Recently, domain adaptation has been widely investigated for solving domain shift problems in mechanical fault diagnosis. Generally, domain adaptation-based diagnosis methods assume that the source and target domains have identical label space. However, a more realistic scenario is that the label space of the target domain is a subset of the source domain, which may introduce two problems: mismatching caused by the occurrence of outlier classes and misalignment caused by overweighting of the uncertain samples near the classification boundary. To address the above problems, a balanced adversarial domain adaptation network (BADAN) is proposed for intelligent fault diagnosis tasks under partial transfer scenarios. A balanced strategy is introduced to augment classes in the target domain using source samples. On this basis, an adversarial domain adaptation method with class-level weight is designed to avoid negative transfer by filtering outlier classes and promote positive transfer by mitigating the distribution discrepancy of shared classes. Moreover, to alleviate the misalignment problem, a complement objective function for ensuring alignment direction toward the support of the source classes is derived by minimizing their predicted scores of the incorrect classes rather than ground-truth classes. Extensive partial transfer diagnosis tasks constructed on two machines are used to demonstrate the robust and superior performance of BADAN.

Failure of cooling fan will have a great impact on the operation of the air-conditioning cooling system. Traditional anomaly detection methods for fans rely on manual feature extraction, which is easy to lose effective information. Deep learning-based methods have potential to overcome such a problem, however, it is difficult to train the network without enough monitoring data. In operation, cooling usually worked at high speed, and its monitoring data contain much high-speed rotational noise. This will confuse the networks to focus on high-energy noise while lose the weak but useful information. To solve these problems, we propose a semi-supervised multi-scale deep adversarial model (SMDAM) for fan anomaly detection. In SMDAM, we propose the Linear Spectral Line Elimination (LSLE) technique to reduce the interference of rotational noise spectral lines on network training. Subsequently, a multi-scale feature extraction module is introduced to enrich diagnostic information, which extracts features from different scales without increasing the depth of the convolutional layer. Finally, two engineering data sets are used to verify the superiority of the proposed method. The results show that SMDAM is superior to other methods.

The signal generated by a diesel engine is considered as relatively intricate mechanical signal due to its susceptibility to interference from both low-pass components and significant background noise. In order to improve the feature extraction performance, a novel method termed LPF/OGS (the simultaneously low-pass filtering and OGS denoising method) is proposed. For filtering low-frequency components, a zero-phase non-causal recursive filter that takes the form of a banded matrix is employed. Meanwhile, the non-convex regularization technique, which incorporates the group sparse structure is adopted to capture the tendency of clustering. Moreover, the whole formulated objective function is constrained as a convex optimization problem. As a result, the sparsity of useful features is maximally promoted while the noise between adjacent pulses is eliminated. The proposed method is applied to a simulated signal and a real signal collected from the diesel engine. The processing results provide compelling evidence that the proposed method outperforms traditional methods in accurately extracting transient impulses for machinery fault diagnosis.

Unsupervised domain adaptation has demonstrated significant success in intelligent mechanical fault diagnosis. However, these methods primarily focus on transferring knowledge from laboratory test rigs or physical entities, which can be costly or unavailable in certain occasions. Moreover, most methods emphasize global distribution alignment while neglecting fine-grained distribution alignment in knowledge transfer, which can lead to misclassification of certain fault categories and subsequently decrease the diagnostic accuracy of the deep model. In response to these issues, a novel simulation-to-real transfer-based deep unsupervised subdomain adaptation method is proposed. Taking bearing fault diagnosis as an example, a high-fidelity 4-DOF dynamic model of rolling bearings is constructed to generate simulation data across various health states and multiple working conditions. The local maximum mean discrepancy is incorporated to achieve a fine-grained distribution alignment between the source and target domains. A deep unsupervised subdomain adaptation transfer learning network is designed to learn knowledge from the sufficiently labeled simulation data and realize fault mode recognition on the unlabeled measured data. Experiments and empirical analyses on three cross-domain diagnostic tasks demonstrate the effectiveness and superiority of the proposed method.

Domain adaptation in fault diagnosis can efficiently handle different data distributions by co-training source and target domain data. However, the source domain data may not be accessible due to privacy or memory issues. An effective solution is to perform unsupervised parameter tuning of the source model using unlabeled target domain data. The focus of this paper is to perform unsupervised domain adaptation using the knowledge of multiple pre-trained source models in Industrial Internet of Things (IIoT) without accessing source data. Almost all of the current researches in this setting requires tuning the entire backbone network, increasing computing costs as model size or source domains grow. To this end, this paper proposes an Efficient Multi-Domain Knowledge Fusion Adaptation (EMDKFA) method, applied to source free cross-domain fault diagnosis. Unlike previous methods, this method greatly reduces the number of trainable parameters and saves computational overhead via low-rank reparameterization strategy. To better fuse and transfer multiple pre-trained models, we propose an unsupervised domain weight initialization method based on nuclear norm, initially focusing credible source models. Furthermore, the multi-domain weighted entropy minimization penalty and noisy label learning are designed to promote correct allocation of target samples. Comprehensive cross-condition and cross-device experiments demonstrate the method’s effectiveness.