Master Theses & Projects (FEAS)

Permanent URI for this collection

https://ontariotechu.scholaris.ca/handle/10155/687

Browse

Now showing 1 - 20 of 478

New frontiers in population-based multi-objective feature selection
(2024-04-01) Zanjani Miyandoab, Sevil; Rahnamayan, Shahryar; Makrehchi, Masoud
Feature selection is a persistent challenge aimed at minimizing the number of features while maximizing accuracy of classification, or any other machine learning and data mining task, by mitigating the curse of dimensionality. We frame feature selection as a multi-objective binary optimization task with the objectives of enhancing accuracy and reducing the feature count. Given that feature selection, along with binary optimization problems in general, is a NP-hard problem since the size of search space increases exponentially by the increase of the number of features, so especially in high-dimensional spaces, it can be very challenging. We propose three innovative approaches to tackle large-scale multi-objective feature selection. The first technique involves an augmentation to the diversity of the population in the well-established multi-objective scheme of the genetic algorithm, NSGA-II, which is achieved through the substitution of the worst individuals with new randomly generated individuals with a limited number of features in each generation. As the second method, a binary Compact NSGA-II (CNSGA-II) algorithm has been introduced for feature selection for the first time, which represents the population as a probability distribution not only to be more memory-efficient but also to accelerate finding a better candidate solution. Additionally, we present a novel binary multi-objective coordinate search (MOCS) algorithm, which, to the best of our knowledge, is the first of its kind, demonstrating effectiveness in solving multi-objective binary optimization problems. A comparative study on the proposed methods and NSGA-II showcases the promising performance of our methods in feature selection tasks involving large-scale datasets, surpassing the renowned NSGA-II algorithm. These methods are not limited to feature selection but can be applied to various binary optimization domains, including the Knapsack problem and training Binary Deep Neural Networks.
Experimental investigations of the thermodynamic properties of Nd-C and Ce-C binary systems for TRISO fuel applications
(2024-04-01) Varga, Ryan Mathew; Atkinson, Kirk; Piro, Markus; Fitzpatrick, Bernie
TRISO fuels are proposed for modular and niche nuclear power reactor technologies, expanding the global nuclear power inventory to meet increasing energy demands through small scale applications. Expansion is dependent upon the improvements to safety through thorough understanding of fission product behaviour, studied and analyzed here using experimental techniques and comparisons to existing literature. Significant knowledge gaps exist in the thermodynamic behaviour of neodymium carbide and cerium carbide fission products that play significant roles in the qualification of TRISO fuels. Thermodynamic investigations of neodymium and cerium carbide isotopic equivalent fission products were performed to improve the knowledge base of TRISO fuels. Various crucible tests, calibrant experimentation, sample generation and sample preparation techniques, and new thermodynamic measurements have been performed. Boron nitride crucibles worked most effectively, providing useful data for comparison to existing literature. Measurements of varying molar compositions of neodymium carbide and cerium carbide provided new data to expand upon predicted phase change boundaries, with experimental phase changes exhibiting comparatively higher transition temperatures.
A hybrid compensation-based misalignment tolerant wireless power transfer system for e-mobility
(2024-04-01) Shrestha, Niranjan; Williamson, Sheldon
The thesis focuses on developing a hybrid compensation with a phase shift control strategy, aiming for misalignment tolerant constant current/constant voltage (CC/CV) charging through a wireless power transfer (WPT) system for e-mobility. The thesis proposes a hybrid multi-resonant compensation network (LCC-LCC and LCC-S) for CC/CV charging during perfect alignment, controlled by the secondary side only. Additionally, the thesis introduces a phase shift control technique in the inverter to maintain the corresponding CC and CV charging mode during the misalignment up to 100 mm between primary and secondary coils. Initially, the theoretical analysis of the proposed system is described in detail. Then, simulation results for 3.7 kW and 270 W peak load were carried out in MATLAB Simulink. Lastly, experimental testing and validation were conducted for the proposed hybrid compensated system for 270 W peak load, applicable to the E-bike. The experimental results show good consistency with theoretical and simulation analysis.
Coupled radiation transport and mobile depletion for health physics and environmental impact applications
(2024-04-01) Sawatzky, Kevin; Atkinson, Kirk D.
The next generation of nuclear reactors pose a large challenge to existing computational methods. Caribou is a MOOSE-based health physics and environmental impact code under development at Ontario Tech which aims to address some of these challenges. This work developed a discrete ordinates radiation transport solver and a radionuclide trace species transport solver in the MOOSE framework for Caribou. This radiation transport solver is compared to several benchmark problems to determine its accuracy with and without ray effect mitigation measures, finding good agreement with all the problems tested. The trace species transport solver is then verified with the method of manufactured solutions. The coupled solvers are then used to analyze the formation of ⁴¹Ar in a containment volume due to ex-core neutron fields, and the photon fields from a ¹³⁷Cs plume. The conclusions of this work indicate that these methods are a valuable addition to Caribou’s suite of capabilities.
Design and development of an LLM-based framework for crime classification and prediction
(2024-04-01) Sarzaeim, Paria; Mahmoud, Qusay H.; Azim, Akramul
Large language models (LLMs), as a subset of generative AI, have been used in different domains, such as financial, medical, legal, and agricultural applications. However, adopting LLMs for smart policing applications remains unexplored. This thesis concentrates on developing a framework based on the transformative potential of the BART, BERT, and GPT models in this domain using methods such as zero-shot prompting, few-shot prompting, and fine-tuning. As a prototype, these methods were used to comprehensively assess the performance of LLMs in crime classification and prediction based on state-of-the-art datasets from two major cities: San Francisco and Los Angeles. The main objective is to illuminate the adaptability of LLMs and their capacity to revolutionize crime analysis practices. Additionally, a comparative analysis of the aforementioned methods on the GPT model and BART with machine learning (ML) techniques is provided. The experimental results demonstrate the feasibility of integrating LLMs into smart policing systems and show that GPT models are more suitable than traditional ML models for crime prediction in most experimental scenarios.
Design and development of a framework for predicting short price jumps in cryptocurrency market
(2024-04-01) Rajaei, Mohammadjavad; Mahmoud , Qusay H.
The cryptocurrency market's volatility offers significant profit opportunities despite its risks. This thesis aims to capitalize on these opportunities by designing and developing a framework to predict whether a coin will experience growth in the next trading candle. To achieve this, we constructed a robust framework that incorporated various input features. We conducted comprehensive analyses by leveraging six machine learning models. Our methodology involves training these models on historical daily data from the Binance Exchange. Subsequently, we evaluate their performance using diverse testing datasets from January 2022 to December 2023. Demonstrating notable precision, especially with a growth rate of 1%, the model has proven effective across various scenarios, consistently yielding profits. Regarding the backward testing results, the XGBoost model combined with the trading strategy made a profit in all three test datasets of Oct 2023, Jul to Sep 2023, and LSK/USDT achieved 24%, 47%, and 98% profits, respectively. This thesis underscores the potential for leveraging well-designed machine learning models to earn significant profits, even in bearish market conditions.
Modeling and testing of an improved 8x8 scaled electric combat vehicle
(2024-04-01) Kim, Junwoo; El-Gindy, Moustafa
The current 8x8 combat vehicle pertaining to multi-wheeled vehicles indicates a lack of effort in developing advanced multi-steering electric drive vehicle. Therefore, novel steering scenarios for an 8x8 scaled electric combat vehicle that features maintaining individual wheel’s steering and speed are developed in this thesis using a scaled 8x8 compat vehicle to introduce a future steering control system for the current combat vehicles. This thesis explains the mechanical improvement of the scaled vehicle compared to the previously developed model. In addition, validation of the scaled electric vehicle model in comparison with a full-size electric drive vehicle model in TruckSim is presented to prove the advantage of using the scaled vehicle in experimental test. Furthermore, development and validation of steering strategies including traditional, fixed 3rd axle, and all-wheel steering scenarios in the scaled vehicle in terms of both experiment and simulation results are conducted to enhance the limitations imposed by conventional vehicle’s steering control system. Active Steering Controller (ASC) is developed and simulated to overcome the limitations of the complete Ackermann steering condition when the scaled vehicle is driving at a relatively high speed. The outcome of this thesis research work is a novel future electric multi-drive and multi-steer combat vehicle.
Modeling and analysis of Regional Haul Steer (RHS) truck tire model
(2024-04-01) Khosravi, Mehran; El-Gindy, Moustafa; El-Sayegh, Zeinab
In this thesis, the RHS truck tire size 315/80R22.5 was developed using Finite Element Analysis and several material properties. The tire model was then validated using static and dynamic testing, against physical measurements provided by the manufacturer. A simulation model of flooded and snow terrain was then developed using the Smoothed-Particle Hydrodynamics technique and hydrodynamic elastic-plastic material model. The tire-terrain interaction characteristics were then evaluated over flooded and snow surfaces. The interaction characteristics included the rolling resistance, cornering force, self-aligning moment, and overturning moment. The analysis was performed at different operating conditions including terrain depth, longitudinal speed, and vertical loads. In general, the results from both surfaces exhibited similar trends, even though the values were not the same. Future work involves the utilization of genetic algorithms to generate semi-empirical relationships, as well as the implementation of temperature and wear models for the RHS tire.
Autonomous UAV-UGV robot collaboration for exploration and mapping of unknown environments
(2024-04-01) Khabbaz, Noor; Nokleby, Scott
This thesis addresses the limitations of existing approaches to autonomous exploration and mapping of unknown environments that use multiple Unmanned Ground Vehicles (UGVs). An Unmanned Aerial Vehicle (UAV) is introduced into the multirobot system to overcome the challenges of relative localization and obstacle detection. A novel method is proposed for autonomously determining the UGVs’ starting poses using ArUco markers visible to the UAV, resulting in the initialization of a global merged map. A second method is developed to overcome UGV obstacle detection limitations. UAV depth camera data is processed to detect and incorporate previously unseen obstacles into the UGVs’ navigation schemes, enabling avoidance. Experimental validation demonstrates the effectiveness of both methods in enhancing system autonomy. The integration of a UAV into multi-robot systems presents a promising solution to address UGVs’ localization challenges and limited field of view to improve their functionality in hazardous environments.
Analysis of microgrid with renewable generation and energy storage system
(2024-03-01) Jadeja, Prachal; Sood, Vijay
The widespread integration of converter-fed renewable energy sources (RESs) and supported by energy storage systems (ESSs), along with their associated challenges, is causing a pressing need to reconsider the operation of distribution networks. Microgrids (MGs) appear as a practical solution to accommodate these RESs and ESSs. This thesis conducts an in-depth analysis of the design and modeling of a generic MG that integrates RESs and ESSs, incorporating a fuzzy logic controller (FLC)-based energy management system (EMS). The MG operation is evaluated in both grid-connected and islanded modes. A frequency analysis reveals that the point of common coupling (PCC) frequency is controlled within the operational range of ± 0.3 Hz, according to IEEE Standard I547. This control is achieved using a combination of battery and supercapacitor ESS in both operational modes. A comparative evaluation of Proportional Integrator (PI) v/s Proportional Resonant (PR) controllers for current controllers for the MG to reduce the amount of total harmonic distortion (THD) at PCC is conducted. Dynamic testing of MG is also conducted under various loading conditions, encompassing small-, medium-, and large-signal changes. The results prove that the PCC voltages are effectively kept within the typical range of ± 0.05 pu through the FLC-based EMS.
Comprehensive integration of safety, optimization, and regulation in ¹⁷⁷Lu-based theranostic radiopharmaceuticals: from production to dose assessment
(2024-03-01) Haghi, Pardis; Waller, Edward
In recent decades, medical radioisotopes have transformed cancer diagnosis and treatment. The advent of theranostic radionuclides, enabling both diagnostic and targeted radiotherapy with a single radionuclide, merges two nuclear medicine fields, enhancing personalized targeted radiotherapy through internal dosimetry. Among these agents, the short-lived β-emitter ¹⁷⁷Lu is a promising theranostic agent for various oncological diseases. This research covers advanced production techniques, radiation safety, transportation, Canadian regulations, and internal dose assessment in radiopharmaceutical clinical applications. The study reviews regulations for obtaining a license from the Canadian Nuclear Safety Commission for a radiopharmaceutical therapy facility. It also examines internal dose assessment in nuclear medicine, explaining each step and providing an introduction to common software codes, along with a comparison of commercial software packages. Two case studies compare software codes and internal dose assessment results, emphasizing the ongoing challenge of determining accurate organ-absorbed doses in the RPT process. The ultimate goal is to comprehensively integrate key aspects of ¹⁷⁷Lu targeted therapy into a unified overview, bridging a crucial gap in understanding.
A Graph Neural Network for pairwise surrogate modeling in population-based algorithms with tournament selection
(2024-04-01) Gharavian, Vida; Makrehchi, Masoud; Rahnamayan, Shahryar
Optimization problems widely arise in various science and engineering fields. Optimisation involves evaluating a candidate solution, which can be computationally intensive. Machine learning-based surrogate models can contribute to learning the specific pattern among the decision variables and objective values to reduce the computation time of fitness evaluation. In this study, we have proposed a novel pairwise surrogate model to identify the superiority between candidate solutions in a pairwise comparison. We demonstrated a Graph Neural Network (GNN) to be trained on number of pairs, then utilized to compare a pair of candidate solutions. To examine the efficacy of our model, we utilized the surrogate model on CEC2017 benchmarks in different dimensions. Moreover, the result of surrogate-assisted and none-assisted form of two well-known optimization algorithms were compared. Results show that the proposed method can significantly reduce the computing cost. In the presence of higher dimensions, our model is more effective than most surrogate models for comparison-based optimizers.
Requirements engineering-driven collaborative software maintenance framework for embedded systems using continual learning
(2024-04-01) Fariha, Asma; Alwidian, Sanaa; Azim, Akramul
Embedded software post-deployment evolutions pose significant threats to the safety and reliability of embedded software if it is not adapted to software maintenance through requirements engineering. To solve this problem, we propose a collaborative framework that enables efficient requirements elicitation and continuously integrates it into maintenance. We designed a requirements forum to enhance elicitation through centralized stakeholder collaboration. This study investigated fault and failure detection in the maintenance phase with continual learning as a mechanism of incremental inclusion. The novel CNNBiLSTM deep-learning model on a public drone dataset outperformed state-of-the-art models, achieving a 100% true positive rate in three scenarios. On the other hand, we experienced a 14% increase in the recall metric for the replay-based method combined with pre-training compared to pre-training when fault detection requirements were integrated incrementally. Our findings support the idea that embedded software safety and security can be greatly enhanced through this collaborative framework.
A novel approach for route generation and real-time scheduling for public services
(2024-04-01) Baghyari, Farhad; Seo, Jaho
Snowplowing and sweeping are essential services to municipalities, which affect travel safety, environment protection, and health to residents. To provide acceptable quality services, route optimization is one of the key strategies that allow for enhancing efficiency, saving costs, and balancing workloads among operational teams. In order to address this issue and reflect on recent research trends in routing problems that require variable conditions and real-time events, this study proposes two heuristic methods: Smart Selective Navigator and a two-stage algorithm for real-time scheduling and route generation. Through two major case studies — winter operations in the City of Oshawa and autonomous street sweeping in Uchi Park —the proposed methods demonstrate superior performance in generating optimal routes that satisfy complex constraints such as turn restrictions and supply limits and handle real-time events like vehicle breakdowns.
Identification of Java lock contention anti-patterns based on run-time performance data using machine learning algorithms
(2024-02-01) Ahmed, Aritra; Liscano, Ramiro; Azim, Akramul
Locks are critical in concurrent Java applications, ensuring synchronized access to shared resources. Mismanagement of locks and threads can result in contention, affecting performance and scalability. Eight Java lock contention anti-patterns have been identified, representing various scenarios of lock contention within intrinsic locks. Traditional methods for identifying these faults rely on legacy tools and experience. This study evaluates machine learning algorithms’ effectiveness in detecting lock contention anti-patterns. When the algorithms were trained and tested with a 70-30 split we obtained an accuracy above 90%. To validate the findings we used the Dacapo benchmark as the testing set and found XGBoost performing well among the rest with an overall accuracy of 87%. Logistic Regression, Random Forest and Support Vector Machine were the best performing models in terms of precision and recall values. We also validated our recommendations provided by comparing three important performance metrics between the original and refactored version of the anti-patterns.
Development and validation of scaled electric combat vehicle virtual model
(2023-11-01) Vaz, Glenn Xavier; El-Sayegh, Zeinab
This research focuses on an 8x8 scaled electric combat vehicle (SECV) and aims to create a virtual model made of the same vehicle on a vehicle dynamics simulation software using parameters from the actual vehicle. In the proposed vehicle, each wheel is independently driven and steered. MATLAB and Simulink software were used to design and implement the electric powertrain while TruckSim Modelling and Simulation software was used to simulate the on-road conditions tests. The simulation data was then compared with the experimental data obtained from the physical test scenarios.
Filtering antennas based on rectangular dielectric resonators
(2023-12-01) Su, Xiangqi; Wang, Ying
Filters and antennas are essential components in communication systems. Being able to simultaneously realize filter and radiation functions, filtering antennas have attracted much attention due to their compactness and low loss. This thesis focuses on different methods of realizing compact filtering antennas using rectangular dielectric resonator antennas (DRAs). First, four designs of DRA to increase bandwidth have been proposed. Low dielectric constant posts are inserted in DRA in different orientations so that the effective dielectric constant can be readily lowered and adjusted. An integrated design of a waveguide dielectric resonator filter and a DRA with an air post shows significant bandwidth improvement. Next, conducting posts are inserted in the DRA to realize radiation cancelation. Radiation nulls, or filtering responses, have been achieved for both linear polarization and dual polarization. All designs have been simulated and optimized using a full wave electromagnetic (EM) simulator, and results are compared and discussed.
A universal centralized protection scheme for feeders of AC microgrids using IEC-61869-9 digital instrument transformers
(2023-10-01) Sharma, Jigyesh; Sidhu, Tarlochan Singh
This thesis underscores the escalating importance of microgrids powered by renewable energy sources and inverters in modern power systems. Despite offering significant environmental and economic advantages, their decentralized and dynamic nature poses unique protection challenges. Traditional protection methods struggle to adapt to the diverse conditions within microgrids. Past protection techniques relying on current or voltage detection have limitations affecting system reliability and security. To address this, the thesis proposes a pioneering protection approach based on ’discrepant impedance.’ This concept calculates impedance disparities using both feeder ends’ positive sequence voltage and current measurements. This value approximates zero during normal operation but deviates during a fault, enabling effective fault detection. The proposed protection philosophy was rigorously tested through simulations, real-time experiments using RTDS®, and validation on the IEEE-9 bus system with inverter-based resources, a benchmark system by the National Renewable Energy Laboratory National Renewable Energy Laboratory (NREL). Comparative assessments with traditional methods underscore the effectiveness and adaptability of the discrepant impedance-based protection scheme. The thesis concludes by discussing practical implementation options, showcasing the approach’s versatility across varied microgrid configurations and control strategies, and ultimately demonstrating the feasibility and effectiveness of discrepant impedance-based feeder protection for modern microgrid systems.
Electroforming of personalized miniature metal parts using additively manufactured molds
(2023-12-01) Hamed, Hazem Fawzi; Abou-Ziki, Jana
In response to evolving manufacturing trends favoring personalized, small-batch production, this thesis centers on the development of additively manufactured molds to facilitate the electroforming of personalized metal parts. The methodology encompasses standardized mold design, experimental procedures for mold development and electroforming, and a simulation model for visualizing and predicting the deposition process. The study provides critical design considerations and guidelines for electroforming within additively manufactured molds, successfully demonstrating the production of composite metal components in 2.5D and 3D configurations. Emphasizing cost efficiency and improved part quality, especially for limited-thickness metal components, the developed technique presents advantages over available metal additive manufacturing processes. Electroforming emerges as a versatile and robust metal additive manufacturing technique, expanding its application beyond traditional limitations of thin-walled hollow structures, 2D components and applications at the nanoscale.
Investigation of signal shape effects on the gas film in spark-assisted chemical engraving
(2023-12-01) Eldiasty, Marwan S; Abou-Ziki, Jana
Spark-Assisted Chemical Engraving (SACE) is a promising method for machining glass micro-parts and devices. However, intricate control requirements linked to the gas film surrounding the tool present a significant challenge in SACE. While several studies have explored the influence of SACE parameters on the gas film, there exists a literature gap regarding the impact of voltage signal shapes on this film. The thesis fills this void by investigating diverse voltage signal shapes designed to enhance the gas film stability. A robust methodology was established linking gas film properties to investigate the effects of signal shapes on the gas film. The research applied these findings to machining applications, establishing correlations between signal shapes and machining outcomes. Key contributions include a refined methodology for gas film evaluation, advancements in understanding signal shapes’ impact on the process, identification of optimal parameters, and potential improvements in machining through a custom signal shape design.

Browse

Recent Submissions