ABSTRACT. The growing adoption of circular economy practices is increasing the need for intelligent recycling systems that combine human expertise with AI-driven automation. While advances in computer vision and robotics have improved waste sorting and disassembly, human operators remain essential for handling uncertainty and supervising complex processes. Extended Reality (XR) technologies offer a promising interface layer for enhancing human-AI collaboration through immersive and context-aware interaction. This paper provides a perspective on the role of XR in recycling applications, including waste identification and sorting, robotic disassembly, training, and collaborative work environments. We discuss the potential of XR to improve situational awareness, transparency, and decision making, while highlighting key challenges related to real-time integration, usability, and trust. Moreover, we discuss the employment of multimedia and XR environments as educational methods to raise awareness about circular economy policies and requirements. Finally, we outline future research directions toward XR-enabled recycling systems aligned with Industry 5.0 and circular economy objectives.

ABSTRACT. Automating the identification of components on printed circuit boards (PCBs) from waste electrical and electronic equipment (WEEE) is a critical step towards making circular economy recycling scalable. Although supervised detectors have demonstrated high accuracy in this task, they require labelled data for every component type and struggle to generalize when new board generations introduce unseen parts. However, recent advances in vision foundation models and vision-language pre-training have enabled open-vocabulary object detection. This allows detectors to recognize arbitrary object categories specified by text prompts without requiring retraining. This paper explores whether such models can address this issue and whether they can operate at the edge, where WEEE sorting lines function without server-class hardware. We benchmark two open-vocabulary detectors: YOLOE and YOLO-World in zero-shot and fine-tuned modes, comparing them with the supervised baselines YOLO26 and RF-DETR-Small on the V-PCB dataset. Our results show that zero-shot open-vocabulary detectors significantly underperform supervised models on domain-specific electronic waste recycling tasks. However, fine-tuned YOLOE achieves 98.4% mAP@50, surpassing all supervised baselines while retaining open-vocabulary flexibility. None of the evaluated open-vocabulary detectors meet the 5 FPS real-time edge requirement, highlighting the need for quantization and model compression for practical WEEE deployment.

ABSTRACT. Abstract The transition towards a circular electronics industry requires a substantial improvement in the recovery of materials from Waste Electrical and Electronic Equipment (WEEE). Current treatment practices are still largely based on bulk shredding followed by downstream sorting, a model that is poorly suited to the growing material complexity of modern electronic products. As a result, many valuable, critical, or low-concentration materials are diluted, downgraded, or permanently lost. This paper presents Ariadne, a data-driven material recovery framework developed by Futuredata Srl. Ariadne translates manufacturer-provided product-level information into operational instructions for selective dismantling and sorting. By linking data on material composition, component location, and disassembly procedures to guided manual and automatic treatment processes, Ariadne shifts product data from passive traceability to active process control. Preliminary applications indicate that Ariadne can increase recoverable material fractions from a conventional baseline of approximately 8–10 to a target range of 25–30 fractions per device, depending on product complexity and data availability. This is particularly relevant given that an average EEE product typically contains more than 60 different materials [1] [2]. Two case studies, involving an air purifier and a coffee machine, illustrate the environmental and economic potential of selective upstream recovery. The paper also introduces the concept of Environmental Additionality Value, arguing that the benefits generated by Ariadne would not occur under standard WEEE treatment practices and are enabled by the voluntary participation of EEE manufacturers.

ABSTRACT. The rapid expansion of artificial intelligence workloads has intensified reliance on centralized data centers, resulting in significant energy consumption and infrastructure constraints. Concurrently, billions of edge devices including smartphones, laptops, and IoT systems operate at low average utilization, representing a vast pool of underutilized computational capacity.

This paper presents Netara, a decentralized AI inference framework that applies circular economy principles by transforming idle edge devices into a distributed computing network without requiring additional physical infrastructure. The proposed system consists of a lightweight device agent that activates during idle periods, a task orchestration engine that decomposes workloads into micro-tasks, a consensus-based verification mechanism ensuring reliability across heterogeneous nodes, and a REST-based interface for seamless integration.

By extending the lifecycle utilization of existing hardware resources, the framework aims to reduce dependence on energy-intensive data centers and improve sustainable compute accessibility. Preliminary prototype analysis suggests the feasibility of parallel task execution and fault-tolerant distributed inference, with projected potential cost reductions relative to comparable centralized inference systems.

The proposed approach contributes to sustainable AI infrastructure aligned with circular economy principles, particularly for resource-constrained regions and the Global South, while introducing the concept of circular compute as a model for reusing underutilized consumer computational resources for AI inference.

ABSTRACT. The growing volume of waste electrical and electronic equipment (WEEE) requires faster, more reliable recycling pipelines. Automated component detection on waste printed cir- cuit boards (WPCBs) is a critical bottleneck in this process. This paper presents a comparative analysis of two distinct real-time object detection models on the V-PCB dataset: YOLO26, a CNN-based method with five variants (n, s, m, l, and x), and RF-DETR, a transformer-based method with five variants (Nano, Micro, Small, Medium, and Large). Each model is trained using three independent random seeds to obtain statistically robust estimates. On the held-out test set, YOLO26x achieves the highest overall accuracy of 94.4% [94.3, 96.2] mAP@50 and 73.3% [73.2, 76.7] mAP@50:95, outperforming the best RF-DETR variant, RF-DETR-Large, which achieves 70.4% [70.3, 70.7] mAP@50. However, at the small-model end, RF-DETR-Nano (65.0% [64.0, 65.2] mAP@50) clearly outperforms YOLO26n (51.7% [40.0, 54.8]), indicating that the DINOv2 backbone provides stronger priors than ImageNet initialization when model capacity is limited. Across its five variants, RF-DETR shows decreased performance and an inconsistent scaling pattern. RF-DETR-Base underper- forms compared to RF-DETR-Medium despite having more parameters, while YOLO26x shows more consistent scaling. A per-class breakdown reveals that both families struggle most with coil and resistor detection and that YOLO26x outperforms the others on every class at the large end of the scale. These results suggest that CNN detectors have a clear accuracy advantage for this application with the given training resources of 150 epochs, while transformer models may benefit from extended training and stronger augmentation.

ABSTRACT. Large-scale metal additive manufacturing (AM) systems require substantial thermal energy during deposition, cooling, and post-processing. A significant portion of this energy is released as excess heat through cooling systems, exhaust streams, and heated products. This paper explores thermal energy recovery and utilization in large-scale AM processes, focusing on Wire Arc Additive Manufacturing (WAAM), Directed Energy Deposition (DED), and Binder Jetting systems. A structured recovery framework is developed by classifying recoverable heat sources into product cooling, printer-level recovery, and furnace-based post-processing recovery. Then, heat transfer and energy balance equations are formulated to estimate recoverable thermal energy from cooling loops, chamber gas streams, conductive substrates, and furnace operations. Finally, an equivalent heat source model is proposed to integrate multiple recovery pathways into industrial thermal networks. Comparative case studies using Stainless Steel 316L and AlSi10Mg alloy analyze the effects of thermal mass, cooling duration, and material properties on heat recovery performance. Results show that product cooling can be the largest source of recoverable heat. Printer-level cooling systems and furnace recovery pathways can also provide substantial recoverable energy, depending on the AM process configuration. Overall, the proposed framework provides a basis for integrating additive manufacturing systems into future energy-efficient and low-carbon industrial thermal infrastructures.

ABSTRACT. This study investigates the impact of electric vehicle (EV) penetration on Oman's Main Interconnected System through 2050, considering simultaneous renewable energy integration targeting 60% capacity share. Using Wien Automatic System Planning (WASP-IV) software, we developed six EV penetration scenarios varying by transition timing (2030, 2035, 2040) and charging behavior (immediate vs. delayed). The base case projects capacity increasing from 9,381 MW (2023) to 27,614 MW (2050) with 87,183 GWh annual generation. Full EV adoption (2.4 million vehicles) requires additional capacity of 2,500-3,300 MW. Early transition with immediate charging demonstrates highest capacity requirement (+3,277 MW) but achieves substantial CO₂ emission reductions of 30.40 million tons annually compared to conventional vehicles. Cumulative costs through 2050 range from $27-29 billion, with Loss of Load Probability maintained at 0.17-0.27%. Results indicate Oman's grid can accommodate full EV transition while achieving renewable targets, with optimal outcomes favoring gradual transition strategies balancing infrastructure costs, reliability, and environmental benefits.

ABSTRACT. Urban congestion remains a major challenge in rapidly growing metropolitan areas. This research provides empirical evidence that Ramadan creates measurable temporal redistribution of network congestion in Riyadh, Saudi Arabia, and introduces the Ramadan Peak Shift Index (RPSI) to quantify that redistribution at corridor level. The work contributes (i) the first network-level probe-data analysis of Ramadan's effect on traffic performance, (ii) a Peak Shift Analysis framework that compares functionally equivalent— rather than clock-identical—peak windows to isolate the redistribution component of the effect, and (iii) the RPSI as a transferable corridor-level indicator. The analysis covers 10 major arterial and ring road corridors representing 20 directional routes, aggregated from more than 4,000 individual road segments. Two complementary analyses are conducted: a direct comparison of identical time windows, and the Ramadan Peak Shift Analysis comparing conventional peak periods with their temporally shifted Ramadan equivalents. Statistical testing using paired t-tests reveals that the pre-iftar window pair (4–6 PM → 3–5 PM) produced the strongest network-level effect: average travel times decreased by 43.5% and speeds increased by 87.3% (both p < 0.001, d > 1.77). The morning commute shift (7–9 AM → 9–11 AM) yielded a 21.4% travel time reduction (p = 0.006). The corridor-level RPSI has a network mean of 30.1% and corridor values ranging from 16.4% to 50.6. These findings suggest that modest schedule shifts can substantially reshape peak-period congestion, highlighting the potential of demand-side temporal management strategies for large metropolitan road networks.

ABSTRACT. Linepack is a critical buffer in gas networks, providing short-term storage and operational flexibility. This paper presents a comparative dynamic analysis of the linepack impact in hydrogen (H2) and natural gas (CH4) networks under compressor contingency conditions. A three-day dynamic simulation is conducted for a CH4 network and a H2 network integrated with a power system. Two cases are investigated: Case 1 with identical and Case 2 with different pipe inner diameters in CH4 and H2 networks. The results in both cases show that H2 exhibits faster transient recovery after the compressor contingency, although its linepack is less than that of CH4. On the other hand, depending on the selected pipe diameter, pressure losses in the two networks can differ significantly. With identical pipe diameters, H2 exhibits lower pressure drop than CH4, and its demand therefore experiences less curtailment. However, when the H2 pipe diameter is reduced, the pressure drop across the H2 pipes increases and the curtailed load becomes higher than in the CH4 network. These results can be used by gas TSOs in designing and operating their grid more efficiently

ABSTRACT. With energy crises always round the corner, the ongoing energy sustainability efforts have significant reliance on citizens. In that respect, citizen awareness around optimal electricity usage and the adoption of energy saving measures is often considered for granted at personal, household, and workplace level. However, this assumption is not always supported in practice and in this work we assess the extent to which this is the case. To achieve this, we carried out a quantitative survey based on a structured questionnaire that explored the habits, knowledge, perceptions, and attitudes of adult inhabitants in Cyprus primarily regarding electricity usage at home while also touching upon other energy-related topics. The results revealed widespread misconceptions and knowledge gaps, including underestimation or overestimation of appliance consumption, behaviors associated with inefficient energy use, and limited familiarity with concepts such as renewable energy sources and electromobility. Based on our findings we designed and developed an IoT-based residential energy monitoring system and deployed two prototypes in real households; one passive consumer household and one prosumer household equipped with a photovoltaic system. Through this work we demonstrate how identified energy-awareness gaps can be translated into a practical monitoring approach that supports more informed electricity use and provides a basis for future work aligned with the Sustainable Development Goals of the United Nations.

ABSTRACT. Power electronics is the enabling backbone of the clean energy transition — every renewable source, storage system, EV charger, and modern grid asset depends on it. Power Electronics Finland forms a complete story in the field, from power semiconductors and their reliability to grid-scale energy systems, and this session presents that value chain as one coherent story. The session is organized by Power Electronics Finland, the inclusive national cluster that unites industry, academia, research institutions, and professionals to strengthen Finland's competitive advantage in power electronics. Following an introduction to the association and its objectives by Matti Kauhanen, five presentations build on one another in scale and scope. ABB (Joonas Leppänen) opens at the foundation with reliability research in power electronics applications — the prerequisite for everything built on top. Danfoss (Nicklas Södö) shows how reliable hardware becomes intelligent, as converters evolve from pure energy conversion toward grid-aware energy systems that actively support the grid. Kempower (Amir Safdarian) extends this intelligence to clean mobility, where energy management systems transform EV chargers into managed energy assets and define the next generation of charging infrastructure. MSc Electronics (Pekka Seppälä) addresses the question of scale through modular solutions for multi-megawatt power conversion systems, and VTT (Ville A. Ollikainen) closes the arc with a research vision of future clean and resilient power grids. Closely aligned with the RTSI 2026 tracks on power electronics and grid integration, clean mobility, and flexible demand, the session offers participants an industrially grounded complement to the academic programme: a condensed view of where research meets real-world deployment, and a direct channel to the Finnish power electronics ecosystem for research collaboration, partnerships, and talent development.

ABSTRACT. In today’s computer networks, network traffic analysis is a very important function in ensuring network security, efficiency, and functionality. In this paper, a network traffic analyser is proposed, which uses Wireshark and Scapy tools to analyse network traffic in real-time by capturing, examining, and evaluating IP packet traffic. Wireshark is a tool with an interactive graphical interface for in-depth examinations of network protocols, while Scapy is a tool enabling users to programmatically manipulate IP packets by writing scripts in a Python language. Based on network traffic analysis, anomalies in a network are detected, malicious network activity is identified, and communication problems in a network are solved. Performance tests conducted on the system prove its efficacy in real-time network traffic analysis of TCP/IP communications to detect intrusion attacks in a network.

ABSTRACT. Machine Learning (ML) models are already being deployed at scale for malware detection. Therefore there is a pressing need to rigorously study adversarial robustness. To that end, we introduce the Adversarial AI Engine (AAIE): a platform for automated security stress-testing. We performed transfer attacks, via EMBER2018, against a randomly initialized Random Forest target. Counterintuitively, we found single-step FGSM to have the highest attack success rate (ASR) of 52.9% compared to iterative PGD with 45.2% (p< 0.002). Compared to iterative attacks which tend to overfit surrogate networks creating brittle perturbations, FGSM created more transferable examples. We benchmarked our attacks using ASR, transferability score, evasion rate, and resilience index. Through these metrics, proposed in concordance with emerging standards in AI governance, we expose shortcomings as our transferability scores peak at ~0.8 with constrained perturbations. We highlight the practical security implications of these findings and demonstrate AAIE for automated adversarial auditing.

ABSTRACT. — Fault-tolerant control (FTC) is a necessity to achieve stability and performance in induction motor drive systems faced to degrade their actuators and have parameter uncertainties. The given paper suggests a hybrid observer-recurrent neural network (RNN) FTC system that brings together classical model-compensated with data-compensated correction. Residual-based detection mechanism presents gated LSTM and GRU correction modules to the condition of fault confirmation only, thus addressing the problem of unnecessary oscillatory compensation. Bounded and smooth control action is achieved by a safety layer of saturation, rate limiting and low-pass filtering that is appropriate to real-time industrial implementation. The given plan is performed in the case of a drastic loss of actuator effectiveness and parameter incompatibility. Comparison to baseline PI control and observer based FTC shows that tracking error (IAE and RMSE) is greatly decreased but with low detection delay and no false alarms. Furthermore, computational analysis indicates the real-time feasibility with a 1 ms sampling period. The findings suggest that, detection-conscious recurrent compensation promotes robustness and tracking accuracy without compromising the smoothness in control or the computational efficiency, in which case detection-conscious recurrent compensation can be implemented in the industry of motor drive.

ABSTRACT. Several goals, including increasing energy resilience, reducing energy and power losses, and enhancing system reliability, have motivated electric companies to connect renewable energy resources locally at the level of distribution networks. In order to realize such goals with the proliferation of the renewable sources, the concept of microgrid was introduced and has evolved into multi-microgrid systems. According to the IEEE Std 1547.4, large distribution systems can be clustered into a number of microgrids to facilitate their control and operation in the future systems with a high penetration of distributed energy units. In this study, a systematic approach is presented to optimally partition a distribution system into multiple microgrids to enhance reliability. In the proposed method, the power produced by distributed generation resources such as wind turbines and photovoltaic panels is modeled based on the capacity factor concept. The proposed optimization problem is implemented on the IEEE-69 bus test system and solved using a genetic algorithm in MATLAB. Also, the sensitivity of the results to the load and distributed generation capacity variations is assessed.

ABSTRACT. One of the main challenges of collaborative transfer learning (TL) is the secure model sharing across distributed edge nodes. In this paper, we propose a novel protocol for secure model publication and controlled access in distributed environments. Any model artifact is encrypted and signed locally by the owner node for integrity and authenticity verification, before publication to the InterPlanetary File System (IPFS) under a unique content identifier (CID). The CID and relevant model metadata are then registered immutably on an Ethereum smart contract via the publisher’s blockchain wallet. The proposed protocol also introduces a blockchain-mediated access-request workflow in which a requesting node registers its public key on-chain, such that it can use its private key to locally recover the symmetric decryption key generated by the model owner, which records the access grant on-chain. The requester then verifies the publisher’s digital signature before decrypting. Our experimental evaluation demonstrates that the protocol supports transparent model lineage, selective access control, and end-to-end confidentiality, making it well-suited to secure TL in trustless distributed environments.

ABSTRACT. • Facilitates the Student-to-YP Transition: Providing information on the hiring process and early-career expectations, directly reducing the friction graduates face when entering the workforce. • High-Value Networking: Provides IEEE Student Members with direct access to key stakeholders in the Finnish and global energy infrastructure sectors. • Enhances Membership Value: Demonstrates the tangible career benefits of IEEE and YP membership by delivering targeted, career-accelerating professional development. • Industry-Academia Feedback Loop: Creates a platform for industry leaders to communicate their current operational needs to the upcoming engineering workforce.

ABSTRACT. The increasing penetration of inverter-based resources has led to a significant reduction in system inertia, resulting in faster and more pronounced frequency deviations in modern low-inertia power systems. In such environments, the dynamic behavior of electrical loads becomes increasingly important in shaping overall system frequency response. This paper presents an enhanced load model that incorporates load-side dynamics in addition to conventional static behavior, thereby augmenting the representation of load–frequency control (LFC) models. This improved formulation increases the accuracy of frequency response studies in power systems. A comparison between the proposed augmented model and the conventional LFC representation demonstrates that relying solely on static load modeling can lead to inaccurate results and potentially misleading conclusions. Therefore, accurate modeling of load-side dynamics is essential for reliable frequency stability assessment in modern power systems.

ABSTRACT. Electric vertical takeoff and landing aircraft (eVTOL) can improve electrified urban mobility, but their end to-end competitiveness depends on road access, transfer burden, departure readiness, aircraft availability, and localized hub energy support. This paper develops an energy-constrained multimodal equilibrium framework for heterogeneous road only EV, pure eVTOL, and EV-to-eVTOL itinerary choice. The model Qinran Hu School of Electrical Engineering Southeast University Nanjing, China qhu@seu.edu.cn aircraft circulation, readiness, and shared charging/storage constraints at hybrid hubs. incorporates value-of-time heterogeneity, road congestion, EV-station waiting, eVTOL departure constraints, aircraft turnaround, and shared charging/storage limits at hybrid hubs. A bounded readiness score represents station-time eVTOL supportability, while hub-side power tightness is reflected through a station-level scarcity signal. A compact asymmetric OD case shows that direct EV remains dominant in most states, but its advantage weakens under higher access burden, lower readiness, and stronger hub-side scarcity. High VOT travelers respond more strongly to favorable airborne conditions, whereas EV-to-eVTOL mainly acts as a state dependent bridging option.

ABSTRACT. This paper deals with economic optimization of a distributed district consisted of buildings that vary in size and type. Since occupants are present at different times of the day in office and residential buildings, coordination among district buildings is driven by differences in heating/cooling energy demand throughout the day. Each building is equipped with battery storage systems that operate as reserve capacity. Thermal models of building zones are based on semi-physical approach, while the walls are modeled using a resistive–capacitive method. A linear optimization problem is associated with each building and solved using local model predictive control. Local constraints ensure that each zone tracks its reference within predefined limits while respecting the maximum building thermal power and the maximum battery charging and discharging limits. Distributed control, based on the asymmetric projection algorithm, is applied to adjust individual solutions of buildings towards coordinated optimal operation and improve overall district performance while adhering to the joint constraint of the local electrical infrastructure capacity. The algorithm is evaluated in a realistic simulation based on real data used for building models, meteorological variables and electricity prices. Distributed control improves privacy compared to centralized control and achieves higher monetary savings than decentralized control. Moreover, by integrating batteries and implementing proposed algorithm, economic benefit is increased by up to 21% compared to conventional control.

ABSTRACT. AI data centers are increasingly becoming tightly coupled compute--energy systems, where workload placement, cooling demand, electricity procurement, storage operation, and carbon emissions interact over time. This paper studies carbon-aware compute--power scheduling for geographically distributed AI data centers with microgrid prosumer capabilities. We propose a mixed-integer linear programming (MILP) framework that jointly schedules rigid training jobs, routes elastic inference workloads, dispatches local generation and battery storage, and manages bidirectional grid interaction under latency, continuity, power-balance, and carbon-budget constraints. The model captures two key features of emerging AI infrastructure: heterogeneous workload flexibility and site-level energy prosumer operation. Experiments on synthetic yet practically motivated instances show that the proposed joint MILP substantially improves total operational benefit over compute-only and energy-only baselines while reducing emissions. The results further indicate that inference-routing flexibility is a major source of value, battery storage provides useful temporal flexibility, and local-generation-rich settings are particularly favorable. The framework provides a tractable optimization abstraction for sustainable and grid-interactive AI data centers.