Existing data-driven fault diagnosis methods assume that the label sets of the training data and test data are consistent, which is usually not applicable for real applications since the fault modes that occur in the test phase are unpredictable. To address this problem, open set fault diagnosis (OSFD), where the test label set consists of a portion of the training label set and some unknown classes, is studied in this article. Considering the changeable operating conditions of machinery, OSFD tasks are further divided into shared-domain open set fault diagnosis (SOSFD) and cross-domain open set fault diagnosis (COSFD) in this article. For SOSFD, 1-D convolutional neural networks are trained for learning discriminative features and recognizing fault modes. For COSFD, due to the distribution discrepancy between the source and target domains, the deep model needs to learn domain-invariant features of shared classes and separate features of outlier classes. Thus, by utilizing the output of an additional domain classifier, a model named bilateral weighted adversarial networks is proposed to assign large weights to shared classes and small weights to outlier classes during the feature alignment. In the test phase, samples are classified according to the outputs of the deep model and unknown-class samples are rejected by the extreme value theory model. Experimental results on two bearing datasets demonstrate the effectiveness and superiority of the proposed method.

Ensuring reliable machine operation is dependent on effective fault detection. Feature indicators construction is considered as the foundation for operational condition assessment among various detection methods. However, the nuclear circulating water pump (NCWP) experiences severe disturbances due to noise and changes in operating conditions. Additionally, the complex movement of the gears causes signal interference, which may result in the feature indicators failing. To address the above issues, this paper proposes a generalized features construction method (GFC) to realize unsupervised planetary gearbox fault detection. First, a feature relevance evaluation method is presented to filter intrinsic mode functions (IMFs) obtained through ensemble empirical mode decomposition (EEMD) for noise reduction. Second, feature indicators are constructed based on interpretable frequency information. Then, global and local features are evaluated respectively to realize fault detection and further detection of faults from the sun or planetary gears. Finally, the effectiveness of the proposed method is demonstrated through data obtained from the NCWP test bench.

Traditional deep learning assumes that the probability distributions of the test set and the training set are exactly the same. When the test samples and the training samples are from different working conditions, the distribution gap between different conditions might dramatically lower the accuracy. To address this problem, domain adaptation (DA) via joint using training and test sets is adopted to increase the final accuracy. The disadvantage of DA is that it requires test samples to participate in the training phase. However, sometimes it is impossible to collect the test samples beforehand. To solve this problem, this article proposes a conditional adversarial domain generalization with a single discriminator, which aims to extract domain-invariant features from different working conditions and generalize the features to unseen working conditions. To realize conditional adversarial training, a new conditional adversarial strategy that the feature extractor can allow the discriminator to distinguish the fault classes, but cannot distinguish the domains, is designed to better confuse the discriminator and generalize the features. Meanwhile, three loss functions, including mean square error (MSE), cross-entropy coupled with entropy, and cross-entropy combined with traditional adversarial methods, are proposed to achieve the new conditional adversarial strategy. The effectiveness of the conditional adversarial strategy is verified by training conditional generative adversarial networks (CGANs) on two image data sets, and the performance of the proposed domain generalization method for bearing fault diagnosis is also verified by bearing data sets. Compared with traditional conditional adversarial domain generalization, the proposed method can save computing resource consumption and improve accuracy that does not appear in the training phase.

Journal bearings are the key components of the nuclear circulating water pump (NCWP), and accurate remaining useful life (RUL) prediction is of great significance for improving the reliability, safety, and maintenance planning of NCWP. However, it is difficult to quantify the uncertainty of bearing RUL based on the current deep learning (DL) model, resulting in a lack of credibility and effective convincing for RUL predicted by the model. Meanwhile, all existing hybrid models are basically simple combinations, and they cannot solve the uncertainty quantification problem of RUL predicted by DL. Hence, the bearing RUL prediction method based on a dynamic interactive hybrid model is proposed. First, a degradation model based on a nonlinear enhanced generalized Wiener process (EGWP) is proposed, which combines gated neural networks and time-varying drift coefficients to describe the nonlinear degradation process of bearing. Then, a corrective gated recurrent unit (CGRU) network is designed to learn and predict real-time degradation increments, and the parameters of the degradation model are dynamically updated through the history and prediction of degradation increments. Finally, the bearing RUL prediction is given by the CGRU network, and the probability density function of RUL is given by the proposed hybrid model. The performance of the proposed method is evaluated using the PHM 2012 bearing dataset and the NCWP journal bearing dataset. The results show that our proposed method can effectively predict bearing RUL and its uncertainty.

Deep learning (DL) has demonstrated splendid performance in fault diagnosis with sufficient samples and ideal operating environments. However, in practice, it is hard to acquire adequate fault samples from high-reliability equipment, thus inducing challenges in applying DL for intelligent diagnosis. Moreover, data distribution shift occurs due to working load variation or environment noise interference, severely degrading DL models’ performance. Aiming at the above problems, this paper proposes a method called Q networks calibrated ensemble (QCE). Specifically, two improved deep Q networks are employed to enhance the attention to critical feature samples and fault samples by quantifying feature distribution and utilizing fault prior knowledge. Afterward, the calibrated Choquet integral is adopted to achieve the models’ ensemble, further enhancing the generalization ability and robustness of the diagnosis results. The proposed method has been verified on the NCWP test bench. Results indicate the superiority and reliability of the QCE method. Besides, results when working load changes demonstrate the outstanding generalization ability of the proposed method for cross-domain fault diagnosis with class imbalance. Furthermore, Grad-CAM visualization verifies that the proposed method effectively learns fault features instead of noise.