RWTH Publications (RWTH Aachen) · 2014 · 15 citations · 0 references
Open access
ComputingHigh-performance ComputingPhysicsComputational PhysicsDistributed SystemsMultiscale ModelingLattice Boltzmann MethodsStatistical MechanicsLattice Boltzmann EquationLattice Boltzmann SimulationLattice ModelsOctrees
This thesis deals with the prediction of aerodynamic noise resulting from the flow through porous media. A porous flat plate silencer of a pneumatic valve terminal serves as a model test case. In an experimental setup, the porous plate was put in the outflow of an air supply pipe and the aerodynamic noise was recorded at five positions on a cylindrical hull surface around the outflow. This thesis aims at correctly reprodudcing sound pressure levels by performing a fully resolved simulation of the flow through the porous pores and a direct computation of the sound generation. This task is approached with the Lattice Boltzmann Method (LBM), which is based on a simplified, statistical model of particle movement allowing the simulation of fluids in a weakly compressible regime. An essential part of this work is the development of the highly parallel, octree-based flow solver Musubi. Musubi is part of the APES-simulation suite, providing tools based on the central octree structure for solving complex fluid problems and allowing the flexible and efficient usage with extremely large data sets. Details and data structures of the flow solver are presented. On modern computing platforms, an efficient solution of large problems can be achieved only through an increased amount of parallelism and heterogeneity in the machines. Software has to adapt to this development and must be specially tailored to make use of the theoretically available computing capacity. This steady increase in complexity also affects programming, by which maintenance, extension and porting of scientific software becomes increasingly difficult. Therefore new programming paradigms are created enabling the compilers in theory to automatically perform large amounts of optimization for these complex machines. The parallel extension of Fortran is such a new model and is evaluated for suitability in the context of the LBM in this thesis. The solver Musubi is then validated by means of suitable test cases for all flow regimes. The latter include the turbulent flow through porous media, the aeroacoustic sound generation as well as a validation of the local grid refinement which is needed to model such a multi-scale problem. Finally, the insights gained are applied to solve the actual problem of the flow through the porous flat plate silencer.