After experiences with the
simple cylinder example, the more complicated model of car geometry with an
open sunroof was analysed. First of all this required very careful meshing of
the domain. The cells near to the sunroof tip, where the vortex shedding
occurs, must be very small to be able to compute all the flow properties,
which influence the noise generation. On the other hand, if the whole domain
would be meshed with this size of cells, it would make the solution
infeasible. Again the details of the solution demands are discussed in the Final Project Report. In figure 8 the finest mesh used for the calculation is presented. In figure 9 the detail of the cells in the region near to the sunroof tip can be seen. The time needed for analysing the whole range of audible sound frequencies was estimated to be 160 hours for this mesh. 






Figure 8 




Figure 9 


However the complete analysis was not performed, because from the
pressure contours plots it is obvious, that the pressure prediction by LES
simulation is not correct in this case. This plot is presented in figure 10 together
with the static pressure plot obtained from the RSM simulation of the steady
state flow for the same case in figure 11. The next contour plots in figures 1217 show other flow properties obtained from the simulations. Click on the pictures for enlarging the plots. 



Figure 10 Static Pressure Contours – LES 
Figure 11 Static Pressure Contours –steady state, RSM 

Figure 12 – Velocity Vectors – LES 
Figure 13 – Velocity Magnitude Contours – RSM 




Figure 14 – Velocity Vectors detail – LES 
Figure 15 – Velocity Vectors – RSM 




Figure 16 – Vorticity Magnitude Contours – LES 
Figure 17 – Vorticity Magnitude Contours – RSM 


There could be more
reasons, why LES did not give desired results. It seems, that the problem
could be caused by the mesh. The cell size must grow in the direction from
the sunroof tip to reduce the total number of cells and thus the
computational time required. Unfortunately, the size change seems to be too
high when concerning the effects of the size on the LES simulation. The size
determines the scale of the turbulent eddies which will be modelled and of
those which will be computed. Because it varies so significantly across the
domain, it may well be the reason of incorrect LES. The second reason could
be using the 2D model together with LES, because turbulence is a 3D
phenomenon and it should also be simulated that way. In comparison of plots in figure 16 and 17, it can be clearly seen that the unsteady LES simulation produces vortices shedding from the sunroof tip. These of course cannot be seen in the time averaged RSM steady state example. The average distance between the vortices was used to calculate the approximate shedding frequency, which corresponds with the dominant noise frequency. This approach however does not allow predicting the noise level. 