Engineering Science - Theses, Dissertations, and other Required Graduate Degree Essays

Receive updates for this collection

Light emission properties of CVD grown 2D monolayer WS2 for optoelectronic applications

Author: 
File(s): 
Date created: 
2020-07-06
Supervisor(s): 
Michael Adachi
Department: 
Applied Sciences: School of Engineering Science
Thesis type: 
(Thesis) M.A.Sc.
Abstract: 

Two-dimensional Transitional Metal Dichalcogenides (TMDs) such as MX2 (M= Mo, W; X= S, Se) have gained tremendous attention for use in optoelectronic applications because of their high carrier mobility and indirect-direct band gap transition for thin layers resulting in light emission. Moreover, monolayer TMDs have exceptional other properties such as piezoelectricity, gate-induced superconductivity, and tunable band structure. Mechanical exfoliation, hydrothermal method, electrochemical exfoliation, chemical vapor deposition (CVD) etc. are the most widely used methods for preparing monolayer TMDs. Among these methods, CVD is regarded as the most promising approach because it can produce large area crystal growth and uniform monolayers. The challenges associated with other methods are either small flake size or low quality with lower carrier mobility limitingperformancein electronic devices. CVD grown TMDs tend to show weak, non-uniform photoluminescence. If we want to use pristine TMDs for optoelectectronics applications, we can use different chemical reagents such as strong acid vapor for passivating surface of pristine TMDs which eventually leads to enhanced photoluminescence. In this study, we first demonstrate growth of monolayer triangular WS2 crystals using a 3-heating zone furnace using a bottom-up CVD process. The average lateral crystal size is ~20-25 µm and the largest crystal size is ~75 µm. Although, several research groups have reported WS2 growth using WO3 and S precursors, specific parameters such as precursor amount, growth substrate, growth pressure and flow rate, temperature, use of gases (e.g. N2, Ar, Ar+H2), growth time, use of promoter (e.g. PTAS, NaCl, KBr), pre-surface treatment of substrate etc. can vary widely from lab to lab,affecting the growth morphology, mechanism, light emission, Raman spectra. Atomic Force Microscopy (AFM) measurements indicate that the thickness of the monolayer WS2 is ~1 nm. We also performed SEM imaging to investigate surface morphology of monolayer WS2 and EDX to perform elemental analysis of monolayer WS2. X-ray Photoelectron Spectroscopy (XPS) has been performed for pristine WS2to reveal its chemical states. Photoluminescence spectroscopy revealed a sharp emission peak at ~626 nm confirming indirect (bulk) to direct band-gap (monolayer) transition in the monolayer. On the other hand, the PL intensity for bi/tri-layer is relatively weak compared to monolayer. Moreover, we investigate the effect of surface passivation using chemical reagents such as H2SO4-vapor for modifying light emission property of pristine WS2 for using in next generation optoelectronics. After H2SO4-vapor treatment, we achieved light emission at ~634 nm corresponding to red-shift with enhanced trion emission. Edges of H2SO4-vapor treated sample shows enhanced biexciton compared to pristine-WS2. We are able to achieve maximum 10-fold PL enhancement from our H2SO4-vapor treated sample and, on an average, we got ~5 fold enhancement. H2SO4-vapor treatment has not been used before for surface passivation. We also studied the laser power dependence PL of pristine and H2SO4-vapor treated monolayer WS2where it shows that with increasing laser power, pristine and H2SO4-vapor treated monolayer WS2shows enhanced PL specially at the crystal edges. In addition, we also focused on investigating photoemission from pristine and H2SO4-vapor treated monolayer WS2along certain lines which eventually shows PL distribution within a specific flake.

Document type: 
Thesis

Evaluation and acceleration of spiking neural networks using FPGAs

File(s): 
Date created: 
2021-11-29
Supervisor(s): 
Zhenman Fang
Jian Li
Department: 
Applied Sciences: School of Engineering Science
Thesis type: 
(Thesis) M.A.Sc.
Abstract: 

Compared to conventional artificial neural networks, spiking neural networks (SNNs) are more biologically plausible and require less computation due to their event-driven nature of spiking neurons. However, the default asynchronous execution of SNNs also poses great challenges to accelerate their performance on FPGAs. In this thesis, we present a novel synchronous approach for rate encoding based Spiking Neural Networks (SNNs), which is more hardware friendly than conventional asynchronous approaches. We first quantitatively evaluate and mathematically prove that the proposed synchronous approach and asynchronous implementation alternatives of rate encoding based SNNs are the same in terms of inference accuracy and we highlight the computational performance advantage of using SyncNN over asynchronous approach. We also design and implement the SyncNN framework to accelerate SNNs on Xilinx ARM-FPGA SoCs in a synchronous fashion. To improve the computation and memory access efficiency, we first quantize the network weights to 16-bit, 8-bit, and 4-bit fixed-point values with the SNN friendly quantization techniques. Moreover, we encode only the activated neurons by recording their positions and the corresponding number of spikes to fully utilize the event-driven characteristics of SNNs, instead of using the common binary encoding (i.e., 1 for a spike and 0 for no spike). For the encoded neurons that have dynamic and irregular access patterns, we design parameterized compute engines to accelerate their performance on the FPGA, where we explore various parallelization strategies and memory access optimizations. Our experimental results on multiple Xilinx ARM-FPGA SoC boards demonstrate that our SyncNN is scalable to run multiple networks, such as LeNet, Network in Network, and VGG, on various datasets such as MNIST, SVHN, and CIFAR-10. SyncNN not only achieves competitive accuracy (99.6%) but also achieves state-of-the-art performance (13,086 frames per second) for the MNIST dataset. Finally, we compare the performance of SyncNN with conventional CNNs using the Vitis AI and find that SyncNN can achieve similar accuracy and better performance compared to Vitis AI for image classification using small networks.

Document type: 
Thesis

Mechanisms of energy optimization in human walking

Author: 
File(s): 
Date created: 
2021-12-17
Supervisor(s): 
J. Maxwell Donelan
Department: 
Applied Sciences: School of Engineering Science
Thesis type: 
(Thesis) Ph.D.
Abstract: 

Humans can learn to move optimally. For many movements, we have a control strategy—or control policy—that optimizes some objective. In walking, we prefer the combination of step widths, step lengths, and speeds that optimizes the amount of energy we need. In familiar contexts, we have had many opportunities to establish this optimal control policy. But in new contexts, the nervous system must quickly learn new control policies in order to continue to move optimally. Our lab has recently demonstrated that humans can continuously optimize energetic cost during walking. This is an impressive feat given that the nervous system has tens of thousands of motor units at its disposal, and it can coordinate these motor units over millisecond timescales, which results in countless combinations of motor unit coordination. The goal of this thesis is to determine how the nervous system navigates this combinatorial problem to learn new energy optimal control policies in new walking contexts. I used three distinct studies to accomplish this goal. For the first two studies, I designed and implemented a simple mechatronic system that applies energetic penalties in the form of walking incline as a function of gait. This creates a new relationship between gait and energetic cost—or new cost landscape—that shifts the energy optimal gait. For the third study, I used exoskeletons that apply assistive torques to each ankle at each walking step to shift the energy optimal gait. The first study tested whether previous findings that people can learn to adapt their control policy when the energy optimum is shifted along step frequency generalize to a different gait parameter and to a different experimental setup. I found that, like step frequency, people can learn to adapt their control policy when the energy optimum is shifted along step width. The second study tested if and how energy optimization extends to multiple gait parameters at the same time. I found that, when the energy optimum is shifted along step width and step frequency, people are limited in their ability to optimize both gait parameters. The third study asked how people learn in which ways to optimize their policy. I found that general variability leads to specific adaptation toward optimal policies. Taken together, these findings provide insight into the mechanisms that underlie energy optimization in walking, as well as the limitations of this optimization.

Document type: 
Thesis

Deep learning applications in non-intrusive load monitoring

Author: 
File(s): 
Date created: 
2020-08-19
Supervisor(s): 
Ivan V. Bajic
Department: 
Applied Sciences: School of Engineering Science
Thesis type: 
(Thesis) M.A.Sc.
Abstract: 

Non-Intrusive Load Monitoring (NILM) is a technique for inferring the power consumption of each appliance within a home from one central meter, aiding in energy conservation. In this thesis I present several Deep Learning solutions for NILM, starting with two preliminary works – A proof of concept project for multisensory NILM on a Raspberry Pi; and a fully developed NILM solution named WaveNILM. Despite their success, both methods struggled to generalize outside their training data, a common problem in NILM. To improve generalization, I designed a framework for synthesizing truly novel appliance level power signatures based on generative adversarial networks (GAN) – the main project of this thesis. This generator, named PowerGAN, is trained using a variety of GAN techniques. I present a comparison of PowerGAN to other data synthesis work in the context of NILM and demonstrate that PowerGAN is able to create truly synthetic, realistic, diverse, appliance power signatures.

Document type: 
Thesis

Filtering in non-Intrusive load monitoring

File(s): 
Date created: 
2021-12-09
Supervisor(s): 
Stephen Makonin
Rodney Vaughan
Department: 
Applied Sciences: School of Engineering Science
Thesis type: 
(Thesis) M.A.Sc.
Abstract: 

Being able to track appliances energy usage without the need of sensors can help occupants reduce their energy consumption. Non-intrusive load monitoring (NILM) is one name for this topic. One of the hardest problems NILM faces is the ability to run unsupervised – discovering appliances without prior knowledge – and to run independent of the differences in appliance mixes and operational characteristics found in various countries and regions. This thesis showcases two filters that are used to denoise power signals, which results in better clustering accuracy for NILM event based methods. Both filters show to outperform a state-of-the-art denoising filter, in terms of run-time. A fully unsupervised NILM solution is presented, the algorithm is based on a hybrid knapsack problem with a Gaussian mixture model. Finally, a novel metric is developed to measure NILM disaggregation performance. The metric shows to be robust under a set of fundamental test cases.

Document type: 
Thesis

Non-parametric modeling in non-intrusive load monitoring

Author: 
File(s): 
Date created: 
2020-12-18
Supervisor(s): 
Stephen Makonin
Ivan Bajic
Department: 
Applied Sciences: School of Engineering Science
Thesis type: 
(Thesis) M.A.Sc.
Abstract: 

Non-intrusive Load Monitoring (NILM) is an approach to the increasingly important task of residential energy analytics. Transparency of energy resources and consumption habits presents opportunities and benefits at all ends of the energy supply-chain, including the end-user. At present, there is no feasible infrastructure available to monitor individual appliances at a large scale. The goal of NILM is to provide appliance monitoring using only the available aggregate data, side-stepping the need for expensive and intrusive monitoring equipment. The present work showcases two self-contained, fully unsupervised NILM solutions: the first featuring non-parametric mixture models, and the second featuring non-parametric factorial Hidden Markov Models with explicit duration distributions. The present implementation makes use of traditional and novel constraints during inference, showing marked improvement in disaggregation accuracy with very little effect on computational cost, relative to the motivating work. To constitute a complete unsupervised solution, labels are applied to the inferred components using a Res-Net-based deep learning architecture. Although this preliminary approach to labelling proves less than satisfactory, it is well-founded and several opportunities for improvement are discussed. Both methods, along with the labelling network, make use of block-filtered data: a steady-state representation that removes transient behaviour and signal noise. A novel filter to achieve this steady-state representation that is both fast and reliable is developed and discussed at length. Finally, an approach to monitor the aggregate for novel events during deployment is developed under the framework of Bayesian surprise. The same non-parametric modelling can be leveraged to examine how the predictive and transitional distributions change given new windows of observations. This framework is also shown to have potential elsewhere, such as in regularizing models against over-fitting, which is an important problem in existing supervised NILM.

Document type: 
Thesis

Addressing the challenges posed by human machine interfaces based on force sensitive resistors for powered prostheses

File(s): 
Date created: 
2020-10-09
Supervisor(s): 
Carlo Menon
Department: 
Applied Sciences: School of Engineering Science
Thesis type: 
(Thesis) Ph.D.
Abstract: 

Despite the advancements in the mechatronics aspect of prosthetic devices, prostheses control still lacks an interface that satisfies the needs of the majority of users. The research community has put great effort into the advancements of prostheses control techniques to address users’ needs. However, most of these efforts are focused on the development and assessment of technologies in the controlled environments of laboratories. Such findings do not fully transfer to the daily application of prosthetic systems. The objectives of this thesis focus on factors that affect the use of Force Myography (FMG) controlled prostheses in practical scenarios. The first objective of this thesis assessed the use of FMG as an alternative or synergist Human Machine Interface (HMI) to the more traditional HMI, i.e. surface Electromyography (sEMG). The assessment for this study was conducted in conditions that are relatively close to the real use case of prosthetic applications. The HMI was embedded in the custom prosthetic prototype that was developed for the pilot participant of the study using an off-the-shelf prosthetic end effector. Moreover, prostheses control was assessed as the user moved their limb in a dynamic protocol.The results of the aforementioned study motivated the second objective of this thesis: to investigate the possibility of reducing the complexity of high density FMG systems without sacrificing classification accuracies. This was achieved through a design method that uses a high density FMG apparatus and feature selection to determine the number and location of sensors that can be eliminated without significantly sacrificing the system’s performance. The third objective of this thesis investigated two of the factors that contribute to increased errors in force sensitive resistor (FSR) signals used in FMG controlled prostheses: bending of force sensors and variations in the volume of the residual limb. Two studies were conducted that proposed solutions to mitigate the negative impact of these factors. The incorporation of these solutions into prosthetic devices is discussed in these studies.It was demonstrated that FMG is a promising HMI for prostheses control. The facilitation of pattern recognition with FMG showed potential for intuitive prosthetic control. Moreover, a method for the design of a system that can determine the required number of sensors and their locations on each individual to achieve a simpler system with comparable performance to high density FMG systems was proposed and tested. The effects of the two factors considered in the third objective were determined. It was also demonstrated that the proposed solutions in the studies conducted for this objective can be used to increase the accuracy of signals that are commonly used in FMG controlled prostheses.

Document type: 
Thesis

Automating assessment of human embryo images and time-lapse sequences for IVF treatment

Author: 
File(s): 
Date created: 
2021-05-21
Supervisor(s): 
Parvaneh Saeedi
Department: 
Applied Sciences: School of Engineering Science
Thesis type: 
(Thesis) M.A.Sc.
Abstract: 

As the number of couples using In Vitro Fertilization (IVF) treatment to give birth increases, so too does the need for robust tools to assist embryologists in selecting the highest quality embryos for implantation. Quality scores assigned to embryonic structures are critical markers for predicting implantation potential of human blastocyst-stage embryos. Timing at which embryos reach certain cell and development stages in vitro also provides valuable information about their development progress and potential to become a positive pregnancy. The current workflow of grading blastocysts by visual assessment is susceptible to subjectivity between embryologists. Visually verifying when embryo cell stage increases is tedious and confirming onset of later development stages is also prone to subjective assessment. This thesis proposes methods to automate embryo image and time-lapse sequence assessment to provide objective evaluation of blastocyst structure quality, cell counting, and timing of development stages.

Document type: 
Thesis

Development of wearable, screen-printable conductive polymer biosensors on flexible and textile substrates

Author: 
File(s): 
Date created: 
2021-06-25
Supervisor(s): 
Bonnie Gray
Department: 
Applied Sciences: School of Engineering Science
Thesis type: 
(Thesis) M.A.Sc.
Abstract: 

Wearable biosensors have great potential for real-time diagnostics, but have been encumbered by costly fabrication processes, rigid materials, and inadequate sensitivity for physiological ranges. Sweat has hitherto been an understudied sample for measurement of components like pH and lactate, which can provide meaningful guidance for wound healing, eczema, and sports medicine applications. This thesis presents the development of a flexible, textile-based, screen-printed electrode system for biosensing applications. Furthermore, a flexible, pH-sensitive composite for textile substrates is developed by mixing polyaniline with dodecylbenzene sulfonic acid and textile screen-printing ink. The optimized composite’s pH response is compared to electropolymerized and drop-cast polyaniline sensors via open circuit potential measurements. A linear response is observed for all sensors between pH 3-10, with the composite demonstrating sufficient response time and a sensitivity better than -20 mV/pH, exceeding existing flexible screen-printed pH sensors. Investigations into a potentiometric, non-enzymatic lactate sensor using polyaminophenylboronic acid are also discussed.

Document type: 
Thesis

Utilization and experimental evaluation of occlusion aware kernel correlation filter tracker using RGB-D

Author: 
File(s): 
Date created: 
2021-02-12
Supervisor(s): 
Shahram Payandeh
Department: 
Applied Sciences: School of Engineering Science
Thesis type: 
(Thesis) M.A.Sc.
Abstract: 

Unlike deep-learning which requires large training datasets, correlation filter-based trackers like Kernelized Correlation Filter (KCF) uses implicit properties of tracked images (circulant matrices) for training in real-time. Despite their practical application in tracking, a need for a better understanding of the fundamentals associated with KCF in terms of theoretically, mathematically, and experimentally exists. This thesis first details the workings prototype of the tracker and investigates its effectiveness in real-time applications and supporting visualizations. We further address some of the drawbacks of the tracker in cases of occlusions, scale changes, object rotation, out-of-view and model drift with our novel RGB-D Kernel Correlation tracker. We also study the use of particle filter to improve trackers' accuracy. Our results are experimentally evaluated using a) standard dataset and b) real-time using Microsoft Kinect V2 sensor. We believe this work will set the basis for better understanding the effectiveness of kernel-based correlation filter trackers and to further define some of its possible advantages in tracking.

Document type: 
Thesis