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

Receive updates for this collection

The design of soft fluid filled actuators driven by conductive nylon

Author: 
Date created: 
2019-09-17
Abstract: 

Soft robots have become increasingly prevalent due to their distinct advantages over traditional rigid robots such as high deformability and good impact resistance. However, soft robotics are currently limited by bulky, non-portable methods of actuation. In this study, we propose a soft actuator driven by conductive nylon artificial muscles which is able to produce forces up to 1.2N. By utilizing nylon artificial muscles, the system does not require sizable pumps or compressors for actuation. The proposed actuator is made up of two main components, a sealed bladder filled with air and an arrangement of nylon artificial muscles. The quasi static behavior of the actuator is characterized using established hyper elastic models and validated against experimental results (maximum error of 5.3%). Using these models, a set of design considerations are formulated which outline the achievable torques for various actuator dimensions.

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

Audio-visual speech processing using deep learning techniques

Author: 
Date created: 
2019-10-02
Abstract: 

Speech separation is the task of segregating a target speech signal from background interference. To differentiate the separation of multiple speech sources from separating speech from non-speech noise, the terms Speaker Separation and Speech Enhancement (or Denoising) are commonly used, respectively. Speech separation can benefit from exploiting different modalities, i.e., audio and visual, and directional information when multiple microphones are available. A new approach is developed for subject-independent speaker separation by incorporating spectral, spatial and visual features. The audio signals have their magnitude and phase modified in the frequency domain for the speaker separation. The key idea is to estimate the target magnitudes from the audio with the Permutation Invariant Training (PIT) technique and then refine these estimates using both visual and spatial audio features. Specifically, visual features are matched to the corresponding audio, and spatial audio features are used as side information and shown to provide drastic improvement for magnitude and phase estimation in terms of output speech intelligibility, quality, and the separation performance. Visual information also provides improvements. Hence, both visual and spatial features are shown to be useful for speaker separation. A monaural speech enhancement model is also developed which incorporates both audio and visual information. In contrast to the audio-visual speaker separation model, the audio-visual speech enhancement model operates in time-domain. Hence, there is no need for a transformation and separate models for estimation of magnitude and phase spectra. According to the results of the objective evaluations, exploiting visual information for enhancement applications improves the performance in terms of both output quality and intelligibility.

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

Developing a machine learning framework for 24-hour data analysis aimed at early detection of cardiac arrhythmias as a guiding tool for physicians

Author: 
Date created: 
2019-05-20
Abstract: 

Cardiovascular diseases (CVD), defined as a spectrum of disorders primarily impacting the heart and the circulatory system, account for a substantial fraction of worldwide morbidity and mortality. Electrocardiograms (ECGs) are routinely implemented in a patient’s diagnosis, both in hospitals and outpatient settings. They serve as one of the primary diagnostic tools as patients encounter medical personnel, particularly in suspected CVD. A cardiac holter monitor is a medical diagnostic device, connected to the patient via several conductive leads placed across the chest, and "worn" on a strap across the shoulder. A holter is applied to record continuous ECG data (typically 24 hours). With recently emerging applications of Machine Learning (ML) in data analysis techniques, the need for human expertise and potential human error could be minimized, and prediction accuracy optimized considerably. Hence, the objective of this research is to develop a machine learning approach to eventually aid physicians with their decisions as a powerful guiding/assisting tool to analyse the ECG information reported by holter monitors. Furthermore, we aim to develop a computer aided diagnostic system that can assist expert cardiologists by providing intelligent, cost effective, and time-saving diagnosis. In this thesis, we implement a deep learning-based solution to analyse readings of 24-holter monitors. In our proposed solution, we train neural networks to extract high-level features from temporal signal recordings of holter monitors. We present a supervised neural network framework to predict the physician’s final interpretations based on the holter recorded signals. The outputs of the network contain the likelihood of four possible scenarios of Normal and three types of arrhythmias. The high classification performance of the proposed methodology emphasizes the capability of this framework to be used as an assisting tool alongside the physicians to interpret holter reports.

Document type: 
Thesis
Supervisor(s): 
Andrew Rawicz
Department: 
Applied Sciences: School of Engineering Science
Thesis type: 
(Thesis) M.A.Sc.

Investigation on compressive devices based on dielectric elastomer actuators

Author: 
Date created: 
2019-11-20
Abstract: 

Dielectric elastomer actuators (DEAs) are an emerging technology from the larger class of artificial muscle actuators, showing interesting properties such as softness and large actuation strains. A DEA can be used to create various types of motions (such as planar, rotary, and bending) and contact forces (such as tension, biaxial compression, and bending moment). Due to its interesting properties, there are a growing number of studies on improving DEA applicability, reliability, and characteristics. Additionally, despite limited commercial use, there is a growing global push toward commercializing DEA in various sectors. In this thesis, in an effort to investigate practical applications of DEAs, their functionality is studied in applications where compressive forces are exerted by the actuator. Two different modalities in which DEAs could apply compressive forces to an encompassed object are identified. To narrow the focus of the thesis, a practical application is introduced and investigated for each modality. The first practical application, based on the first modality, introduces the use of DEAs as a compression bandage to improve blood circulation in the human leg. The proposed compression bandage could potentially enable a controlled variable compression around the lower leg. The second practical application, based on the second modality, proposes a novel gripper that uses DEAs as a mean to apply a soft touch on objects. The proposed gripper can apply up to 2N of grasping force to select objects. The gripper may be adopted as an end effectors for collaborative robotic arms, where a human operator is collaborating with the robotic arm to handle a delicate object. Typically, DEAs are actuated using high electrical voltages of several kilovolts. This operating voltage gives rise to a safety concern for practical applications where DEAs are sought to be operated in proximity to the human body. Since the two aforementioned applications require, or may require the operation of DEAs close to the human body, it is very important to study the electrical safety of DEAs and investigate methods to manage the risk of their high operating voltages. The last part of this thesis discusses the electrical safety of DEAs in scenarios where DEA circuitry is in contact with the human body, and shows that they can be safely used. Critical parameters for identification of the electrical safety of DEAs are also introduced to assist in designing safer DEAs for such applications.

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

Differences in neuroimaging, clinical, cognitive score measures between males and females in Alzheimer's Dementia and Healthy Aging

Author: 
Date created: 
2019-07-18
Abstract: 

In this thesis, differences in the brain structure, clinical and cognitive scores as a function of sex (male/female) were studied for the control (healthy aging) and Alzheimer's Dementia (AD) group. Analysis of brain structure volumes, cortical thickness, CSF measures and cognitive scores were performed on the entire ADNI dataset. In healthy aging, the male brain structure shows greater atrophy with age. However, in case of AD, differences in brain structure that could be explained due to sex differences are found to be reduced. This suggests that female brains showed greater atrophy in AD as compared to male brains. Also, the role of APOE4 gene status on brains of control and AD group was studied. APOE4 gene status did not have any significant effect on baseline measurements in the control group, but showed significant difference for left hippocampus volume in the AD group. In addition, APOE4 carriers having a thicker cortex than non carriers for regions in right temporal lobe for AD patients. Role of APOE4 towards AD has been well studied, but the role played by the absence of even a single APOE4 allele in Alzheimer's Dementia must be explored next.

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

Enhanced fireworks algorithm to optimize extended Kalman filter speed estimation of an induction motor drive system

Author: 
Date created: 
2019-08-15
Abstract: 

To maximize efficiency, the speed of an induction motor (IM) is controlled to match the load. Often an extended Kalman filter (EKF) estimates the speed of the IM, eliminating the speed sensor. The EKF requires a mathematical model of the IM and system and noise covariances, typically determined by optimization using trial-and-error or a genetic algorithm (GA). My research objective was to investigate a relatively new algorithm, the enhanced fireworks algorithm (EFWA) and its ability to determine covariances compared to current methods. I used a Simulink model of a system comprised of an IM controlled by a variable frequency drive (VFD) to experiment with the EKF using trial and error, genetic algorithm and EFWA optimization methods. My results indicated that in this application, using selected parameters, the EFWA provides a good solution in fewer iterations than the GA, which may be required for online adaptive tuning of the EKF.

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

Towards the Development of an Adaptive Compression System

Author: 
Date created: 
2016-11-16
Abstract: 

Regarded as the mainstay for treatment of venous insufficiency and the associated complications, compression therapy aims at assisting with venous return through the exertion of external pressure on the limbs. Compression is achieved by medical bandages and stockings, which hold promise only during supine and walking conditions, or mechanical pumps, which are usually bulky and limited to non-ambulatory use. Hence, the purpose of this study was to develop an improved compression system that eliminates the flaws of the existing products. To attain this goal, a motorized compression bandage was designed that takes advantage of force-sensing resistors (FSRs®) to exert reproducible, controlled pressure on the lower extremities. The performance of the device in enhancing venous return was explored in a pilot experiment, wherein graded lower body negative pressure (LBNP) was employed as a surrogate of standing erect. The results revealed a significant reduction in the mean cardiovascular changes to LBNP.

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

Clustering and identification of body extremities for pose recognition through a network of calibrated depth sensors

Author: 
Date created: 
2019-08-06
Abstract: 

This thesis presents a framework of a marker-less human pose recognition system by identifying key body extremity parts through a network of calibrated depth sensors. The depth sensors can overcome challenges related to low illuminations which usually compromises the information from the RGB cameras. The thesis proposed a novel approach for calibrating multiple depth sensors using retro-reflective (RR) marked spheres. The calibrated parameters are then used to align the point cloud data of the human body associated with multiple depth sensors with respect to a common coordinate frame. This fusion of point clouds facilitates in overcoming the self-occlusion problems from body parts without incurring disjointedness in the fused point cloud data. The second part of the thesis introduces a novel algorithm for the identification of key body extremities such as head, hands, and feet of a human subject. A geodesic mapping is applied on the fused point cloud to produce a set of distinct topological clusters of 3D points. From these clusters, a hierarchical skeleton tree graph is generated and used for key extremities classification which finally leads to pose recognition. The thesis presents the assessment of each proposed part and its comparison with other available techniques in a succession of experimental configurations.

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

Evaluation of Support Vector Machine kernels for detecting network anomalies

Author: 
Date created: 
2019-04-18
Abstract: 

Border Gateway Protocol (BGP) is used to exchange routing information across the Internet. BGP anomalies severely affect network performance and, hence, algorithms for anomaly detection are important for improving BGP convergence. Efficient and effective anomaly detection mechanisms rely on employing machine learning techniques. Support Vector Machine (SVM) is a widely used machine learning algorithm. It employs a set of mathematical functions called kernels that transform the input data into a higher dimensional space before classifying the data points into distinct clusters. In this Thesis, we evaluate the performance of linear, polynomial, quadratic, cubic, Gaussian radial basis function, and sigmoid SVM kernels used for classifying power outage such as Moscow Power Blackout, BGP mis-configuration, and BGP anomalies such as Slammer, Nimda and Code Red I. The SVM kernels are compared based on accuracy and the F-Score when detecting anomalous events in the Internet traffic traces. Simulation results indicate that the performance heavily depends on the selected features and their combinations.

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

Deep learning-based computer-aided diagnosis for brain and retinal diseases

Author: 
Date created: 
2019-08-15
Abstract: 

Deep neural network has achieved excellent performance for many recognition tasks. Despite its recent wide application on medical imaging tasks, the requirement of large amount of manually labeled samples limits its performance on medical image recognition tasks. Comparing with natural images, medical image is difficult and expensive to acquire and requires specialized training for its labeling. However, the data samples for a specific clinical task shares much less heterogeneity comparing with most image recognition tasks. Exploring advanced network architecture and incorporate it with a-priori, domain-specific knowledge has a great potential to deliver superior recognition performance and better computer-aided diagnosis system. In this thesis, we presented the development of four novel deep learning based frameworks regarding four medical image recognition tasks, early diagnosis of Alzheimer's disease, differential diagnosis of multiclass dementia, OCT retinal fluid segmentation and OCT retinal layer segmentation. Comprehensive experiments proved that the proposed frameworks out-performed state-of-the-art methods in each individual task.

Document type: 
Thesis
File(s): 
Supervisor(s): 
Mirza Faisal Beg
Department: 
Applied Sciences: School of Engineering Science
Thesis type: 
(Thesis) Ph.D.