The transfer learning method, based on unsupervised domain adaptation (UDA), has been broadly utilized in research on fault diagnosis under variable working conditions with certain results. However, traditional UDA methods pay more attention to extracting information for the class labels and domain labels of data, ignoring the influence of data structure information on the extracted features. Therefore, we propose a domain-adversarial multi-graph convolutional network (DAMGCN) for UDA. A multi-graph convolutional network (MGCN), integrating three graph convolutional layers (multi-receptive field graph convolutional (MRFConv) layer, local extreme value convolutional (LEConv) layer, and graph attention convolutional (GATConv) layer) was used to mine data structure information. The domain discriminators and classifiers were utilized to model domain labels and class labels, respectively, and align the data structure differences through the correlation alignment (CORAL) index. The classification and feature extraction ability of the DAMGCN was significantly enhanced compared with other UDA algorithms by two example validation results, which can effectively achieve rolling bearing cross-domain fault diagnosis.

Aiming to extract useful features from bearing signals for fault classification, an intelligent fault diagnosis method is proposed with a stacked denoising auto-encoder (SDAE) and adaptive deep hybrid kernel extreme learning machine (ADHKELM). The deep network architecture of the SDAE is used automatically to extract deeply important features, and a new HKELM is constructed by combining a polynomial with a wavelet kernel function to overcome a single kernel function not being universal. After that a DHKELM, from stacking multiple HKELMs, and the sparrow search algorithm are introduced to iteratively determine the optimal value of core hyper-parameter combinations of the DHKELM to generate the final fault classifier ADHKELM to enhance the performance of the model. Two experimental verification results show that the SDAE-ADHKELM has better fault classification precision, robustness and generalizability than other related methods.

Detecting road cracks is crucial for ensuring road traffic safety and stability. However, currently existing detection methods do not usually pay close attention to the global, local and multi-scale feature information of road crack images, resulting in poor detection effects. To overcome this limitation, we propose a road crack detection model that combines the global and local multi-scale attention network (GLMANet) and enhanced feature pyramid network (EFPN). The GLMANet can effectively extract useful global, local and multi-scale feature information of crack images through a novel global and local multi-scale attention mechanism. Meanwhile, EFPN utilizes global adaptive attention module and multi-receptive field feature enhancement module to mitigate information loss during feature map generation, enhancing the representation of advanced semantic features in the feature pyramid. Leveraging faster region-based convolutional neural networks (Faster R-CNN) as the object detection architecture, we conduct experimental evaluations on both publicly available crack datasets and a dataset we collected. The proposed model achieved optimal detection performance across all three datasets. In comparison experiments with the current state-of-the-art crack detection methods, the proposed model outperformed other models, with AP and AP50 increasing by up to 16% and 11%, respectively, validating its effectiveness and superiority. Additionally, the model achieved a good balance between complexity and detection performance with 79.2 GFLOPs and 72.60 million parameters.