No description available.

Better understanding of the role of vegetation in the transport of fluid and pollutants requires improved knowledge of the detailed flow structure within the vegetation. Instead of spatial averaging, this study uses discrete measurements at multiple locations within the canopy to develop velocity and turbulence intensity profiles and observe the changes in the flow characteristics as water travels through a vegetation array simulated by rigid dowels. Velocity data were collected with a one dimensional laser Doppler velocimeter (LDV) under single layer emergent and submerged flow conditions, and through two layers of vegetation. The effects of dowel arrangement, density, and roughness are also examined under the single layer experiments. The results show that the velocity within the vegetation array is constant with depth and the velocity profile is logarithmic above it. The region immediately behind a dowel, where the vorticity and turbulence intensity are highest, is characterized by a velocity spike near the bed and an inflection point near the top of the dowel arrays. With two dowel layers, the velocity profile in the region behind a tall dowel exhibits multiple inflection points and the highest turbulence intensities are found there. / Master of Science

(English) Perfluorocarbons (PFC) are hydrocarbons in which some or all of the hydrogen atoms are replaced with fluorine. PFC have a very high capacity for dissolving oxygen. They are chemically and biologically inert. The most successful clinical application of PFC is the "two-layer method" for pancreas preservation before islet isolation. The two-layer organ preservation method (TLM) is based on oxygenated perfluorocarbon overlaid with University of Wisconsin (UW) solution. In experiment it has been successfully used for heart and intestine transplantation. We tested whether this technique would prevent tissue damage and improve results of kidney, pancreas and islets of Langerhans transplantation with prolonged ischemia time in an experimental model of syngenic rats. In kidney and islets of Langerhanse transplantation model we used TLM preservation method. In pancreas transplantation model we used perfluorohexyloctane (PFH) as a new generation of less lipophilic PFC. 1. Kidneys were stored for 24 hours either in UW solution (n = 16), with TLM (n = 16) or transplanted immediately (control group, n = 12). In half of the animals, survival was observed and in the other animals grafts were procured for semiquantitative histological scoring and TUNEL apoptosis assessment 24 h after transplantation....

The present understanding of the structure and dynamics of turbulent boundary layers on aerodynamically smooth walls has been clarified over the last decade or so. However, the dynamics of turbulent boundary layers over rough surfaces is much less well known. Nevertheless, there are many industrial and environmental flow applications that require understanding of the mean velocity and turbulence in the immediate vicinity of the roughness elements.</p> <p>This thesis reports the effects of surface roughness on the flow characteristics in a turbulent boundary layer. Both experimental and numerical investigations are used in the present study. For the experimental study, comprehensive data sets are obtained for two-dimensional zero pressure-gradient turbulent boundary layers on a smooth surface and ten different rough surfaces created from sand paper, perforated sheet, and woven wire mesh. The physical size and geometry of the roughness elements and freestream velocity were chosen to encompass both transitionally rough and fully rough flow regimes. Three different probes, namely, Pitot probe, single hot-wire, and cross hot-film, were used to measure the velocity fields in the turbulent boundary layer. A Pitot probe was used to measure the streamwise mean velocity, while the single hot-wire and cross hot-film probes were used to measure the fluctuating velocity components across the boundary layer. The flow Reynolds number based on momentum thickness, , ranged from 3730 to 13,550. The data reported include mean velocity, streamwise and wall-normal turbulence intensities, Reynolds shear stress, triple correlations, as well as skewness and flatness factors. Different scaling parameters were used to interpret and assess both the smooth- and rough-wall data at different Reynolds numbers, for approximately the same freestream velocity. The appropriateness of the logarithmic law and power law proposed by George and Castillo (1997) to describe the mean velocity in the overlap region was also investigated. The present results were interpreted within the context of the Townsends wall similarity hypothesis. </p> <p>Based on the mean velocity data, a novel correlation that relates the skin friction to the ratio of the displacement and boundary layer thicknesses, which is valid for both smooth- and rough-wall flows, was proposed. In addition, it was also found that the application of a mixed outer scale caused the velocity profile in the outer region to collapse onto the same curve, irrespective of Reynolds numbers and roughness conditions. The present results showed that there is a common region within the overlap region of the mean velocity profile where both the log law and power law are indistinguishable, irrespective of the surface conditions. For the power law formulation, functional relationships between the roughness shift, and the power law coefficient and exponent were developed for the transitionally rough flows. The present results also suggested that the effect of surface roughness on the turbulence field depends to some degree on the specific characteristics of the roughness elements and also the component of the Reynolds stress tensor being considered. </p> <p>In the case of the numerical study, a new wall function formulation based on a power law was proposed for smooth and fully rough wall turbulent pipe flow. The new formulation correctly predicted the friction factors for smooth and fully rough wall turbulent pipe flow. The existing two-layer model realistically predicted the velocity shift on a log-law plot for the fully rough turbulent boundary layer. The two-layer model results also showed the effect of roughness is to enhance the level of turbulence kinetic energy and Reynolds shear stress compared to that on a smooth wall. This enhanced level extends into the outer region of the flow, which appears to be consistent with present and recent experimental results for the boundary layer.

The present understanding of the structure and dynamics of turbulent boundary layers on aerodynamically smooth walls has been clarified over the last decade or so. However, the dynamics of turbulent boundary layers over rough surfaces is much less well known. Nevertheless, there are many industrial and environmental flow applications that require understanding of the mean velocity and turbulence in the immediate vicinity of the roughness elements.</p> <p>This thesis reports the effects of surface roughness on the flow characteristics in a turbulent boundary layer. Both experimental and numerical investigations are used in the present study. For the experimental study, comprehensive data sets are obtained for two-dimensional zero pressure-gradient turbulent boundary layers on a smooth surface and ten different rough surfaces created from sand paper, perforated sheet, and woven wire mesh. The physical size and geometry of the roughness elements and freestream velocity were chosen to encompass both transitionally rough and fully rough flow regimes. Three different probes, namely, Pitot probe, single hot-wire, and cross hot-film, were used to measure the velocity fields in the turbulent boundary layer. A Pitot probe was used to measure the streamwise mean velocity, while the single hot-wire and cross hot-film probes were used to measure the fluctuating velocity components across the boundary layer. The flow Reynolds number based on momentum thickness, , ranged from 3730 to 13,550. The data reported include mean velocity, streamwise and wall-normal turbulence intensities, Reynolds shear stress, triple correlations, as well as skewness and flatness factors. Different scaling parameters were used to interpret and assess both the smooth- and rough-wall data at different Reynolds numbers, for approximately the same freestream velocity. The appropriateness of the logarithmic law and power law proposed by George and Castillo (1997) to describe the mean velocity in the overlap region was also investigated. The present results were interpreted within the context of the Townsends wall similarity hypothesis. </p> <p>Based on the mean velocity data, a novel correlation that relates the skin friction to the ratio of the displacement and boundary layer thicknesses, which is valid for both smooth- and rough-wall flows, was proposed. In addition, it was also found that the application of a mixed outer scale caused the velocity profile in the outer region to collapse onto the same curve, irrespective of Reynolds numbers and roughness conditions. The present results showed that there is a common region within the overlap region of the mean velocity profile where both the log law and power law are indistinguishable, irrespective of the surface conditions. For the power law formulation, functional relationships between the roughness shift, and the power law coefficient and exponent were developed for the transitionally rough flows. The present results also suggested that the effect of surface roughness on the turbulence field depends to some degree on the specific characteristics of the roughness elements and also the component of the Reynolds stress tensor being considered. </p> <p>In the case of the numerical study, a new wall function formulation based on a power law was proposed for smooth and fully rough wall turbulent pipe flow. The new formulation correctly predicted the friction factors for smooth and fully rough wall turbulent pipe flow. The existing two-layer model realistically predicted the velocity shift on a log-law plot for the fully rough turbulent boundary layer. The two-layer model results also showed the effect of roughness is to enhance the level of turbulence kinetic energy and Reynolds shear stress compared to that on a smooth wall. This enhanced level extends into the outer region of the flow, which appears to be consistent with present and recent experimental results for the boundary layer.

The physical processes in lakes remain only partially understood despite successful data collection from a variety of sources spanning several decades. Although numerical models are already frequently employed to simulate the physics of lakes, especially in the context of water quality management, improved methods are necessary to better capture the wide array of dynamically important physical processes, spanning length scales from ~ 10 km (basin-scale oscillations) - 1 m (short internal waves). In this thesis, high-order numerical methods are explored for specialized model equations of lakes, so that their use can be taken into consideration in the next generation of more sophisticated models that will better capture important small scale features than their present day counterparts. The full three-dimensional incompressible density-stratified Navier-Stokes equations remain too computationally expensive to be solved for situations that involve both complicated geometries and require resolution of features at length-scales spanning four orders of magnitude. The main source of computational expense lay with the requirement of having to solve a three-dimensional Poisson equation for pressure at every time-step. Simplified model equations are thus the only way that numerical lake modelling can be carried out at present time, and progress can be made by seeking intelligent parameterizations as a means of capturing more physics within the framework of such simplified equation sets. In this thesis, we employ the long-accepted practice of sub-dividing the lake into vertical layers of different constant densities as an approximation to continuous vertical stratification. We build on this approach by including weakly non-hydrostatic dispersive correction terms in the model equations in order to parameterize the effects of small vertical accelerations that are often disregarded by operational models. Favouring the inclusion of weakly non-hydrostatic effects over the more popular hydrostatic approximation allows these models to capture the emergence of small-scale internal wave phenomena, such as internal solitary waves and undular bores, that are missed by purely hydrostatic models. The Fourier and Chebyshev pseudospectral methods are employed for these weakly non-hydrostatic layered models in simple idealized lake geometries, e.g., doubly periodic domains, periodic channels, and annular domains, for a set of test problems relevant to lake dynamics since they offer excellent resolution characteristics at minimal memory costs. This feature makes them an excellent benchmark to compare other methods against. The Discontinuous Galerkin Finite Element Method (DG-FEM) is then explored as a mid- to high-order method that can be used in arbitrary lake geometries. The DG-FEM can be interpreted as a domain-decomposition extension of a polynomial pseudospectral method and shares many of the same attractive features, such as fast convergence rates and the ability to resolve small-scale features with a relatively low number of grid points when compared to a low-order method. The DG-FEM is further complemented by certain desirable attributes it shares with the finite volume method, such as the freedom to specify upwind-biased numerical flux functions for advection-dominated flows, the flexibility to deal with complicated geometries, and the notion that each element (or cell) can be regarded as a control volume for conserved fluid quantities. Practical implementation details of the numerical methods used in this thesis are discussed, and the various modelling and methodology choices that have been made in the course of this work are justified as the difficulties that these choices address are revealed to the reader. Theoretical calculations are intermittently carried out throughout the thesis to help improve intuition in situations where numerical methods alone fall short of giving complete explanations of the physical processes under consideration. The utility of the DG-FEM method beyond purely hyperbolic systems is also a recurring theme in this thesis. The DG-FEM method is applied to dispersive shallow water type systems as well as incompressible flow situations. Furthermore, it is employed for eigenvalue problems where orthogonal bases must be constructed from the eigenspaces of elliptic operators. The technique is applied to the problem calculating the free modes of oscillation in rotating basins with irregular geometries where the corresponding linear operator is not self-adjoint.

Large Eddy Simulation (LES) is a relatively more accurate and reliable alternative to solution of Reynolds Averaged Navier Stokes (RANS) equations in simulating complex turbulent flows at a lesser computational cost than a direct numerical simulation (DNS). However, LES of wall-bounded flows still requires a very high grid resolution in the inner wall layer making its widespread use difficult. Different attempts have been made in the past time to overcome this problem by modeling the near wall turbulence instead of resolving it. One such approach is a two-layer wall model that solves for a reduced one-dimensional equation in the inner wall layer, while solving for the filtered Navier-Stokes equations in the outer layer. The use of such a model allows for a coarser grid resolution than a wall resolved LES. This work validates the performance of a two-layer wall model developed for an arbitrary body fitted non-orthogonal grid in the flow over a wall mounted hump at Reynolds number 9.36x105. The wall modeled large eddy simulation (WMLES) relaxes the grid requirement compared to a wall resolved LES (WRLES) by allowing the first off-wall grid point to be placed at a y+ of approximately 20-40. It is found that the WMLES results are general good agreement with WRLES and experiments. Surface pressure coefficient, skin friction, mean velocity profiles, and the reattachment location compare very well with experiment. The WMLES and WRLES exhibit some under prediction of the peak values in the turbulent quantities close to the reattachment location, with better agreement with the experiment in the separated region. In contrast, a simulation that did not employ the wall model on the grid used for WMLES failed to predict flow separation and showed large discrepancies with the experimental data. In addition to the relaxation of the grid requirement in the wall normal direction, it was also observed that the wall model allowed a reduction in the number of computational cells in the span-wise direction by half. However an LES calculation on a grid with reduced number of cells in span-wise direction turned unstable almost immediately, thereby highlighting the effectiveness of the wall model. Besides reducing the number of grid points in the spatial domain, the relaxed grid resolution for the WMLES also permitted the use of a larger time step. This resulted in an order of magnitude reduction in the total CPU time relative to WRLES. / Master of Science

Oceanic waves registered by satellite observations often have curvilinear fronts and propagate over various currents. In this thesis, we study long linear and weakly-nonlinear ring waves in a stratified fluid in the presence of a depth-dependent horizontal shear flow. It is shown that despite the clashing geometries of the waves and the shear flow, there exists a linear modal decomposition, which can be used to describe distortion of the wavefronts of surface and internal waves, and systematically derive a 2+1-dimensional cylindrical Korteweg-de Vries (cKdV)-type equation for the amplitudes of the waves. The general theory is applied to the case of the waves in a two-layer fluid with a piecewise-constant shear flow, with an emphasis on the effect of the shear flow on the geometry of the wavefronts. The distortion of the wavefronts is described by the singular solution (envelope of the general solution) of the nonlinear first order differential equation, constituting generalisation of the dispersion relation in this curvilinear geometry. There exists a striking difference in the shape of the wavefronts: the wavefront of the surface wave is elongated in the shear flow direction while the wavefront of the interfacial wave is squeezed in this direction. We solve the derived 2+1-dimensional cKdV-type equation numerically using a finite-difference scheme. The effects of nonlinearity and dispersion are studied by considering numerical results for surface and interfacial ring waves generated from a localised source with and without shear flow and the 2D dam break problem. In these examples, the linear and nonlinear surface waves are faster than interfacial waves, the wave height decreases faster at the surface, the shear flow leads to the wave height decreasing slower downstream and faster upstream, and the effect becomes more prominent as the shear flow strengthens.

Cette thèse s'inscrit dans un contexte d'étude de la dynamique des fronts atmosphériques et océaniques et de l'origine des ondes de gravité dans l'atmosphère. Pour cela on s'intéresse à un front composé de deux couches de fluides miscibles en milieu tournant et soumis à un cisaillement vertical. Un travail à la fois expérimental et numérique met en évidence des phénomènes agéostrophiques qui vont au-delà de la configuration équilibrée usuelle d'un front barocline. D'abord, l'étude des différents régimes instables d'un front en configuration annulaire en terme de nombre de Rossby et de Burger révèle une instabilité agéostrophique couplant des mouvements équilibrés et divergents grâce à la résonance entre une onde de Rossby et une onde de Kelvin. Cette instabilité Rossby-Kelvin a été confirmée numériquement grâce aux structures des perturbations en champs de vitesse dans chaque couche. Ensuite, des structures de petite échelle présentes sur le front ont aussi été observées expérimentalement. Les caractéristiques de l'interface en terme de nombre de Richardson et épaisseurs de l'interface en vitesse et en densité suggère une instabilité de cisaillement de Hölmböe. Une simulation directe numérique axisymmétrique avec un nombre de Schmidt valant 700 confirme cette conjecture. Des ondes inertie-gravité supplémentaires sont observées numériquement sur un mode instable Rossby-Kelvin et le mécanisme de génération de ces ondes est discuté. Enfin l'étude numérique d'un front stable a mis en évidence la présence de couches d'Ekman internes avec une structure additionnelle pour des valeurs élevées de nombre de Schmidt et un faible nombre de Rossby. Dans le cas de front en in/outcropping, la dynamique est modifiée par l'interaction du front avec les couches d'Ekman au niveau du point singulier d'épaisseur nulle. Elle dépend à la fois de la circulation verticale et du mélange sur le nez du front et des nombreuses instabilités possibles associées à des résonances d'ondes horizontalement et verticalement. / This thesis has to be seen within the general study of atmospheric and oceanic fronts and the origin of gravity waves in the atmosphere. In this context we focus on a front in a rotating two-layer miscible fluid under vertical shear. Both experimental and numerical study highlights ageostrophic phenomena going beyond the usual geostrophic equilibrated configuration of a baroclinic front. First, the classification of different instability regimes of a front in an annular configuration as a fonction of Rossby number and Burger number reveals an ageostrophic instability coupling equilibrated and divergent motions due to a resonance between a Rossby wave and a Kelvin wave. This Rossby-Kelvin instability is confirmed numerically by the structure of the perturbation velocity fields in each layer. Second, small-scale structures have also been observed experimentally. Caracteristics of the interface in function of Richardson number and density and velocity interface thicknesses suggests the presence of the Hölmböe shear instability. A direct numerical simulation with an axisymmetric configuration and with a Schmidt number 700 confirms this conjecture. Other smalle-scale perturbations compatible with inertia-gravity waves have been observed numerically superimposed on an unstable Rossby-Kelvin front and the wave generation mechanism is discussed. In addition, a numerical study of a stable front highlighted the presenec of internal Ekman layers with an additional interfacial structure in the case of high Schmidt number and small Rossby number. For fronts in in/outcropping, front dynamics is modified by interaction with Ekman boundary layer at the location of the intersection zero-thickness singular point. It depends on both vertical circulation and mixing on the nose of the front and the various possible instabilities associated to horizontal or vertical wave resonances.

Neural networks was shown to be very susceptible to small and imperceptible perturbations on its input. In this thesis, we study perturbations on two-layer piecewise linear networks. Such studies are essential in training neural networks that are robust to noisy input. One type of perturbations we consider is `1 norm bounded perturbations. Training Deep Neural Networks (DNNs) that are robust to norm bounded perturbations, or adversarial attacks, remains an elusive problem. While verification based methods are generally too expensive to robustly train large networks, it was demonstrated in [1] that bounded input intervals can be inexpensively propagated per layer through large networks. This interval bound propagation (IBP) approach lead to high robustness and was the first to be employed on large networks. However, due to the very loose nature of the IBP bounds, particularly for large networks, the required training procedure is complex and involved. In this work, we closely examine the bounds of a block of layers composed of an affine layer followed by a ReLU nonlinearity followed by another affine layer. In doing so, we propose probabilistic bounds, true bounds with overwhelming probability, that are provably tighter than IBP bounds in expectation. We then extend this result to deeper networks through blockwise propagation and show that we can achieve orders of magnitudes tighter bounds compared to IBP. With such tight bounds, we demonstrate that a simple standard training procedure can achieve the best robustness-accuracy tradeoff across several architectures on both MNIST and CIFAR10. We, also, consider Gaussian perturbations, where we build on a previous work that derives the first and second output moments of a two-layer piecewise linear network [2]. In this work, we derive an exact expression for the second moment, by dropping the zero mean assumption in [2].