XSPH model in new DualSphysics V4.0

mac198833

Hi,

I am trying to simulate the Sloshing and Swirling behavior in Sphere water Tank by DualSPHysics. It was noticed that the simulation results shows larger damping and the water motion of experiment is much severer than the simulation (according to the video). The radius of the tank is 0.6m and Dp=0.01m is utilized in my calculation. Is there any idea to improve the sloshing calculation?
I also found the XSPH model is eliminated in the new DualSPHysicsV4. As the XPSH model is efficient to stable the calculation, is there any special reason to omit it in the new version code?

Thank you.

Alex

New version of the code includes other improvements such as shifting or delta-SPH, so that, XSPH is not needed.

In order to improve your simulation you should use a low value of ALPHA (visco in artificial viscosity). You should start with 0.01 and increase in case the simulation is not stable. Also note that varying "Dp" alpha should also vary.

Regards

mac198833

Thank you a lot for the kind reply.

In fact, I have used a small artificial coefficient as alpha=0.000001 for the case Dp=0.01m. I am not sure if it is still helpful to use smaller alpha than 0.000001 in my calculation. Also, is there any materials which discussed the relationship between the alpha and Dp?

In fact, the experiments shows a significant swirling when the total force in perpendicular direction of tank motion is almost identical with the force in the same direction of tank motion. But I did not get the same behavior by the DualSphysics.

I tried once with Dp=0.02 (same coefh=0.91924 with Dp=0.01) and the calculation results for swirling is a little better. So I guess the interaction between particles in larger range may be helpful to get the swirling. Also, I found a previous discussion about the “Reduce Damping” between kfield and Slongshaw, and the conclusion about the smoothing length is similar. Now, I am trying a larger smoothing length (1.7) for Dp=0.01 and see what will happen.

Another question is about the calculation speed. The present model for Dp=0.01 has totally 542235 particles (bound=108976, fluid=433259). With single-precision and coefh=0.919, it takes 430.35s to calculate one physical second. My GPU is GTX980Ti with 2816 cuda cores. It will double the time when I use coefh=1.7 or triple the time when I use double precision. As it is necessary to simulation a long enough physical time (more than 100s) to get a stable swirling, is there any good idea to improve the calculation efficiency?

Thank you.

Alex

Your value of alpha=0.000001, basically, means no viscosity. Minimum value is suggested to be 0.01 as I told you before.
Increasing coefh you should also obtain better propagation

Regards

mac198833

Thank you a lot for the comments. I will try again for alpha=0.01.