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

Receive updates for this collection

Biomechanical analysis and simulation of backward falls with head impact in older adults

Author: 
Date created: 
2020-08-05
Abstract: 

This thesis examined the dynamics of backward falls in older adults involving head impact. Time-varying kinematics were extracted from digitizing videos of 11 real-life falls by residents of long-term care. The pelvis always impacted the ground before the head. On average, the head descended 1.2 m, and had a vertical velocity of 1.7 m/s just before it struck the ground. A novel dummy was used to examine how fall mechanics and compliant flooring affect head acceleration. Landing with a curved versus flat torso decreased peak rotational acceleration by 27% (4633 versus 5901 rad/s2). Landing with fixed versus freely rotating hips lowered peak translational accelerations by 36% (101.5 versus 158.7 g) and peak rotational accelerations by 38% (4168 versus 6366 rad/s2). The protective benefit of compliant flooring depended on torso curvature and hip stiffness. These results show that unexplored aspects of fall mechanics strongly influence head impact severity.

Document type: 
Thesis
File(s): 
Supervisor(s): 
Stephen Robinovitch
Department: 
Applied Sciences: School of Engineering Science
Thesis type: 
(Thesis) M.A.Sc.

Deep learning-based multimedia content processing

Author: 
Date created: 
2020-07-24
Abstract: 

In the last few years, deep learning has revolutionized many applications in the field of multi-media content processing such as music information retrieval (MIR) and image compression, which are addressed in this thesis. In order to handle the challenges in acoustic-based MIR such as automatic music transcription, the video of the musical performances can be utilized. In Chapter 2, a new learning-based system for visually transcribing piano music using the convolutional neural networks and support vector machines is presented that achieves an average improvement of ~0.37 in terms of F1 score over the previous works. Another significant problem in MIR is music generation. In Chapter 3, a semi-recurrent hybrid model combining variational auto-encoder and generative adversarial network for sequential generation of piano music is introduced that achieves better results than previous methods. Auto-encoders have also been used as a perfect candidate for learned image compression, which has recently shown the potential to outperform standard codecs. Some efforts in integrating other computer vision tasks and image compression to improve the compression performance have also been made. In Chapter 4, a semantic segmentation-based layered image compression method is presented in which the segmentation map of the input is used in the compression procedure. Most learned image compression methods train multiple models for multiple bit rates, which increase the implementation complexity. In Chapter 5, we propose a variable-rate image compression model employing two novel loss functions and residual sub-networks in the auto-encoder. The proposed method outperforms the standard codecs and also previous learned variable-rate methods on Kodak image set. The state-of-the-art image compression has been achieved by utilizing joint hyper-prior and auto-regressive models. However, they suffer from the spatial redundancy of the low frequency information in the latents. In Chapter 6, we propose the first learned multi-frequency image compression approach that uses the recently developed octave convolutions to factorize the latents into high and low frequencies. As the low frequency is represented by a lower resolution, their spatial redundancy is reduced, which improves the compression rate. Our experiments show that the proposed scheme outperforms all standard codecs and learning-based methods in both PSNR and MS-SSIM metrics, and establishes the new state of the art for learned image compression on Kodak image set.

Document type: 
Thesis
File(s): 
Supervisor(s): 
Jie Liang
Department: 
Applied Sciences: School of Engineering Science
Thesis type: 
(Thesis) Ph.D.

Propagation and narrow cylindrical antennas for non-line-of-sight links

Author: 
Date created: 
2020-08-12
Abstract: 

Marconi’s century-old commercialization of wireless has grown to billions of radio links. In cities there can be thousands of cellular base stations, usually mounted on buildings, to link millions of terminals, and there are many WiFi devices in homes and offices. These links are nearly all non-line-of-sight (NLOS), with signal processing at the terminals striving to cope with the degradation from the propagation and the antennas. The signal processing, antenna design, and the propagation, are now separate disciplines as a result of their expansion. The limitations from the propagation channels and the antennas are often blindly accepted by signal processing. Innovations become most likely when there is an in-depth understanding of each discipline, an increasingly difficult prospect. But no matter how powerful or innovative the electronic signal processing, the propagation and antenna performance remain the biting constraint for communications performance. This motivates a hypothesis: improving the understanding of the bottleneck mechanisms - the propagation and antennas - enables innovation for better link performance. The approach is to select topics in propagation and antennas which bottleneck the link performance. For NLOS, diffraction is the critical mechanism. The thesis therefore opens with a look at diffraction, in the context of two applications: classical around the corner propagation, where simple arrangements of passive dipoles are demonstrated to drastically improve a diffraction-limited link; and through-forest propagation, where a new model, combining diffraction across the tree tops and direct transmission, is demonstrated to fit the full range of short- to long-distances established from recent experiments. For the antennas, tubular platforms offer challenges which have not been widely addressed, and yet such platforms are ubiquitous in the form of bicycle frames, drone struts, and masts. Designs are investigated where compactness is a critical requirement: externally-mounted, small narrowband antennas for where the curvature of the cylindrical tube is too small for planar groundplane principles to guide the design; and configurations that deploy the tubular structure as a compact coaxial cylindrical waveguide, to feed slot elements in the cylinder. These are demonstrated to be extremely wideband and have low loss at microwave frequencies.

Document type: 
Thesis
File(s): 
Supervisor(s): 
Rodney G. Vaughan
Department: 
Applied Sciences: School of Engineering Science
Thesis type: 
(Thesis) Ph.D.

Technical development of the brain vital signs framework as an objective and practical test for concussion

Author: 
Date created: 
2020-06-24
Abstract: 

The engineering design criteria for an objective test for concussion dictate that it should be rapid, portable, practical, robust, and sensitive over the time-course of injury. This is an important technical challenge, which is yet to be solved. Currently, sports medicine professionals who treat concussions are limited by a lack of access to objective measurement tools for evidence-based treatment. Electroencephalography-based technologies present a unique opportunity to address these criteria. The brain vital signs framework uses electroencephalography to rapidly assess electrical brain responses to auditory stimuli and make interpretations on cognitive changes. However, applications in electroencephalography are typically recorded under controlled laboratory conditions and have not been validated under the uncontrolled, noisy environments necessary to evaluate concussions. These clinical and athletic environments are fundamentally different to those found in the laboratory and have a unique set of constraints that make traditional methods impractical. This thesis addresses the technical challenges to demonstrate that the brain vital signs framework can be used as an objective, practical tool for monitoring concussion. First, the brain vital signs framework was successfully deployed in a clinical environment. Markers of brain function demonstrated significant concussion-related changes in athletes that were undetected by standard concussion protocols. This is the first demonstration of a portable brain technology being implemented immediately at the point of care for concussion. However, there are additional technical challenges to improve the practicality of this framework. Subsequently, an automated framework for numerically assessing the signal quality of electroencephalography is presented. This framework is highly sensitive and specific to classifying artifacts and can approximate the signal-to-noise-ratio of a recorded signal. Finally, a new approach for recording event-related potentials from distant sensor locations is presented to optimise the speed and practicality of portable brain technologies. The presented body of work incorporates novel engineering developments to provide robust, technical solutions for a clinical problem. These are impactful and important results that will allow for improved medical applications in concussion.

Document type: 
Thesis
File(s): 
Supervisor(s): 
Ryan D'Arcy
Department: 
Applied Sciences: School of Engineering Science
Thesis type: 
(Thesis) Ph.D.

Optical coherence tomography and deep learning for ophthalmology

Author: 
Date created: 
2020-04-06
Abstract: 

Robust quantitative tools require large data sets for testing efficacy and accuracy, which is especially true when using machine learning and neural networks. However, large datasets with corresponding manual annotations are uncommon with state-of-the-art imaging systems, particularly in the medical field. Ophthalmology is one such field, for which recent imaging advances allow clinicians to use multiple imaging modalities to diagnose and monitor patients. Optical coherence tomography (OCT) has become an integral imaging modality in ophthalmic clinics due to its non-invasive nature and ability to acquire micrometer scale sub-surface images of ophthalmic tissue. In this thesis, several different techniques to mitigate the need for large annotated datasets when translating machine learning tools to an ophthalmic clinic are evaluated. First, the concept of transfer learning is assessed through fine-tuning networks trained on a different domain (adaptive optics scanning laser ophthalmoscopy) to the domain of interest (adaptive optics OCT) to detect cone photoreceptors. Second, both adversarial and semi-supervised learning are investigated which allow for unlabelled data to be used in the training process. Finally, the more challenging task of diagnostics with limited data was investigated using diabetic retinopathy OCT Angiography data and an ensemble of networks. Through these investigations, the utility of transfer learning, adversarial and semi-supervised learning, and ensembling is shown for small ophthalmic datasets.

Document type: 
Thesis
File(s): 
Supervisor(s): 
Marinko Venci Sarunic
Department: 
Applied Sciences: School of Engineering Science
Thesis type: 
(Thesis) Ph.D.

Antenna preselection for massive MIMO, multi-channel sounder concept, and fibre-fed test-bed, for 5G

Author: 
Date created: 
2019-12-10
Abstract: 

The fifth generation (5G) of wireless connectivity is a global research effort for providing a significant jump in communications capacity. This will improve existing services for personal and business communications, navigation, media distribution, etc. New applications will include wearable terminals, smarter homes, better vehicular safety and other critical infrastructure, and products that are currently unimagined. The 5G capabilities include higher reliability and data rates with lower latency, realized through new technologies such as (i) massive MIMO (multiple input, multiple output) - meaning the use of a large or massively large number of antennas, and (ii) higher carrier frequencies – meaning tens to hundreds of GHz - in order to have physically smaller antennas. The massive MIMO is the most visible and compelling technology for 5G. The idea is to use arrays with a massive number of antenna elements for serving mobile terminals simultaneously. With 5G a cornerstone goal of current research in communications theory, radio-wave propagation, antennas, and electronics, new paradigms are being sought in many aspects of communications design and implementation. This particularly motivates a study of massive MIMO, with the aim of understanding and contributing to the knowledge pool for the communications performance expected from 5G. The breadth of technologies is too overwhelming to address within a single cover and so the following projects were selected and are presented in this thesis as contributions to 5G: (i) a signal processing algorithm for massive MIMO antenna selection combining, evaluated in a modelled, realistic propagation environment; (ii) a design concept for a distributed antenna channel sounder, demonstrated for magnitude-only indoor channel sounding; (iii) a MIMO test-bed for proof-of-concept demonstration of FPGA-based MIMO signal processing algorithms

Document type: 
Thesis
File(s): 
Supervisor(s): 
Rodney Vaughan
Department: 
Applied Sciences: School of Engineering Science
Thesis type: 
(Thesis) Ph.D.

Use of functional correlation tensors for fMRI monitoring of neuroplasticity during motor learning

Author: 
Date created: 
2020-05-11
Abstract: 

The development of Functional Correlation Tensors (FCT) is driving novel investigations into whole-brain functional magnetic resonance (fMRI) signal synchronicity. FCTs are mathematically analogous to the established structural diffusion modality. Unlike conventional fMRI analysis, FCTs examine functional signal independently of hemodynamic response assumptions. In this work participants trained on a motor task for two weeks with fMRI and diffusion scans collected at baseline and endpoint. Using only baseline data, a significant correlation was detected for the fractional anisotropy of the diffusion data with the signal synchronicity anisotropy of the FCT data. Previous work on this data detected white matter (WM) neuroplasticity in motor regions between baseline and endpoint. As FCT is sensitive to WM function, it was hypothesized that WM neuroplasticity could be further detected. Significant increases in signal synchronicity were detected in areas of motor task planning and execution. This represents the first instance of this novel methodology for identifying neuroplasticity.

Document type: 
Thesis
File(s): 
Supervisor(s): 
Ryan D'Arcy
Department: 
Applied Sciences: School of Engineering Science
Thesis type: 
(Thesis) M.A.Sc.

3D convolutional neural networks for Alzheimer’s disease classification

Author: 
Date created: 
2020-01-21
Abstract: 

Dementia of the Alzheimer’s type (DAT) is a neurodegenerative disease characterized by abnormal brain metabolism and structural brain atrophy. These functional and structural changes can be observed in images acquired using 18F-fluorodeoxyglucose positron emission tomography (FDG-PET) and structural magnetic resonance imaging (MRI). Traditional machine learning framework for DAT classification often involves time-consuming segmentation of brain images as part of the feature extraction process, while deep neural networks can learn DAT-related patterns directly from brain images to generate DAT probability scores. In this thesis, we design 3D convolutional neural networks (CNN) for two applications: classification and segmentation. We design classification networks for single modality use and perform comprehensive evaluation by measuring the performance of our networks on images along the entire DAT spectrum. To support traditional DAT classification framework, we design a fast and accurate segmentation pipeline. We propose a hemisphere-based approach where we train networks to localize and segment hemispheres.

Document type: 
Thesis
File(s): 
Supervisor(s): 
Faisal Beg
Department: 
Applied Sciences: School of Engineering Science
Thesis type: 
(Thesis) M.A.Sc.

Magnetic composite polymer membrane actuators with applications to microfluidic devices

Author: 
Date created: 
2020-04-08
Abstract: 

This thesis demonstrates new materials and microfabrication techniques for integrating membrane-type magnetic composite polymer (M-CP) actuators into microfluidic devices and systems. A membrane actuator with a powerful stroke volume that displaces 7.4 µL of water under a 110mT external magnetic field is developed and demonstrated in a hybrid M-CP/thermoplastic microfluidic device and in an all-PDMS microfluidic device. To achieve injection mouldable M-CP devices, a new M-CP is developed that consists of an injection mouldable off-thiol-ene-epxoy (OSTE+) polymer resin that is embedded with 25 weight-% rare earth magnet particles to be permanently magnetized. To support the rapid prototyping of PDMS and OSTE+ polymer microfluidic devices, a new type of micromould is developed that uses laser ablation of tape to deliver low cost, ultra-rapid moulds. These developments facilitate future commercial mass production through integration with thermoplastic polymers favored by the microfluidics industry in a scalable, “design-to-manufacture” scheme.

Document type: 
Thesis
File(s): 
Supervisor(s): 
Bonnie Gray
Department: 
Applied Sciences: School of Engineering Science
Thesis type: 
(Thesis) M.A.Sc.

Tensor completion methods for collaborative intelligence

Author: 
Date created: 
2020-03-20
Abstract: 

In the race to bring Artificial Intelligence (AI) to the edge, collaborative intelligence has emerged as a promising way to lighten the computation load on edge devices that run applications based on Deep Neural Networks (DNNs). Typically, a deep model is split at a given layer into edge and cloud sub-models. The deep feature tensor produced by the edge sub-model is transmitted to the cloud, where the remaining computationally intensive workload is performed by the cloud sub-model. The communication channel between the edge and cloud is imperfect, which will result in missing data in the deep feature tensor received at the cloud side, an issue that has mostly been ignored by existing literature on the topic. In this thesis I study four methods for recovering missing data in the deep feature tensor. Three of the studied methods are existing, generic tensor completion methods, and are adapted here to recover deep feature tensor data, while the fourth method is newly developed specifically for deep feature tensor completion. Simulation studies show that the new method is 3 − 18 times faster than the other three methods, which is an important consideration in collaborative intelligence. For VGG16’s sparse tensors, all methods produce statistically equivalent classification results across all loss levels tested. For ResNet34’s non-sparse tensors, the new method offers statistically better classification accuracy (by 0.25% − 6.30%) compared to other methods for matched execution speeds, and second-best accuracy among the four methods when they are allowed to run until convergence.

Document type: 
Thesis
File(s): 
Supervisor(s): 
Ivan Bajic
Department: 
Applied Sciences: School of Engineering Science
Thesis type: 
(Thesis) M.A.Sc.