CFD Online Discussion Forums

CFD Online Discussion Forums (http://www.cfd-online.com/Forums/)
-   Main CFD Forum (http://www.cfd-online.com/Forums/main/)
-   -   CFD VS SPH (http://www.cfd-online.com/Forums/main/14399-cfd-vs-sph.html)

James Downing November 19, 2007 08:38

CFD VS SPH
 
Hello

i would like to have your ideas about how to simulate hydrodynamic losses of a ship,

i try cfx and ls dyna via sph, as i know the cfd codes are better to simulate this kind of analysis but recentrly i hear that the sph method is better,

so is there anyone that try to see that , or anyone with any idea about that?

thanks james

Fab November 21, 2007 13:59

Re: CFD VS SPH
 
Hi James,

I actually would like to know more about SPH too; esp. concerning the calculation time and accuracy :)

Greetings! Fab

rt November 24, 2007 07:32

Re: CFD VS SPH
 
SPH is meshless method and simply it can be said that: its computational time is considerably higher than any mesh based method and due to irregularity of spatial stensil^1, its accuracy is also lower than mesh based methods (accuracy of boundary condition is also in question). e.g. read this paper:

http://dx.doi.org/10.1016/S0309-1708(03)00030-7

it is shown that convergence rate is below first order !

Also SPH in its formal version is weakly compressible and so time step size is severly limited by courant number based on sound speed. In incompressible format you have to slove a sparse unstrcutured poisson equation with large number of non-zero which is expensive to invert.

But when you have severe splashing e.g. atomization ... (presence of sub grid scale free surface) SPH is good candidate and could be superior than mesh based methods (in terms of CPU and accuracy).

1- To compare meshless vs. meshbase: assume 2d case, to reach to 2nd order accuracy of second derivative 3 point is sufficient on unifirm cartesian grid, but if sample points do not have reqularity it is needed to write multidimensional taylor seris and then for computation of second order derivative 9-sampe point is needed (this is more in 3d) also singularity of this matrix could be issue.



All times are GMT -4. The time now is 04:25.