School of Electrical Engineering and Technology (SEET)

Permanent URI for this communityhttp://197.211.34.35:4000/handle/123456789/34

School of Electrical Engineering and Technology (SEET)

Browse

Search Results

Now showing 1 - 10 of 13
  • Item
    Classification and Severity Estimation of Eccentricity Faults in Salient Pole Synchronous Machine using Deep Learning
    (IEEE, 2025) Yusuf, Latifa; Moa, Belaid; Ilamparithi, Thirumarai Chelvan
    The presented research work is focused on the classification and severity estimation of eccentricity faults in Salient Pole Synchronous Machines. Building on our comparative study of Artificial Neural Network and Convolutional Neural Network for eccentricity fault classification, we propose an end-to-end deep learning model, namely Hierarchical Convolutional Neural Network, for eccentricity classification and severity estimation. The deep learning model inherently consists of an eccentricity detection component for fault classification and a severity estimation component for fault quantification. The deep learning model is built using the experimental data of a 3-phase, 2-kW, salient pole synchronous machine. The machine is subjected to 20%, 40%, and 60% severities of static and dynamic eccentricity faults under different loading conditions. Stator line currents and line-to-line voltages obtained from different operating conditions are used to train, validate and test the proposed model. To enhance the model's performance, time delay construction was incorporated to augment the datasets and carefully evaluate the impact of selected raw input features, specifically stator currents and voltages, as well as the load. Among the evaluated scenarios, the use of voltage with time delay (V, TD) as input features produced the best results, achieving 100% classification accuracy and a root mean square error of 0.0046 for static eccentricity and 0.0188 for dynamic eccentricity estimation. Results indicate that the model performs excellently in fault classification and severity estimation. Compared to traditional machine learning models, the presented model is an end-to-end deep learning architecture devoid of manual feature extraction and is robust to load variations.
  • Item
    Development of stability charts for double salience reluctance machine modeled using hill’s equation
    (Bulletin of Electrical Engineering and Informatics, 2024-06-10) Enesi Asizehi Yahaya; Emenike Chinedozi Ejiogu
    The paper presents a novel algorithm for the development of stability charts. The second-order differential homogeneous equation describing a double salient reluctance machine with a capacitance connected to its stator winding is transformed into hill’s equation. The circuit components are the stator coil time-varying inductance of a double salient reluctance machine, capacitance and resistance. All these are modeled by hill’s equation. The double salient reluctance machine acts as an energy conversion system. The maximum and minimum inductance of the energy conversion system is measured in laboratory by inductance, capacitance, and resistance (LCR) meter. These values help to determine the inductance modulation index. The inductance modulation indetx, the characteristic constant and the characteristic parameter obtained from modeling equations are used in the MATLAB/Simulink model. The MATLAB/Simulink simulations generate stable and unstable oscillations to form stability charts. The proposed stability charts are in good agreement with the Ince-Stritt stability chart, which is widely applied in physics, mechanics and in electrical engineering, especially where the state of stability of a system or an electric oscillatory circuit is to be determined
  • Item
    Investigating the Thresholding Effect and Fingerprint Transformation Using Cross-Correlation Similarity Matching
    (Faculty of Science Lafiya, 2025-01-17) Garuba O.R., Abdullahi, I.M., Dogo, E.M., & Maliki, D
    One of the leading diseases globally is cancer and breast cancer is not exempted. The objective of the WHO Global Breast Cancer Initiative (GBCI) is to reduce global breast cancer mortality by 2.5% per year, thereby averting 2.5 million breast cancer deaths globally between 2020 and 2040. The three pillars toward achieving these objectives are: health promotion for early detection; timely diagnosis; and comprehensive breast cancer management. In this study we propose an early and comprehensive detection technique in combating breast cancer diagnosis by combining the strength of both PSO (Particle Swarm Optimization) and BPSO (Binary Particle Swarm Optimization) to achieve optimal solution. The results obtained indicated the superiority of the Hybrid PSO-BPSO model in detection over an existing solution by achieving an accuracy of 98.82% on both the WBCD and WDBC datasets.
  • Item
    An Electronic Voting System with Directed Acyclic Graph (DAG)- Based Blockchain Using ShimmerEVM
    (El-Amin University Journal of Computing (EAUJC)., 2024-04-07) Maliki. D., C. Oruche, I. M. Abdullahi, B. G Najashi, O. R. Isah
    This research introduces an innovative electronic voting system that enhances transparency, anonymity, and reliability, aiming to revolutionize both traditional and existing electronic voting methodologies. The system increases accessibility, security, and efficiency in the electoral process. Advanced web development technologies, including NextJs, TailwindCSS, TypeScript, and JWT tokens, are integrated for an improved e-voting experience. This system employs encryption and cryptographic hashes to secure sensitive information, alongside smart contracts on ShimmerEVM— a Directed Acyclic Graph (DAG)-based blockchain—to ensure data persistence and immutability. A user-friendly front-end interface serves as a portal to the web application, enabling seamless interaction with the ShimmerEVM network. A critical feature of the system is the activation of a biometric hardware component, essential for voter registration and participation. ShimmerEVM facilitates the execution of smart contracts, offering a decentralized, transparent, and secure environment without relying on traditional blockchain technology. The focus of this system is on the implementation of security-centric smart contracts, which are pivotal in maintaining voting data integrity and mitigating the risks of vote count manipulation.
  • Item
    ANALYSIS OF SPECTRUM OCCUPANCY PREDICTION RESULTS FOR MAITAMA ABUJA
    (International Conference on Communication and Information Science (ICCIS), 2024) Ajiboye, Johnson Adegbenga; Mary Adebola Ajiboye; Babatunde Araoye Adegboye; Daniel Jesupamilerin Ajiboye; Jonathan Gana Kolo; Abiodun Musa Aibinu
    This research investigates the efficacy of Artificial Neural Networks (ANN) in predicting spectrum occupancy in Maitama, Abuja, Nigeria, focusing on frequency bands ranging from 30 MHz to 300 MHz. The primary objective was to evaluate the accuracy of ANN-based predictions of spectrum usage and compare these predictions with actual measurements. The study employed ANN to forecast spectrum occupancy across various frequency bands, and the predicted data were then compared with empirical measurements to assess the performance of the model. The analysis revealed that prediction errors were generally low across all frequency bands, with most errors falling below 1.5%. Specifically, the 30-47 MHz sub-band demonstrated an average percentage difference between the actual and predicted value of 0.087%, with a maximum error of 1.12% occurring at frequency of 44.65 MHz. For the 47.05-68 MHz band, the average percentage difference was slightly higher at 0.106%, and the maximum error was 2.18% occurring at frequency of 50.2 MHz. In the 68.05-74.8 MHz band, the average percentage error was 0.040%, but with highest error of 0.232% at frequency of 73.95 MHz. The 74.85-87.45 MHz band showed the most accurate predictions with an average error of just 0.010%, and a maximum error of 0.174% at 75.1 MHz. Overall, the highest prediction error was 0.106% in the 47.05-68 MHz band, whereas the lowest was 0.010% in the 74.85-87.45 MHz band. These results highlight the high accuracy of ANN in predicting spectrum usage, demonstrating its potential for effective spectrum management and planning in Maitama, Abuja.
  • Item
    STATE OF THE ART ON PATH LOSS MODEL DEVELOPMENT
    (Humminbird Publications and Research International, 2024-01-29) Ibukun Aderele Adeyemi; Jonathan Gana Kolo; Ajiboye, Johnson Adegbenga
    This is a study of path loss prediction modelling. Path loss modelling is widely applied in determining mobile wireless signal propagation in a given environment. This helps radio network planners to have an accurate view of requirements to obtain good quality of service when deploying radio networks. The empirical models are exhaustively analysed and compared with the emerging machine learning models. Also, mention is made of RIS models which are beginning to gather some attention due to their focus on the programmable electromagnetic properties. The study was able to establish empirical models as the most simple and efficient method of path loss prediction models. Attention is paid to the application of these models in both 900MHz and 1800MHz in urban, suburban and rural areas. This is due to the wide application of these frequencies in mobile wireless communication. The machine learning models present better results and give a high level of accuracy for diverse environments. However, they require large volume of data and environmental features extraction at both 2D and 3D to get reliable model. This makes it imperative to carry out field measurement tasks that are basically synonymous with methodologies employed in empirical approach to modelling. The variation in vegetation determines the best fit model for each particular case as well as the derivation of path loss exponent. The RIS modelling approach gives positive views especially at higher frequencies. The tuneable properties of the surfaces give a wide berth in application across different frequency spectrum. Complex and large volume of computation required in use of RIS implies that machine learning models, especially deep learning models will be better off incorporated into the process. It is thus beneficial to the researcher to ensure that a good grasp of the different approaches highlighted is obtained such that the benefits available are merged to produce finer results.
  • Item
    Design and Development of An IoT – Based Multi – Health Vital Signs Monitoring System
    (I3C 2024, 2024-04-22) Adegboye N.J; Dauda U.S; Ajiboye, Johnson Adegbenga; Ohize H.O; Adegboye B.A
    This study presents the design and development of an IoT-based multi-health vital signs monitoring system that can monitor a patient’s basic health physiological parameters in real-time. In this system, four (4) sensors were used to capture the data from the patient. These are body temperature sensor, electrocardiogram (ECG) sensor, accelerometer sensor and the eye blink sensor. The hardware modules were interfaced with the liquid crystal display (LCD) to display the required data. Memory modules stored the designated phone numbers. The GSM module retains connectivity with the cellular networks acting as SMS receiver, which sends data on the patient’s vital signs. The LCD displays the data, which can be seen through the IoT. The microcontroller was programmed using C++ programming language and connects all sensors. This enabled conveyance of data on the patient’s health condition via IoT to the doctor for further processing and analysis.
  • Item
    Artificial intelligence model for prediction of cardiovascular disease: An empirical study
    (AccScience Publishing, 2024) Umar, B. U.; Ajao, L. A.; Dogo, E. M.; Ajao, F. J.; Atama, M.
    Cardiovascular disease (CVD) is a disease related to the heart and blood vessels. Prediction of CVD is essential for early detection and diagnosis, which is however compounded by the complex interplay between medical history, physical examination outcomes, and imaging results. While the existing automated systems are fraught with the usage of irrelevant and redundant attributes, artificial intelligence (AI) helps in the identification of potential CVD populations by prediction models. This work aims at developing an AI model for predicting CVD using different classifications of machine learning techniques. The CVD dataset was obtained from the UCI repository containing about 76 cardiac attributes for training in various machine learning models, which include a hybrid of artificial neural network genetic algorithm (ANN-GA), artificial neural network, support vector machine (SVM), K-means, K-nearest neighbor (KNN), and decision tree (DT). The performance of the models was measured in terms of accuracy, means square error, sensitivity, specificity, and precision. The results showed that the hybrid model of ANN-GA performs better with an accuracy of 86.4%, compared to the SVM, K-means, KNN, and DT measured at 84.0%, 59.6%, 79.0%, and 77.8%, respectively. It was observed that the system performs better as the number of datasets increases in the database, with a fewer selection of attributes using genetic algorithm for selection. Thus, the ANN-GA model is recommended for CVD prediction and diagnosis.
  • Item
    Investigating the Thresholding Effect and Fingerprint Transformation Using Cross-Correlation Similarity Matching
    (2024) Garuba, O. R.; Abdullahi, I. M.; Dogo, E. M.; Maliki, D.
    This research presents a cross-correlation similarity matching method to study the fingerprint transformation and thresholding impact. This work directly compares the impact of various transformations (rotation, translation, elastic deformation, and scaling) on the fingerprint matching performance at different threshold values, in contrast to the standard minutiae-based systems. In order to compare the template positions of the two fingerprints using plots, the cross-correlation similarity matching of fingerprints first selects suitable templates in the primary fingerprint and then uses template matching to assess the impact of each transformation on matching accuracy, FRR, and FAR in the secondary print. The findings highlight the potential of thresholding in developing reliable and practical fingerprint recognition systems.
  • Item
    Robust Multi-Target Tracking with a Kalman-Gain CPHD Filter: Simulation and Experimental Validation
    (2025-04-19) Abdullahi Daniyan
    We introduce a novel cardinalized implementation of the Kalman-gain-aided particle probability hypothesis density (KG-SMC-PHD) filter, extending it to form the Kalman-Gain Particle Cardinalized Probability Hypothesis Density (KG- SMC-CPHD) filter. This new approach significantly enhances multi-target tracking by combining the particle-based state correction mechanism with the propagation of both the PHD and target cardinality distribution. Unlike conventional particle filters that require a large number of particles for acceptable performance, our method intelligently corrects selected particles during the weight update stage, resulting in a more accurate posterior with substantially fewer particles. Through comprehensive evaluations on both simulated and experimental datasets, the KG-SMC-CPHD filter demonstrates superior robustness and accuracy, particularly in high-clutter environments and nonlinear target dynamics. Notably, it offers improved cardinality estimation and maintains the computational efficiency and performance advantages of its predecessor, the KG-SMC-PHD filter, making it a powerful tool for advanced multi-target tracking applications.