Abstract:
The advent of cheap massively parallel computer architectures in accelerators such as Graphical Processing Units (GPUs) has provided an impetus for computing fluid flow and heat transfer problems using particle methods such as Lattice Boltzmann Method (LBM) where, instead of solving the continuum Navier Stokes equations, the discrete Boltzmann equation is solved with collision models such as Bhatnagar-Gross-Krook (BGK) and the moving particle semi-implicit method based on Lagrangian formulation of computing thermo-fluid problems. The LBM is an efficient parallel algorithm for simulating single-phase and multi-phase fluid flows and for incorporating additional physical complexities. It is especially useful for modeling complicated boundary conditions and multi-phase interfaces. It is traditionally defined on two- or three- dimensional discrete structured lattices to model transport processes for different boundary conditions and has been established as a reliable flow solver and prediction tool in recent years. The current work explores and demonstrates the use of two different open source solvers implementing the LBM, namely, Sailfish (which runs on GPU cores) and Palabos (which runs on Multi-Cores). Various benchmark problems such as driven lid cavity problem, flow past bluff and streamlined bodies are addressed for validation and an insight into the reliability of the LBM for problems in applied aerodynamics. Further, to understand the parallel efficiency of implementing the LBM algorithm on high performance computing (HPC) platforms, a serial LBM code written in Matlab for flow past an airfoil has been parallelized using parallel Matlab (P-Matlab) on a Multi-Core and GPU based HPC system to accelerate the computations while ensuring high order accurate results. Traditional implementations of LBM uses Cartesian mesh of lattices, which require special treatment on irregular boundaries (which are typically encountered in applied aerodynamics) for the specification of boundary conditions and this is often cumbersome to implement in a code. The implementation of this special boundary condition treatment on the computed results is verified on selected benchmark problems and the results compared with the corresponding results from finite volume continuum based flow modeling techniques which is the standard practice in Computational Fluid Dynamics (CFD). The work also describes an attempt at the implementation of overlapping meshes inside the open-source solver Stanford University Unstructured (SU2) as a prelude to implementing overlapping mesh system of lattices for LBM and laying a starting point for addressing moving boundary problems in applied aerodynamics in the future.