Cette thèse présente une étude numérique des régimes turbulents au sein d'un écoulement de Poiseuille plan forcé par un gradient de pression constant. L'effort numérique a porté principalement sur le concept d'Unité Minimale. Dans la première partie, des simulations en régime turbulent ont été conduites en géométrie périodique. Les DNS en Unité Minimale montrent que, l'activité turbulente se trouve localisée à proximité d'une des parois, et que la dynamique aux temps longs s'organise autour de renversements abrupts. Dans la seconde partie, on recherche par le calcul les états cohérents exactes en particulier les états dits frontière. Ces états frontière, obtenus par dichotomie, sont caractérisés par tourbillons longitudinaux et une paire unique de stries toujours localisées à proximité d'une seule paroi. Des représentations de la dynamique dans l'espace des phases sont reconstruites à l'aide de divers observables. La dynamique d'un renversement s'articule autour de visites transitoires vers un espace de solutions quasi-symétriques. Une onde progressive exacte, instable et quasi-symetrique a ainsi été identifiée. L'analyse de stabilité révèle que ses vecteurs propres séparent l'espace des phases en deux basins distincts. La dernière partie remet en question l'auto-similarité des différents régimes d'équilibre d'écoulement. Contrairement aux études récentes qui se concentrent sur les solutions à structure symétrique imposée, nos résultats suggèrent que les unités de parois sont également pertinentes pour les états frontière lorsqu'ils sont localisés près d'une paroi, meme si l'auto-similarité n'est pas aussi flagrante que pour les régimes turbulents. / This thesis numerically investigates the dynamics of turbulence in plane Poiseuille flow driven by a fixed pressure gradient. The focus is especially on computations carried out within the minimal flow unit (M.F.U.). In the first part, turbulent simulations are carried out in spatially periodic channels. In the M.F.U. simulations, the turbulent activity appears to be localised near one wall and the long term dynamics features abrupt reversals. In the next part, we look numerically for exact coherent states in the M.F.U. system. Edge states, which are computed using bisection exhibit streamwise vortices and a single pair of streaks localised near only wall at all times. Different state space representations and phase portraits were constructed using appropriately chosen variables. The dynamics along a turbulent reversal is organised around transient visits to a subspace of (almost) symmetric flow fields. A nearly-symmetric exact travelling wave (TW) solution was found in this subspace. Stability analysis of the TW revealed that its unstable eigenvectors separate the state space into two symmetric basins. In the last part of this thesis, the self-similarity of the different non-trivial equilibrium flow regimes computed in this work, is addressed. Contrarily to most studies focusing on symmetric solutions, the present study suggests that inner scaling is relevant for the description of edge regimes as well although the self-similarity is not as satisfactory as for the turbulent regimes.

Recently, a new kind of turbulence has been discovered in the flow of concentrated polymer melts and solutions. These flows, known as purely elastic flows, become unstable when the elastic forces are stronger than the viscous forces. This contrasts with Newtonian turbulence, a more familiar regime where the fluid inertia dominates. While there is little understanding of purely elastic turbulence, there is a well-established dynamical systems approach to the transition from laminar flow to Newtonian turbulence. In this project, I apply this approach to purely elastic flows. Laminar flows are characterised by ordered, locally-parallel streamlines of fluid, with only diffusive mixing perpendicular to the flow direction. In contrast, turbulent flows are in a state of continuous instability: tiny differences in the location of fluid elements upstream make a large difference to their later locations downstream. The emerging understanding of the transition from a laminar to turbulent flow is in terms of exact coherent structures (ECS) — patterns of the flow that occur near to the transition to turbulence. The problem I address in this thesis is how to predict when a purely elastic flow will become unstable and when it will transition to turbulence. I consider a variety of flows and examine the purely elastic instabilities that arise. This prepares the ground for the identification of a three-dimensional steady state solution to the equations, corresponding to an exact coherent structure. I have organised my research primarily around obtaining a purely elastic exact coherent structure, however, solving this problem requires a very accurate prediction of the exact solution to the equations of motion. In Chapter 2 I start from a Newtonian ECS (travelling wave solutions in two-dimensional flow) and attempt to connect it to the purely elastic regime. Although I found no such connection, the results corroborate other evidence on the effect of elasticity on travelling waves in Poiseuille flow. The Newtonian plane Couette ECS is sustained by the Kelvin-Helmholtz instability. I discover a purely elastic counterpart of this mechanism in Chapter 3, and explore the non-linear evolution of this instability in Chapter 4. In Chapter 5 I turn to a slightly different problem, a (previously unexplained) instability in a purely elastic oscillatory shear flow. My numerical analysis supports the experimental evidence for instability of this flow, and relates it to the instability described in Chapter 3. In Chapter 6 I discover a self-sustaining flow, and discuss how it may lead to a purely elastic 3D exact coherent structure.