Volume 14 | Issue 7 |

 Abstract

wireless charging is becoming popular all over the world to charge the electric vehicle (EV). But an EV cannot go too far with a full charge. It will need more batteries to increase its range. Dynamic wireless charging is introduced to EVs to capitally increase their driving range and get rid of heavy batteries. Some modern EVs are getting off this situation. But with Dynamic WPT the need of plug-in charge and static WPT will be removed gradually and the total run of an EV can be limitless. If we charge an EV while it is driven, we do not need to stop or think for charging it again. Eventually, in the future the batteries can be also removed from EVs by applying this method in everywhere. Wireless charging needs two kinds of coils named the transmitter coil and the receiver coil. The receiver coil will collect power from the transmitter coil while going over it in the means of mutual induction. But the variation of distance between two adjacent coils affects the wireless power transfer (WPT). To see the variation in WPT, a system of two Archimedean coils of copper is designed and simulated for vertical and horizontal misalignment in Ansys Maxwell simulation software. The transfer power for 150 mm air gap is 3.74 kW and transfer efficiency are gained up to 92.4%. The charging time is around 1 hour and 39 minutes to fully charge its battery from 0 state for a 150mm air gap for an EV with 6.1 kW power may gain efficiency up to 94%- 96% and take 50-60 minutes. Also, a charging lane is designed for dynamic charging. Then the power transfer is calculated from mutual inductance when the EV is driven on a charging lane. From the load power, it can be calculated how further an EV can go with this extra power.

Licence: creative commons attribution 4.0

  License

Creative Commons Attribution 4.0 and The Open Definition

 Keywords

Wireless charging, Electric Vehicle (EV), Wireless Power Transfer (WPT)

  License

Creative Commons Attribution 4.0 and The Open Definition

 Abstract

The voltage quality is one of the major concerns for industrial and distribution consumers. In this work, a series active filter for voltage compensation has been verified using vectorial power theory for a three-phase system with non- linear load. The simulation has been designed using MATLAB/Simulink. The control scheme provides better unbalance voltage compensation and voltage regulation for three phase system with non-linear loads which operates in different conditions. The design and simulation of series active filter performance results were presented in this paper to show effectiveness of controller.

Licence: creative commons attribution 4.0

  License

Creative Commons Attribution 4.0 and The Open Definition

 Keywords

Series Active Filter (SAF), Power Quality Improvement (PQI), Voltage unbalance, Power vectorial theory

  License

Creative Commons Attribution 4.0 and The Open Definition

 Abstract

Wireless Power Transfer (WPT) is an emerging technology that enables electric vehicle (EV) charging without physical connectors, improving convenience, safety, and reliability. This project presents the design and analysis of three major WPT technologies: Inductive Power Transfer (IPT), Capacitive Power Transfer (CPT), and Hybrid Power Transfer (HPT). The systems are evaluated based on efficiency, waveform performance, misalignment tolerance, electromagnetic field (EMF) exposure, and practical implementation using MATLAB/Simulink models. The study focuses on the use of resonant compensation techniques to improve power transfer capability under varying operating conditions. The effect of air-gap variation and coil misalignment on system performance is also analyzed. In addition, the project investigates system losses, voltage and current characteristics, and safety considerations associated with wireless charging. Comparative analysis of the three methods is carried out to identify their advantages and limitations for EV applications. The research concludes that IPT provides higher efficiency, stable operation, lower EMF exposure, and simpler implementation compared to CPT and HPT systems. CPT and HPT offer advantages in compactness and coupling improvement but face limitations related to high-frequency operation and increased complexity. Overall, the study identifies IPT as the most practical and reliable wireless charging technology for present electric vehicle applications.

Licence: creative commons attribution 4.0

  License

Creative Commons Attribution 4.0 and The Open Definition

 Keywords

Wireless Power Transfer (WPT), Inductive WPT (IPT), Capacitive WPT (CPT), Hybrid WPT (HPT).

  License

Creative Commons Attribution 4.0 and The Open Definition

 Abstract

This paper presents SignO, a novel real- time bidirectional sign language translation system that seamlessly integrates Convolutional Neural Net work (CNN) based hand tracking, transformer-based natural language processing, and artificial intelligence- driven avatar animation to support multiple sign languages including American Sign Language (ASL), British Sign Language (BSL), and Indian Sign Language (ISL). The system addresses the fundamental challenges in sign language translation through innovative implementations of self-supervised sign pose learning, multilingual transformer architectures, and photorealistic 3D avatar rendering. Our approach tackles gesture ambiguity, lighting variance, and data scarcity through pose-conditioned Generative Adversarial Networks (GANs), multistream graph neural networks, and low resource pretraining methodologies. SignO achieves 94.7% accuracy in sign-to-text translation and 91.2% accuracy in text-to-sign conversion across evaluated languages, maintaining real- time performance at 30 frames per second. The system incorporates advanced features including voice-to-sign conversion, offline mode functionality, and automatic dialect detection, significantly advancing accessibility solutions for the deaf and hard-of-hearing community worldwide.

Licence: creative commons attribution 4.0

  License

Creative Commons Attribution 4.0 and The Open Definition

 Keywords

sign language translation, computer vision, transformer networks, CNN, accessibility technology, human-computer interaction

  License

Creative Commons Attribution 4.0 and The Open Definition

 Abstract

Music genre classification is a challenging task in the field of audio signal processing and machine learning. As the volume of digital music continues to grow exponentially, the need for automated systems to organise and categorise this vast array of music becomes crucial. This research presents a comprehensive Music Genre Classification System (MGCS) that leverages advanced machine learning techniques to accurately categorise music into predefined genres. The proposed MGCS employs a feature extraction process to capture key characteristics of audio signals, including spectral features, rhythm patterns, and temporal information. These features serve as input to a well-designed machine learning model, such as a convolutional neural network (CNN) which is trained on a diverse and extensive dataset of labelled music samples. The evaluation of MGCS involves testing its performance on various benchmark datasets, comparing its accuracy, precision, recall, and F1 score with existing state-of-the-art methods. The results demonstrate the efficacy of the proposed system in achieving high classification accuracy across a wide range of music genres. Furthermore, the research explores the interpretability of the model's predictions, shedding light on the features that contribute most to genre classification decisions.

Licence: creative commons attribution 4.0

  License

Creative Commons Attribution 4.0 and The Open Definition

 Keywords

Harmonising Genres by Exploring Music Genre Classification through Comparative ML Models

  License

Creative Commons Attribution 4.0 and The Open Definition

 Abstract

The growing need for real-time intelligent services has expedited the transition of Artificial Intelligence workloads from centralized cloud infrastructures to edge settings. Although edge computing markedly decreases latency and bandwidth consumption, it engenders substantial privacy issues stemming from the sensitive data produced by Internet of Things devices. Traditional centralized learning methods jeopardize privacy by necessitating the aggregation of raw data, while typical Federated Learning, although avoiding direct data sharing, is nevertheless vulnerable to inference-based assaults. Moreover, many existing privacy-preserving techniques introduce substantial computational and communication overhead, limiting their applicability in resource-constrained edge settings. This research introduces an effective distributed AI architecture for privacy-preserving edge computing that simultaneously enhances privacy, model performance, and communication efficiency. The proposed approach integrates an adaptive, device-aware Differential Privacy technique that modifies noise injection according to client capabilities, in conjunction with gradient sparsification to minimize communication overhead. A hierarchical aggregation approach is utilized with edge servers to mitigate synchronization bottlenecks. Experimental findings in a simulated heterogeneous edge environment indicate that the proposed framework attains performance akin to centralized learning while markedly decreasing communication costs, thus ensuring practical viability for implementation on resource- limited IoT devices.

Licence: creative commons attribution 4.0

  License

Creative Commons Attribution 4.0 and The Open Definition

 Keywords

Edge Computing, Federated Learning, Privacy Preservation, Differential Privacy, Distributed AI, IoT, Communication Efficiency.

  License

Creative Commons Attribution 4.0 and The Open Definition

 Abstract

Crime detection and public safety monitoring in urban environments are traditionally dependent on manual surveillance systems, which are often inefficient, time-consuming, and prone to human error. This project proposes a deep learning-based crime detection system for public safety using Convolutional Neural Networks (CNNs) to automate the identification of violent and non-violent activities from real-time video streams. The system leverages computer vision techniques to analyze video frames by performing preprocessing steps such as frame extraction, resizing, normalization, and data augmentation, followed by classification using CNN architectures such as AlexNet and LeNet. The proposed system is capable of detecting suspicious or aggressive behavior in real time and generating alerts to authorities through an automated notification mechanism. Additionally, a web-based interface built using Django enables users to upload videos, visualize detection results, and monitor crime alerts efficiently. The system also supports location- based alert tracking and dashboard visualization to enhance decision-making for law enforcement agencies. Experimental results demonstrate high accuracy in classifying violent and non-violent activities, along with efficient real-time performance. The proposed framework reduces dependency on manual monitoring, improves response time, and provides a scalable solution for intelligent surveillance in smart cities, public transport systems, and high-security areas.

Licence: creative commons attribution 4.0

  License

Creative Commons Attribution 4.0 and The Open Definition

 Keywords

Crime Detection, Convolutional Neural Networks (CNN), Deep Learning, Video Surveillance, Public Safety, Computer Vision, Real-Time Monitoring, Automated Alert System, Smart City Security

  License

Creative Commons Attribution 4.0 and The Open Definition

 Abstract

The increasing deployment of intelligent surveillance systems has created a need for efficient real-time processing while ensuring data privacy and security. Traditional cloud- centric architectures are often limited by high latency, bandwidth constraints, and risks of sensitive data exposure. This study presents a novel framework that combines edge artificial intelligence, transformer-based deep learning models, and federated learning to enable real-time anomaly detection in surveillance environments. The proposed system processes data at the edge nodes, reducing latency and minimizing the dependency on centralized servers. Federated learning allows collaborative model training without sharing raw data, thereby preserving the privacy of the user. Transformer models enhance detection performance by capturing temporal dependencies in video streams more effectively than conventional methods. The proposed framework demonstrates improved accuracy, reduced response time, and strong privacy guarantees, making it suitable for applications such as smart cities, public safety, and critical infrastructure monitoring.

Licence: creative commons attribution 4.0

  License

Creative Commons Attribution 4.0 and The Open Definition

 Keywords

Edge AI, Federated Learning, Transformer Models, Privacy Preservation, Surveillance Systems, Anomaly Detection, Deep Learning

  License

Creative Commons Attribution 4.0 and The Open Definition

 Abstract

Internet-exposed services attract steady background noise: credential guesses, parameter tampering, and injection-style probes. Firewalls and IPS can drop packets, yet they rarely explain what a probe tried or how often a single address repeats a pattern. We built a small deception stack on Node.js/Express that answers that need without touching production data.

Licence: creative commons attribution 4.0

  License

Creative Commons Attribution 4.0 and The Open Definition

 Keywords

Honeypot, deception, attack logging, rule engine, JSON analytics, brute-force heuristics, real-time dashboard, cyber-physical signaling

  License

Creative Commons Attribution 4.0 and The Open Definition

 Abstract

The rapid advancement in computing power presents challenges for current asymmet- ric cryptography schemes such as RSA and ECC. As a result, there is a growing need to implement quantum cryptography as a replacement. However, due to the slow speed and high computational demands of quantum cryptography algorithms like ML-KEM, they cannot be utilized in applications that require real-time responses. Furthermore, modern cryptographic solutions lack adaptive mechanisms; current cryptographic configurations are applied uniformly across entire systems and cannot adjust to emerging requirements, such as rapid changes in the network environment. The proposed Software-Defined Cryptographic Agility Framework (SD-CAF) employs Deep Reinforcement Learning to select optimal cryp- tographic configurations based on actual network parameters, such as current latencies and packet losses. Thus, our proposed solution enables dynamic configuration changes without disrupting secure connections. SD-CAF is implemented in Python using network simulation to measure the performance of various cryptographic settings. Our results demonstrate that employing an intelligent approach can reduce latencies without compromising cryptographic security, thereby proving the effectiveness of our idea.

Licence: creative commons attribution 4.0

  License

Creative Commons Attribution 4.0 and The Open Definition

 Keywords

Post-Quantum Cryptography, Cryptographic Agility, Deep Reinforcement Learn- ing, IP Fragmentation, Software-Defined Networking.

  License

Creative Commons Attribution 4.0 and The Open Definition