21793 research outputs found
Sort by
Parametric Hydrogen Tank Sizing Model for Aircraft Conceptual Design
Get PDFHydrogen-based propulsion is a promising research area for reducing aviation’s CO2 emissions. However, hydrogen’s low volumetric efficiency makes this alternative fuel challenging in terms of storage and integration into the aircraft. Developing parametric models becomes essential to allow sizing and exploration of possible hydrogen storage tank configurations and layouts within the aircraft. This thesis introduces a model and illustrates analysis capabilities, such as the effect of the variation in geometry on the overall tank volume and mass, which are key aspects to consider in conceptual design. The work presented here combines various methods from the literature into a new parametric approach considering filling and venting pressures, tank geometrical constraints, material properties, and thermal conditions. It supports the analysis of various tank configurations and layouts and allows trade-off studies on gravimetric and volumetric efficiencies to be conducted at the aircraft level. Validation was performed with industry storage tank data and showed an acceptable error range. Case studies illustrate the model capabilities for future hydrogen-powered aircraft configurations. These include tank geometry, thermal insulation, and tank layout-focused studies, allowing the exploration of various aspects of the hydrogen storage system design. Overall, the proposed tank sizing model supports the definition of a feasible design space for future, more environmentally friendly aircraft designs
Developing UMAC: A Unified Model-Agnostic Computation Process for Enhanced Machine Learning Explainability
Get PDFThe rapid evolution of Convolutional Neural Networks (CNNs) has produced increasingly efficient and versatile algorithms, but the factors driving their superior performance remain underexplored. While previous research has primarily focused on explaining Semi-Supervised Machine Learning (SSML) algorithms in a model-specific manner, this thesis aims to generalize those findings, making them applicable across a wider range of CNNs. The challenge lies in achieving a method that can both enhance performance and improve interpretability, while remaining adaptable to various models.
This thesis introduces a post-hoc Explainable Artificial Intelligence (XAI) method, called Unified Model Agnostic Computation (UMAC), designed to generalize common components of CNNs by drawing insights from SSML and Self-Supervised Learning (SSL) algorithms. Our research begins by focusing on two primary aspects: (1) the effect of parameter updates during training on both labeled and unlabeled data in SSML and SSL, and (2) the transition from model-specific SSML frameworks to a more generalized, model-agnostic approach using SSL.
In the first phase, we used SSML as a foundation, breaking down their components into preprocess-centric and classifier-centric elements, which led to the creation of SSCPs (Semi-Supervised Computation Processes). These processes were tested across five state-of-the-art SSML algorithms and three SSL algorithms, using various Deep Neural Networks (DNNs). Although this phase acted as a testing ground to understand the mechanics of SSML, it allowed us to identify key drivers of performance, especially in relation to parameter updates and data handling.
Through 45 rigorous experiments, we observed an 8% reduction in training loss and a 6.75% increase in learning precision using the Shake-Shake26 classifier with the RemixMatch SSML algorithm. A key observation was the positive correlation between labeled data and training time, showcasing the importance of label quantity in enhancing model efficiency.
In the second phase, we transitioned from SSML to SSL to prove that the methodology could be generalized to a model-agnostic approach. By integrating SSL components, we aimed to develop a unified framework that worked across various DNN architectures. Building upon this analysis, we developed a UMAC process for SSL, tailored to complement modern self-supervised learning algorithms. UMAC serves as a model-agnostic XAI methodology that explains models by composition, systematically integrating and enhancing state-of-the-art algorithms. Through UMAC, we identified key computational mechanisms and crafted a unified framework for self-supervised learning evaluation. Our systematic approach yielded a 17.12% improvement in training time complexity and a 13.1% boost in testing time complexity, with notable improvements observed in augmentation, encoder architecture, and auxiliary components within the network classifier. This phase demonstrated that UMAC could enhance accuracy and reduce training loss under different data conditions, showing its adaptability to different models and datasets.
In the third phase, we applied the UMAC framework to the field of medical image classification. Medical imaging tasks often suffer from data scarcity, making it challenging to achieve both high performance and model interpretability. By leveraging the UMAC methodology, we integrated it into CNNs and Transformers to generate high-quality representations, even with limited data. Experiments across five 2D medical image datasets showed that UMAC outperformed traditional augmentation methods by 1.89% in classification accuracy. Additionally, incorporating explainable AI (XAI) techniques ensured that the models provided transparent and reliable decision-making processes, enhancing their interpretability in critical medical applications.
Throughout this process, UMAC served as an XAI method based on explaining models by composition, systematically breaking down computational processes to reveal how model components contribute to overall performance. This approach enabled us to create a unified, model-agnostic framework that enhanced both transparency and efficiency in CNNs.
Ultimately, this thesis contributes a structured and generalizable approach for Machine Learning (ML) developers, offering step-by-step guidelines to improve model performance and interpretability. By generalizing the computation processes of SSML and SSL through the UMAC framework, we provide developers with the tools needed to optimize their models across various domains, particularly in fields where transparency and accuracy are critical
Design of a Systematic Comprehensive Integrative Indicator-based Sustainability Assessment Framework for Organizations
No full textIn response to the urgent need for organizations to comprehensively assess their sustainability performance, this research introduces a novel approach through the development of the Systematic Comprehensive Integrative Indicator-based Sustainability Assessment Framework (SCII-SAF). This framework addresses the limitations of existing sustainability assessment frameworks (SAFs) by providing a systematic and integrative method for selecting, evaluating, and understanding sustainability indicators (SIs). A thorough literature review of SAFs across diverse industries reveals a critical gap: the lack of a universal method for identifying the most relevant indicators and assessing their interdependencies without heavy reliance on subjective expert judgment. Current multi-criteria decision-making models, such as AHP, DEMATEL, TOPSIS and VIKOR, offer valuable insights but suffer from their dependence on expert opinion, leading to subjective assignments of importance. To address this, this research proposes an innovative data-driven approach that integrates correlation analysis and network analysis for evaluating static relationships and interdependencies, and system dynamics simulation to capture dynamic changes and the performance impact of SIs over time. This hybrid methodology, combining graph theory with machine learning, represents a unique approach not previously applied in academic literature. By minimizing reliance on expert judgment, the SCII-SAF enhances objectivity and analytical rigor. The SCII-SAF model provides a groundbreaking, data-driven framework that aligns SIs with long-term sustainability objectives, offering decision-makers valuable insights. It facilitates informed decision-making by quantifying trade-offs between economic, social, and environmental dimensions of sustainability, tailored to specific industry needs. By integrating Network Analysis for structural understanding and System Dynamics Simulation for temporal and dynamic insights, this research not only addresses significant limitations in existing SAFs but also represents a transformative contribution to sustainability assessment practices
Mechanochemically enabled methods for sustainable and scalable synthesis of small molecules and functional materials
No full textMechanochemistry, which induces chemical reactions via mechanical agitation, offers a promising alternative to traditional solution-based synthesis by reducing solvent use and chemical waste. This thesis investigates mechanochemistry-specific methods for the sustainable and scalable synthesis of small molecules and functional materials.
Chapter 2 employs resonant acoustic mixing (RAM) to perform mechanoredox diazonium couplings, with BaTiO3 as the piezoelectric catalyst. RAM proceeds without formal grinding or impact media and is readily scalable compared to ball-milling. X-ray diffraction and spectroscopy indicate that reusability of BaTiO3 as a mechanoredox catalyst under ball-milling or RAM is limited by boration.
Chapter 3 introduces the concept of direct mechanoredox catalysis by utilizing a piezoelectric organic polymer, polyvinylidene difluoride (PVDF). PVDF, transformed into a piezoelectric phase, is used to create catalytically active reaction vessels, coatings, and inserts, enabling mechanoredox reactions under ball-mill and RAM conditions. Using aryldiazonium salt borylation as a testbed, five methods for PVDF-based catalysis are explored: (i) PVDF as a powder additive, (ii) a milling jar, (iii) a coating on conventional jars, and a strip insert under (iv) ball-milling or (v) RAM conditions.
Chapter 4 demonstrates a rapid, room-temperature mechanochemical synthesis of 2 and 3 dimensional boroxine covalent organic frameworks (COFs), enabled by using trimethylboroxine as a dehydrating reagent to overcome the hydrolytic sensitivity of boroxine based COFs. This approach achieves high-porosity, crystalline COF, including the first 3-D COF synthesized mechanochemically. Compared to solvothermal methods, this method reduces solvent consumption 20-fold and reaction time 100-fold, delivering quantitative yields with minimal workup. Straightforward scale-up by RAM enables synthesis of multi-gram amounts of the target COFs.
Chapter 5 investigates RAM-assisted topochemical [2+2] photocycloaddition reactions of trans 1,2 bis(4 pyridyl)ethylene to synthesize cyclobutane moieties in the solid state. Appending ultraviolet light-emitting diodes to the RAM instrument enables templated cocrystallization and photocycloaddition to proceed in situ in one pot. This study specifically reports the supramolecular catalysis potential of four templates, including citric acid. Notably, we show that the reaction works when using lemon juice as catalyst, thus elevating the green chemistry aspects of the methodology.
Collectively, this thesis advances mechanochemistry methods and demonstrates potential for sustainable and scalable synthesis
A Decision-Making Framework for the Built Facilities’ End-of-Life from Sustainability and Circular Economy Viewpoints
Get PDFThe construction industry generates over a third of global waste, most of which is produced at the End-of-Life (EoL) stage of built facilities and disposed of in landfills. Existing decision-making models often focus on a limited range of EoL decisions and lack stakeholder diversity, leading to biased outcomes. This study develops an inclusive decision-making model for the EoL stage. It defines a framework of four EoL sub-phases, related decision problems, and alternatives based on a systematic literature review. A set of 25 criteria was compiled, validated through interviews with 25 stakeholders, and weighted using an Analytical Hierarchical Process (AHP) survey with 52 participants. The model was tested on a two-story, 59,360 SF resort facility in Ontario using data from reports, government sources, quantitative tools, and industry partners. Four scenarios were analyzed: Full Deconstruction, Partial Deconstruction, Full Demolition with landfill disposal, and Full Demolition with recycling. Full Deconstruction emerged as the optimal solution, outperforming Full Demolition with landfill disposal by ~40%. Regulations were identified as the most critical factor, followed by Environmental aspects, ranking higher than Economics—contrary to prior studies. These findings highlight the vital role of government regulations in driving sustainable practices. The resulting model provides robust guidance for decision-makers, addressing EoL complexities and promoting sustainability and circularity in the construction industry
Comprehensive Facial Attractiveness Analysis with Stacked Regression and Geometric Angles
Get PDFThe analysis of facial beauty holds considerable importance in social interactions and has emerged as a prominent research topic in various fields. Automatic assessment of facial attractiveness has garnered significant interest due to its wide range of potential applications. This research aims to pinpoint the primary attributes contributing to beauty and assess the influence of various facial angles, such as those of the eyes, nose, lips, chin, eyebrows, and jaws. Additionally, it examines the significance of geometric facial measurements, encompassing distances between facial landmarks and ratios, in the context of beauty evaluation. The study also employs two techniques, namely Principal Component Analysis (PCA) and stacked regression, to predict the attractiveness of faces. The experimental data set used for evaluation is the well-known SCUT-FBP benchmark database (consisting of 500 facial images). The obtained results demonstrate the superiority of our method, which achieved the highest Pearson's Correlation Coefficient (PCC) of 0.8043 and the lowest Mean Absolute Errors (MAE) of 0.3068 among all the evaluated methods. Our method incorporates six angles and leverages PCA for dimensionality reduction, along with 23 distance features, resulting in improved accuracy. Furthermore, the utilization of the stacking ensemble approach instead of individual machine learning methods contributes to the impressive performance of our method. This unequivocal success underscores the pivotal role of machine learning and pattern recognition in providing an insightful and precise assessment of facial beauty
A simulation-based comparison of confidence interval coverage, bias, and variance of alternative spatial biomass estimation methods used for the evaluation of Northern Shrimp biomass
Get PDFIn fisheries management, reliable estimates of population abundance metrics such as standing biomass are crucial for sustainability and balanced decision-making. The OGive MAPping (OGMAP) method, used by the Department of Fisheries and Oceans (DFO) in Newfoundland and Labrador for biomass estimation, addresses non-normally distributed populations but raises concerns about handling spatial data variations. I conducted a simulation-based comparative analysis comparing OGMAP against Generalized Additive Models (GAMs) and STRAtified Programs (STRAP) to answer the following question: “are the uncertainties of the estimates calculated from these different methods reliable?”
Using Northern Shrimp, Pandalus borealis, as a reference, I simulated biomass landscapes, exploring parameters like landscape roughness, sampling intensity, and model settings. The analysis consistently showed OGMAP's failure to capture nominal confidence intervals (CIs) compared to alternatives, regardless of the treatment. OGMAP exhibited tighter intervals, raising concerns about overfitting and its inability to reflect the true landscape biomass. However, halving the automatically optimized bandwidths for OGMAP's probability distribution fields significantly improved its realized coverage.
These findings underscore OGMAP's variability, shedding light on its limitations in decision-making by the Department of Fisheries and Oceans. I stress the pivotal role of reliable estimates in fisheries management. Additionally, I suggest that alternative methods, like GAMs, may offer more dependable forecasts given OGMAP's underperformance. This research prompts a review of the fisheries management framework relying on OGMAP, suggesting potential inadequacies in capturing the true uncertainty associated with spatially distributed stocks
The Impact of Foreign Migration to Canada on REIT Operating Performance
Get PDFThis thesis explores the influence of international migration on the operational performance of Real Estate Investment Trusts (REITs) in Canada. This study focuses on the relationship between international migrants and the operational performance of Canadian REITs. An exposure to international migration measure is built using the weight of properties held by REITs in different provinces and the international migrants in the respective province. The findings reveal that an increase in the exposure to net non-permanent residents will increase the operational performance of Canadian REITs. Furthermore, the implementation of Bill-28 in British-Colombia and NRST in Ontario results in REITs residential exposure to net non-permanent residents decreasing the operational performance in each respective region. This thesis contributes to the understanding of how demographic shifts, specifically international migration, affect REITs operational performances
Model-Based Predictive Control Strategies and Renewable Energy Integration for Energy Flexibility Enhancement in School Buildings
Get PDFThis thesis investigates methods to enhance the energy flexibility potential of school buildings through simulation and experimental studies. It contributes a general methodology for the development of data driven grey-box thermal models and the implementation of model-based predictive control (MPC). The methodology is applied to an archetype fully electric school building near Montréal, Québec, Canada. This approach is scalable and transferable to other institutional or mid-size commercial buildings.
To streamline the implementation of MPC, the proposed approach employs grey-box low-order resistance-capacitance (RC) thermal network models, a clustering of weather conditions to identify typical anticipated scenarios, and several near-optimal setpoint profiles corresponding to each cluster. Archetype control-oriented models for zones with convective systems and zones with radiant floor systems are developed and calibrated with measured data. The calibrated models are used to apply MPC to the school building using the established dynamic tariffs for morning and evening peaks. For the experimental study, the developed MPC framework is applied in six classrooms, and the results are compared with four classrooms with the reactive control system as reference cases. The energy flexibility is quantified based on a proposed building energy flexibility index (BEFI). Results indicated that the school building can provide 45% to 95% energy flexibility (load shifting relative to reference) during on peak hours while satisfying thermal comfort constraints.
Finally, this thesis presents an MPC methodology for the integration of air-based photovoltaic/thermal (PV/T) systems to further enhance the energy flexibility in school buildings so that in addition to the production of solar electricity, they can be used to preheat fresh air for the classrooms during the heating season. A data-driven grey box model for the classrooms is calibrated with measured data, and a PV/T model as a renewable energy retrofit measure for energy efficiency and flexibility is developed. These models are integrated to apply MPC and reduce peak demand during morning and evening. Results show that using an MPC along with PV/T integration can significantly reduce peak demand during morning and evening high demand periods for the grid. The proposed methodology helps institutional buildings to facilitate their integration into future smart grids and smart cities
Confirmation of Genes Involved in the Degradation of Protocatechuate in Aspergillus niger through Characterization of their Encoded Enzymes
Get PDFLignocellulosic biomass is an abundant and renewable source of aromatic compounds and carbohydrates. Microbial lignin catabolism leads to the formation of seven central aromatic intermediates that have potential markets valuated in the billions. Protocatechuate is one of the most common central intermediates and is the precursor to chemical building blocks, including cis,cis-muconic acid, used for production of bioplastics, cosmetics, food preservatives and antioxidants. Understanding the full extent of the microbial aromatic catabolic pathways is necessary to design the most efficient strains to produce valuable chemicals from protocatechuate. Aromatic catabolic pathways have been mapped out in bacteria, however, the equivalent pathways in fungi have not been as well characterized. In this study, the candidate genes NRRL3_01405, NRRL3_02586 and NRRL3_01409 encoding protocatechuate-3,4-dioxygenase, 3-carboxy-cis, cis-muconate cyclase and β-carboxymuconolactone hydrolase/decarboxylase in the Aspergillus niger protocatechuate catabolic pathway were expressed in Escherichia coli. The target proteins were purified by column chromatography and characterized by enzyme activity assays.
The products of each enzyme were characterized by time-of-flight mass spectrometry, nuclear magnetic resonance spectroscopy, and ultraviolet-visible spectroscopy. NRRL3_00837 was determined to not participate in the in vitro catabolism of protocatechuate to β-ketoadipate despite having been previously reported to participate in the in vivo protocatechuate catabolism. Mapping out the central aromatic catabolic pathways in fungi may lead to the refinement of strains engineered for lignin valorization and bioremediation