It is difficult to obtain expensive labeled data in industrial fault diagnosis applications, which easily leads to overfitting of deep learning and restricts its extensive usage. Aiming at this issue, this article proposed an improved few-shot semisupervised learning method, called semisupervised momentum prototype network (SSMPN), to realize gearbox fault diagnosis under limited labeled samples. First, the proposed SSMPN utilizes the powerful few-shot learning ability of the prototype network to learn the feature mapping and obtains prototypes by using limited labeled samples. Then, a threshold selection based on Monte Carlo uncertainty is adopted in pseudo label learning to increase the confidence of pseudo labels. Finally, the expended labeled dataset is utilized to optimize feature extraction and the momentum prototype method is proposed to fine-tune the prototype of each category. The experiments on both test-bench and wind turbine gearbox fault diagnosis demonstrated that SSMPN is more effective than the comparable methods under the same situation.

Partial domain adaptation (PDA) for fault identification has been widely researched to help construct self-monitoring systems in the era of the Industrial Internet of Things (IIoT). However, the existing PDA fault identification methods neglect the influence of uncertainty of the target domain on the identification performance. To solve this problem, this work developed a prototype-guided partial domain adaptation method with momentum weight for fault diagnosis. Specifically, to reduce the risk of ruling out the outlier by the output of a classifier or a discriminator, a class-wise selectively source weighting strategy that follows the number of the target pseudo labels is proposed. The target instances’ pseudo labels, which are obtained by calculating the distance between the target instance and the source prototypes, are irrelevant to the classifier and discriminator. Further, the momentum algorithm, by which the historical weights information could be retained, is employed in the source weights calculation procedure to alleviate the fluctuation and more closely to the global optimal. Experiments demonstrated the effectiveness and superiority of the developed method.

Domain adaptation (DA) methods are widely used in rolling bearing remaining useful life (RUL) prediction to align source and target domains. However, DA relies on previously collected target datasets, which are not always available in practical applications. To address this problem, an uncertainty-weighted domain generalization (UWDG) method is proposed for RUL prediction of rolling bearings under unseen conditions. Firstly, a knowledge distillation framework is established to capture internally-invariant features, which helps keep the features from varying with the existence of other domains. Secondly, correlation alignment is applied to acquire mutually-invariant feature distributions across domains, while regularization is employed to increase the disparity between the internally-invariant and mutually-invariant features. Then, multiple domain-generalization-task losses are weighted by homoscedastic uncertainty, ensuring consistency among separate task outputs. Finally, the effectiveness and superiority of UWDG have been validated through comparative experiments using the experimental dataset.

In recent years, deep learning-based fault diagnosis methods have witnessed significant advancements and successful applications in engineering practice. However, the increasing complexity of network structures demands higher computational resources in terms of floating point operations per second (FLOPs) and parameters, which poses challenges when deploying diagnostic models in edge computing scenarios with limited run resources. To address this issue, this study proposes a novel lightweight network, namely cheap ghost network (CGhostNet), incorporating fine-grained feature knowledge distillation (FFKD). FFKD-CGhostNet leverages CGhostNet, a lightweight architecture, and transfers diagnostic knowledge from ResNet, a complex yet high-performing network, through FFKD. Extensive experiments are conducted on two test benches to demonstrate that FFKD-CGhostNet achieves comparable diagnostic performance to ResNet while significantly reducing parameter count by nearly 88 times and computational requirements by almost 14 times. These findings highlight the effectiveness of FFKD-CGhostNet in achieving superior diagnostic performance in resource-constrained edge computing scenarios.

Compared with traditional machine learning algorithms, the convolutional neural network (CNN) has an excellent automatic feature learning ability and can complete the nonlinear representation from original data input to output by itself. However, the CNN does not sufficiently mine the tool wear information contained in the multi-sensor data due to disregard of the differences in the contribution of different features when extracting features. In this paper, a tool wear prediction method based on a multi-scale convolutional neural network with attention fusion is proposed, which fuses the tool wear degradation information collected by different types of sensors. In the multi-scale convolution module, convolution kernels with different sizes are used to extract the degradation information of different scales in the wear information, and then the attention fusion module is constructed to fuse the multi-scale feature information. Finally, the mapping between tool wear and multi-sensor data is realized through the feature information obtained by residual connection and full connection layer. By comparing the multi-scale convolutional neural network with different attention mechanisms, the experiments demonstrated the effectiveness and superiority of the proposed method.