Search Results (831)

Rotating machinery is a key component in modern industry, and its operating condition directly affects equipment safety and production reliability. However, discrepancies among different machines cause source–target distribution shifts, while fault annotation for target machines is costly, limiting the performance of deep learning-based diagnosis under cross-machine scenarios with limited labels. To address these issues, this paper proposes a multi-scale time–frequency semantic consistency model based on self-supervised transferable representation learning, termed MTFSC. First, augmented waveform views and multi-scale frequency-domain views are constructed from unlabeled source-domain vibration signals for self-supervised pre-training without source labels. Then, a time-domain impulse-aware feature extractor and a time–frequency decoupled spectral feature extractor are designed to enhance local impulsive responses and emphasize fault-sensitive time–frequency patterns. Furthermore, a semantic-aware soft contrastive loss is developed to mine potential semantic neighbors from multi-scale frequency-domain structural similarity, reducing false-negative effects in conventional hard-label contrastive learning. Finally, the pre-trained time-domain extractor is transferred to the target machine and fine-tuned with limited labeled samples. Experimental results show that MTFSC outperforms comparison methods under different labeled sample ratios and achieves an average accuracy of 97.5% across four cross-machine diagnostic tasks. Full article

Industrial systems increasingly rely on MQTT-based message streaming to enable automated, data-driven production processes at the network edge. While semantic models such as the SSN/SOSA ontology enable machine-interpretable descriptions of observations and actuations, an explicit model of message transport is rarely considered. Consequently, MQTT-based communication remains opaque, particularly regarding information processing, hindering the semantic analysis of application-specific topic structures and the behavior of transport protocols. To close this gap, this work introduces the revised MQTT4SSN ontology as a key contribution, extending existing semantic models with protocol-aware representations of MQTT entities, control packets, and transport-level interactions. MQTT4SSN enables end-to-end semantic traceability, from sensor observations and actuator controls to the underlying message transmission within distributed systems. Building on this contribution, the MQTT2RDF integration framework incorporates MQTT4SSN as its core to capture live MQTT traffic and represent both payload meaning and transport-level provenance within an RDF knowledge graph. This work presents a novel approach for representing edge computing and information processing over MQTT, addressing two key challenges. First, the framework supports semantic interpretation of topic hierarchies and provides configurable mappings between MQTT topics, payload structures, and observation or actuation semantics. This approach facilitates the setup of edge computing systems and enables context-aware subscription management and structured data formatting, thereby improving interoperability between heterogeneous deployments. Second, transport-level provenance analytics provide a semantic basis for query-based detection, classification support, and diagnostic analysis of malformed or incomplete MQTT communication. The approach provides explainable, traceable information processing through transport provenance, which is essential for safety-critical industrial environments. The contributions are validated through an industrial use case from a production environment, demonstrating its applicability for system monitoring, troubleshooting, and semantic analytics of MQTT-based infrastructures. Full article

Climate change is reshaping agricultural systems by altering temperature and rainfall regimes, increasing the frequency of extreme events, and intensifying risks to crop productivity, water use, and farm decision-making. As climate–agriculture research expands rapidly, it becomes increasingly difficult to identify consolidated knowledge domains, emerging priorities, and evidence gaps. This study maps the structure and evolution of this literature using 219,261 Scopus-indexed documents selected from 290,560 records published between 1990 and 2025. A text-mining workflow combined BERTopic-based semantic modeling with supervised thematic classification into 18 macro-themes, while annual shares, z-scores, and document-level primary–secondary co-framing were used to assess temporal salience and cross-theme coupling. The results show sustained growth in research output, with 53.67% of publications produced between 2016 and 2025, and strong geographical concentration in the United States and China, which together account for 41.98% of the corpus. Hydrology and water management, crop production, impact assessment, and atmospheric processes remain central pillars, while socio-economic vulnerability, food security, sustainability, biotechnology, and greenhouse gas mitigation have gained prominence. The resulting evidence map provides a reproducible overview of the climate–agriculture knowledge landscape and can support research prioritization and policy design for climate-resilient agrifood systems. Full article

The rapid expansion of artificial intelligence in the educational field has configured a broad, dynamic, and constantly evolving research domain. Nevertheless, there remains a need to systematically analyze the evolution of its pedagogical approaches and to identify the conceptual dimensions that structure recent scientific production. For this purpose, a systematic literature review was conducted following the PRISMA protocol, based on searches in Web of Science and Scopus. The final corpus consisted of 235 articles, analyzed using bibliometric and semantic techniques in R, including bibliometrix, tidyverse, and ggplot2, complemented by co-occurrence maps developed with VOSviewer. The thematic classification was carried out through an inductive analysis based on clusters and emerging patterns. The results reveal a progressive transition from technocentric approaches toward more complex and integrative pedagogical perspectives. The semantic analysis made it possible to identify four structuring dimensions of the field: critical, ethical, literacy-oriented, and humanistic. Recent literature also shows a growing emphasis on teacher education, academic integrity, and cognitive coexistence between humans and intelligent systems. These findings indicate that artificial intelligence not only introduces technological innovations but is also reconfiguring the epistemological and pedagogical foundations of contemporary education, demanding conceptual frameworks capable of articulating its ethical, cognitive, and formative implications. Full article

Steel surface defect inspection plays an essential role in maintaining product quality and production safety in industrial manufacturing. However, existing detection methods still encounter difficulties in accurately identifying tiny defects, suppressing interference from complex backgrounds, and balancing detection accuracy with computational cost. To address these challenges, this paper proposes CNS-YOLOv8, an improved defect detection model based on YOLOv8n. First, a C2f_SCConv module is introduced to enhance multi-scale feature extraction and spatial representation capability. Second, a Normalization-based Attention Module (NAM) is embedded after the high-level semantic feature layer to improve the model’s sensitivity to critical defect regions. Third, a SlimNeck structure is adopted to strengthen feature fusion while reducing computational overhead. Experimental results on the NEU-DET dataset demonstrate that CNS-YOLOv8 achieves 83.1% mAP@0.5 and 49.6% mAP@0.5:0.95, surpassing YOLOv8n by 3.9 and 1.2 percentage points, respectively. In addition, comparative experiments show that CNS-YOLOv8 outperforms Faster R-CNN and YOLOv7 in terms of mAP@0.5 while requiring substantially fewer GFLOPs. In general, the proposed method balances detection accuracy and computational efficiency effectively, highlighting its potential for real-time industrial surface defect detection. Full article

Multi-source heterogeneous data in discrete manufacturing shop floors, including vibration signals, equipment logs, visual monitoring data, and handwritten production reports, exhibit significant differences in modality and semantic representation. Traditional fusion methods often fail to bridge the semantic gap between low-level sensing signals and high-level manufacturing cognition, limiting intelligent anomaly analysis and decision-making capability. To address this issue, this paper proposes a semantic translation and fusion framework for industrial heterogeneous data based on Knowledge Graph (KG), Retrieval-Augmented Generation (RAG), and Large Language Models (LLMs). First, a unified semantic translation mechanism is developed to convert multimodal industrial data into structured semantic representations for cross-modal alignment. Second, an industrial knowledge graph and RAG mechanism are introduced to integrate process knowledge, maintenance manuals, and historical fault records into the reasoning process. Third, an LLM-driven reasoning framework is designed for multimodal semantic fusion, anomaly identification, causal analysis, and optimization recommendation generation. In addition, a digital twin-based visualization interface is constructed to realize real-time interaction between production lines, industrial data, and intelligent cognitive reports. Experimental results demonstrate that the proposed framework significantly improves industrial reasoning accuracy, anomaly analysis correctness, and response efficiency compared with general-purpose LLMs, providing an effective solution for intelligent cognition and decision-making in discrete manufacturing systems. Full article

The study aimed to examine the role of artificial intelligence in preservice science teachers’ analogical reasoning by comparing the features of analogy designs produced with and without artificial intelligence. The research was conducted with 133 preservice science teachers at a public university in Türkiye. Participants were divided into two conditions: those who designed analogies using artificial intelligence (n = 62) and those who designed analogies without artificial intelligence (n = 71). Analogy design products were analyzed using descriptive analysis, and categorical data derived from these analyses were examined through Pearson’s chi-square tests. In addition, qualitative data obtained from structured interviews with the AI-supported condition were analyzed using content analysis. The results revealed significant differences between the groups in several dimensions of analogy design, presentation format, semantic distance, analogical association, wealth level, and the identification of limitations. Analogies designed with artificial intelligence were more frequently pictorial–verbal, involved both close and remote semantic distance, integrated structural–functional associations, and exhibited extended analogy characteristics. Interview results indicated that preservice science teachers primarily used AI for idea generation, visualization, and creative exploration rather than for generating factual knowledge. These results contribute to the literature by highlighting the potential role of AI in supporting representational transformation processes within science teacher education. Full article

Product bundling boosts e-commerce revenue by recommending complementary item combinations. However, existing methods face two critical challenges: (1) collaborative filtering approaches struggle with cold-start items owing to dependency on historical interactions, and (2) LLMs lack inherent capability to model interactive graph directly. To bridge this gap, we propose a dual-enhancement method that integrates interactive graph learning and LLM-based semantic understanding for product bundling. Our method introduces a graph-to-text paradigm, which leverages a Dynamic Concept Binding Mechanism (DCBM) to translate graph structures into natural language prompts. The DCBM plays a critical role in aligning domain-specific entities with LLM tokenization, enabling effective comprehension of combinatorial constraints. Experiments on three benchmarks (POG, POG_dense, Steam) demonstrate 6.3–26.5% relative improvements in HitRate@1 over the strongest baseline. Full article

The canopy closure stage is a critical phase of rice (Oryza sativa L.) development that influences canopy structure and final grain yield. Accurate and continuous monitoring of canopy closure dynamics is therefore essential for variety screening and cultivation optimization. This study combines unmanned aerial vehicle (UAV) remote sensing technology with deep learning-based semantic segmentation to establish an efficient framework for quantifying rice canopy closure dynamics. UAV RGB images were acquired for 198 hybrid rice varieties during early growth stages and used to build a canopy segmentation dataset. Three semantic segmentation models, i.e., DeepLabv3+, U-Net, and PSPNet, were systematically evaluated. Results show that DeepLabv3+ performed the best and enabled precise extraction of rice canopy features, obtaining a mean intersection over union (mIoU) of 0.86. Based on the extracted canopy coverage, the Gompertz model was utilized to characterize temporal canopy closure trajectories for all varieties, achieving an average R² of 0.978. Subsequently, five key dynamic indicators were derived, including canopy closure limit value (K), initial growth coefficient (a), growth rate coefficient (b), maximum instantaneous growth rate (MGR), and days to maximum growth rate (Tm). K-means clustering analysis was performed on these indicators to categorize all rice varieties into three clusters, disclosing pronounced differences in early-stage canopy development characteristics. Correlation analysis further demonstrated that canopy closure dynamics were closely associated with grain yield. Overall, while acknowledging the limitations of a single-season and single-site dataset, this study provides a scalable and objective framework for quantifying rice canopy closure dynamics, offering valuable support for variety selection, cultivation optimization, and high-yield rice production. Full article

The battery cell production, a cornerstone of the net-zero vision, is a multifaceted process chain involving diverse processes, spanning from batch to continuous to single-unit steps. The quality of the battery cell as the final product is affected by various product and process parameters along this process chain. In the era of Industry 4.0, data-driven approaches have emerged as a promising solution to navigate these complexities and derive effective quality management practices. A key prerequisite for the successful implementation is the availability of accurate data. A tracking and tracing system in battery cell production provides the foundation to acquire such data. It supports the development of a digital twin of the product, enabling real-time monitoring of key performance indicators, in-line quality control, resource optimization, and compliance fulfillment, among others. This article presents an implementation methodology and discusses the key aspects to consider for upscaling such a system focusing on data management, including relevant parameters, data acquisition, and storage, as well as data structuring and mapping. It highlights the advantages of using ontology-based data descriptions, enabling semantically mapped production environments. Lastly, this article explores potential use cases facilitated by a traceability system, emphasizing its potential to realize intelligent, data-driven production. Full article

We study positive–negative guarded systems of language equations over a fixed finite alphabet. The ambient space is the complete ultrametric space of all formal languages equipped with a length-based distance, where two languages are close whenever they agree on all words up to a sufficiently large length. The systems considered here contain both positive recursive dependencies and negative dependencies expressed through language complements. To handle this mixed structure, we introduce a suitable product order on pairs of languages and prove that the associated system operator has the weak monotone property. We show that the complement is an isometry for the length-based ultrametric and establish a signed wrapping estimate for guarded positive and negative language terms. These estimates lead to an ordered contraction principle for comparable pairs. As a consequence, the canonical lower and upper Picard iterations converge to the same limit, which is the unique fixed pair of the system. We also derive an explicit convergence rate and a finite-depth certification result: after a prescribed number of iterations, the approximants agree with the fixed-point semantics on all words below a given length. Additional symmetry assumptions are shown to force the unique fixed pair to be diagonal, reducing the system to a single language equation. Finally, we discuss an application to trace-based policies for tool-using AI agents. In this interpretation, finite executions of an agent are represented as words over an alphabet of observable tool-events, and the two components of the fixed point provide a stable semantics for policy-defined admissible and risky trace classes. The resulting framework gives a mathematically certified method for finite-depth analysis of recursive trace-based policies based on ultrametric fixed-point techniques. Full article

Frequent engineering changes in manufacturing require worker instructions to be updated quickly and reliably. In many production environments, however, update handling still depends on manual comparison procedures, delayed communication, or repeated traversal of large document collections, limiting responsiveness during ongoing production changes. This paper presents a hierarchical hash-based method for change detection in structured manufacturing documents as the computational core of a worker assistance system for near-real-time instruction updates in the context of in-line qualification. Heterogeneous instruction data are transformed into canonical hierarchical document structures, from which SHA-512 digests are generated at multiple structural levels. During repeated comparison operations, document-state evaluation is reduced to digest comparison, while structural differences can be localized through hierarchical refinement of affected substructures. The method is integrated into a system architecture that combines predecessor-linked version management with role-specific filtering for controlled dissemination of relevant instruction updates. The approach was implemented in an automotive assembly use case involving structured work instructions and evolving production documentation. The evaluation demonstrates that the proposed approach reduces repeated comparison effort relative to conventional field-wise traversal methods while maintaining the ability to localize structural changes through hierarchical refinement. The reported results focus on computational behavior and implementation feasibility in structured manufacturing environments rather than hardware-specific throughput benchmarks. Overall, the results indicate that hierarchical comparison of structured instruction states provides a practical basis for change-aware worker assistance and controlled propagation of instruction updates in evolving manufacturing environments. The evaluation focuses on repeated-comparison scenarios in structured manufacturing settings and does not address semantic interpretation of detected changes or large-scale distributed deployments. Full article

Complex products continue to face low iterative-design efficiency and poor cross-generation data compatibility, while existing product-gene research is still constrained by the predominance of qualitative approaches, ambiguous representations of hierarchical associations, and insufficient standardization. Based on the principles of decomposition and reconstruction and the systems thinking of genetic engineering, this study develops a generic three-level framework for product genes at the platform, assembly, and component levels. Hierarchical mapping functions and parameter-constraint equations are introduced to enable quantitative representation, and a quantitative product-gene information system is established, including a core-parameter quantification model and inter-/intra-level association-strength models. By integrating multiple international standards, the study further constructs a tripartite standardized description system covering metadata, semantics, and format, and proposes a mathematical mapping method from product information to standardized formats. A case study of Company A’s Platform B and Concept Vehicle C shows that the association-strength model achieves the required adaptation threshold, thereby validating the proposed framework. This study provides quantitative theoretical support for the platform-based and intelligent development of complex products and offers an implementable technical solution for product-gene reuse and data sharing, particularly in the new energy vehicle industry. Full article

Intelligent autonomous driving systems must not only predict the appropriate driving manoeuvre but also provide human-interpretable evidence that justifies the decision. However, existing methods typically address these objectives separately, leading to three practical limitations: multi-stage perception-to-language pipelines can propagate upstream perception errors into downstream explanations; post hoc saliency methods often produce pixel-level highlights that are difficult to interpret semantically; and decoupled decision and explanation modules cannot guarantee that the explanation reflects the same scene evidence used for behaviour prediction. In this paper, we propose a unified framework that jointly performs vehicle behaviour prediction and human-centric interpretation from a shared visual backbone. Specifically, a hierarchical Swin Transformer encodes the driving scene into a sequence of spatial tokens, which are processed by two complementary branches. The first branch, termed the Object Selection Module (OSM), learns a compact scene-level semantic representation through query-guided cross-attention, while the second branch extracts a small set of class-agnostic object-centric tokens without requiring bounding-box or segmentation supervision. These two representations are subsequently integrated by a Semantic–Object Fusion (SOF) module based on scaled dot-product attention, residual connections, and a feed-forward network. The behaviour prediction head operates on the fused representation, whereas the interpretation head leverages the semantic representation through a skip connection to preserve decision-relevant context. For surround-view perception, learnable per-camera embeddings are introduced to maintain viewpoint identity with negligible additional parameter cost. Furthermore, a compact language model fine-tuned via Low-Rank Adaptation (LoRA) generates fluent, label-conditioned natural-language justifications. Extensive experiments on two public benchmarks, BDD-OIA and nu-AD, demonstrate that the proposed framework consistently delivers superior performance and provides effective, human-readable interpretations of driving decisions. Full article

While social media platforms are primary vectors for misinformation, automated detection systems remain largely confined to English. This paper presents a transferable, three-stage framework for fine-tuning transformer models to detect domain-specific deceptive content in Spanish. The pipeline comprises: (1) corpus unification, merging fragmented datasets into a 61,674-article resource mapped into three classes (Real, Fake, Satire) to prevent stylistic confounding; (2) systematic model optimization, extensively benchmarking classical metaheuristics against eight transformer architectures (including mBERT, XLM-RoBERTa, and BETO) using strong regularization to mitigate overfitting; and (3) production deployment, encapsulating the optimized model as a containerized web application for real-time inference. Through rigorous experimentation, the Spanish-specific BETO encoder emerged as the strongest model for this task, achieving 89.18% overall accuracy. The model attains a near-perfect in-source F1-score on the satire class; however, a strict source-held-out test reveals that this performance is highly source-dependent—recall on satire from an unseen outlet drops to 0.08—indicating that single-source class construction leads the model to recognize the source rather than a generalizable category. We report this finding as a central methodological result: corpus design, and in particular the source diversity of each class, is the primary determinant of whether the framework generalizes. Adversarial robustness tests using named-entity masking and typo injection provide complementary evidence on the model’s reliance on semantic versus surface cues. The methodology is designed to be adaptable across domains: by substituting the training corpus, the same framework may in principle be retargeted to other digital threats, such as investment scams and phishing, provided that suitable labeled corpora are constructed and validated for each new domain. The complete framework, dataset, and application are released as open-source resources to support reproducible research and practical countermeasures against online misinformation. Full article