Neural Networks are at the core of computer vision solutions for various applications. With the advent of deep neural networks Facial Expression Recognition (FER) has been a very ineluctable and challenging task in the field of computer vision. Micro-expressions (ME) have been quite prominently used in security, psychotherapy, neuroscience and have a wide role in several related disciplines. However, due to the subtle movements of facial muscles, the micro-expressions are difficult to detect and identify. Due to the above, emotion detection and classification have always been hot research topics. The recently adopted networks to train FERs are yet to focus on issues caused due to overfitting, effectuated by insufficient data for training and expression unrelated variations like gender bias, face occlusions and others. Association of FER with the Speech Emotion Recognition (SER) triggered the development of multimodal neural networks for emotion classification in which the application of sensors played a significant role as they substantially increased the accuracy by providing high quality inputs, further elevating the efficiency of the system. This thesis relates to the exploration of different principles behind application of deep neural networks with a strong focus towards Convolutional Neural Networks (CNN) and Generative Adversarial Networks (GAN) in regards to their applications to emotion recognition. A Motion Magnification algorithm for ME's detection and classification was implemented for applications requiring near real-time computations. A new and improved architecture using a Multimodal Network was implemented. In addition to the motion magnification technique for emotion classification and extraction, the Multimodal algorithm takes the audio-visual cues as inputs and reads the MEs on the real face of the participant. This feature of the above architecture can be deployed while administering interviews, or supervising ICU patients in hospitals, in the auto industry, and many others. The real-time emotion classifier based on state-of-the-art Image-Avatar Animation model was tested on simulated subjects. The salient features of the real-face are mapped on avatars that are build with a 3D scene generation platform. In pursuit of the goal of emotion classification, the Image Animation model outperforms all baselines and prior works. Extensive tests and results obtained demonstrate the validity of the approach.

oTUS is a Long Term Underwater Sensing, bottom landing, node for observations of ocean water temperatures. LoTUS measures temperature (moored to the seafloor) according to a spec-ified time schedule until, at the end of the mission, it surfaces to transmit the collected data to on shore recipients using an Iridium link. The paper presents an extension of the sensing capability to include water current velocity (speed and direction) using a robust, reliable and inexpensive Eulerian method. The method is based on the "tilting stick" principle where a combination of inertia measurement data and magnetic sensor data is used. The paper discusses the principal technique, the modeling of the system, practical considerations, and optimization of the setup for specific flow conditions along with verifying experimental data.

京都大学 / 新制・課程博士 / 博士(工学) / 甲第23614号 / 工博第4935号 / 新制||工||1771(附属図書館) / 京都大学大学院工学研究科機械理工学専攻 / (主査)教授 黒瀬 良一, 教授 花崎 秀史, 教授 岩井 裕 / 学位規則第4条第1項該当 / Doctor of Philosophy (Engineering) / Kyoto University / DFAM

No description available.

N/A / Thesis / Master of Applied Science (MASc)

Because of its thermochemical qualities, ammonia is an attractive alternative to carbon-based fuels. Indeed, the lack of carbon atoms in its molecular structure and the ease of storage make its widespread use desirable. However, there are a number of technological challenges that must be overcome due to the slow burning rate and its large latent heat. The objective of the dissertation is to model ammonia spray flames because direct liquid fuel injection in a combustion chamber is an essential aspect of the design of practical devices. The topic has been divided into a number of sub-problems, which are examined in each chapter of the thesis, due to the lack of fundamental physical details of the individual processes occurring and modeling considerations that cannot be ignored anymore.To better understand how the large latent heat affects the spray dynamics, a campaign of direct numerical simulations is initially performed at various ambient temperatures. Then, conducting large eddy simulations is preferred to lower the computational cost. The assessment of the dispersion models showed that the available options, however, are unable to reproduce the averaged droplet distribution across the entire domain and an improved model is proposed. Droplet evaporation causes local inhomogeneities in the mixture, which simultaneously induces multiple combustion modes. The Darmstadt Multi-Regime Burner (MRB) was the ideal candidate to investigate the physical aspects in advance. The best option for capturing its flame structure was the physically-derived multi-modal manifold and a regime classification index is formulated and tested on the MRB.Then, a machine learning strategy based on neural networks is suggested to quicken the look-up procedure, and preliminary validation of the methodology revealed that a time reduction of 30% is achieved without affecting the results' accuracy.

Sickle cell disease (SCD) is a severe genetic disease, affecting over 100,000 in the United States and millions worldwide. Individuals suffer from stroke, acute chest syndrome, and cardiovascular complications. Much of these associated morbidities are primarily mediated by blockages of the microvasculature, events termed vaso-occlusive crises (VOCs). Despite its prevalence and severity, the pathophysiological mechanisms behind VOCs are not well understood, and novel experimental tools and methods are needed to further this understanding. Microfluidics and computational fluid dynamics (CFD) are rapidly growing fields within biomedical research that allow for inexpensive simulation of the in vivo microenvironment prior to animal or clinical trials. This study includes the development of a CFD model capable of simulating diseased and healthy blood flow within a series of microfluidic channels. Results will be utilized to further improve the development of microfluidic systems.

No description available.

<p>Finite element analysis (FEA) employs a science-based approach in which the complete machining process can be simulated and optimized before resorting to costly and time-consuming experimental trials. In this work, cutting coefficient of AISI 1045 steel will be estimated using finite element modelling using Arbitrary Lagrangian Eulerian formulation (ALE). The estimated values are then experimentally validated. A parametric study is carried out after in order to investigate how some cutting parameters can affect the cutting coefficients. The process parameters to be varied include feed rate, cutting speed, and cutting edge radius.</p> / Master of Applied Science (MASc)

Nanoparticle-based technologies are an emerging field with the promise to impact a wide range of application areas. However, that potential is somewhat married to a host of research questions that remain to be answered. This work explores the surface modification of magnetic nanoparticles in a controlled fashion to produce hybrid nanoparticle (metal/polymer) systems with different morphologies, understand in-situ behavior of stimuli-responsive polymers grafted to a substrate, and obtain better computational methods for particle-tracking and -deposition. Nanoparticle surface modification was performed using ATRP, obtaining homo-, block-co-, and ‘twoaced/biphasic’ polymer structures on the nanoparticle surfaces. Biphasic Janus nanoparticles (JPs) were formed using a magnetic nanoparticle core and an innovative technique combining non-covalent solid protection with sequential controlled radical polymerization to form the two surface-grafted polymer phases. Surface-confined polymerizations were conducted using pH- and thermo-responsive materials. Poly(methacrylic acid) (PMAA) and a series of (aminoalkyl) methacrylate polymers were used as pH responsive polymers. Additionally, poly(N-isopropylacrylamide) (PNIPAM) was selected as the thermo-responsive material for this study. In-situ characterization techniques, including atomic force microscopy (AFM), dynamic light scattering (DLS), and ellipsometry, were used to evaluate the thermo- and pH-responsiveness of these stimuli responsive materials. A new general-oscillator (GENOSC) model was used to determine swelling ratio, thickness, and optical constant changes in the polymer brush as pH was changed in-situ. AFM was used to study morphological changes due to changes in pH and temperature. Nanoparticle temperature responsiveness was investigated using DLS. A related effort involved the use of computational fluid dynamic (CFD) methods to track (micron-sized) particles in certain geometries, including a human lung morphology. Predicted particle transport and deposition was compared to Lagrangian computational approaches and available experimental data. The Eulerian particle phase modeling method developed resulted in the accurate prediction of both near-wall particle tracking and wall deposition. This Eulerian-Eulerian model is a new tool that has potential for particle tracking in physiological morphologies. This combination of experimental and computational research has led to new nano- and micro-particle surface modification methods and particle transport modeling.