The lack of labeled data and variable working conditions brings challenges to the application of intelligent fault diagnosis. Given this, extracting labeled information and learning distribution-invariant representation provides a feasible and promising way. Enlightened by metric learning and semi-supervised architecture, a triplet-guided path-interaction ladder network (Tri-CLAN) is proposed based on the aspects of algorithm structure and feature space. An encoder–decoder structure with path interaction is built to utilize the unlabeled data with fewer parameters, and the network structure is simplified by CNN and an element additive combination activation function. Metric learning is introduced to the feature space of the established algorithm structure, which enables the mining of hard samples from extremely limited labeled data and the learning of working condition-independent representations. The generalization and applicability of Tri-CLAN are proved by experiments, and the contribution of the algorithm structure and the metric learning in the feature space are discussed.

Due to the lack of fault data in the daily work of rotating machinery components, existing data-driven fault diagnosis procedures cannot accurately diagnose fault classes and are difficult to apply to most components. At the same time, the complex and variable working conditions of components pose a challenge to the feature extraction capability of the models. Therefore, a transferable pipeline is constructed to solve the fault diagnosis of multiple components in the presence of imbalanced data. Firstly, synchrosqueezed wavelet transforms (SWT) are improved to highlight the time-frequency feature of the signal and reduce the time-frequency differences between different signals. Secondly, we proposed a novel hierarchical window transformer model that obeys a dynamic seesaw (HWT-SS), which compensates for imbalanced samples while fully extracting key features of the samples. Finally, a transfer diagnosis between components provides a new approach to solving fault diagnosis with imbalanced data among multiple components. The comparison with the benchmark models in four datasets proves that the proposed model has the advantages of strong feature extraction capability and low influence from imbalanced data. The transfer tests between datasets and the visual interpretation of the model prove that the transfer diagnosis between components can further improve the diagnostic capability of the model for extremely imbalanced data.