Multi-access Edge Computing (MEC) is pivotal for smart factories, yet a significant challenge remains in efficiently offloading heterogeneous tasks, which requires balancing deterministic tasks with strict time windows and nondeterministic tasks that demand minimized latency. Traditional optimization methods and standard Deep Reinforcement Learning (DRL) algorithms often struggle to address the inherent hybrid discrete-continuous action space in joint task offloading and resource allocation. To address this challenge, a Cascaded Dual-Branch Network-based Deep Reinforcement Learning (CDBN-DRL) method is proposed. First, a dual-branch network that combines convolutional and self-attention mechanisms is constructed for the robust extraction of environmental state features. Then, the extracted feature vector is fed into a cascaded two-layer DRL architecture, where an upper-layer Double Deep Q-Network (DDQN) handles discrete offloading decisions, while a lower-layer Soft Actor-Critic (SAC) network manages continuous resource allocation based on the upper-layer decisions. This cascaded structure is designed to effectively manage and co-optimize the hybrid discrete-continuous action space. Experimental results indicate that the CDBN-DRL method achieves notable performance, achieving completion rates of 99.4% for deterministic tasks and 98.3% for nondeterministic tasks while outperforming existing state-of-the-art offloading schemes in terms of latency, power consumption, and generalization capability.

Domain generalization-based fault diagnosis (DGFD) has emerged as a promising approach for addressing mechanical fault diagnosis under unseen working conditions. However, mainstream DGFD methods based on statistical dependencies typically model only explicit relationships between temporal data and labels. They often fail to uncover implicit causal connections across different conditions, impairing the reliability and interpretability of the diagnosis results. To address this issue, an interpretable causal filtering and disentanglement domain generalization network is proposed. Specifically, discrete wavelet prior knowledge is embedded in the model to expand the signal from the time domain to the wavelet space, a causal filter is then designed to mine and filter causal features in the wavelet domain. Furthermore, a causal clustering loss and a noncausal discrimination loss are introduced to guide the network in disentangling causal information from latent representations within the causal feature space, thereby enhancing causal disentanglement. Extensive generalization experiments conducted on four machines demonstrate that the proposed method achieves superior generalization performance and interpretability.

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 PDA 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 classwise 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. Furthermore, 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.

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.