DalSpace Institutional Repository
DalSpace is a digital service that collects, preserves, and distributes digital material produced by the Dalhousie community.
- To learn about content guidelines, policies, and how to deposit, view the Help documents.
- Contact us to get started submitting content to DalSpace at dalspace@dal.ca
Communities in DSpace
Select a community to browse its collections.
Recent Submissions
TRAINING AND EVALUATING THE USE OF LARGE LANGUAGE MODELS (LLMS) IN THE DOMAIN OF CANADIAN NUCLEAR INDUSTRY
(2025-07-10) Anwar, Muhammad Saleh; Not Applicable; Master of Science; Department of Engineering Mathematics & Internetworking; Not Applicable; N/A; Not Applicable; Dr. Guy Kember; Dr. Kamal El-Sankary; Dr. Issam Hammad
This thesis addresses the challenges of accuracy, reliability, data privacy, and resource constraints in applying Large Language Models (LLMs) to the Canadian nuclear industry. It presents a multi-faceted approach by evaluating existing models, developing synthetic data generation techniques, and training a secure, domain-specific LLM from scratch.
The research first demonstrates that while general-purpose LLMs are prone to factual inaccuracies on nuclear-specific topics, their reliability is significantly improved by integrating a Retrieval-Augmented Generation (RAG) framework. This approach enhances factual accuracy by grounding responses in verified, domain-specific documents.
To overcome data scarcity and confidentiality barriers, the thesis pioneers a methodology for generating synthetic, structured question-and-answer pairs from unstructured nuclear texts using LLMs. This scalable and privacy-preserving approach creates valuable, model-ready datasets for training and evaluation without exposing sensitive information.
Furthermore, the work validates the feasibility of developing a secure, private LLM from scratch. By training a compact model on a single GPU using the "Essential CANDU" textbook, it demonstrates a practical path for creating in-house models that mitigate cybersecurity risks and can learn specialized terminology within a resource-constrained and secure environment.
Collectively, this research provides a comprehensive framework for integrating LLM technology safely and effectively into the nuclear industry, establishing a foundation for advanced AI tools that enhance knowledge management and operational support.
Model Predictive Control for Autonomous Sailboat with Sail Angle Optimization and Trajectory Planning Under Stochastic Winds
(2025-07-14) Wu, Junzhuo; Not Applicable; Master of Applied Science; Department of Mechanical Engineering; Not Applicable; N/A; Not Applicable; Dr. Baafour Nyantekyi-Kwakye; Dr. Hamed Aly; Dr. Ya-Jun Pan; Dr. Chao Shen
This thesis investigates the development and implementation of model predictive control (MPC) frameworks for autonomous sailboat control under different wind conditions. The research is motivated by the need for robust, constraint-aware control strategies for sailboats, that can effectively handle the nonlinear dynamics and environmental uncertainties.
The study begins with a comprehensive modelling of sailboat dynamics, incorporating sailboat kinematic and kinetics and wave disturbances. Two simulation-based studies both using MPC are then conducted. The first study focuses on sail angle optimization and trajectory tracking, where the control objective is to maximize sailing speed with safety considerations and then achieve accurate trajectory tracking. The second study addresses stochastic wind sailing, where an NMPC-based path planning and tracking controller is designed and evaluated. This controller integrates the controllability analysis to improve performance under both deterministic and stochastic wind conditions. Comparative simulations highlight the advantages of the proposed approach. Then, an experimental platform is instrumented on a small-scale physical autonomous sailboat. This platform integrates the ArduPilot control stack, hardware components and supporting software infrastructure. System identification techniques are applied to extract sailboat’s dynamic model, and experimental trials are conducted to assess the controller performance in real-world conditions.
The results demonstrate that MPC offers a viable and effective solution for autonomous sailboat control, capable of addressing both operational constraints and environmental variability.
Beyond Monocular Vision: Assessing LLaVA's Performance on an Augmented CLEVR-like Dataset with Binocular Images
(2025-07-07) Devesh, Sagar; Not Applicable; Master of Computer Science; Faculty of Computer Science; Not Applicable; n/a; Not Applicable; Vlado Keselj; Hassan Sajjad; Frank Rudzicz
This thesis investigates how binocular vision impacts the spatial reasoning capabilities of Large Language and Vision Assistant (LLaVA) models in visual question answering tasks. By developing BiCLEVR, an augmented CLEVR-like dataset featuring stereoscopic image pairs and expanded visual attributes, we systematically evaluate the effect of different visual inputs across varying model sizes. Our experiments compare two LLaVA variants (7B and 13B parameters) across three dataset configurations: standard CLEVR, monocular BiCLEVR, and binocular BiCLEVR. Results reveal a nuanced relationship between model capacity and the ability to leverage stereoscopic information. The larger model demonstrated significant performance improvements with binocular input, while the smaller model showed degraded performance, suggesting insufficient capacity to process the additional visual information effectively. Particularly notable were improvements in numerical comparison and counting tasks for the larger model, indicating that stereoscopic cues enhance object individuation abilities. These findings contribute to our understanding of how vision-language models process spatial information and provide a pathway toward more robust visual reasoning systems capable of understanding 3D relationships in complex environments.
Active Removal of Tumbling Orbital Debris Using an Autonomous Chaser Vehicle in the Presence of Perturbations
(2025-07-02) Adolph, Mathew; Not Applicable; Master of Applied Science; Department of Mechanical Engineering; Not Applicable; n/a; Not Applicable; Dr. Guy Kember; Dr. Darrel Doman; Dr. Mae Seto
To address and abate the issues related to uncontrolled debris in orbit around Earth, active debris removal (ADR) is imperative. To perform ADR, a chaser satellite is used to rendezvous and capture the debris. Docking with debris is difficult as it is rotating about multiple axes; this is known as tumbling. This thesis details the development of an autonomous flight control strategy that attempts to synchronize and dock with tumbling debris in orbit to facilitate capture. The control strategy employs a model predictive control algorithm to predict the future state of the chaser and debris while simultaneously layering a separate closed-loop attitude controller to orient the chaser towards the debris. The control strategy was shown to successfully dock to debris tumbling at 2.6 deg/s with a 0% failure rate when at an initial separation of less than 25 m from the debris. The chaser was also able to reliably dock in less than 5 minutes to debris tumbling up to 4.5 deg/s when initially positioned 2.45 m from the debris. This research demonstrates that a single control scheme can potentially be used to successfully service a wide variety of candidate debris missions and docking conditions. The control strategy also demonstrates disturbance rejection, including recovery from a simulated micro-asteroid impact. The implemented algorithms were developed to integrate with a planar air-bearing testbed for future validation of the control strategy using hardware.
FRONTAL PLANE LANDING MECHANICS AND GLUTEUS MEDIUS MUSCLE ACTIVITY IN INDIVIDUALS RETURNED TO SPORT POST ANTERIOR CRUCIATE LIGAMENT RECONSTRUCTION
(2025-07-09) Galloway, Ewan; Not Applicable; Master of Science; School of Physiotherapy (Rehabilitation Research); Received; Dr. Christopher MacLean; Not Applicable; Dr. Rebecca Moyer; Dr. Scott Landry; Dr. Derek Rutherford
Knee injuries, particularly to the anterior cruciate ligament (ACL), are common and can negatively impact joint health. These injuries often occur during high-impact movements, such as landing from a jump. While jump landings are used in rehabilitation and return-to-sport protocols, current assessments lack sensitivity in identifying individuals at risk. This thesis investigated differences in peak frontal plane projection angle (FPPA) and gluteus medius (GMed) activity during single-leg landings, as well as the relationship between FPPA, GMed activity, and hip abductor strength. Twenty-five individuals post-ACL reconstruction (ACLR) and 25 asymptomatic controls performed single-leg drop landings and maximal voluntary isometric contractions. No significant differences in peak FPPA were found between groups, but the affected limb in the ACLR group showed greater FPPA than the unaffected limb. No group differences were observed in GMed activity. However, lower GMed activity and hip abductor strength were significantly associated with greater FPPA.
Enhancing the Monogastric Gut Microbiome Through Innovative Nutritional Strategies
(2025-07-05) Lu, Jing; No; Doctor of Philosophy; Department of Animal Sciences and Aquaculture; Received; David Huyben; Yes; Renee Petri; Beth Mason; Vasantha Rupasinghe; Stephanie Collins
The host and microbiome can be viewed as one integrated system, which highlights the need to optimize nutrition for both the host and its gut microbiome. This thesis explored the use of nutritional strategies as selective forces to support gut microbiome diversity and resilience and ultimately improve host health and adaptability, particularly in those facing loss of host genetic and microbial diversity. Two model species, chickens (Studies 1 and 2) and polar bears (Study 3), were used to investigate gut microbiome modulation from developmental (early life vs. adult stage) to ecological (agriculture vs. conservation) contexts. Study 1 investigated reinforcing deterministic selection during early life to facilitate a lasting gut microbiome modulation through the priority effect in broiler chickens. In ovo delivery of seaweed polyphenols significantly reduced the abundance of a necrotic enteritis-causing genus, Clostridium sensu stricto 1, in the ileum of broiler chickens by day 28 post-hatch, similar to long-term supplementation of in-feed antibiotics. Study 2 examined the role of diet as a strong selection force during the production phase in two commercial strains of laying hens. Dietary inclusion of black soldier fly larvae meal (BSFLM; 0%, 6.5%, and 13%) significantly increased cecal microbial diversity and shifted short-chain fatty acid profiles toward higher acetic acid production. The 13% BSFLM inclusion increased nitrogen and ammonia excretion, which was alleviated by protease supplementation, suggesting that increased microbial diversity may reflect suboptimal protein utilization and proliferation of proteolytic taxa. Study 3 shifted focus to wildlife and compared the fecal microbiome of wild and captive bears to understand the influence of environmental factors. Captive bears had greater fecal microbial diversity, and a distinct community structure compared to wild bears. Individual variation was the main driver of microbial differences among captive bears. Captive bears fed seaweed, a natural dietary item for wild bears, showed minimal change in fecal microbiome. Together, these findings demonstrated the potential of gut microbiome-informed nutritional strategies to promote animal gut microbial resilience across life stages and in both agricultural and conservation settings. Microbial diversity should be interpreted with functionality and interaction with the host to fully understand its implications.