mechanical failure increment cases. Experimental analysis demonstrates that the designed CDCIBN has significant advantages in the variable working condition class incremental application.

Rotating machinery often runs under large-speed-fluctuation (LSF) conditions, which results in severe data distribution domain shift for intelligent fault diagnosis methods. However, this challenge is rarely discussed in current studies. Hence, in this article, motivated by the active imagination of a human being, a new tool named actively imaginative data augmentation (AIDA) is constructed to solve machinery intelligent diagnosis under LSF conditions. The two adversarial training steps, namely, knowledge learning and sample imagining, are included in AIDA. In knowledge learning, a deep model is trained to learn the classification knowledge. In sample imagining, the parameters of the deep model are fixed and samples are generated via inversely training the model. As a result, diversified samples and an intelligent deep model adapting to the LSF condition are obtained by alternately carrying out the two steps. Moreover, a detailed discussion is given to interpret the process of actively imagining samples in the proposed AIDA in which some measures are designed, and the feature visualization is conducted. Experimental results show the effectiveness and superiority of AIDA in machinery diagnosis under LSF conditions, and the good performance of AIDA is due to the diversified dataset generated by changing the degrees and directions of each sample imagining.

Recently, various fault diagnosis methods based on domain adaptation (DA) have been explored to solve the problem of discrepancy between the source and target domains. However, given complex industrial scenarios, DA-based methods usually fail when the working conditions of machines are unseen, i.e., target data are unavailable during model training. In this article, a generic domain-regressive framework for fault diagnosis, namely, adversarial domain-invariant generalization (ADIG), is proposed. ADIG leverages multiple available domain data to exploit domain-invariant knowledge through adversarial learning between the feature extractor and the domain classifier. Simultaneously, the fault classifier generalizes the knowledge from the source-related domain to diagnose the unseen but related target domain signals. Moreover, customized strategies of feature normalization and adaptive weight are proposed to promote diagnosis performance. Comprehensive case studies show that ADIG achieves satisfactory diagnosis accuracy and robustness under unseen conditions, indicating that ADIG is a remarkably potential diagnosis tool for real-case industrial machines.

Cross-domain fault diagnosis methods based on transfer learning attempt to leverage knowledge from a domain with sufficient labeled samples to a different but related domain with few or even nonlabeled samples. These methods have been widely investigated in the past years. Notwithstanding the efficacy, most existing approaches assume that the label spaces of training and testing data are the same. However, this assumption is not practical in actual applications because new fault category usually happens in the testing stage. A cross-domain open-set transfer diagnosis method is presented in this article to manage the aforementioned problem. Domain adversarial model is employed to discriminate known from unknown target instances. Moreover, multiple auxiliary classifiers introduce a weighting module to evaluate the distinguishing domain knowledge to provide target instances with representative weights. The new adversarial domain adaptation network with diverse supplementary classifiers can effectively identify the unknown and known fault categories in the target domain and bridge the domain shift between the common fault category of the source and target domain. Experiments on two bearing datasets show the effectiveness and advantage of the proposed method.

Fault diagnosis based on data-driven methods are widely investigated when enough supervised samples of the target machine are available to build a reliable model. However, the labeled samples in practical operated machine are usually scarce and difficult to collect. If the model is built based on the sufficient labeled samples from different source machines, the diagnosis performance will degenerate owing to the domain discrepancy. To solve this issue, in this article, transfer learning (TL) is proposed by leveraging knowledge learned from source domain to target domain. While TL methods for fault diagnosis have been actively studied, most of them focus on learning from a single source. Since the labeled samples can come from multiple domains, more general diagnosis knowledge can be learned, which is beneficial to the prediction for the target domain. Therefore, a new TL approach based on multisource domain adaptation is proposed. A multiadversarial learning strategy is utilized for obtaining feature representations, which are invariant to the multiple domain shifts and discriminative for the learning goal at the same time. Extensive experimental analysis on four different bearing datasets is performed to illustrate the effectiveness and advantage of the proposed method.

Due to the complexity of industrial intelligent diagnosis, transfer learning-based fault diagnosis has become an evolving focus of the research field. Transfer learning uses knowledge of the source domain to identify faults in the target domain, which is a powerful tool to solve the problem of fault signal domain shift. However, existing methods have a limitation on multiple target domains. In other words, for different domains, respective transfer tasks are necessary. To seek a breakthrough, a adversarial multi-domain adaptation (AMDA) fault diagnosis method is proposed, realizing the fault diagnosis of multiple target domains by using the knowledge of a single source domain. AMDA is divided into three parts, namely, feature extractor, fault classifier and domain classifier. Through multi-domain adversarial learning, feature extractor and domain classifier mine the knowledge shared by multiple domains, and fault classifier can identify fault features distributed in different domains. The proposed AMDA method can surpass some traditional transfer learning fault diagnosis methods. Furthermore, as feature visualization result revealed, AMDA has significant advantages in multi-domain and broad research prospects.

Owing to distribution discrepancy between source training and target testing data, the performance of fault diagnosis by traditional supervised learning models will degenerate. Though domain adaptation methods for diagnosis have been actively investigated recently, most of them are devoted to learning from a single source. However, in reality, the supervised samples can be collected from different sources such as various working conditions. These sources are not only different from target but also from each other. The way of effectively fusing these sources to contribute the prediction of target remains a challenge. In this work, a new framework of multi-source domain adaptation is proposed for cross-domain fault diagnosis under different working conditions. Specially, this framework is realized by two alignment stages. At the first stage, multiple specific feature spaces are obtained, then the distributions of each pair of source and target domain are aligned since it is difficult to extract the common domain-invariant features for all domains. At the second stage, by considering the domain specific decision boundaries, the probabilistic outputs of classifiers are also aligned. Various experimental analysis on four different bearing working conditions is conducted to show the effectiveness of the proposed method. The performance of the proposed method is superior to state-the-art cross-domain fault diagnosis methods.