Automotive - external aerodynamics and water management

Any car, wether it is a racing car or a standard road car, is strongly influenced by the surrounding airflow and can therefore be considered as a highly complex aerodynamic device. Finding an accurate and reliable way to design the shape of a car or certain components of it in order to use the influence of the airflow for an increase in performance is a great engineering challenge. In this context, fast and efficient flow field predictions, that can already be used in the very early design stage of the vehicle, are highly appreciated. E.g., elbe was used for the three-dimensional turbulent numerical simulation of a Formula Student race car to estimate the drag forces and down forces of several different underbody configurations.


EGN 2014 in the numerical wind tunnel.

Due to the increasing interest in all-terrain cars and the interest in developing cars for emerging markets, water-wading simulations become more and more relevant for the automotive industry. On the other hand, this type of simulation is very demanding, since it involves multiphase flow physics, a moving geometry and the need for a sharp interface capturing approach. For conventional CFD solvers, this is a highly demanding task, due to the mesh motion or re-meshing to capture the moving geometry, the cell refinement around the interface in order to reconstruct an sharp interface. This leads to huge amounts of grid cells and very small time steps, which results in computational times of weeks rather than days on a CPU cluster. Opposite to that, the elbe code highly benefits from combining the efficient LBM framework with GPGPU hardware: The inherent massive parallel layout of the graphical processing device combined with the local nature of the Lattice Boltzmann approach lead to computational and hardware costs that are only a fraction of those of a standard CFD simulation on a distributed memory cluster. The simulation scenario presented here is the water-wading of a car through a channel filled with water. Here, the car geometry is the DrivAer model (17) from the TU München, which is comparable with a real production car. This geometry is explicitly meant to provide free accessible detailed car geometry comparable for validation. For this simulation we choose the DrivAer Estate Back geometry with detailed underbody, mirrors and wheels (E_D_wM_wW).

The simulation is conducted in a domain with a length of 20 meters, a width of 5 meters and a height of 3.75 meters. The computational domain represents a channel with water height h=0.4 meters. Since we operate on a uniform grid with cell size meter, the total mesh size is 24 million cells. The simulated physical time is 3 seconds. The time step size is 0.0000325 seconds. The velocity of the car during the water-wading is 4 m/s. The computational time for 24 million cells and 3 seconds simulation time is 6 hours. The measured amount of million node updates per seconds (MNUPS) yields 100 MNUPS. We present the results for the simulation for some distinguished time steps: t=0, 0.5, 1.0, 1.5, 2.0, 2.5 seconds (see Figure 4). The visualization of the results shows the movement of the car through the channel. At time t=0 seconds we see the undisturbed water level. When the car starts to move, the deflection of the interface can immediately be seen. Due to the absence of surface tension effects in the present code, the fluid builds a water film on the car rather than a spray development. The development of the interface reproduces available reference data very well. The resulting computational time (6 hours for an intermediate grid) is very encouraging and shows that numerical simulations on GPGPUs with the elbe environment definitely allow to incorporate complex numerical simulations directly into the product development process.
  • C. Bieler, "Numerische Analyse und strömungstechnische Optimierung des Unterbodens eines Formula Student Rennwagens", Bachelor thesis, 2014.
  • L. Reichert, "Fluiddynamic Analysis of Front- and Rear Wings of a Formula Student Racing Car", Master Thesis, November 2015.
  • C. Janßen and T. Grahs, "Hochleistungsrechnen auf Grafikkarten für innovative Anwendungen in der Automobilindustrie", NAFEMS Magazin, accepted for publication.
  • C. Janßen and T. Grahs, "High performance computing on General Purpose Graphical Processing Units for innovative automotive application", Proc. of the NAFEMS Seminar “Innovative Applications of Computational Fluid Dynamics (CFD) in Product Development”. March 2013.