ABSTRACT. As the global energy corrective shifts towards sustainability, microgrids have surfaced as a viable solution to integrate renewable energy sources with increased energy resilience. Efficient energy management within microgrids assists with appropriate power distribution, cost minimization, and grid stabilization. This paper discusses the role of machine learning and the Internet of Things, through real-time monitoring, predictive analytics, and intelligent decision-making, in enhancing the energy management of microgrids. Machine learning algorithms help in demand forecasting, fault detection, and adaptive control strategies; IoT enables seamless collection and transmission of data across the microgrid ecosystem. The paper discusses several advantages of integrating machine learning and IoT with energy management, such as greatly increased efficiency, reliability, and sustainability. We discuss the scope of extension from small microgrid solutions to larger energy networks and provide indications for future work, notably refining machine learning models for increased accuracy along with operational issues ensuring security in IoT-based solutions. The results demonstrate how machine learning and IoT technology will transform the future of intelligent, autonomous, and sustainable energy management systems.

ABSTRACT. Credit risk prediction is both a difficult and of great interest problem, due to inherently unbalanced nature of such data and continuous interest in performing the prediction with high precision. We improve previous results on credit risk prediction and present an ensemble of decision Artificial Neural Networks architecture for credit risk classification. The extensive experimental results we present show improvements of previous work on metrics including accuracy, precision, sensitivity and specificity. Unlike previous methods, our method is completely automated, eliminating the need of manual processing and selection of data features, which improves generalization and scalability. While the main focus of this work is on credit risk prediction, our analysis shows that the model we propose can be used successfully for dimensionality reduction and classification of unbalanced data, in general.

ABSTRACT. In the digital age, analysing the discursive tone in medical articles has become essential for evaluating scientific objectivity and credibility. This paper explores an automated method for tone classification in medical abstracts, using a BERT-based model adapted as "Bio_ClinicalBERT" and modern natural language processing techniques. Data were automatically extracted from PubMed, and a labelled subset was used for fine-tuning. The labels were generated heuristically and used to train the model in a Databricks environment. The paper evaluates the model’s performance and highlights the method’s limitations, offering recommendations for extending the research through manual labelling and GPU (Graphics Processing Unit) usage. The results demonstrate the potential of tone classification in supporting critical analysis of medical literature.

ABSTRACT. In the fast-evolving landscape of industrial automation, maintaining the performance and reliability of mechatronic systems has become increasingly critical. Traditional maintenance strategies, such as reactive and time-based preventive approaches, often lead to inefficiencies, unplanned downtime, and excessive operational costs. This research examines how artificial intelligence (AI) optimizes predictive maintenance (PdM) for mechatronic systems, aiming to enhance uptime, extend asset life, and reduce maintenance costs.The study adopts a mixed-methods approach, combining a detailed literature review with case study analysis and model development. By examining real-world implementations from industries such as manufacturing and rail transport, the research identifies key enablers and obstacles in the transition to AI-driven PdM. A structured framework is proposed, guiding organizations through the steps of data acquisition, model selection, integration, and continuous improvement. Findings suggest that AI, particularly machine learning and deep learning techniques, significantly enhances the accuracy and timeliness of failure predictions when supported by high-quality sensor data. However, challenges such as data scarcity, system integration, model transparency, and ethical concerns remain barriers to widespread adoption. This work contributes a practical, scalable roadmap for organizations aiming to move beyond traditional maintenance paradigms. It emphasizes the value of aligning technological innovation with human expertise to create more resilient, efficient, and intelligent maintenance ecosystems in the era of Industry 4.0.

ABSTRACT. This paper explores the ethical integration of Artificial Intelligence (AI) into IT risk management processes. With a focus on transparency, fairness, and accountability, the work investigates how modern AI technologies can be deployed in various dimensions of risk identification, analysis, prevention, and control. Referencing ISO 31000 and NIST AI RMF, the paper presents key domains where AI intersects with IT risk management and proposes a structured ethical model supported by real-world examples.

ABSTRACT. The increasing pressure to reduce carbon footprint in the logistics sector has driven significant interest in sustainable practices within container terminal operations. This paper focuses on modelling and optimizing the carbon footprint of a container terminal operator by analyzing key operational processes. Existing literature extensively covers carbon footprint reduction strategies in logistics; however, specific optimization models tailored to container terminal operations remain limited. This study employs a quantitative approach, integrating existing simulation and optimization models to assess the impact of different operational strategies on carbon emissions. Key research questions refer to the primary sources of carbon emissions in container terminal operations and how can operational adjustments and technological interventions optimize carbon footprint reduction. Findings suggest that electrification of handling equipment, process automation, and energy-efficient scheduling significantly reduce emissions. The study provides a structured framework for terminal operators to assess and enhance sustainability efforts, contributing to the broader discourse on green port operations.

ABSTRACT. The electromagnetic diaphragm pump system, which contains electromagnets, permanent magnet, membrane, valves, chamber, has electromagnetic forces that act to change the position of the permanent magnet located on membrane driven by the polarization of magnets and has the pressure to pump the liquid under the control of digital cabinet. This study proposes a novel approach that uses the temperature sensor probe and an infrared thermal imaging camera to measure the temperatures of an electromagnetic diaphragm pump. The results show good agreement between the infrared and temperature sensor techniques confirming that the proposed technique could accurately estimate the temperature. The temperature varies as a function of liquid pumped in the electromagnetic pump system, and an equation model is proposed to express the formulation dependence between temperature, the switching speed of the permanent magnet and the pumped liquid.

ABSTRACT. This paper introduces an optimal control approach for a palletizing manipulator operating within a load-unload cycle. By employing the matrix representation of the Lagrange dynamics, the motion equations of the manipulator are systematically derived. An energy-based objective function, together with terminal state constraints, is incorporated into the Hamilton formulation. By applying Pontryagin’s Principle to this formulation, a set of optimality conditions is obtained in the form of a two-point boundary value problem. This problem is then addressed through numerical simulation using computer software, thereby validating the effectiveness and applicability of the proposed control strategy.

ABSTRACT. In this paper we describe the development of a terrestrial robotic system that uses specialized sensors to create a map of the radiation intensity in the environment, predominantly in buildings or underground - salt mines. The data collected by the sensors are used to calculate the radiation intensity. A control system involves moving the robot along the contours of constant intensity. The purpose of this paper is to highlight the kinematic and dynamic characteristics of the terrestrial robot on wheels. Potential beneficiaries of the research are Research groups, in particular those in radiation physics, high energies, seismology, with needs to secure the information transmitted from the governmental, industrial and commercial areas.

ABSTRACT. In the context of Industry 4.0, the adoption of Digital Twins is an essential solution for the optimization, monitoring and control of complex industrial processes. This article presents the development and implementation of a real-time Digital Twin for an automated Pick-and-Place system, using the Unity simulation environment. The physical system is composed of two electric axes and one pneumatic axis, as well as a suction cup system for handling parts, controlled by a Siemens S7-1200 PLC. Communication between the physical and digital systems is achieved through industry-standard PROFINET and IO-Link protocols, ensuring a bidirectional exchange of data in real time. The digital model faithfully reproduces the mechanical structure and dynamic behavior of the physical system, including precise synchronization of movements on all axes and the state of the handling system. The complete simulation of the Pick-and-Place cycle in Unity allows for real-time visualization and monitoring of the process, providing a safe and flexible platform for testing, training and performance analysis without risk to the physical equipment. Experimental validation indicates a high correlation between the physical and simulated system performances, with differences below 5% for critical parameters such as cycling time and positioning accuracy. This solution brings major advantages, including reduced picking time, minimized downtime, improved process quality, and increased industrial digitalization. Furthermore, the implemented Digital Twin provides a solid foundation for expansion towards smart factories, with the possibility of integrating VR/AR technologies and adaptive control in the future. The article details the modeling methodology, communication and synchronization architecture, as well as the practical benefits of real-time digital simulation for Pick-and-Place systems in the context of Industry 4.0.

ABSTRACT. Digital Twin technology offers new opportunities for optimizing industrial operations. In the context of Pick-and-Place operations, a real-time framework is proposed that integrates a Unity-based digital twin, synchronized with a Siemens S7-1200 PLC, and a Q-Learning algorithm to reduce cycle time without compromising positional accuracy. The Digital Twin continuously collects kinematic and state data (positions, velocities, accelerations, vacuum states) via a PLC web server. The Q-Learning algorithm is trained offline in the simulated environment to learn an optimal X- and Y-axis movement policy, Z-axis control, and vacuum management. The learned policy is then transferred to the live system, avoiding wear and tear on the physical equipment. Experimental results indicate an approximately 10% reduction in total cycle time and maintain positional deviations below 2%, compared to conventional control. This solution demonstrates the feasibility of real-time optimization through reinforcement learning in industrial environments and opens up prospects for adaptive, data-driven control strategies within Industry 4.0.

ABSTRACT. System simulation is already state-of-the-art in the era of digitalization. In order to keep up with the latest megatrends, this work aims to present a Simulink model of the Serial Peripheral Interface used in a wide spread of automotive applications, supporting fast design-in. The main benefit consists in a fully customizable solution that combines all modeling components: µC (Microcontroller) ↔ SPI ↔ PROFET™ (PROtected FET) into a single Simulink project. Once the set-up is implemented for the pilot project, it can be reused for many other products, being a cost-effective method. This customer’s demand appeared because Spice based simulators are analog oriented and the digital part cannot be integrated. The SPI Simulink model is completely configurable in terms of: register’s bit length, MSB/LSB first, clock phase/polarity, being compatible and easy to integrate in use cases for different types of applications.

ABSTRACT. The nonsinusoidal regime is present in almost every power network, triggering implications that negatively affect the connected equipment and the power system. Consequently, researchers allocate significant resources to treat the effects of a nonsinusoidal regime. The quality of energy directly results from the variation of the quantities which define the nonsinusoidal regime. This paper proposes a new method for the determination of the dependency between the real power (P), total harmonic distortion of voltage (THDu), total demand distortion of current (THDi) and the magnitudes of the voltage and current harmonics. Once implemented into the MATLAB software package, the calculation algorithm presented in the method addresses a real-life application. The errors are minimal while evaluating the real power P, THDu, and THDi by applying the sensitivities and comparing them with the values obtained from the classical definitions. Consequently, one regards the method proposed in this paper as correct.

ABSTRACT. In the context of Romania’s expanding nuclear energy program and global initiatives focused on Generation IV reactors, the testing of nuclear fuel handling machines (F/Ms) is critical for safe and efficient reactor operation. This paper analyzes the design requirements, standards - International Atomic Energy Agency (IAEA) SSG-63, REGDOC-2.4.5, and testing procedures for fuel handling equipment, highlighting validated solutions for Canada Deuterium Uranium (CANDU), Pressurized Water Reactor (PWR), Boiling Water Reactor (BWR), Sodium-cooled Fast Reactor (SFR), and Lead-cooled Fast Reactor (LFR) systems. The specific features of CANDU reactors—enabling online refueling under high-pressure, high-temperature conditions—are detailed along with the benefits of out-of-pile testing. Romania’s experience with the full-scope testing of F/Ms for Unit 2 at the Cernavodă Nuclear Power Plant (NPP) forms the basis for this work. Building on this experience, the original contribution is the development and experimental validation of an integrated testing and simulation platform based on advanced automation and digital twin technologies, supporting the commissioning of fuel handling machines for Units 3 and 4. This platform also serves as the foundation for the HANDS-ON facility’s control system, under development at the Institute for Nuclear Research (INR) Pitești, with direct involvement of the authors. Comparative insights on fuel handling in PWR, BWR, SFR, and LFR reactors are provided, highlighting the challenges of vertical fuel positioning. Finally, emerging research directions are discussed, aiming to enhance equipment reliability, reduce maintenance time, and enable realistic validation of future nuclear fuel handling technologies.

ABSTRACT. Nowadays, the quantitative and quality characteristics analysis subjecting a nonsinusoidal regime has become one of the most significant preoccupations for researchers involved in power systems and is triggering considerable interest. The drawbacks determined by the nonsinusoidal regime negatively affect the electrical power systems and their components. The time variation of the quantities that define the nonsinusoidal regime directly influences electrical energy quality. Following this reality, this paper proposes a new method for assessing the dependency between the relative weightings of the voltage, respectively current harmonics and the reactive and apparent power recorded in the nonsinusoidal regime. Such a method relies on using sensitivity as a tool for analysis. The present material includes the mathematical formulation of the problem through new calculation relationships, including the reactive, apparent power expressed with respect to the harmonics’ weightings, voltage and current, respectively sensitivities. The implementation of the logical flow diagram utilized the MATLAB/Simulink software package to run a case study. Following the reactive and apparent power determination using voltage and current harmonics’ weightings formulas versus their evaluation based on sensitivities, one can figure out extremely low differences between the values obtained with both methods. Considering the first method based on weightings as the reference, one can observe minimal errors, whereas using the second one utilizes sensitivities, rightfully claiming the correctness of the second method as well.

ABSTRACT. The Control Room plays a central role in the operation and supervision of the Fuel Handling System in CANDU-600 nuclear reactor, being responsible for critical decision-making during system exploitation. The current state of the workforce and the limited number of highly trained personnel in this field highlight the need to optimize and simplify operator workload by developing systems capable of taking over parts of their tasks. Romania must develop new electricity generation capacities to support the transition toward a low-carbon economy while ensuring stability, security, and availability within the National Energy System, as well as generating socio-economic benefits. The completion of Units 3 and 4 Project at the Cernavodă Nuclear Power Plant is positioned to deliver these outcomes. This paper aims to explore new solutions for improving the Human-Machine Interaction in Fuel Handling Systems by investigating the implementation of modern display, data acquisition, and control systems suitable for use in the Fuelling Machine Testing Rig located in the Institute for Nuclear Research Pitești. This work addresses the selection of relevant design requirements and standards for nuclear Human-Machine Interfaces, the identification of operational issues encountered in the Control Room, the proposal of improvement solutions, as well as the definition of objectives, while considering the necessity of preparing and optimizing the testing facility for the acceptance and pre-acceptance tests of the Fuelling Machines for Units 3 and 4.

ABSTRACT. The study aims to develop an application in MATLAB that uses fuzzy logic to analyze and interpret different features of navigation maps before they are released into production. The paper proposes an original approach to the process of creating, testing, and validating navigational maps, using fuzzy logic to analyze factors such as map size, complexity, and risk of corruption prior to release into production. Unlike existing studies, this method provides a predictive framework that allows you to anticipate problems and optimize the production process. Applied to real cases, it could reduce launch time by up to 30% and validation costs by about 20%, as it allows early identification of maps with a high risk of corruption or slow implementation, thus avoiding repeated testing and reconfiguration cycles.

ABSTRACT. This paper proposes a control system based on fuzzy logic for optimizing the operation of blowers in a wastewater treatment plant. The system uses ammonium (NH₄⁺) and nitrate (NO₃⁻) concentrations as input variables to decide whether to turn the blowers on or off in order to maintain efficient nitrification and reduce energy consumption. The implementation was carried out in MATLAB/Simulink, using a modular architecture and fuzzy Mamdani rules. The proposed method was tested in two scenarios: one with synthetic data and one with real data from the Pitesti Wastewater Treatment Plant. The results indicate a good correlation between the decisions of the fuzzy system and those of the existing SCADA system, confirming the practical applicability of the solution. The approach is flexible and can be extended to integrate additional parameters such as dissolved oxygen or temperature, thus providing intelligent support for the control of biological purification processes.

ABSTRACT. Abstract— In this study, we perform an in-depth analysis of Digital Imaging and Communications in Medicine (DICOM) security. The study begins by offering an overview of DICOM standards from a security standpoint. A comparative evaluation of selected DICOM techniques is provided, considering not only technical performance metrics but also implementation feasibility and complexity. Beyond the protocol-level analysis, we propose an experimental framework that leverages Software-Defined Networking (SDN) to optimize the transport of DICOM traffic across hospital networks. As the volume of medical imaging data continues to grow and the demand for real-time diagnostic access increases, traditional network architectures face limitations in delivering timely and reliable performance. In response, our experiment simulates a healthcare network environment in which DICOM traffic is dynamically prioritized using an SDN controller based on OpenFlow rules. Performance metrics including end-to-end latency, packet loss, and delivery time are collected and compared with and without SDN flow-based prioritization, under controlled background congestion. The results demonstrate how SDN-driven traffic engineering can significantly enhance Quality of Service (QoS) for critical healthcare applications. By combining DICOM security considerations with programmable network optimization, this work contributes a holistic approach to strengthening both the protection and performance of imaging workflows in clinical environments.

ABSTRACT. This article studies the problem of SDN controller placement in the network, as well as the deployment of multiple controllers in large-scale networks. Given the SDN technology’s principle of centralized organization, this issue is highly relevant, which justifies the importance of this study. On the other hand, optimizing the placement of SDN controllers within the network is a complex multi-criteria problem; therefore, numerous studies and/or experiments are focused on this research subdomain, which remains open—especially under dynamic conditions. We propose an SDN architecture that provides the functional support required for optimizations within the SDN control plane. In particular, the goal is to achieve optimal placement of the SDN controller(s) in the network (Controller Placement Problem – CPP) as well as dynamic adaptation of control in response to traffic and/or topology changes. The functional blocks of the system implementing the architecture are defined and configured, followed by the development of optimization experiments in the control plane.

ABSTRACT. This paper presents a comparative study of two popular JavaScript-based technology stacks, the MERN Stack and MEAN Stack, used in web application development. The comparative study highlights the significant similarities and differences between the two stacks, considering factors such as performance, scalability, security, community support, learning curve, and suitability. Additionally, this paper presents the MERN and MEAN stack architectures, providing information on the functionality and integration of each component.

ABSTRACT. This study allows for an analysis of the factors affecting the functioning of the RORIS (Romanian River Information Services) system, the Danube Ship Traffic Management System, and the Inland Waterway Transport Information System. The identification of vulnerabilities is analyzed from the perspective of the impact of environmental conditions on equipment and from the standpoint of the performance of communication networks. The study is conducted based on data provided by the Lower Danube River Administration, and it allows for the offering of solutions to maintain the resilience of the Roris system. By analyzing the specifics of the equipment, the operating conditions in critical situations, and identifying alternatives for field data communication, the study identifies the critical points of these components and proposes strategies to improve the system’s resilience. The results impact the assurance of data reliability in dynamic conditions and the prevention of failures that could compromise both navigation security and environmental conservation efforts.

ABSTRACT. This work is aimed at presenting a solution for automatic collection of travelers’ data, intended to measure passenger flowing and/or density in specific indoor environments. The proposed solution integrates cost-effective sensing technologies, data processing algorithms, and real-time analytics to efficiently track and quantify the movement patterns of travelers in confined spaces such as subway stations, airports, or other transit hubs. The test bed consisted of strategically positioned sensors and a centralized processing unit deployed in a specific indoor facility. Key performance indicators (KPIs), such as number of detected devices, and RSSI (Received Signal Strength Indicator) were evaluated under varying environmental conditions. Results demonstrated that the proposed solution achieved satisfactory accuracy in detecting and counting individuals, with minimal latency and consistent performance across different scenarios. Conclusions drawn from the field tests indicate that the proposed solution offers significant advantages in automating traveler flow measurement, providing valuable insights for facility management and decision-making processes. Future work will focus on optimizing the system's scalability and integration with broader smart building frameworks.

ABSTRACT. This paper presents 5G antenna design with a approach machine learning-based to predict the S11 parameter . The input data, generated using ANSYS HFSS, includes the final geometric parameters such as patch antenna dimensions and slots. Three regression models used, Fourier Series Regression (FSR), Sum Sine Regression (SSR), and Gaussian Process Regression (GPR), were used to predict the S11 parameter. for the fisrt model FSR with Coefficient of Determination (R2) of 0.9444, Root Mean Squared Error (RMSE) egual 0.9801, Sum of Squared Errors (SSE) of 108.5365. than the second model SSR, R2 of 0.9131, RMSE of 1.2593, SSE of 169.6838. the final model GPR is the best model for predicting S11 with R2 of 0.9886, RMSE of 0.4374, and SSE of 22.1961 . FSR and SSR also showed excellent performance, qualifying them for regular data analysis. The study demonstrates how machine learning can accurately predict antenna performance and provides a reliable alternative to traditional simulation techniques.

ABSTRACT. Printed Circuit Boards (PCBs) are essential components in modern electronic assemblies, where even minor defects can lead to significant performance degradation or total system failure. This paper presents a novel, automated framework for defect detection and classification in PCBs, based on a reference comparison approach integrated with deep learning techniques. The methodology involves aligning each test PCB image with a defect-free reference, performing image pre-processing to highlight discrepancies, and applying a YOLO-based convolutional neural network for defect localization and categorization. The model is trained to identify various defect types, such as missing holes, spurs, open circuits, and short circuits, by learning from annotated visual patterns. Preprocessing steps include grayscale transformation, morphological operations, image differencing, and bounding box annotation for supervised training. Experimental results demonstrate the system's high detection accuracy and robustness across multiple defect classes, confirming its potential for real-time application in industrial quality control processes.

ABSTRACT. The automatic classification of "fire" and "fireless" sounds plays a crucial role in the development of intelligent fire warning and detection systems. This paper explores advanced methods of audio signal processing, with a focus on extracting relevant acoustic characteristics (Mel-Frequency Cepstral Coefficients (MFCC), Spectral Centroid, Zero Crossing Rate (ZCR)), used to train automatic classification models. Challenges such as ambient noise, spectral variability and feature redundancy are analyzed. The study demonstrates the potential of acoustic processing technologies in improving early fire detection systems in real-world environments.

ABSTRACT. In recent years, the automatic detection of fire events using acoustic analysis has gained momentum as a viable alternative or complement to traditional sensor-based systems. This paper proposes a modern approach for the classification of fire and non-fire audio smears using principal component analysis (PCA) for dimensionality reduction and supervised machine learning classifiers. Feature vectors are extracted from real-world audio datasets using Mel-frequency cepstral coefficients (MFCCs), Chroma, Spectral Centroid, and Zero-crossing rate (ZCR), they are then compressed via PCA to preserve the most relevant features. These findings highlight the potential of PCA-enhanced audio classification models in real-time sensing systems, including Internet of Things (IoT) and smart city infrastructures. The study concludes with recommendations for the integration of autoencoders, Uniform Manifold Approximation and Projection (UMAP) and audio-visual learning architectures in future research.

ABSTRACT. This paper deals with reversible contrast enhancement for images with a limited range of graylevels. Image enhancement is directly approached by the classical contrast stretching that linearly expands image histogram over the entire graylavel scale. The original image is recovered by embedding in the enhanced version the data necessary for contrast stretching reversal. The sparsity of the enhanced image histogram ensures reversible embedding at the cost of least significant bit substitution. Experimental results for classical low contrast test images are provided.

ABSTRACT. Single Pixel Cameras (SPC) deploy the Compressive Sensing (CS) theory to sequentially acquire data directly in a compressed format (measurements). This paper proposes a method to perform reversible data hiding (RDH) with partial encryption in CS measurements by encrypting part of them using a secret key and then inserting the resulting values on-the-fly in the following eligible measurements. The insertion results in visible distortion, the images reconstructed based on the marked measurements still having discernible content. Without performing a decryption an unauthorized user can only obtain a low quality version of the original image. The simulation results show that the capacity and distortion are directly linked, while the number of insertion levels also determines the image distortion. The proposed modifications to the prediction error expansion algorithm adapt it for usage in a SPC scenario while limiting the data expansion.

ABSTRACT. Harnessing wind power for hydrogen production enables a carbon-free energy conversion pathway, offering a key solution for advancing China’s low-carbon energy transition. This study presents an optimal dispatch strategy for a wind-hydrogen joint system (WHJS) under a stepped carbon trading constraints. Initially, the WHJS model is constructed, incorporating real-time electricity price fluctuations derived from load and price forecasting. A stepped carbon trading framework is then introduced to guide emission limitations within the market-based regulatory environment. Based on this model, an optimization strategy is developed to minimize total system operating costs. The proposed strategy is constructed as a linear programming and solved in Python using CBC. Simulation results demonstrate that, in comparison to the conventional carbon penalty mechanism, the proposed approach reduces system-wide carbon emissions by 1456.98 kg-equivalent to a 27.5% decreasing- highlighting its effectiveness in promoting both economic and environmental performance.

ABSTRACT. This paper presents an optimal dispatch model for a Virtual Power Plant (VPP) that integrates carbon trading and green certificate trading mechanisms. The primary objective function is to maximize the net profit of the VPP by reducing operational costs and enhancing revenue from carbon and green certificate trading. The dispatch model incorporates a diverse mix of energy sources, including gas turbines, solar power, wind farms, and storage systems, to achieve a balanced and sustainable operation. To address the complexities of the optimization, the Coronavirus Search Algorithm (CVSA) is applied for its efficacy in handling renewable energy uncertainties. The optimization is conducted under two scenarios: a deterministic and a stochastic scenario, enabling the model to account for uncertainties in renewable energy production and market prices. Simulation results indicate that both deterministic and stochastic optimizations enhance profitability, reduce operational costs, mitigate penalty risks from renewable shortfalls, and support a low-carbon approach, making the VPP framework viable and adaptable in real-world applications.

ABSTRACT. Solid-State transformers (SSTs) are emerging as next-generation substitutes for the traditional Line-Frequency transformers due to the rapid expansion of power electronics, which has evolved the distribution network operation. Among various SST configurations, the three-stage topology is characterized by a considerable number of semiconductor switches and bulky passive components such as capacitors. These elements give rise to complex nonlinear dynamics, posing significant challenges for system modeling and control. Therefore, achieving robust and optimal system performance requires implementing an efficient control mechanism. In this paper, a constant-frequency dual-loop control method is applied to a power factor correction (PFC) rectifier in the SST rectification stage. The coefficients of both controllers are optimized using the great Wall construction algorithm, with the ISTSE index as the objective function, to ensure voltage and current quality during load or reference variations and to maintain clean power delivery for the DC-DC stage. The proposed control strategy is evaluated in comparison with the conventional single-loop control method under various conditions, and the results indicats a significant improvement in transient performance while maintaining acceptable output voltage ripple.

ABSTRACT. Spoken language recognition (SLR) is a pivotal challenge in speech processing, serving a variety of practical applications such as cross-lingual communication platforms, speech-based authentication systems, and real-time transcription that adapts to multiple languages. This study evaluates the effectiveness of self-attention-driven transformer models in automatically identifying spoken languages, with a particular emphasis on five distinct languages: German, Turkish, French, Spanish, and English. To build a diverse and representative dataset, speech samples are systematically gathered from YouTube using API integration. This approach ensures a broad range of speakers, accents, and environmental conditions, enriching the model training process. The collected data undergo essential preprocessing steps, including noise reduction and normalization, to improve audio quality and standardize input. These refined datasets are used to train and assess the performance of several advanced transformer-based models, including HuBERT, Wav2Vec2, and WavLM, along with their specific variants. The experimental results reveal that HuBERT leads with an accuracy of 99.30%, achieving near-perfect results. These outcomes emphasize the efficacy of transformer-based architectures in distinguishing between linguistically diverse languages. Furthermore, the findings point to the substantial potential of these models in real-world multilingual applications, where precise and effective spoken language recognition is essential for seamless interaction with automated systems.

ABSTRACT. Blockchain innovation, best embodied by Ethereum, has revolutionized online transactions by making them more transparent and secure. However, the demand for more sophisticated fraudulent schemes increases with wider adoption, calling for more sophisticated fraud detection methods. Therefore, this paper contributes to the area of blockchain security by providing insights to regulators and stakeholders in Ethereum through an analysis of the Machine Learning (ML) models. We compare traditional approaches like logistic regression and decision trees with more advanced techniques like neural networks and ensemble methods. The performance of the model is measured using accuracy, precision, recall, and the ROC curve. The best accuracy of 0.98 is achieved by the optimized XGBoost framework.

ABSTRACT. Efficient path planning is essential for Unmanned Aerial Vehicles (UAVs) to navigate complex environments while avoiding obstacles. This paper presents a novel global path planning algorithm that integrates a simplified visibility graph (VG) method with the Dijkstra algorithm to enhance computational efficiency and adaptability. Unlike traditional polygon-based approaches, the proposed method models obstacles as rectangles with safety buffer, reducing graph node complexity. The algorithm optimizes the graph by considering only obstacles intersected by the M-Line connecting start and goal points, minimizing unnecessary computations. A smart obstacle merging strategy ensures path existence by addressing narrow gaps between adjacent obstacles and accounting for UAV size and sensor noise-induced position errors. Simulation results show that the proposed method outperforms the traditional VG approaches as well as another popular global path planning algorithm, the Rapidly exploring Random Tree Connect (RRT-Connect), achieving shorter paths and faster computation time. These advantages make the proposed algorithm well-suited for real-time UAV navigation in obstacle-rich environments.

ABSTRACT. The effective treatment of brain tumors can be accelerated by the early detection of the disease. Machine learning techniques can be used to provide an accurate diagnoses with minimal prediction error. Instead of using traditional techniques like hand-crafted features, in recent years, deep learning models proved to be successful in generating reliable classification results offering automated solutions thereby revolutionizing medical image analysis. In this paper, we propose a framework of combining multiple deep features extracted from two efficient Convolutional Neural Networks (CNNs): EfficientNet and MobileNet and combined to form a robust neural network. This network achieved high classification accuracy by utilizing the efficient features extracted from the two networks. To evaluate the new network, we applied the framework on a brain tumor dataset with two classes. Augmentation techniques were used to increase the number of images per class during training. The accuracy of the new model for detecting brain tumor cases is 98.04\%. This accuracy is better than using features extracted from a single network.

ABSTRACT. The face recognition systems are the most widely deployed biometric infrastructure for secured human authentication. However, these systems are vulnerable to a variety of spoof attacks, where assaulters utilize artificially created fake replicas of the human face. To mitigate these attacks, a variety of face anti-spoofing mechanisms are used, where generalization and accuracy of these algorithms are crucial performance protocols. In this research work, we expound an efficient and accurate face anti-spoofing model (i.e. HyFaNet) that integrates feature maps of a pre-trained model with Vision Transformer (ViT). The proposed face anti-spoofing model exploits the potency of the pre-trained model to generate face feature maps and global-level singularities are explored via ViT to yield a robust model. The HyFaNet is trained and evaluated on a benchmark face anti-spoofing dataset and it demonstrates a remarkable performance under unseen scenario with an EER of 0.83%. Moreover, the model exhibits a comparable performance with state-of-the-art (SOTA) face liveness detection methods.

ABSTRACT. Traditional vermicomposting assessment methods rely on labor-intensive manual sampling and subjective visual inspections, which introduce delays in process optimization, compromise compost quality, and hinder scalability in agricultural waste management. Current techniques lack non-invasive tools for real-time monitoring of subsurface parameters such as moisture content and pH levels, leading to inefficient resource utilization and suboptimal nutrient retention. To address these limitations, this study introduces a novel integration of impulse radar technology and the VGG-16 convolutional neural network (CNN) for the automated classification of three vermicomposting phases, Pre-decomposition, Curing, and Maturation. The methodology employs impulse radar (1.3–4.4 GHz) to generate wide bandwidth signals and convert the reflected signals into high-resolution grayscale radargrams, which are analyzed via a pre-trained VGG-16 model. Experimental results show that CNN with pre-trained model VGG-16 demonstrates 90.16% overall classification accuracy, with phase-specific accuracies of 85.34% (pre-decomposition), 89.10% (Curing), and 96.28% (Maturation), outperforming ResNet50 (78.02%) and in actual testing’s 8/10 or 80% were correctly predicted by the system. By eliminating physical sampling, the non-invasive design preserves compost bed integrity. It supports real-time decision-making while enhancing humus consistency and nutrient retention. Combining impulse radar with deep learning establishes a new benchmark for precision in organic waste management, demonstrating significant potential to advance sustainable agriculture and circular economy initiatives.

ABSTRACT. This paper investigates the use of a stacking en semble model for predicting hospital readmissions in diabetic patients, focusing on the urgent issue of hospital readmission rates, which are an important indicator of healthcare quality and cost control. We analyze the effectiveness of different machine learning algorithms and their integration using a dataset from the UCI Irvine Machine Learning Repository. The dataset includes information from 130 US hospitals between 1999 and 2008. The evaluated machine learning models consist of Decision Tree, Random Forest, Gradient Boosting, XGBoost, LightGBM, Logistic Regression, and MLP Classifier(Neural net work). Every model passed through hyperparameter tuning using GridSearchCV for optimal performance, specifically targeting metrics such as accuracy, precision, recall, and F1 score across training, validation, and testing datasets. The stacking ensemble technique, which combines predictions from multiple models, has been modified using Gradient Boosting as the meta-learner to improve predictions even more. The results of our study highlight the small improvements in performance that achieved through hyperparameter tuning and ensemble learning. The stacked ensemble model achieved an accuracy of 89%, higher than other models such as the decision tree with an accuracy of 86% and the neural network with an accuracy of 84%. . These f indings offer important perspectives into how these techniques might be strategically used in healthcare settings. The findings contribute to the continuing discourse on using powerful machine learning methods to improve diagnostic accuracy and the delivery of healthcare for chronic diseases such as diabetes.

ABSTRACT. The increasing deployment of nonlinear loads (NLLs) such as electric vehicle (EV) chargers and renewable energy inverters has raised serious power quality concerns in low-voltage microgrids (MGs), particularly due to harmonic distortions. To address these challenges, this paper proposes a coordinated control strategy that integrates droop-controlled inverters with a decentralized active power filter (APF). The inverters employ a virtual voltage reference generation technique in the αβ-frame, avoiding the need for direct voltage measurements and enabling autonomous operation. The APF is responsible for compensating harmonic currents and reactive power, improving the power quality at the point of common coupling (PCC). A comprehensive stability analysis of the droop control scheme is presented, and damping conditions are derived to ensure robust operation. The proposed system is evaluated under various conditions including islanded and grid-connected modes, using PSCAD/EMTDC simulations. Results demonstrate that the integrated control approach ensures seamless transition between modes, reduces total harmonic distortion (THD), and maintains voltage and frequency stability even under highly nonlinear EV charging conditions.

ABSTRACT. This paper presents a novel integral continuous sliding mode control (ICSMC) strategy designed to enhance the dynamic performance of speed servo systems in permanent magnet synchronous motors (PMSM) while effectively mitigating external disturbances. The proposed controller incorporates an integral sliding surface structure that successfully eliminates steady-state error and maintains robust system stability under varying operating conditions. To address the fundamental trade-off between rapid convergence and chattering suppression inherent in conventional sliding mode controllers, we develop an innovative reaching law featuring a high-slope saturation function that seamlessly replaces the traditional discontinuous sign function. Rigorous stability analysis based on Lyapunov theory demonstrates that the closed-loop system maintains global stability, while mathematical proof confirms finite-time convergence of tracking error to zero. The superiority of the proposed ICSMC methodology over conventional control techniques is validated through comprehensive numerical simulations implemented in MATLAB environment, demonstrating significant improvements in terms of response time, steady-state error elimination, and robustness against external disturbances.

ABSTRACT. This study focuses on the Inverse kinematic modeling of a differential drive robot. A common design in mobile robotics with (CoppeliaSim) is used for modeling the robot's physical structure. The model controls MATLAB Simulink to run simulations. To tackle the challenges of establishing inverse kinematics, a neural network-based method is provided. The neural network model is created and tested with simulated data to forecast wheel velocities for various robot trajectories. Its performance is then compared to that of the theoretical inverse kinematic model in terms of accuracy, robustness, and computing time. The findings indicate that the neural network achieves equivalent accuracy while being more responsive to variations in robot parameters and ambient variables. Our paper deals essentially with the need to combine certain approaches centered on artificial intelligence with classical tools.

ABSTRACT. The development of variable-speed wind energynnconversion systems is essential for the next power system generation. This new resarch article presents and compares two control techniques for a wind turbine: Backstepping (BS) control and Fuzzy Logic Control (FLC), an artificial intelligence-based method. The goal is to maximize power extraction under variable wind conditions. The system is based on a Doubly Fed Induction Generator (DFIG). Both control strategies are first designed and tested in simulation using MATLAB/Simulink, and then implemented in real time using a Processor-in-the-Loop (PIL) test on the dSPACE 1104 board. Tests are carried out under a real wind speed profile. The results show that the BS controller offers a response time of approximately 0.023 seconds, while the FLC stabilizes faster at around 0.012 seconds, confirming a 48% improvement in dynamic behavior. Additionally, the Tip Speed Ratio (TSR) is better maintained around its optimal value of 8.17 ± 0.01 with FLC, compared to 8.14 ± 0.02 for BS, resulting in more efficient aerodynamic performance. Both controllers demonstrate good performance in tracking the angular speed of the wind turbine. Furthermore, their real-time implementation on the dSPACE 1104 platform confirms the practical feasibility of integrating both control strategies into real wind energy conversion system prototypes.

ABSTRACT. There is an increasing recognition of microgrids as fundamental to the modern configuration of energy infrastructures because of their localized generation, which increases the resilience of energy distribution and seamlessly integrates renewable energy sources. The article reviews new approaches toward increased microgrid reliability and resilience and takes a close look at the transformative contribution of Energy Management Systems in that regard. Thus, EMS intelligently copes with energy distribution, storage, and backup systems toward great reductions of blackout duration and minimization of energy imbalances, hence a notable improvement in the performance of microgrids. Taking a case study from a Hypothetical Microgrid of 100 Homes, the implementation of robust EMS increased the resilience score by almost 15% from 5.5/10 to 7/10. It also covers the use of comprehensive monitoring, predictive maintenance, and data analytics in those upgrades. The future of microgrid reliability looks even more promising with the integration of newer technologies, such as artificial intelligence a new era for energy independence and sustainability.

ABSTRACT. This paper presents a comparative review of four main algorithms used for Maximum Power Point Tracking (MPPT) in wind energy conversion systems namely Perturb and Observe (P&O), Tip Speed Ratio (TSR), Fuzzy Logic (FLC) and Artificial Neural Networks (ANN). These techniques were compared in terms of efficiency, simplicity, adaptability and cost. Each method has its advantages and its limitations, based on wind variations. And a particular attention is given to the offshore wind systems, where the conditions are more challenging because of the structure movements, the strong wind and the maintenance difficulties. The research shows that these environments require more intelligent and more adaptive control solutions, and it concludes that the hybrid approaches that combine conventional and intelligent methods may offer better performance in the offshore environments.

ABSTRACT. With the global push toward maximizing renewable energy efficiency, effectively extracting wind power remains a challenge. This study presents the design and implementation of two modern control techniques—Sliding Mode Control (SMC) and Artificial Neural Network (ANN)—aimed at maximizing turbine output through Maximum Power Point Tracking (MPPT) methods in Wind Energy Conversion Systems (WECS). Both control strategies were developed using MATLAB/Simulink under dynamic wind profiles, with a focus on improving response time, minimizing steady-state error, and increasing the power coefficient (Cp). Simulation results show that the ANN-based controller achieves a faster response time of 1.6 milliseconds and a minimal steady-state error of 0.3069 rad/s, outperforming the SMC controller in terms of speed and power extraction. The ANN controller also achieved a higher power coefficient of 0.4799, indicating more energy capture. While the sliding mode MPPT controller maintained a steady-state error of approximately 0.0749 rad/s, demonstrating strong adaptability and consistent performance. Finally, both control strategies were implemented on the dSPACE 1104 to validate them on real-time.

ABSTRACT. Transition to sustainable mobility could include advanced and reliable solutions for storing energy to cover the fundamental issues related to electric and hybrid vehicles. The problem in such vehicles remains limited energy issues of capacity, power delivery, and battery degradation. Battery energy storage systems (BESS) remain the specific but expensive component in such vehicles. Also, extreme temperatures and high dynamic loads may disturb the battery's chemistry, resulting in irreparable damage, shorter life-cycle, and high overall cost. This paper introduces dual-level HESS with Lithium batteries, supercapacitors, and photovoltaic sources. Fully active design is considered to improve energy management and overall performance. These early gains in efficiency, through to the present, have shown the positive benefits of this source with smoother power delivery, extended battery life, and better efficiency. The integration of microgrid (MG) technologies further enables coordinated control of distributed energy and storage units, ensuring real-time power balancing and scalability. Further analysis is conducted on multi-source HESS setups within new technologies, like fuel cells and flywheels, to see how they measure up and lock in even more gains.

ABSTRACT. The integration of renewable energy into smart grids requires intelligent battery storage systems that can adapt to fluctuating loads and intermittent generation. This paper presents a machine learning-based approach for predicting battery’s state-of-charge (SoC) and optimizing storage capacity in solar-powered smart grids. A random forest regressor (RFR) is trained on time-series data, including energy load, time, and previous load features, to forecast SoC with higher accuracy. The model achieved a mean absolute error (MAE) of 8.31% and a mean squared error (MSE) of 106.40, demonstrating its effectiveness in SoC estimation. These predictions are used to determine optimal battery capacities for different seasonal demands using a storage capacity of 233 kWh is sufficient for summer, while 200 kWh meets requirements. This dynamic sizing approach helps avoid battery oversizing and under utilization, improving reliability and suggests potential cost-efficiency under idealized conditions. The proposed system supports real-time adaptability and outperforms traditional rule-based energy management strategies. Overall, this study highlights the potential of integrating machine learning with smart grid infrastructure to enhance operational efficiency and enable more sustainable and intelligent energy storage solutions.

ABSTRACT. This research article studied and simulated a packed bed thermal energy storage system to elucidate its thermodynamic behavior. A transient mathematical model for turbulent flow in a hybrid media, comprising both porous and transparent components, incorporating forced and natural convection, has been formulated. Comsol Multiphysics CFD software was employed for the numerical solution. The Local Thermal Non-Equilibrium (LTNE) methodology was utilized to assess heat transfer in both solid and fluid phases and to ascertain the thermal exchange coefficient between them, which is a critical characteristic for evaluating stored energy. Simulations were conducted on an axisymmetric cylindrical tank filled with pebbles (solid material) and circulated by a thermal oil, serving as a heat transfer fluid (HTF). The presented model was validated and approved using experimental data. The velocity distribution and temperature profiles of the solid phase and fluid phase in the system during the charging process were ascertained and shown. The influence of porosity and particle size on the thermal performance of the TES system was assessed and reported. The findings indicated that the duration needed to charge the tank diminishes with an increase in both porosity and particle size

ABSTRACT. To solve the fluctuating nature of wind energy systems, this paper aims to design, analyze, and simulate a wind energy conversion system based on a Doubly Fed Induction Generator (DFIG) connected to a wind turbine. A wide range of wind speed fluctuations is possible with a Doubly Fed Induction Generator (DFIG) due to its ability to handle a wide range of rotation speed variation. Direct Field Oriented Control (DFOC) has been frequently used to decouple the active and reactive power control. Enhancing performance can be attained directly using controllers such as Proportional Integral (PI). However, this kind of controller heavily depends on parameter variations and frequently has difficulty dealing with nonlinear dynamics. Sliding Mode Control (SMC) provides robust performance. This work presents a comparative analysis between Proportional Integral (PI) control and Sliding Mode Control (SMC) in (DFOC) of a DFIG; the performances of both controllers are examined in terms of rapidity, precision, robustness, and power quality under parameter variations and strong disturbances. The findings confirmed that integration of SMC in DFOC improves robustness and power tracking even under uncertain conditions, whereas PI control allows good performance only under nominal conditions.

ABSTRACT. Grid-connected inverters are evaluated under operational challenges such as grid disturbances, nonlinearities, and parameter variations. This paper investigates the Internal Model Control (IMC) principle for inverter control in renewable energy integration, microgrids, and smart grids. A literature review that is followed by a detailed analysis on mathematical base of IMC examines existing control strategies and their limitations. The study highlights the robustness of IMC in managing system uncertainties and disturbances by highlighting its advantages over conventional control methods. Mitigation strategies including advanced filtering and adaptive control are discussed. To validate the proposed IMC-based control strategy, simulations assess its performance under different grid conditions. Results confirm the ability of IMC to enhance stability, reduce harmonic distortion, and improve dynamic response. The study also explores future advancements, focusing on weak grid environments and optimized filter topologies to improve power quality. Outcomes contribute to developing more resilient and adaptive inverter control strategies for modern energy systems that reinforce the role of IMC in ensuring stable and efficient grid integration.

ABSTRACT. This paper proposes a Fixed Switching Frequency Direct Model Predictive Control (FSF-MPC) strategy for a two-level grid-tied inverter with an LCL filter. The control method is tested with 100 kVA grid-tied inverter that is implemented to manage active and reactive power rates individually to ensure precise power regulation. The proposed FSF-MPC method maintains a fixed switching frequency that enables the inverter to generate a discrete grid current harmonic spectrum for effectively reducing total harmonic distortion (THD) and improving the power quality. Various simulation studies have been conducted under different grid conditions to assess the performance of the proposed control strategy. The results demonstrate the robustness and efficiency of the FSF-MPC approach in handling dynamic grid disturbances including frequency and phase shifts, voltage sags, swells, and sudden load variations. The ability of controller to maintain stable operation, fast transient response, and low harmonic distortion highlights its suitability for medium-power grid-tied applications by ensuring compliance with grid codes and enhancing overall system efficiency.

ABSTRACT. The paper presents a PC-based system for control and monitoring of a PV solar system. An architecture based on the AI principle in the control loop is proposed. The developed AI system is designed to improve the use of control efficiency by adapting the parameters, as well as providing new optimization opportunities. These functionalities are included in the developed software, and a detailed description is provided. The system is tested by implementing communication between a RAD application and a simulation model in a MATLAB/Simulink environment. The simulation model includes an expanded set of subsystems in order to study the dependencies and influence of each of them. The simulation results show that the combination of a GPT model for dynamic determination of the controller parameters leads to better stabilization of the PV power and the overall system performance under dynamic changes in solar radiation, achieving an efficiency of 98.12%. The development and research confirm the potential of reproducible AI for intelligent management of renewable energy sources, offering guidelines for future improvements in industrial automation and energy systems, by upgrading to self-aware intelligent control systems.

ABSTRACT. In this article, approaches for implementing Artificial Intelligence (AI) functionalities into Rapid Application Development (RAD) environments are explored. The study emphasizes the advantages of building PC-based control systems through RAD applications, considering both speed of development and flexibility. Various strategies for integrating Generative Pretrained Transformers (GPT) models into PC-based industrial control systems are reviewed and proposed. Specific programming implementations in high-performance computing (HPC) languages such as C++ and Python are presented, illustrating practical methods for enabling AI-driven functionalities within industrial environments. A comparative analysis is conducted based on key criteria, including implementation complexity and the execution time for sending and receiving responses from GPT models. Validation is performed through tasks related to technological process control, focusing on PID controller tuning and system identification using example processes. The paper discusses the challenges and benefits of different integration methods and highlights practical guidelines for choosing the most effective approach in real-world applications.

ABSTRACT. Effective inventory tracking plays a crucial role in facility management by optimizing resource allocation, reducing operational costs, and minimizing manual effort. This study conducts a comparative analysis of multiple YOLO-based deep learning models—YOLOv8, YOLOv9, YOLOv11, and YOLOv12—for office inventory detection and quantification. The primary objective is to assess the performance of these models in accurately identifying and counting office supplies from image data. Since standard YOLO architectures do not include predefined categories for office inventory, a custom dataset is developed. A total of 10,000 images of common office inventory items are gathered through web crawling, and 2,000 images are manually annotated using Roboflow to facilitate model training. The dataset comprises 14 distinct office inventory classes, ensuring a broad representation of essential items. Each YOLO version is evaluated based on detection accuracy, processing speed, and computational efficiency. Comparative experimental results reveal the strengths and limitations of each model, highlighting the trade-offs between precision and inference time. The findings provide valuable insights into the most suitable YOLO architecture for real-world facility management applications, contributing to the advancement of automated inventory tracking systems.

ABSTRACT. In recent years, Large Language Models (LLMs) have excelled in different domains to an expert’s level by training them on downstream tasks. However, re-training them on tasks where the information is constantly being updated can be computationally expensive and incurs additional costs and time. To address this problem, end-task knowledge is added through common techniques, such as Retrieval Augmented Generation (RAG) and Fine-tuning with Parameter Efficient Fine Tuning (PEFT). This paper critically examines the existing solutions and methodologies to address such limitations. In particular, challenges including hallucination, computational overhead, and underperformance on critical tasks can lead to unreliable results, affecting both user experience and trust in AI-driven applications. As solution, we propose a two staged framework, utilizing Facebook AI Similarity Search (FAISS) and Hierarchical Navigable Small World (HNSW) to optimize RAG system and fine-tuning with LoRA Hyperparameter Tuning under limited computational setting. An evaluation was conducted on four different GPU units, demonstrating outstanding performance across various evaluations, including accuracy, memory usage, and output diversity.

ABSTRACT. High-capacity wireless communication needs driven by the emergence of 5G technology have triggered advancements in sophisticated antenna array designs. Multi-beam antenna arrays function as core components in these systems due to their capabilities of precise beam steering and their contributions to enhanced spectral efficiency and better spatial coverage. This study introduces a new Fourier-based approach for designing multi-beam antenna arrays. Unlike traditional methods, our technique gives precise control over beam shaping while significantly lowering unwanted side lobes. A more optimized radiation pattern that boosts performance for 5G and beyond. Through theoretical framework development and numerical simulations supported by experimental validation we demonstrate the method's potential to enhance antenna array optimization. The experimental and numerical findings validate that the introduced method based on Fourier synthesis considerably boosts beamforming performance through enhanced directivity and side lobe suppression. The demonstrated results reveal the method's potential to enhance multi-beam antenna arrays in 5G networks by enabling efficient spectrum utilization and interference control.

ABSTRACT. This paper explores use of the Least Mean Squares (LMS) algorithm, based on stochastic gradient descent, for implementation in adaptive beamforming systems in 5G millimeter-wave (mmWave) multiple-input multiple-output (MIMO) systems. An 8-element array designed for use at 28\,GHz is integrated with the LMS algorithm to give the radiation beam dynamic steering capability in multiple directions during operation. The mathematics of the LMS algorithm is demonstrated before applying this to lead to optimization and full-wave electromagnetic validation of the LMS applied to a patch antenna array prototype built with a RO4003C substrate. Simulated results show its success to achieve solid beam steering capability with increased directivity and sidelobe suppression over four different steering angles of 0°, ±30°, and ±60°. The results presented in this paper demonstrate the algorithm’s potential for near-real-time spatial filtering and spectral efficiency enhancement for massive MIMO implementations and exhibit robust performance, fulfilling the requirements for next generation communication systems in a dynamic and high-performance way.

ABSTRACT. This paper presents a novel deep learning-based approach for anticipating workplace accident risks through artificial intelligence-driven stress monitoring. Our method focuses on the analysis of physiological signals, specifically electrocardiogram (ECG) data, using the publicly available Wearable Stress and Affect Detection (WESAD) dataset. We introduce a comprehensive framework that includes feature extraction from ECG segments and utilizes the combined strengths of one-dimensional convolutional neural networks (1D-CNNs) and recurrent neural networks (RNNs), particularly bidirectional long short-term memory (BiLSTM) and bidirectional gated recurrent unit (BiGRU) architectures, to capture temporal patterns in the data. Two hybrid models are proposed, both incorporating attention mechanisms that dynamically focus on the most informative parts of the input sequence. We further investigate adversarial robustness through perturbation experiments to assess model reliability under challenging conditions. Experimental results demonstrate strong performance and robustness, with the CNN-BiLSTM model with attention achieving superior results. This work contributes to the development of more effective and resilient stress monitoring systems for enhancing occupational safety.

ABSTRACT. This paper studies the introduction of fault tolerance into Flexible Manufacturing System (FMS) design. The focus is on workcell controller failures. The proposed solution does not require any additional hardware. Riverbed simulations are used to prove that the proposed solution satisfies the required timing constraints. To keep the packet delay within acceptable limits, it is necessary to operate some workcells at a reduced speed. To quantify the effect of speed reduction (in case of a controller failure) on the production rate, a performability model is developed which takes into account controller lifetime and the speed of operation of the different workcells in the FMS over time.

ABSTRACT. This research aims to optimize object detection capabilities on the FPGA Arria 10 SoC platform using the YOLO-tiny model. The system integrates hardware and software to optimize data flow with OpenCL. The study uses half-precision floating-point data to balance performance and hardware resource savings, significantly reducing resource requirements without compromising accuracy. Additionally, the system incorporates a data flow pipeline approach. Experiments on a standard dataset demonstrate that the system achieves an average processing speed of 220ms per frame with performance improvements ranging from 15% to 20% when we implemented on-board FPGA Arria 10. These results validate the system's efficiency and highlight its potential applications in AI systems designed for edge devices, which demand high performance while conserving energy and meeting the requirements of real-time applications.

ABSTRACT. The fast spread of Internet of Things (IoT) devices has created major difficulties to uphold data security and maintain its reliability and integrity for interconnected systems. Traditional centralized systems cannot protect electronic evidence sufficiently in zero-trust environments because they lead to evidence that becomes vulnerable to unauthorized tampering. Blockchain technology solves these issues effectively through decentralized data processing and unalterable databases that remain easily viewable to all users. The research investigates blockchain smart contracts as a solution to improve IoT electronic evidence reliability by implementing automated access verification and data integrity assessment and event activation. Automated smart contract technology establishes transparent data security through policy enforcement which happens without third-party organizations. The research demonstrates how blockchain technology and smart contracts assist different industries like healthcare facilities and supply chains and industrial Internet devices and smart cities to operate. This research describes the principal obstacles within the field like scalability problems together with resource constraints and legal complications while offering recommendations about possible future investigations. Through blockchain technology alongside smart contracts this research develops a framework which enhances the reliability and security of IoT electronic evidence while benefiting IoT ecosystem reliability and safety.

ABSTRACT. This paper presents a novel application of Gener ative Adversarial Networks in the culturally significant domain of Indian ethnic wear fashion. Leveraging the Generative Adver sarial Contrastive Language–Image Pretraining approach, Our Proposed system is capable of synthesizing high-quality Indian Men’s traditional attire such as kurtas, sherwanis, Indo-Western outfits, and Nehru jackets , based on text descriptions.The model is trained on a comprehensive dataset of ethnic wear, encompassing diverse categories and intricate details such as fabric type, color,embroidery work, and recommended occa sions.The proposed system allows users to visualize personalized traditional outfits from descriptive text. A robust image similarity retrieval system is utilized that connects generated images with actual images of products. Evaluation results using the Fr´ echet Inception Distance and Cosine similarity metrics of contrastive models demonstrate that the system produces coherent and visually consistent images.This work highlights the potential of Generative Artificial Intelligence solutions in such specific fashion settings, giving designers and consumers an enriched, interactive experience of ethnically inspired outfits.

ABSTRACT. Automatic vehicle classification is a crucial aspect of intelligent transportation systems for monitoring traffic flow, security enforcement, automated parking, driver assistance, and autonomous vehicle control. These systems face challenges such as limited computational resources, limited sensor range, and communication channels, which demand fast and lightweight algorithms for accurate and reliable performance. This paper presents an accurate, fast, and lightweight convolutional neural network-based vehicle image classification model for intelligent traffic management applications using the urban scene database. The dataset includes multiple scenes captured during morning, evening, and night, presenting diverse lighting and environmental conditions. The performance of the proposed model is evaluated in terms of class-wise accuracy, overall accuracy, and computational resources. The proposed CNN model achieves an overall accuracy of 99.33% with an inference speed of 188fps and a model size of 16.6MB, and outperforms most of the benchmark pretrained models. The classification accuracy highlights the model’s reliability and adaptability to real-world challenges such as variable lighting conditions.

ABSTRACT. It is essential to minimize model size while preserving high detection accuracy in order to optimize storage and processing efficiency, particularly for real-time applications on devices with limited resources. The deterministic block binary diagonal (DBBD) sensing matrix for image compression is used with a compressed YOLOv8 model in this study to offer a floating debris detection approach. The DBBD matrix efficiently minimizes memory use while maintaining essential characteristics when reducing image dimensions to 128×128 pixels. This makes it a highly efficient processing methodology. A unique data set of 16,000 identified debris images divided into four categories: dense, sparse, single, and piece of debris was used to train the YOLOv8 model. The accuracy, precision, and recall of the model are 95.96%, 97.94% and 99.77% respectively, indicating high quality performance. The values illustrate the reliability with which it can locate floating debris in a variety of environmental scenarios. The final model size is under 5 MB, which allows it to function effectively on devices with constrained computational resources despite its superior performance. Fast processing and effective memory utilization are made possible by the inclusion of DBBD compression, which makes this method appropriate for real-time monitoring systems, especially when used in internet of underwater things (IoUT) applications. For the purpose of observing and managing marine pollution and protecting fragile aquatic ecosystems, this innovative approach offers a practical solution for floating debris identification that strikes a balance between high detection accuracy and computing economy.

ABSTRACT. The proliferation of drones, particularly in unauthorized or hostile deployments, has necessitated the development of automated anti-drone systems capable of accurate real-time detection. However, the effectiveness of vision-based drone detection remains hindered by challenges such as cluttered backgrounds, varying environmental conditions, and the visual similarity between drones and benign aerial objects like birds or aircraft. Existing datasets are often insufficiently diverse, limiting model generalization. This study addresses these challenges by constructing a high-resolution, square, rectangular and polygon-shaped annotation drone detection dataset comprising 3359 images that capture diverse real-world aerial scenes. A lightweight, you only look once 11 small (YOLO11s) model, detection framework was trained and evaluated under both ideal and noisy conditions. On noise-free test images, the model achieved a strong recall of 95.76% and mAP of 83.53%, though the moderate precision of 71.31% suggested over-detection. Under high Gaussian noise (σ = 50.0), recall dropped to 57.80% while precision increased to 81.97%, with mAP falling to 69.89%. Salt-and-pepper noise (p = 0.5) caused complete failure across all metrics. Gaussian and median filtering techniques were applied as pre-processing steps to enhance robustness. Median filtering demonstrated superior resilience under impulsive noise, preserving a F1-score of 27.02% at p = 0.5, compared to 10.56% using Gaussian filtering. Overall, the proposed pipeline combining noise-resilient filtering and YOLO11s detection offers a reliable framework for drone identification in both ideal and degraded visual environments.

ABSTRACT. This paper presents an deep learning approach for fruit recognition to enhance accuracy and computational efficiency for real-world applications. Fruit recognition is challenging due to variations in background, illumination and color, high processing latency, large model sizes, and the need for effective classification across multiple categories. To address these issues, the method integrates discrete cosine transform (DCT), discrete wavelet transform (DWT), and compressed imaging techniques with convolutional neural networks (CNNs) to improve feature extraction, reduce dimensionality, and enhance model robustness. The models were evaluated using the fruits 360 dataset, consisting of 141 fruit categories, and benchmarked using accuracy, precision, recall, and f1-score. Results show that the wavelet transform-based model achieved an accuracy of 97.53%, while the baseline CNN reached an accuracy of 98.05%(existing model), but with higher computational complexity. The approach improves noise resilience and reduces computational demands without significant accuracy loss. Additionally, the models were deployed in an Android application for real-time fruit recognition, demonstrating their efficiency in resource-constrained environments. This study highlights the effectiveness of integrating image processing techniques with deep learning for efficient and scalable fruit recognition, enabling applications in automated retail, precision agriculture, and dietary management.

ABSTRACT. In this study, we propose an automatic photoplethysmogram (PPG) Fourier magnitude spectrum (FMS) quality assessment (PPG-FMS-QA) using deep convolutional neural network (CNN/ConvNet) architectures to reduce false alarm rates. Acceptable and unacceptable quality raw PPG segments are multiplied with a Hamming window (HW) and further computed the FMS using a fast Fourier transform (FFT) to avoid the leakage problem. The noise-free (NF) or acceptable PPG segments are collected from different standard databases. The noisy or unacceptable PPG segments are collected using three sets of noise sources such as wrist-cup database (N-WDB01), random noise corrupted PPG segments (N-RDB02), and acceleration noise corrupted PPG segments (N-ADB03). The proposed method includes a total of six 1D-ConveNet architectures, such as 2 and 4-convolutional layers (CLs) with 16, 32, and 64 kernels, and rectified linear unit (ReLU) activation function (AF). From the evaluation results, the proposed 4-layer with 16 filters and ReLU outperformed other models in terms of benchmark metrics. For the known dataset, the optimal model achieved an accuracy of 70.10% for NF versus N-WDB01, 99.16% for NF versus N-RDB02, and 91.62% for NF versus N-ADB03. Further, the method was tested using unseen databases and achieved an accuracy of 91.36\% for vitalDB, 51.53% for PulseDB.

ABSTRACT. Today, the internet is extensively utilized across numerous domains. With indispensable applications ranging from education to healthcare, and from banking systems to e-commerce, it also attracts the attention of malicious actors. In the first quarter of 2024 alone, approximately 10 million attacks were recorded. Therefore, the detection and prevention of internet-based attacks is an increasingly critical issue that demands resolution.

In this study, we propose three Deep Learning (DL) models, namely Deep Neural Network (DNN), Densely Connected Deep Neural Network (DenseNet), and Convolutional Neural Network (CNN), for the detection of URL-based phishing attacks. Additionally, we examine the impact of Word Encoding (WE) and Character Encoding (CE) approaches on the performance of these models. The results demonstrate that the WE approach yields superior performance on large-scale datasets. Conversely, the CE approach achieves better results on smaller datasets that are insufficient for effective model training. In all experiments, the CNN model got the most successful, achieving an accuracy of 0.99732 on the first dataset and 0.83447 on the second dataset.

ABSTRACT. Maritime transport is the backbone of global trade, accounting for about 80% of the volume of international trade. However, the sector generates a significant proportion of greenhouse gas (GHG) emissions and contributes to marine pollution. In this context, the development of a regulatory framework that encourages the transition to renewables and clean technologies is essential. This paper analyzes the main international, regional and national regulations, with a special focus on the integration of renewable energy sources in maritime transport, highlighting the associated challenges and opportunities. At the same time, a case study is presented on the sustainability, from an energy point of view, of hybrid fuel cell-battery energy systems, taking into account the conditions of use of these systems on maritime transport vessels.

ABSTRACT. This article investigates advanced monitoring solutions for hybrid fuel cell and battery-based systems, with applications in hybrid electric systems and shipping. Emphasis is placed on presenting intelligent energy performance monitoring strategies, using technologies such as artificial intelligence (AI), machine learning (ML) algorithms to optimize energy flows, increase efficiency and extend the life of components. At the same time, an optimized energy management strategy is proposed, integrating real-time monitoring with predictive analytics and optimization algorithms. The study highlights the contribution of advanced diagnostic and prognostic algorithms, as well as innovative sensors, in optimizing component performance and safety. The paper supports the development of an integrative framework for energy management in complex systems, highlighting promising prospects for the decarbonization of transport and the transition to the use of renewable energy sources. The focus is on the need to integrate these technologies into real applications, with additional testing to validate the performance and scalability of the proposed solutions.

ABSTRACT. This paper proposes an integrated architecture designed to enhance the performance of logistics systems through the convergence of embedded systems, the Internet of Things (IoT), and Artificial Intelligence (AI) applications. A layered structure is introduced, combining real-time data acquisition from embedded devices and sensors, intelligent orchestration through edge computing, and advanced analytics powered by machine learning algorithms. The novelty of the architecture lies in its intelligent middleware layer, semantic data modeling, distributed learning capabilities, and context-aware adaptive behavior in dynamic logistic environments. A simulated environment will be developed, , in a future research paper, to validate the architecture, including embedded IoT devices, edge gateways, and a cognitive layer implementing predictive models such as Random Forest and LSTM. Key performance indicators such as delivery time, prediction accuracy, anomaly detection rate, and system responsiveness will be monitored. The expected results suggest significant improvements in operational efficiency, decision accuracy, and resilience compared to conventional systems. This contribution supports the development of scalable, intelligent logistics infrastructures aligned with Industry 4.0 paradigms. Future work will address deployment in real-world scenarios and integration with ERP and WMS systems.

ABSTRACT. The present paper provides an analysis of the technologies used for the detection of persons hidden in cargo vehicles and containers, in the context of preventing human trafficking. The need for effective control at border points is highlighted, given the transnational nature of these crimes and the major humanitarian impact. The study reviews three major categories of technologies: sensors for the detection of volatile compounds (CO₂, VOCs), imaging technologies (X-rays, terahertz, thermal), as well as systems for detecting vibrations generated by the human body. Each method is analysed from the perspective of operational advantages, technical limitations and conditions of applicability. It also stresses the importance of a multimodal approach combining multiple technologies into an integrated system to maximise accuracy and minimise false alarms. The paper argues that the future of effective detection of hidden persons is based on technological interoperability, standardization, portability, and compliance with ethical and legal norms. Thus, an applied research direction is proposed oriented towards smart, fast and sustainable solutions, capable of meeting the current requirements of border security

ABSTRACT. This paper presents a comparative analysis of two architectures for low-power urban electric vehicles: one with a central electric motor and the other with in-wheel motors. Both systems are designed with a total power of 4.5 kW, 48 V supply voltage, and 6 kWh Li-ion batteries. The study includes energy and cost evaluations, estimating real-world urban autonomy, average consumption, and operating time under full load. Architectural differences are examined in terms of energy efficiency, mass distribution, drivetrain complexity, and controller integration. The acquisition costs of key electrical components are compared, and investment payback is assessed based on the savings generated compared to internal combustion engine vehicles. Results indicate that the central motor configuration reaches full amortization after approximately 31,900 km, while the in-wheel motor configuration requires over 36,000 km, both under exclusive urban usage. This study provides an objective foundation for selecting the appropriate configuration based on application context, performance, and cost criteria.

ABSTRACT. Today, there is a strong focus on improving the efficiency of both energy consumption and its sources. To this end, renewable energy generated by photovoltaic panels presents an effective solution. On the one hand, it can be used for both individual and shared consumption, while any surplus can be stored in batteries or injected into the grid, leading to a significant long-term reduction in consumers' energy bills. On the other hand, efficient and transparent energy management can be achieved through advanced technologies such as blockchain. This research validates, through the implemented application, the feasibility of using blockchain technology to optimize electricity consumption within a homeowners' association. Based on an analysis of the operating principles of blockchain and an analysis of specialized studies proposing its use in the energy sector, the study introduces mechanisms for trading electricity through blockchain in a small community, with the aim of optimizing consumption. A suite of software tools is implemented to facilitate the trading of renewable electricity and provide statistical insights into energy consumption, both at the organizational and individual levels. The use of photovoltaic panels and blockchain in a small homeowners' association brings multiple benefits, including cost reduction, increased trust among members due to the transparency provided by blockchain, and increased sustainability through reduced energy consumption and greater energy independence. This integration transforms a homeowners' association into a smart, efficient, and sustainable community. In addition, based on the rewards offered within the application, it encourages responsible consumption of electricity.

ABSTRACT. The scope of this paper is to present a hybrid power system primarily based on renewable energy sources, and to study the sizing of the system using the iHOGA software. The hybrid system is composed of: wind turbines, a Diesel generator (D.G.), a converter, a battery, and an electric load. Two case study on the design of a systems were performed using IHOGA: first case for the city Rm. Valcea (County Valcea) and second case for the city Mioveni (County Arges), for the same load. The results from the simulation are presented both in tabulated and graphical forms in order to provide a better way to visualize any differences present between them. To further help an investor in choosing the most convenient design, the pros and cons of each city were presented in a comparative analysis. The analysis is made in the southern part of Romania, the locations chosen are the city Rm. Valcea situated at Latitude: 45.1 and Longitude: 24.36 and the city Mioveni situated at Latitude: 45.9 and Longitude: 24.95. Simulation results show that the share of renewable energies for the city Mioveni is higher, sitting at a value of 16.36%, in comparison to the city Rm. Valcea which is currently proving a renewable energy share of 7.06%. The operating costs are higher for the city Rm. Valcea with 24.37% compared to the city Mioveni (1189 Euro/ year - Rm. Valcea, 899,6 Euro/ year - Mioveni). The cost of energy supplied by the city Rm. Valcea is 6% more expensive than the same cost for the city Mioveni. Regarding the pollutant emissions as expected, they are with 9.83% higher for city Rm. Valcea, the city Mioveni produces 432 kg/year less carbon dioxide than Rm. Valcea.

ABSTRACT. In the last three years, many research papers have been published in the field of Hybrid Energy Systems based on Fuel Cells/Batteries, which highlight that the Zero Emissions objective for transport can be achieved soon, according to the requirements of transport regulations that have been recently revised relative to the proposed objectives for pollutant emissions, carbon footprint, etc. The main objectives of this paper are to analyze the new Hybrid Energy Systems proposed for vehicles, the new Management Systems proposed for these systems, highlighting their advantages and disadvantages compared to previously proposed solutions, emission reduction, etc. compared to previously proposed solutions. The contribution and novelty of the analysis refers to the critical analysis of the new proposed solutions and the presentation of current solutions and challenges for this field.

ABSTRACT. With hydrogen becoming an attractive energy carrier, the generation of green hydrogen through electrolysis to decarbonize sectors with high greenhouse gas (GHG) emissions such as the energy generation and transportation sectors has started to gain more traction as an important research focus. Electrolysis, as a process that is not fully efficient, results in the generation of waste heat. The scope of this paper is in increasing the overall efficiency of an electrolyzer system by utilizing the waste heat generated. The recovery of waste heat can increase the efficiency of a PEM electrolyzer system from 75-80% up to 86-95%. By adjusting the cooling system to facilitate the inclusion of a puffer vessel ~1.1kWh of thermal energy were recovered from the 10kWh PEM electrolyzer system used in the experiment.

ABSTRACT. Railway interlockings represent the systems on which the safety of all people using the railways depends on. Legacy systems, especially relay based interlockings, persist on the railway network and are foreseen to remain in operation in the long term. This paper presents the development of an electronic simulator of relay interlocking designed to be used for the training of future engineers. The system reproduces the functions of a typical Romanian relay interlocking. The simulation software was developed in LabVIEW, while hardware development involved the design of special interfaces with a traditional mosaic control panel and a scale railway model.

ABSTRACT. Automated baggage handling networks are critical to airport operations, yet their reliability and interdependencies make them vulnerable to disruptions ranging from electrical power outbreaks to mechanical failures or to cyber-attacks. This study proposes a systems engineering framework to model and quantify vulnerabilities in automated airport baggage handling systems (“BHS”), integrating classical models, failure mode analysis, and simulation techniques. By mapping BHS components (e.g., conveyor systems, electrical motors, sorting mechanism, IoT sensors and data) and their interactions, the framework identifies cascading failure risks and vulnerable points which threaten operational continuity. Using a combination of SysML (Systems Modelling Language) diagrams and agent-based simulations, the research evaluates scenarios such as ransomware attacks on control systems, sensor malfunctions, and peak-load congestion. Metrics like node criticality, recovery time, and redundancy efficiency are developed to prioritize mitigation strategies. The study analyses how adaptive redundancy, predictive maintenance solutions, and decentralized control architectures can reduce system-wide vulnerability exposure by up to 30%. By bridging systems engineering principles with real-world baggage logistics, this approach provides actionable insights for designing reliable and resilient, self-healing networks. The findings advocate holistic vulnerability assessments in aviation infrastructure, ensuring compliance with evolving safety standards like ICAO’s cybersecurity guidelines. This methodology is scalable to other automated transport systems, fostering safer, more reliable air travel ecosystems.

ABSTRACT. This paper presents a novel methodology for enhancing distribution network reliability – often characterized as power quality markers – by optimally placing remotely-operated switches (RCSs) in medium-voltage feeders. Our approach formulates a multi-objective optimization problem targeting reductions in System Average Interruption Duration Index (SAIDI), System Average Interruption Frequency Index (SAIFI), Momentary Average Interruption Frequency Index (MAIFI), Energy Not Supplied (ENS), and improvements in Average Service Availability Index (ASAI). A mixed-integer model was developed, that computes the reliability indices as functions of switch placements and solves it using a custom genetic algorithm (GA). Unlike previous works that consider only single objectives (e.g. minimizing ENS) or use game-theoretic allocation, the proposed approach simultaneously optimizes multiple indices and explicitly incorporates ASAI and MAIFI. The GA-based solution effectively identifies switch configurations that balance cost and reliability gains. In a benchmark case study (inspired by a real radial network), inserting eight optimally chosen RCSs reduced SAIDI by ~27% and ENS by ~48%, with negligible impact on SAIFI (since RCSs do not prevent faults but only limit their scope). Our results demonstrate that smart RCS deployment can significantly improve supply reliability, with performance supported by simulation data and graphics.

ABSTRACT. The technical issues faced by Distribution Network Operators (DNOs) regarding the distribution infrastructure, particularly at the low-voltage level, are becoming increasingly urgent due to the rising number of integrated prosumers and the growing demand for electric vehicles, which necessitate additional load for charging. To address these challenges, DNOs can take proactive measures by optimising the placement of electric vehicle charging stations (EVCSs) as close as possible to the power injections of prosumers. In this context, the paper proposes an engaging and effective clustering-based decision-making support system (DMSS) specifically designed to locate electric vehicle charging stations within low-voltage electrical distribution networks (LV-EDNs) in rural and peri-urban zones. The proposed DMSS integrates the K-means clustering algorithm, which identifies candidate zones based on hourly power flows and average phase voltages at each node, calculated for reference operating regimes, prioritising both practicality and reliability in LV-EDNs. Strategically, the EVCSs are positioned in nodes located within "candidate" zones, focusing on those with minimal average percentage errors for both hourly power flows and phase voltages. The optimal location is achieved by comparing them to a virtual node that represents the zone's characteristics. The objective is to determine the optimal number of EVCSs and their placement by minimising energy losses and voltage deviations, taking into account various typical power demand profiles for the EVCSs. The proposed DMSS has been tested in an LV-EDN with 36 nodes in a peri-urban zone of northwestern Romania. The findings were quite promising; four EVCSs were optimally located without requiring additional investment from the DNO for network reinforcement.