Industry 5.0 (I5.0) represents a paradigm shift toward resilient, human-centered, and sustainable industrial systems. In this context, agri-food supply chains (AFSCs) have emerged as a critical domain for applying digital technologies to enhance traceability and performance. This study conducts a systematic literature review of 115 articles published between 2020 and 2025 to analyze how traceability technologies are integrated within AFSCs under the I5.0 paradigm. The results identify blockchain and IoT as the dominant technologies, supported by complementary solutions such as Artificial Intelligence and QR/RFID systems. The analysis reveals that traceability is inherently multi-actor and data-intensive, with significant variability in the variables tracked across supply chain stages. Furthermore, three main integration approaches are identified: data acquisition-driven, blockchain-based, and system-level approaches, highlighting a transition toward more integrated but still fragmented implementations. Building on these findings, the study proposes a multi-layer, matrix-based conceptual framework that integrates functional layers with AFSC stages, enabling traceability to be conceptualized as a system-level capability. The framework provides a conceptual basis for advancing traceability toward more integrated, adaptive, and human-centered systems aligned with I5.0 principles. The findings contribute to the literature by providing a structured understanding of traceability integration and offer practical insights for developing more resilient and sustainable AFSCs.
The rapid expansion of complex network infrastructures and the widespread adoption of end-to-end encryption have rendered traditional Network Intrusion Detection Systems (NIDS) increasingly ineffective. While conventional Machine Learning methods address the challenge of encrypted traffic by analyzing statistical features, their tabular nature fundamentally ignores the relational and topological semantics inherent in cyberattacks. To overcome this limitation, Geometric Deep Learning, specifically Graph Neural Networks (GNNs), has emerged as a state-of-the-art solution. This paper presents a Systematic Literature Review (SLR) guided by the PRISMA 2020 framework, synthesizing recent advancements (2020–2026) in GNN-based NIDS. By analyzing a curated corpus of 70 primary studies, we identify three critical algorithmic gaps constraining real-world deployment: extreme reliance on labeled data, lack of topological explainability (XAI), and severe vulnerability to Domain Shift phenomena due to insufficient Out-Of-Distribution (OOD) evaluation protocols. Consequently, we propose a strategic roadmap for future research, advocating for the transition towards edge-centric, self-supervised architectures leveraging Graph Contrastive Learning (GCL) to achieve robust and inductive generalization against Zero-Day threats in heterogeneous environments.
The virtualization of laboratory experiments offers an alternative for addressing infrastructure, safety, and resource availability constraints in university chemistry laboratories. This study presents the development and preliminary evaluation of a virtual prototype for the laboratory practice of physical and chemical changes. The prototype was developed by virtualizing real laboratory procedures through three-dimensional modeling, interactive simulation, and a modular software architecture designed to support component reuse and the progressive incorporation of additional experimental practices. Nine experimental trials involving physical and chemical phenomena were implemented within a virtual representation of a real chemistry laboratory. A preliminary perception-based evaluation was conducted with 30 participants, including students, chemistry faculty members, and laboratory staff. The prototype obtained an overall mean perception score of 4.13 out of 5, with the highest ratings corresponding to acceptance and adoption potential (4.53) and perceived experimental fidelity (4.33), while usability received the lowest score (3.63). These findings indicate favorable initial perceptions of the prototype and identify usability as the main area for further improvement. The implemented architecture provides a technical basis for extending the virtual environment with additional experimental chemistry practices.
Vehicular Ad Hoc Networks (VANETs) and Vehicle-to-Everything (V2X) systems require simulation workflows that combine realistic mobility, wireless propagation, routing behavior, and packet-level performance evaluation. However, using OpenStreetMap, SUMO, and NS-3 together often requires a fragmented and error-prone sequence of manual tasks, limiting accessibility and reproducibility. This paper presents a web-based application that transforms this conventional workflow into an end-to-end, trace-driven simulation environment for VANET/V2X studies. The proposed framework integrates a React frontend, a Django REST backend, SUMO-based mobility generation, and a customized NS-3 pipeline. Through a unified interface, users configure map regions, vehicular densities, communication scenarios, routing protocols, wireless modes, and traffic parameters, while the backend automatically generates mobility traces, FlowMonitor reports, JSON summaries, execution logs, and cumulative CSV records. The system was validated with urban scenarios derived from Quito, Ecuador, considering three density levels, V2V, V2I, and V2X communication patterns, and four routing protocols: OLSR, AODV, DSDV, and DSR. Results show that the application reveals protocol-dependent differences in Packet Delivery Ratio, throughput, delay, jitter, and packet loss. Beyond individual experiments, the framework provides a reproducible basis for simulation campaigns, V2X dataset generation, and future AI-assisted communication analysis.
3D object pose estimation is required when a robotic manipulator must locate an object before grasping, placing, or interacting with it. Training deep-learning models for this task usually requires large annotated datasets that are costly to collect in physical environments. This paper presents an integrated synthetic-data pipeline for a custom plastic container. It combines 3D modeling, photorealistic rendering, domain randomization, a DOPE-inspired convolutional network for projected-keypoint detection, Perspective-n-Point pose recovery, and robotic simulation. The 4,000-image dataset contains 2,000 photorealistic and 2,000 domain-randomized images and was divided into 3,200 training and 800 test images. Evaluation comprised 25 virtual trials in CoppeliaSim and 10 controlled webcam trials. The retained measurements support camera-to-object radial-distance evaluation rather than full vector-translation evaluation: the virtual trials produced a mean absolute error of 0.889 cm, an RMSE of 1.039 cm, and a mean absolute percentage error of 1.69
