Search

Extracting flow statistics from flow simulations requires generally long in- tegration intervals in which a good deal of the time is spent time-averaging the results. In order to avoid this, ensemble averaging can be applied to reduce the overall simulation time by simulating multiple statistically independent flow states for a smaller period of time and later averaging among them. In order to optimize the process, a cross-platform portable parallel-in-time method is presented in which the overall simulation is solved making use of sparse matrix-matrix products (SpMM), a high arithmetic intensity counterpart for sparse matrix-vector products, so that the efficiency of the simulation is increased., J.P-R., F.X.T., A.A-B. and A.O. are supported by the Ministerio de Econom a y Competitividad, Spain, RETOtwin project (PDC2021-120970-I00). J.P- R. is also supported by the Catalan Agency for Management of University and Research Grants (AGAUR), and A. A-B. is also supported by the predoctoral grants DIN2018- 010061 and 2019-DI-90, by MCIN/AEI/10.13039/501100011033 and AGAUR. Calcula- tions are being performed on the MareNostrum 4 supercomputer at the BSC. The authors thankfully acknowledge these institutions., Postprint (published version)

An increasing amount of sensors is being placed in cars in order to monitor their vicinities, which performance is highly dependant on the driving conditions, in which weather has a big role, and thus water contamination is a topic that will increase its importance in the future of the car industry. One of these sources of contamination is self-soiling: how cars get dirty because of their own tires. Thus, historically speaking this phenomena has been treated computationally by the use of droplet rakes in the tires which do not consider how this water has been picked up, or how it detaches from the tire itself. In this thesis the method to apply results obtained from Smoothed-particle Hydrodynamics (SPH) simulations of the tire droplet adhesion and release, to vehicle soiling simulations in order to improve the accuracy of these studies has been developed. In order to develop the method, first the SPH simulation data has been treated by filtering it according to the droplet location and the tire characteristics which has lead to the development of a geometrical and a kinematical filter for the data set. Afterwards, the results for the coarsest data set have been applied in the external aerodynamics simulation of a DrivAer bluff car model, as a proof-of-concept simulation, in which a Lagrangian Multiphase + Fluid film simulation has been developed for three different regimes: an unsteady injection with data extracted from the SPH simulations (USPH), a steady injection (SSPH) with the same data set origin and eventually a variable mass flow rake simulation (VMFR), with particles being injected in a rake pattern following the classic method which has been the most extended method in industry and academia to simulate vehicle soiling. These three simulations have been compared among each other so that the main similarities and differences are presented in the report. It has been seen that the injection pattern of USPH and SSPH is similar compared to the classic VMFR simu

Direct Numerical Simulations of the incompressible Navier-Stokes equations for relatively high Reynolds numbers, as required for airfoils, are extremely expensive in terms of number of CPUs as well as processing time. Thus, small-scale modelling is a clever way to reduce this cost by introducing an extra dissipation in the form of a turbulent viscosity. In this thesis, the turbulence phenomenon is reviewed, from both theoretical and technical points of view, and applied to a turbulent Lid-Driven Cavity. In order to do so, different eddy viscosity models are applied in a Large Eddy Simulation formulation melded into a symmetrypreserving discretization that presents the optimal conditions for turbulence simulation. In fact, S3PQ model developed by CTTC is used and tested in a Lid-Driven Cavity at Re = 10000, provided its remarkable turbulent properties, in which the properties of a turbulent boundary layer are studied and compared to the theoretical approach, previously developed.