The potential flow initial-boundary value problem describing fluid-structure interaction with fully nonlinear free surface boundary conditions has been studied using a mixed Lagrangian-Eulerian formulation. The boundary-value problem has been solved in the physical domain by means of a Bubnov-Galerkin formulation of the Laplace equation. The initialvalue problem related to the behavior of some of the moving boundaries has been discretized using various numerical techniques. Among these is a series of predictor-corrector methods. These methodologies proved to require considerable numerical smoothing to maintain stability of the numerical scheme. In turn, dissipation led to inaccuracies in the solution of the problem. In order to avoid this negative effect, a semi-implicit semi-Lagrangian two-time level iterative scheme that is almost free from smoothing has been developed. A Bubnov-Galerkin formulation of an elliptic system for the generation of boundary fitted curvilinear coordinates has been used. When solved iteratively, this method provides orthogonal meshes of very good characteristics for both symmetric and non-symmetric domains. Previous publications concluded that the present system was inadequate for non-symmetric regions leading to lack of convergence in the iterative process. Solutions described in this work show that this limitation has been overcome. Fluid responses to periodic excitation of surface-piercing and submerged bodies have been calculated. Both linear and nonlinear cases show agreement with published results. Very low total energy/work error has been obtained which demonstrates accuracy, good stability and convergence characteristics of the numerical scheme. The impulsive response of tanks of various shapes has also been simulated. Resulting natural frequencies show good agreement with available data. A slender body representation of the flow around a hull advancing with forward speed in otherwise calm water has also been simulated. Numerical calculations of a number of quantities of engineering interest are presented for different length Froude numbers. Results compare favorably with experimental data. / Applied Science, Faculty of / Mechanical Engineering, Department of / Graduate

Engineering problems frequently require solving a sequence of dual linear systems. This paper introduces recycling BiCG, that recycles the Krylov subspace from one pair of linear systems to the next pair. Augmented bi-Lanczos algorithm and modified two-term recurrence are developed for using the recycle space. Recycle space is built from the approximate invariant subspace corresponding to eigenvalues close to the origin. Recycling approach is extended to the CGS and the BiCGSTAB algorithms. Experiments on a convection-diffusion problem give promising results. / Master of Science

A new combination of techniques for the numerical computation of incompressible flow is presented. The temporal discretization bases on the discontinuous Galerkin-formulation. Both constant (DG(0)) and linear approximation (DG(1)) in time is discussed. In case of DG(1) an iterative method reduces the problem to a sequence of problems each with the dimension of the DG(0) approach. For the semi-discrete problems a Galerkin/least-squares method is applied. Furthermore a non-overlapping domain decomposition method can be used for a parallelized computation. The main advantage of this approach is the low amount of information which must be exchanged between the subdomains. Due to the slight bandwidth a workstation-cluster is a suitable platform. Otherwise this method is efficient only for a small number of subdomains. The interface condition is of the Robin/Robin-type and for the Navier-Stokes equation a formulation introducing a further pressure interface condition is used. Additionally a suggestion for the implementation of the standard k-epsilon turbulence model with special wall function is done in this context. All the features mentioned above are implemented in a code called ParallelNS. Using this code the verification of this approach was done on a large number of examples ranging from simple advection-diffusion problems to turbulent convection in a closed cavity.

A new combination of techniques for the numerical computation of incompressible flow is presented. The temporal discretization bases on the discontinuous Galerkin-formulation. Both constant (DG(0)) and linear approximation (DG(1)) in time is discussed. In case of DG(1) an iterative method reduces the problem to a sequence of problems each with the dimension of the DG(0) approach. For the semi-discrete problems a Galerkin/least-squares method is applied. Furthermore a non-overlapping domain decomposition method can be used for a parallelized computation. The main advantage of this approach is the low amount of information which must be exchanged between the subdomains. Due to the slight bandwidth a workstation-cluster is a suitable platform. Otherwise this method is efficient only for a small number of subdomains. The interface condition is of the Robin/Robin-type and for the Navier-Stokes equation a formulation introducing a further pressure interface condition is used. Additionally a suggestion for the implementation of the standard k-epsilon turbulence model with special wall function is done in this context. All the features mentioned above are implemented in a code called ParallelNS. Using this code the verification of this approach was done on a large number of examples ranging from simple advection-diffusion problems to turbulent convection in a closed cavity.

Cette thèse porte sur l'estimation simultanée du coefficient de diffusion et du terme source régissant le modèle de transport de la température dans les plasmas chauds. Ce phénomène physique est décrit par une équation différentielle partielle (EDP) linéaire, parabolique du second-ordre et non-homogène, où le coefficient de diffusion est distribué et le coefficient de réaction est constant. Ce travail peut se présenter en deux parties. Dans la première, le problème d'estimation est traité en dimension finie ("Early lumping approach"). Dans la deuxième partie, le problème d'estimation est traité dans le cadre initial de la dimension infinie ("Late lumping approach"). Pour l'estimation en dimension finie, une fois le modèle établi, la formulation de Galerkin et la méthode d'approximation par projection sont choisies pour convertir l'EDP de transport en un système d'état linéaire, temps-variant et à entrées inconnues. Sur le modèle réduit, deux techniques dédiées à l'estimation des entrées inconnues sont choisies pour résoudre le problème. En dimension infinie, l'estimation en-ligne adaptative est adoptée pour apporter des éléments de réponse aux contraintes et limitations dues à la réduction du modèle. Des résultats de simulations sur des données réelles et simulées sont présentées dans ce mémoire. / This work deals with the diffusion and source term estimation in a heat transport model for tokamaks plasma . This phenomenon is described by a second-order linear parabolic partial differential equation (PDE) with distributed diffusion parameter and input. Both "Early lumping" and "Late lumping" approaches are considered in this thesis. First, once the heat model is chosen, the Galerkin formulation and the parameter projection method are combined to convert the PDE to a set of ordinary differential equations (ODEs). Then, two estimation methods able to give optimal estimates of the inputs are applied on the reduced model to identify simultaneously the source term and the diffusion coefficient. In the infinite dimensional method, the adaptive estimation technique is chosen in order to reconstruct "freely" the unknown parameters without the constraints due to the model reduction method. Simulation results on both simulated and real data are provided to attest the performance of the proposed methodologies.