Federated Learning (FL) has emerged as a machine learning approach able to preserve the privacy of user’s data. Applying FL, clients train machine learning models on a local dataset and a central server aggregates the learned parameters coming from the clients. The training of the global machine learning model runs without sharing user’s data. However, the state-of-the-art shows several approaches to promote attacks on FL systems. For instance, inverting or leaking gradient attacks can find, with high precision, the local dataset used during the training phase of the FL. This paper presents an approach, called FAST Improved Deep Leakage from Gradients (FAST-iDLG), which is able to improve the inverting gradient attack, considering the spatial correlation that typically exists in a set of images stored in the local FL dataset. Instead of initiating the attack of an image using dummy data, FAST-iDLG blends the previous reconstructed images using a sliding window and an exponential function to weight the blend. The performed evaluation shows an improvement of 66% - 26% in terms of attack success rate and reduces by 55% - 25% the number of iterations per attacked image.
The rapid growth of electric vehicle (EV) adoption has intensified the need for secure, reliable, and interoperable charging infrastructures. The Open Charge Point Protocol (OCPP) 2.0.1 introduces advanced features to support secure and intelligent EV charging. However, critical fields like the state of charge (SOC) still remain highly unexplored and optional in practice. The selective adoption of the SOC within the OCPP protocol introduces risks related to the fair scheduling, grid stability, and billing accuracy. In this paper, we introduce a case study on SOC spoofing to demonstrate the implications of treating this field as optional within the OCPP protocol. The priority hack and the overstay hack are studied as two different types of attacks, where the adversaries manipulate the SOC values in order to gain unfair charging priority or extend the station occupancy, respectively. An autoencoder-based anomaly detection mechanism is designed to identify the spoofing attacks, achieving an F1 score of 87%. Our testing and validation demonstrate that even when only the initial and final SOC values are transmitted, the spoofing can still be detected. Our analysis concludes to the proposal of making the SOC reporting mandatory, at least at session start and end, in order to significantly strengthen the reliability and security of the OCPP 2.0.1-enabled infrastructures.
Recent advances in Machine Learning (ML) are revolutionizing industrial processes, particularly in the context of Industry 5.0, which promotes sustainable and intelligent manufacturing practices such as Zero Waste Manufacturing. However, the application of ML in industrial settings is often constrained by limited labeled data and the high dimensionality of sensor inputs, which hinders both model performance and interpretability. This study addresses the challenges of high-dimensional industrial data by evaluating conventional Feature Selection and Feature Extraction methods, such as ANOVA, Mutual Information, and PCA, as well as more advanced techniques based on Explainable AI (XAI) and Causal AI. In addition to benchmarking traditional methods, we experiment with a novel hybrid feature selection approach that combines SHAP-based feature importance analysis with causal discovery techniques. Experimentation was conducted using a real-world dataset from Carpigiani, a leading manufacturer of ice cream production machines. We aim to predict mixture filling levels to prevent critical freezing events during production. Our results demonstrate that integrating XAI and Causal AI enables the development of more interpretable and reliable ML models, thereby enhancing computational efficiency and decision transparency, a key requirement in Industry 5.0 applications.
In recent years, deep learning models for computer vision applications like YOLO (You Only Look Once) have become widely used for object detection applications in real time. Mobile GPUs such as the NVIDIA Jetson series are highly promising for deploying these models in resource-constrained edge computing platforms. The objective of this work is to examine and compare the performance for object detection (including detection accuracy, model size, execution time, and power consumption) of different versions of YOLO models deployed on different NVIDIA Jetson mobile GPUs using the COCO8 dataset. This allows for evaluation of the four performance metrics, emphasizing very different aspects, across combinations of various YOLO versions, edge devices, and datasets. The experimental results offer important insights into the trade-off among the choices of different YOLO versions and mobile GPUs and their practical potential. Our findings demonstrate that the YOLOv8 models maintain similar accuracy levels across all Jetson devices, where model accuracy achieves up to 0.779 in the COCO8 dataset, underscoring their effectiveness for real-time applications in edge computing. Additionally, our analysis reveals significant variations in power consumption, with the Jetson AXG Orin showing an efficient balance between performance in inference time and energy usage, where the power consumption is 11.05 W and the inference time is 78.1 ms even while running the most complex YOLOv8x model. This study will help researchers and practitioners to gain insight and select the best combination of mobile GPU and YOLO version for future object detection tasks at the edge.
The rapid adoption of Internet of Medical Things (IoMT) devices integrated with the proposed novel idea of Ceiling-Mounted Systems (CMS) significantly enhances healthcare data analytics and patient monitoring. However, this integration presents critical challenges related to data security, privacy, and efficient data transmission. Therefore, this study introduces an innovative cryptographic framework specifically designed to secure data exchange between IoMT devices and CMS infrastructures. The proposed system combines advanced encryption algorithms, circular matrix-based cryptographic methodologies, and deep reinforcement learning-based resource allocation to ensure data confidentiality, integrity, and timely transmission. Simulation results demonstrate that the proposed approach significantly outperforms traditional methods in reducing latency, optimizing energy consumption, and enhancing overall system security.
Enhancing reliability and energy efficiency is a key objective of sixth-generation (6G) wireless communication systems. To support this goal, we propose a novel ambient backscatter-assisted NOMA (AmBC-NOMA) system that jointly supports device-to-device (D2D) and cellular uplink communications, aiming to improve both energy efficiency and spectral utilization for low-power IoT networks. The system introduces a passive backscatter device (BSD) that enhances data forwarding without requiring active RF transmission. We derive closed-form expressions for the outage probabilities and SINR of both D2D and cellular links, considering realistic channel impairments and interference. Numerical results reveal that our architecture achieves up to 40% improvement in D2D outage performance and 10% better cellular reliability compared to OMA counterparts under practical SINR thresholds. These findings demonstrate the feasibility of integrating AmBC with NOMA for future green 6G wireless systems.
In this paper, we consider performance of virtual network embedding algorithms and their applicability to inter-connection topologies used for inter-processor communication networks. Virtual Network Embedding (VNE) is a process of assigning virtual network requests to a substrate network (physical infrastructure) in data centers. VNE algorithms are applied to interconnection networks employing Butterfly and Hypercube topologies that are distinct from network topologies used in data center networks. Simulation results indicate that these topologies lead to higher acceptance ratios than topologies used in data center networks. The Butterfly topology leads to the highest acceptance ratio while having lower revenue to cost ratio. In contrast, the Hypercube topology offers similar acceptance ratio to Butterfly topology while having a higher revenue to cost ratio.
Emerging paradigms such as cloud-edge continuum, Software-Defined Networking (SDN), and Digital Twin (DT) technologies are transforming smart manufacturing systems by enabling intelligent automation, real-time monitoring, and scalable orchestration. These capabilities are particularly critical in additive manufacturing (AM), where latency-sensitive control and predictive maintenance are essential. However, current architectures often require dynamic network programmability, synchronized twin management, and secure telemetry pipelines. This paper presents an SDN-Integrated Cloud-Edge DT Framework tailored for real-time AM monitoring. The framework integrates KubeEdge’s DeviceTwin module for edge-local twin representation, a telemetry agent for structured data streaming, and SDN-controlled Open vSwitch (OVS) for adaptive traffic control. IoT-enabled AM devices, including 3D printers, CNC machines, and robotic arms, interface with the edge node for local state caching, while KubeEdge’s CloudCore aggregates device states for analytics, visualization, and policy enforcement. Experimental validation on a Kubernetes cluster demonstrates sub-100-ms twin synchronization, SDN enforcement under 312 ms, and streaming latency breakdowns across MQTT-Kafka stages. This work establishes a scalable, resilient, and programmable foundation for next-generation, Industry 5.0 manufacturing ecosystems.
Communication between assets and systems is one of the foundational principles of the Industry 4.0 paradigm, enabling increased automation, data exchange, and real-time decision-making across industrial environments. However, the growing integration between Operational Technology (OT) networks and core Information Technology (IT) infrastructures —and their progressive exposure to the Internet — introduces a broad spectrum of cybersecurity vulnerabilities. These threats range from unauthorized access and data exfiltration to lateral movement attacks and system-level disruptions, which can significantly impact safety, production continuity, and system integrity.Traditional security models often fall short in this context due to the many OT components’ unique constraints and legacy nature. This paper explores how Software-Defined Networking (SDN), with its centralized control and programmable architecture, offers a flexible and robust solution to improve the security posture of OT networks. By decoupling the control and data planes, SDN enables fine-grained traffic monitoring, dynamic policy enforcement, and rapid threat mitigation — essential in protecting heterogeneous industrial systems. The paper highlights the key advantages of SDN for OT-IT convergence. It discusses concrete use cases where SDN principles help detect, isolate, and respond to cybersecurity incidents in modern industrial environments.
Modern IoT and Fog computing environments consist of numerous interconnected and often powerful devices. Many of these devices are capable of processing not only local tasks but also tasks offloaded to them. Efficient handling of offloaded tasks is essential to maintain devices performance and responsiveness. A key challenge lies in optimizing how the offloaded tasks are scheduled and executed within devices while respecting their local tasks demands for CPU, RAM, and energy. This work proposes the application of Federated Learning (FL) to enable collaborative and privacy-preserving training of task scheduling models across distributed IoT devices. Each device learns to optimize task execution locally, while contributing to a global model that benefits the entire system. By leveraging FL, devices can adapt their scheduling strategies in real time without exposing sensitive local data. The proposed approach aims to: improve offloaded task waiting time, balance resource utilization, minimize energy consumption and the overall negative impact of offloaded tasks on a device’s functionality. The contribution of this research lies in exploring FL as a scalable and privacy-aware framework for optimizing in-device scheduling of offloaded tasks in heterogeneous IoT environments.
The radio access network (RAN) is responsible for nearly 75% of the total network energy use. Advanced capabilities introduced by Fifth-generation and Beyond (B5G) are likely to increase this energy foot-print. For example, in Open RAN (O-RAN) architecture Radio Intelligent Controllers (RICs) execute autonomous decision-making steps through AI/ML-based xApps and rApps increasing computational overhead and energy consumption. Existing studies focus on energy consumption in the core and conventional RAN components while the energy profiling of O-RAN RICs remains largely unaddressed. This paper integrates software-based power monitoring tools with two prominent RIC platforms. i.e., O-RAN Software Community (OSC) Near-Real-Time RIC and OpenAirInterface (OAI) FlexRIC. The energy profiles across these two RICs are discussed while running xApps under multiple workloads. The results offer actionable data to pursue energy-efficient and sustainable B5G network design.
This paper presents a framework designed to significantly enhance the performance and visibility of Autonomous Aerial Vehicles (AAVs) deployed for maritime border surveillance. Traditional approaches often struggle with limited coverage and inefficiencies due to static deployment and isolated optimization strategies. To address these issues, the proposed approach integrates 3D Voronoi-based node deployment with hybrid optimization techniques. Specifically, it combines a modified Genetic Algorithm (GA) to strategically refine sensor node placements, along with a Twin-Delayed Deep Deterministic Policy Gradient (TD3) algorithm enhanced by Particle Swarm Optimization (PSO) for dynamically optimizing AAV trajectories and computational task allocations. Simulation results demonstrate that the proposed method achieves significantly enhanced area coverage, reduced energy consumption, improved task completion rates, and overall better Quality of Service (QoS) compared to conventional methods.
A honeypot is a defense technique that complements defense systems, which are generally composed of firewalls and intrusion detection systems. A honeypot consists of a “bait” system that simulates attractive and vulnerable targets for potential attackers. In this article, we address the problem of the lack of dynamism in traditional honeypots, which are based on static scripts, with a special focus on botnet-type threats. To tackle this problem, the proposed solution integrates a large language model into a honeypot, capable of interacting with the attacker by dynamically imitating a target device. The open-source tool Cowrie was used as the traditional reference honeypot to receive functionalities from the large language model, including command response generation, log data analysis, and the emission of alerts and actions. Regarding the evaluation, tests were conducted to measure the accuracy of the command responses produced by the proposed honeypot, the time required for response generation, and the number of tokens. The obtained results reveal the viability of the proposed integration, showing that a large language model is a promising alternative for honeypots to achieve a higher level in deceiving attackers.
This demo paper presents a testing and validation platform for validating 5G Core Network and RAN functions through robotic process automation. Built with the Robot Framework, the platform enables reproducible verification of end-to-end signaling procedures and interface configurations across N1, N2, N3, N4, and F1 interfaces according to the 3rd Generation Partnership Project (3GPP). While fault injection is used for demonstration purposes only, the platform’s primary objective is to support scalable and standards-driven testing of the Open-Source 5G stack from OpenAirInterface.
Advanced applications emerging from distributed cloud and edge computing and Fifth-generation and Beyond (B5G) networks increasingly require high-capacity, low-latency, and highly reliable optical transport networks. Accurate fault localization and isolation are therefore of the essence in optical transport networks, including timely identification of both correlated alarms and the identification of the root alarm triggering further fault propagation. In this paper, a Graph Neural Network with Multi-Head Attention (GNN-MHA) architecture is applied to jointly address alarm correlation and root alarm detection within a unified framework. To support realistic training and evaluation, we develop a configurable Reconfigurable Optical Add-Drop Multiplexer-based (ROADM) simulator that generates timestamp-aware alarm propagation graphs under diverse fault scenarios. This simulator enables a comprehensive assessment of both alarm correlation and root cause detection capabilities. Simulation based on synthetic datasets shows that the proposed GNN-MHA model achieves an F1 score of 0.9202 for alarm correlation and an accuracy of 97.34% for root alarm detection, outperforming all baseline methods in overall performance. These results corroborate the effectiveness of the proposed model for topology-aware and time-sensitive fault analysis in optical transport networks.
The rapid growth of Electric Vehicle (EV) charging infrastructure has introduced critical cybersecurity challenges. Particularly, stealthy attacks that manipulate billing data by exploiting the Open Charge Point Protocol (OCPP) 2.0.1 have become a real threat. This paper identifies two novel attack vectors, stealthy under-billing, where the adversaries disguise energy consumption to reduce their payments, and stealthy over-billing, where they inflate costs by falsifying session data. In contrast to the traditional attacks, these exploits mimic legitimate charging behavior, thus, rule-based detection is not applicable. To address this challenge, we propose an autoencoder (AE)-based anomaly detection framework, deploying two specialized models: Under-bill-AE for non-concurrent sessions and Over-bill-AE for concurrent sessions. Our proposed solution is trained on real-world and synthetic datasets. The testing and validation result in exceptional detection performance, with Under-bill-AE attaining a 97.9% F1 score and Over-bill-AE reaching 80.84%. Also, a detailed comparative evaluation of the proposed solution to unsupervised approaches, like Isolation Forest, demonstrates its superiority and robustness in terms of identifying subtle billing manipulations while minimizing false positives.
A group of unmanned aerial vehicles (UAVs) cooperating to complete predetermined tasks is referred to as a swarm of UAVs (S-UAVs). Clustering, which divides UAVs into clusters, is one of the routing techniques that S-UAVs use the most. A cluster head (CH) and cluster members (CM) make up each cluster. Due to its critical importance in routing packets to their destination, the CH selection process is an ongoing study area. Any UAV is vulnerable to damage in emergency situations, such as a fire. To guarantee end-to-end communication in the event of a non-functional CH, we suggest an optimized clustered weighted approach based on multiple redundancy. The CH, redundant CHs, and CMs are chosen using an optimized weighted measure that combines separation, speed, energy, and performance parameters. The proposed method automatically assigns a redundant node for every UAV to ensure a functional CH despite the number of non-functional UAVs that might be. According to the outcomes of the simulation that was run using MATLAB, this is a promising approach that ensures data delivery with a minimal delay in any emergency.
The integration of Information Technology (IT) and Operational Technology (OT) in modern production systems requires flexible, scalable, and deterministic control systems. Virtual Programmable Logic Controllers (vPLCs), deployed in containerized environments, are gaining traction as a cost-effective and adaptable alternative to traditional hardware PLCs. However, ensuring real-time performance in virtualized environments remains a critical challenge, especially under varying system loads and resource constraints. This paper presents a comparative evaluation of two vPLC platforms, CODESYS Control and OpenPLC, running within Docker and Podman containers on a Linux system with the PREEMPT-RT real-time kernel. The analysis covers both default and real-time-optimized configurations, focusing on timing precision, jitter, and deadline violations under increasing computational workloads. Experimental results reveal that Podman offers more predictable and stable real-time behavior than Docker, particularly in untuned scenarios. These findings emphasize the importance of running-time selection and system-level tuning to achieve deterministic execution for industrial automation. The results provide practical guidance for the deployment of containerized control workloads on the industrial edge and contribute to the broader goal of reliable hardware-agnostic systems in time-sensitive environments.
This paper introduces a high-fidelity electromagnetic simulation tool based on the Partial Element Equivalent Circuit (PEEC) method for the analysis of Reconfigurable Intelligent Surfaces (RIS). The tool supports the simulation of RIS under both plane wave illumination and more complex configurations involving transmitter–RIS–receiver interactions. It also enables the modeling of radiating elements connected to arbitrary external impedances. Implemented in Julia, the proposed framework demonstrates significant improvements in computational efficiency compared to traditional MAT-LAB implementations, particularly in matrix assembly and solution time. In the example presented in this paper, this translates into a reduction of computation time by approximately 84%, highlighting the practical benefits of the Julia-based approach for large-scale problems. The flexibility and precision of the tool make it a valuable asset for the optimized design and analysis of RIS in next-generation wireless communication systems.
Cyber-physical systems (CPS) are increasingly integral to the operation of critical infrastructure, blending physical processes with computational intelligence to enable advanced functionality. However, the interconnected nature of CPS makes them vulnerable to sophisticated cyber threats and physical anomalies, necessitating robust anomaly detection mechanisms. In this paper, we explore the IEEE 123 bus system to detect a stealthy insider threat resulting to a volt-var attack targeting photovoltaic (PV) systems, where the volt-var control is disabled to disrupt grid voltage operations. We propose GraphLLM-CPS, a novel framework leveraging Large Language Models (LLMs) to generate node embeddings for anomaly detection in CPS. By converting graph-based representations of CPS data into textual formats, LLMs are employed to extract meaningful node embeddings that capture the structural and semantic relationships within the system. These embeddings are then utilized to train supervised and unsupervised classifiers for detecting and locating anomalous nodes and events. The proposed approach is evaluated on CPS datasets derived from the IEEE 123 bus system, demonstrating its effectiveness in locating anomalies with high accuracy. Furthermore, we visualize the detected anomalies and compare them with ground truth impacted nodes, providing insights into the system’s behavior during the stealthy volt-var attack. The results highlight the potential of LLM-driven embeddings to enhance the security and resilience of cyber-physical systems against emerging threats.
