The increasing complexity and interconnectivity of modern process industry systems, particularly in critical infrastructure such as circulating water pump systems in nuclear and chemical plants, pose significant challenges for ensuring operational security and system-level situational awareness. Conventional deep learning models, while effective in temporal pattern recognition, exhibit limited capability in capturing spatial dependencies across heterogeneous sensor networks, thereby restricting their ability to represent dynamic system behavior. To address these limitations, this paper presents a novel method titled graph attention-based deviation network (GADN4SA) for circulating water system situational awareness. The proposed method dynamically constructs task-aware sensor graphs based on signal correlations, integrates both structural and semantic representations, and leverages attention-guided mechanisms to model long-range spatial-temporal dependencies. In addition, a structured deviation scoring module is incorporated to enable interpretable anomaly quantification and attribution. Extensive experiments on both simulated and real-world datasets demonstrate that the proposed method consistently outperforms state-of-the-art baselines in terms of robustness, accuracy, and generalization, highlighting its effectiveness for deployment in safety-critical industrial environments.

Mining excavators, transport vehicles, and other construction machinery, as well as energy equipment like nuclear power plant units, operate in complex and harsh environments, where high reliability and continuous operation are crucial. However, existing health monitoring methods have difficulty adapting their thresholds, leading to false alarms and missed detections. Additionally, the prediction accuracy of current health prognostics methods is still not up to expectations, and their reliability is insufficient. To address these issues, a dynamic adaptive Transformer-based method is proposed. In the health monitoring module, a dual-stage attention mechanism is introduced, which includes spatial and temporal attention, aimed at capturing the spatiotemporal features in the multi-source data and optimizing anomaly detection through dynamic threshold adjustments. To further improve prediction accuracy, a multi-scale multi-head attention mechanism is used, and Monte Carlo Dropout is applied to estimate the uncertainty of the prediction results. Furthermore, a dynamic adaptive fine-tuning mechanism solves the issue of prediction bias, making the predictions more aligned with actual operational trends. Experimental validation demonstrates that the proposed method performs excellently in anomaly detection and trend prediction, offering high engineering application value.

The advancement of large models has initiated a transformation in the field of time-series forecasting. Both the repurposing of existing large models and the development of large models tailored for time-series analysis have exhibited impressive performance. In industrial applications, challenges, such as limited data availability and constrained computational resources, render the first approach viable. However, it is important to note that this approach is still in its infancy and lacks both a thorough technical analysis and a unified effective framework. Meanwhile, as large models become a mainstream artificial intelligence paradigm, it is urgent to discuss typical industrial scenarios, such as how automated systems can transition from intelligent to collaborative operation and maintenance. In light of this premise, this article endeavors to advance a generalized technical framework for large model-driven time-series forecasting, under which existing methods can be subsumed. Then, within this overarching technical paradigm, the technical advancements facilitated by diverse methods will be systematically elucidated and analyzed, along with a comparative evaluation conducted across seven benchmark datasets. Concluding this analysis, the implementation pathway for the industrial automation system is delineated that integrates operator action commands to forecast post-action trends to assess action correctness in advance. Finally, the challenges and future directions of large model-based time-series forecasting are outlined.

Existing domain adaptation methods strive to align all domains equally under a single domain shift dimension, which poses two problems. On the one hand, multiaspect domain transferring factors and homogenous alignment may lead to suboptimal results in more distant domains. On the other hand, such a global alignment ignores local discriminatory information, making class boundary samples susceptible to misclassification. Hence, the three-types-of-graph-relational guided domain adaptation (TGGDA) is proposed. First, thedomain graphis formed based on condition-dependent slow variables. The domain discriminator is redesigned to reconstruct the domain graph. Second,intrinsicandpenalty graphsare integrated to draw the same class but different domains sample closer and vice versa. The TGGDA is a system-assisted cross-domain diagnosis method that enables multidimensional domain information measurable, and the adjacency alignment allows for more accurate diagnostic results. Finally, experiments on gearbox fault diagnosis in circulating water pumps show that TGGDA can improve diagnosis accuracy.

Insufficient spatio-temporal feature extraction in normalizing flows (NFs) based anomaly detection (AD) method impedes their performance improvement. Moreover, it is worth noting that the multioperational nature of the process poses a challenge for most AD methods, including those based on NF, rendering them largely ineffective. Hence, this article introduces a new method called spatial-temporal graph conditionalized normalizing flows (STGNFs). First, multiscale dilation convolutional layers and mix-hopping graph convolutional layers are interleaved to form a spatio-temporal feature extractor. Second, spatio-temporal features are employed as conditional information for NF, while scheduling variables are factored in to adapt to operating conditions. Then, tracing the anomaly variables through the conditional density magnitude allows for interpretable AD results. Finally, experimental results on four datasets, including high-fidelity experimental bench data and real nuclear power plant data, demonstrate the performance of STGNF. STGNF enables the detection and precise localization of anomalies in various power modes, including nuclear plant shutdown and peaking, transcending the limitations of existing methods.

Existing self-supervised medical image segmentation usually encounters the domain shift problem (i.e., the input distribution of pre-training is different from that of fine-tuning) and/or the multimodality problem (i.e., it is based on single-modal data only and cannot utilize the fruitful multimodal information of medical images). To solve these problems, in this work, we propose multimodal contrastive domain sharing (Multi-ConDoS) generative adversarial networks to achieve effective multimodal contrastive self-supervised medical image segmentation. Compared to the existing self-supervised approaches, Multi-ConDoS has the following three advantages: (i) it utilizes multimodal medical images to learn more comprehensive object features via multimodal contrastive learning; (ii) domain translation is achieved by integrating the cyclic learning strategy of CycleGAN and the cross-domain translation loss of Pix2Pix; (iii) novel domain sharing layers are introduced to learn not only domain-specific but also domain-sharing information from the multimodal medical images. Extensive experiments on two publicly multimodal medical image segmentation datasets show that, with only 5% (resp., 10%) of labeled data, Multi-ConDoS not only greatly outperforms the state-of-the-art self-supervised and semi-supervised medical image segmentation baselines with the same ratio of labeled data, but also achieves similar (sometimes even better) performances as fully supervised segmentation methods with 50% (resp., 100%) of labeled data, which thus proves that our work can achieve superior segmentation performances with very low labeling workload. Furthermore, ablation studies prove that the above three improvements are all effective and essential for Multi-ConDoS to achieve this very superior performance.

Existing self-supervised learning methods based on contrastive learning and masked image modeling have demonstrated impressive performances. However, current masked image modeling methods are mainly utilized in natural images, and their applications in medical images are relatively lacking. Besides, their fixed high masking strategy limits the upper bound of conditional mutual information, and the gradient noise is considerable, making less the learned representation information. Motivated by these limitations, in this paper, we propose masked patches selection and adaptive masking strategy based self-supervised medical image segmentation method, named MPS-AMS. We leverage the masked patches selection strategy to choose masked patches with lesions to obtain more lesion representation information, and the adaptive masking strategy is utilized to help learn more mutual information and improve performance further. Extensive experiments on three public medical image segmentation datasets (BUSI, Hecktor, and Brats2018) show that our proposed method greatly outperforms the state-of-the-art self-supervised baselines.

Analysis of X-ray images is one of the main tools to diagnose breast cancer. The ability to quickly and accurately detect the location of masses from the huge amount of image data is the key to reducing the morbidity and mortality of breast cancer. Currently, the main factor limiting the accuracy of breast mass detection is the unequal focus on the mass boxes, leading the network to focus too much on larger masses at the expense of smaller ones. In the paper, we propose the multi-head feature pyramid module (MHFPN) to solve the problem of unbalanced focus of target boxes during feature map fusion and design a multi-head breast mass detection network (MBMDnet). Experimental studies show that, comparing to the SOTA detection baselines, our method improves by 6.58% (in AP@50) and 5.4% (in TPR@50) on the commonly used IN-breast dataset, while about 6-8% improvements (in AP@20) are also observed on the public MIAS and BCS-DBT datasets.