Coherent structures of turbulence in wall-bounded turbulent flows

Abstract

Direct Numerical Simulation (DNS) of a fully developed turbulent channel flow represents a powerful tool in turbulence research: it has been carried out to investigate the main characteristics of wall-bounded turbulence. It consists of solving numerically the Navier-Stokes equations with physically-consistent accuracy in space and time. The major difficulty in performing turbulence calculations at values of the Reynolds number of practical interest lies in the remarkable amount of computational resources required. Recent advances in high performance computing, especially related to hybrid architectures based on CPU/GPU, have completely changed this scenario, opening the field of High Performance Direct Numerical Simulation of turbulence (HPDNS), to which new and encouraging perspectives have been associated with the development of an advanced numerical methodology for studying in detail turbulence phenomena. The research activities related to the Ph. D. Program concerns the high performance direct numerical simulation of a wall-bounded turbulent flow in a plane channel with respect to the Reynolds number dependence in order to investigate coherent structures of turbulence in the wall region. The objectives of the research have been achieved by means the construction and the validation of DNS turbulent flow databases, that give a complete description of the turbulent flow. The Navier- Stokes equations that governs the flow of a three-dimensional, fully developed, incompressible and viscous fluid in a plane channel have been integrated and a computational code based on a mixed spectral-finite difference scheme has been implemented. In particular, a novel parallel implementation of the Navier-Stokes solver on GPU architectures have been proposed in order to perform simulations at high Reynolds numbers. In order to deal with large amount of data produced by the numerical simulation, statistical tools have been developed in order to verify the accuracy of the computational domain and describe the energetic budgets that govern the energy transfer mechanisms close to the wall. Flow visualization has been provided in order to identify and evaluate the temporal and morphological evolution coherent structures of turbulence in the wall region. The objectives of the research have been achieved by means the construction and the validation of DNS turbulent flow databases, that give a complete description of the turbulent flow. The Navier- Stokes equations that governs the flow of a three-dimensional, fully developed, incompressible and viscous fluid in a plane channel have been integrated and a computational code based on a mixed spectral-finite difference scheme has been implemented. In particular, a novel parallel implementation of the Navier-Stokes solver on GPU architectures have been proposed in order to perform simulations at high Reynolds numbers. In order to deal with large amount of data produced by the numerical simulation, statistical tools have been developed in order to verify the accuracy of the computational domain and describe the energetic budgets that govern the energy transfer mechanisms close to the wall. Flow visualization has been provided in order to identify and evaluate the temporal and morphological evolution threedimensional, fully developed, incompressible and viscous flow. The second part is devoted to the study of the numerical method for the integration of the Navier-Stokes equations. A mixed spectral-finite difference technique for the numerical integration of the governing equations is devised: Fourier decomposition in both streamwise and spanwise directions and finite difference method along the wall-normal direction are used, while a third-order Runge-Kutta algorithm coupled with the fractional-step method are used for time advancement and for satisfying the incompressibility constraint. A parallel computational codes has been developed for multicore architectures; furthermore, in order to simulate the turbulence phenomenon at high Reynolds numbers, a novel parallel computational model has been developed and implemented for hybrid CPU/GPU computing systems. The third part of the Ph. D. thesis concerns the analysis of numerical results, in order to evaluate the relationship between turbulence statistics, energy budgets and flow structures, allowing to increase the knowledge about wall-bounded turbulence for developing new predictive models and for the control of turbulence