Fault diagnosis of aero engine bearings under speeds changing by combining attention mechanism and adversarial network

Gearboxes represent a key feature of wind turbines, fulfilling the function of connecting and transmitting power. The analysis of gearbox health assessment has the capacity to predict the failure time of the wind turbine with both timeliness and accuracy, thus providing guidance for subsequent maintenance and repair. In this paper, a gearbox health assessment method based on an adversarial network is proposed. The transition from normal to fault conditions facilitates the early detection of faults. The residual between the original signal and the predicted signal is utilised to indicate whether the gearbox operating state is normal. The deviation from the normal data to the abnormal data is quantified through the deviation index and the technique of change. This is then converted into a probability value to obtain the health index of a single variable. The overall health index (HI) of the gearbox is composed of a single variable index. The experimental findings demonstrate the efficacy of the proposed method in conducting health assessments of wind turbine gearboxes, thereby substantiating its wide-ranging applicability.

The rapid growth of short video platforms has fueled interest in automated short video generation, a task that demands high fidelity, temporal coherence, and semantic alignment with input data such as text, audio, or static images. This task poses important issues since to the requirement for accurate semantic understanding, spatial-temporal alignment, and realistic visual synthesis. In this research, a novel deep learning-based cross-modal short video generation framework is developed to address these challenges effectively. The Microsoft Research Video to Text (MSRVTT) dataset is gathered from the Kaggle, has 10,000 video clips from 20 categories undergoes extensive preprocessing, tokenization, Audio Modality using Mel-Frequency Cepstral Coefficients (MFCCs), Video Modality using Temporal-Spatial Synchronization and Normalization (TSSN). Then feature extraction is performed with a semantic decoupling module that utilizes encoders to extract and disentangle high-level semantic components, such as object appearance, motion dynamics, background context, and emotional tone from the input. A Namib Beetle Optimized Self-Attention based Generative input in various Adversarial Networks (NBO-SAGAN) is employed to generate short videos and to enhance fine details and correct visual artifacts. Experimental evaluations using Python shows that the NBO-SAGAN approach outperforms traditional methods Frechet Video Distance (FVD) of 230.74 and Kernel Video Distance (KVD) of 12.58 highlighting its effectiveness for expressive cross-modal video generation. This integrated methodology effectively combines modeling to produce controllable, expressive, and visually rich cross-modal video content.

MediFlora-Net: Quantum-enhanced deep learning for precision medicinal plant identification.

The purpose of this study is to investigate how homophily, memory constraints, and adversarial disruptions collectively shape the resilience and adaptability of complex networks. To achieve this, we develop a new framework that integrates explicit memory decay mechanisms into homophily-based models and systematically evaluate their performance across diverse graph structures and adversarial settings. Our methods involve detailed experimentation on synthetic datasets, where we vary decay functions, reconnection probabilities, and similarity measures–primarily comparing cosine similarity with traditional metrics such as Jaccard similarity and baseline edge weights. The results show that cosine similarity achieves up to a 30% improvement in stability metrics in sparse, convex, and modular networks. Moreover, the refined value-of-recall metric demonstrates that strategic forgetting can bolster resilience by balancing network robustness and adaptability. The findings underscore the critical importance of aligning memory and similarity parameters with the structural and adversarial dynamics of the network. By quantifying the tangible benefits of incorporating memory constraints into homophily-based analyses, this study offers actionable insights for optimizing real-world applications, including social systems, collaborative platforms, and cybersecurity contexts.

Swift and accurate semantic segmentation of underwater images is key for precise object recognition in complex underwater environments. Nonetheless, the inherent complexity of these environments and the limited availability of labeled data pose significant challenges to underwater image segmentation. Traditional deep learning methods struggle to cope with limited and noisy annotations. In this paper, we delineate the formulation of a novel semi-supervised paradigm with dynamic mutual adversarial training for the semantic segmentation of underwater images. This paradigm identifies the sources of inaccuracies in pseudo-labeling by analyzing different confidence maps, generated by models with unique prior knowledge. A dynamic reweighing loss function is then employed to orchestrate the mutual instruction of two divergent models. Furthermore, the delineation of confidence map is facilitated via adversarial networks, which involves simultaneous adversarial refinement of the discrimination network and the segmentation model, using the predictions with high-confidence maps as pseudo-labels. Experimental results on public underwater datasets verify that the proposed method can effectively improve semantic segmentation performance under the condition of a small amount of labeled data.

Secure distributed estimation algorithms have garnered widespread attention for their stability and security. However, in the harsh adversarial network environment, the network is subjected to multiple attacks, and a large number of nodes are compromised, leading to a significant reduction in the estimation performance of traditional secure algorithms. To tackle this problem, a divide-and-conquer secure distributed estimation with density peak (DCSDP) algorithm is proposed. Specifically, a divide-and-conquer attack defense (DCAD) strategy is proposed, in which two specialized attack detection steps are designed to divide and process the compromised data in the network. Considering the presence of multiple attacks in the network, a density peak-based attack detection scheme is proposed and embedded into the DCAD strategy. Additionally, by constructing adaptive thresholds and introducing a trust mechanism, the robustness of the algorithm is enhanced. Finally, the performance of the proposed algorithm is theoretically analyzed in terms of the mean and mean-square. Simulation experiments demonstrated that DCSDP maintains excellent robustness and effectiveness in different adversarial network environments.

With the rapid expansion of IoT and the advent of the 6G era, ensuring efficient and secure communication for the edge large language model (LLM) servicing has become a critical priority. However, malicious relay nodes injecting delays pose a significant threat to network performance and security. This paper proposes a novel framework combining Stackelberg Game theory with Multi-Agent Reinforcement Learning (MARL) to mitigate communication delay injection attacks in edge networks. By modeling the interaction between edge devices and malicious nodes as a Stackelberg Game, where edge devices act as followers optimizing relay strategies and malicious nodes as leaders seeking disruption, our approach enables edge devices to cooperatively identify and avoid malicious nodes. The distributed MARL framework allows each edge device to learn their optimal strategies independently, enhancing network resilience. Experimental results demonstrate the effectiveness of our scheme in effectively reducing the impact of malicious nodes and improving the deployment capability of edge computing services.

Web3.0 aims to foster a trustworthy environment enabling user trust and content verifiability. However, the proliferation of fake news undermines this trust and disrupts social ecosystems, making the effective alignment of visual-textual semantics and accurate content verification a pivotal challenge. Existing methods still struggle with deep cross-modal interaction and the adaptive calibration of discrepancies. To address this, we introduce the Bidirectional Semantic Enhancement and Adversarial Network (BSEAN). BSEAN first extracts features using large pre-trained models: a hybrid encoder for text and the Swin Transformer for images. It then employs a Bidirectional Modality Mapping Network, governed by cycle consistency, to achieve preliminary semantic alignment. Building on this, a Semantic Enhancement and Calibration Network explores inter-modal dependencies and quantifies semantic deviations to enhance discriminative capability. Finally, a Dual Adversarial Learning framework bolsters event generalization and representation consistency through adversarial training with event and modality discriminators. Experiments on public Weibo and Twitter datasets validate BSEAN’s superior performance across all metrics, demonstrating its efficacy in tackling the complex challenges of deep cross-modal interaction and dynamic modality calibration within Web3.0 social networks.

ABSTRACTThe increasing complexity of long‐term evolution (LTE) networks presents challenges in fault detection and automated diagnosis. These traditional supervised learning approaches need extensive labeled datasets, which are challenging to obtain in live network environments. In order to address this issue, this paper proposes an automatic root cause analysis model based on an unsupervised learning mechanism (ARCA‐ULM) for self‐healing LTE networks. The proposed ARCA‐ULM adopts the twin attentional‐generative adversarial network (TA‐GAN) model for synthetic data generation, addressing class imbalance issues in training datasets. Then, the root cause pattern is identified using a lightweight walrus‐based deep coupling gated recurrent autoencoder (WDCGRA) model, which incorporates coupling autoencoders (AEs), gated recurrent units (GRUs), and the walrus optimization algorithm (WOA) to distinguish hardware faults from software misconfigurations. The radio link control (RLC) layer optimization scheme is employed to improve network performance. The ARCA‐ULM model is implemented in Python and evaluated in terms of different performance indicators. Simulation outcomes demonstrate superior performance, achieving 99.17% accuracy, a diagnosis error rate (DER) of 0.14, and an undetected error rate (UER) of 2.29, outperforming existing methods. The outcomes confirm the effectiveness of the ARCA‐ULM model in enhancing the self‐healing abilities of LTE networks, allowing more reliable fault detection and resolution.

ABSTRACT Plant disease is accountable for majority of economic losses in agricultural industry across the globe. Hence, plant disease detection at an earlier phase is highly important to provide food safety and enhancement of farming systems. The manual detection techniques are challengeable and consume much time for classifying plant leaf diseases. Here, Wolf‐Bird Skill Optimizer based Optimal Pooling SqueezeNet (WBSO_OptimalPool SqNet) is presented for a multiclass plant disease detection. Initially, plant leaf image is pre‐processed employing Kuwahara filter. Afterwards, plant leaf disease segmentation is conducted by Spine‐Generative Adversarial Network (Spine‐GAN). Next, image augmentation is done and next, Convolutional Neural Network (CNN) features and DAISY with statistical features are extracted. Thereafter, two levels of classification namely plant type classification and plant disease classification are conducted. The classification of plant type and plant disease are accomplished by SqueezeNet, wherein pooling layers are modified by optimal pooling layer based on weights. The weights are calculated utilising WBSO, which is devised by incorporating Wolf‐Bird Optimizer (WBO) with Skill Optimization Algorithm (SOA). In addition, WBSO_OptimalPool SqNet achieved maximal accuracy and True Positive Rate (TPR) about 91.897% and 90.815% as well as minimal False Positive Rate (FPR) about 7.186%.

Abstract Over recent years, adversarial transfer learning (TL) has gained extensive application in fault diagnosis of rotating machinery. However, traditional adversarial TL is relatively generalized in feature extraction, which makes it difficult to capture the features that are crucial for fault diagnosis. This will cause the model to be inaccurate in identifying fault categories, especially when faced with intricate fault features in the signal. Therefore, a novel convolutional sparse filtering (SF) adversarial network is proposed in this paper. Initially, convolutional SF combined with an attention mechanism is utilized to extract sparse features from the input data. Subsequently, these features are fed into both the domain discriminator and the label classifier. To achieve domain invariance, a gradient reversal layer is incorporated, which maps and blends samples from the source and target domains. Additionally, the joint maximum mean difference is incorporated into the adversarial training process to significantly improve domain adaptation (DA). The combination of convolutional SF and adversarial network can effectively perform feature extraction and DA in the scene of cross-domain TL, thus enhancing the generalization of the model in the target domain. Experimental results on two planetary gearbox datasets show that the method achieves excellent fault diagnosis performance under varying conditions, with an average accuracy of 96.85%.

ABSTRACT Integrating institutional theory from a network perspective, this study explores the network dynamics among NGOs, governmental organizations, and corporations for Corporate Social Responsibility (CSR) practices in South Korea, a state-led market economy. By exploring the relationships among three types of centralities of NGOs in their governmental and corporate networks, the study finds that the type of strategic position (i.e. eigenvector and betweenness centralities) the NGOs develop in their governmental network is significantly related to and congruent with the same strategic positions in their corporate network, within the same type of relationship (collaborative/adversarial). When developing networks within their own sector, on the other hand, the NGOs benefit the most from developing a large number of ties (degree centralities), as opposed to strategic positions, in developing strong positions in both collaborative and adversarial networks with the governmental and corporate sectors. The implications for NGO roles and international strategies for cross-sector CSR institutionalization are discussed.

ABSTRACT The rapid advancement of remote sensing technologies and the increasing availability of satellite and aerial imagery have created new opportunities for change detection (CD). The existing deep learning (DL) methods often focus on single-source images, such as very high-resolution (VHR) RGB images, but struggle due to the limited spectral information and imaging inconsistencies within them. To overcome these limitations, this paper proposes an innovative end-to-end adversarial network with dual-encoders that extracts spectral, spatial, and textural features from multiple image sources for CD in built-up areas. The generator network of the proposed network employs a dual-encoding architecture that simultaneously processes VHR RGB and multispectral bi-temporal images and a decoder architecture that incorporates a channel attention module to improve the construction of the change map (CM). The discriminator network of the proposed network evaluates the authenticity of the generated CM, refining feature differentiation between changed and unchanged regions. Conducting experiments on two public datasets demonstrates that our method significantly outperforms existing DL approaches, highlighting the effectiveness of adversarial training in improving multi-source CD reliability in urban environments.

We propose denoising diffusion variational inference (DDVI), a black-box variational inference algorithm for latent variable models which relies on diffusion models as flexible approximate posteriors. Specifically, our method introduces an expressive class of diffusion-based variational posteriors that perform iterative refinement in latent space; we train these posteriors with a novel regularized evidence lower bound (ELBO) on the marginal likelihood inspired by the wake-sleep algorithm. Our method is easy to implement (it fits a regularized extension of the ELBO), is compatible with black-box variational inference, and outperforms alternative classes of approximate posteriors based on normalizing flows or adversarial networks. We find that DDVI improves inference and learning in deep latent variable models across common benchmarks as well as on a motivating task in biology-inferring latent ancestry from human genomes-where it outperforms strong baselines on 1000 Genomes dataset.

Credit card fraud detection is a critical issue in the financial industry, requiring robust and efficient methods to identify and prevent fraudulent transactions. This survey paper reviews various machine learning (ML) and deep learning (DL) techniques employed in credit card fraud detection, highlighting the performance and limitations of each method. Identifying credit card fraud remains a pressing concern for financial systems, demanding innovative and reliable detection strategies. Classic machine learning techniques, including Random Forest, Support Vector Machines, and K-Nearest Neighbors, have delivered encouraging outcomes, with accuracy peaking at 94.84% in controlled settings.generative adversarial networks (GANs), have demonstratedsuperior performance. Moreover, the paper discusses the robustness of hybrid models like AdaBoost and ensemble learning in real world scenarios, even under noisy conditions.

The development of artificial intelligence enables computers to not only simulate human artistic creations, but also synthesize fine art works with deeper meanings based on natural images. This study digitally parses the fusion of fine art and philosophy visual expressions, and develops a visual expression system based on the fusion of fine art and philosophy by utilizing a variety of key big data algorithms for visual expressions such as adversarial networks. Research on pattern recognition of this system in art creation is carried out through model training, recommendation performance evaluation, pattern recognition strategy application and regression analysis. The model in this paper works best when the number of nearest neighbors k=15, and the recommendation model in this paper can provide a personalized list of artwork recommendations for different people. The recognition of the system in this paper in the five dimensions of “spiritual level”, “value level”, “philosophical level”, “aesthetic level” and “technical level” is distributed between 4.24 4.55. The results of regression analysis indicated that the system in this paper can improve the artistic creation as well as pattern recognition.

Abstract In the age of collaboration and shared governance, paradoxically, distrust manifests frequently in government and political institutions and is seen as dysfunctional to democracy, making governing networks challenging. Yet, previous studies emphasize the significance of promoting trust more than addressing distrust in networks. Distrust differs from the absence of trust. It involves relationships characterized by doubt, suspicion, or opportunism. Relatively little is known about why distrusting relationships occur and how they develop in adversarial interorganizational governance networks. Using quantitative network surveys and qualitative interview data from organizations involved in an adversarial local hydraulic fracturing governance network in New York, our mixed-method analyses fill this gap. We found evidence of cognitive distrust from different policy beliefs and identity-based subgroups and two sources of behavioral distrust (competition and non-collaboration), as well as the interactions between cognitive and behavioral sources of distrusting relationships. We further identified underexplored sources of endogenous relational distrust: strong and negative reciprocity, non-transitivity, and Simmelian ties (meaning mutual third-party ties). These relational sources suggest that the distrust networks mutually reinforce each other but are less clustered and more indirect. Our study advances network management scholarship by showing why distrusting relationships occur and how they escalate within adversarial networks.

The exponential growth of the Internet of Things (IoT) has heightened the need for secure, privacy-preserving, and efficient cyber-attack detection mechanisms. This study introduces the Customized Temporal Federated Learning through Adversarial Networks (CusTFL-AN) framework, which combines Temporal Convolutional Networks (TCNs) and Generative Adversarial Networks (GANs) for robust and personalized attack detection. CusTFL-AN enables clients to train local models while maintaining data privacy by generating synthetic datasets using GANs and aggregating these at a central server, thereby mitigating risks associated with direct data sharing. The framework's effectiveness is demonstrated on three benchmark datasets—UNSW-NB15, BoT-IoT, and Edge-IIoT—achieving detection accuracies of 99.2%, 99.5%, and 99.25%, respectively, significantly outperforming state-of-the-art methods. Key enhancements include addressing data heterogeneity through federated aggregation, minimizing overfitting using GAN validation and cross-validation techniques, and ensuring interpretability to support practical adoption in real-world IoT scenarios. Privacy mechanisms are strengthened to prevent potential data leakage during aggregation, and ethical considerations surrounding the use of synthetic datasets are acknowledged. Furthermore, the impact of computational constraints, network latency, and communication overhead on resource-constrained IoT devices has been carefully analyzed. While the results affirm the robustness and scalability of CusTFL-AN, future work will focus on extending evaluations to more diverse datasets and addressing the challenges of adversarial attacks. CusTFL-AN represents a significant step forward in privacy-preserving federated learning, offering practical solutions to real-world IoT cybersecurity challenges.

Accurately extracting organs from medical images provides radiologist with more comprehensive evidences to clinical diagnose, which offers up a higher accuracy and efficiency. However, the key to achieving accurate segmentation lies in abundant clues for contour distinction, which has a high demand for the network architecture design and its practical training status. To this end, we design auxiliary and refined constraints to optimize the energy function by supplying additional guidance in training procedure, thus promoting model’s ability to capture information. Specifically, for the auxiliary constraint, a set of convolutional structures are involved into a conventional network to act as a discriminator, then adversarial network is established. Based on the obtained architecture, we further build adversarial mechanism by introducing a second discriminator into segmentor for refinement. The involvement of refined constraint contributes to ameliorate training situation, optimize model performance, and boost its ability of collecting information for segmentation. We evaluate the proposed framework on two public databases (NIH Pancreas-CT and MICCAI Sliver07). Experimental results show that the proposed network achieves comparable performance to current pancreas segmentation algorithms and outperforms most state-of-the-art liver segmentation methods. The obtained results on public datasets sufficiently demonstrate the effectiveness of the proposed model for organ segmentation.