Pressure Fluctuations

JuliaB

Hi,

I have been testing some parameters using the dambreak testcase from SPHeric (similar to the CaseDambreak) and compared the results with the experimental results by Kleefsman (2005). The water height data fit rather well in the end. However, when looking at the resulting pressure, very strong fluctuations could be observed. I did measure the pressure at a distance of 1.5 h from the boundary and did not manage to improve this behavior by changing any of the following parameters: resolution (dp), viscosity, kernel function, time-step, delta-SPH, smoothing length. The results for P1(front of obstacle) and P5 (top of obstacle) can be seen in the following figures:

https://www.dropbox.com/s/qh8j43fmc0mhmnw/P1.png?dl=0
https://www.dropbox.com/s/1xp94wqz4ahyrbn/P5.png?dl=0

It can bee seen, that the general behavior is pretty good, but the fluctuations are very strong.

Does anyone have any idea, what else could help? As far as I can see, Canelas et al. (2015) reached better results. However, they used an earlier version of DualSPHysics, where the Shepard filter was still included. Might this be the reason?

I would appreachiate any thoughts on this matter!

Best regards
Julia

musm

What happens when you decrease dp?

moh

Try SPS viscosity and set the viscosity to 0.000001
Also Symplectic scheme for time-step and CFL number equal to 0.12

JuliaB

Thank you for the answers so far!
I don't think that decreasing dp would solve the problem, since e.g. Canelas (2015) use an even larger dp than I am. I have tried it before and it resulted in even stronger fluctuations.

I started a simulation using the values recommended by moh, even though I did already try these values separately from each other.

JuliaB

As I expected from the experience altering the Viscosity and time-step schemes separately, the suggestions by moh even increased the oscillations and made the general behavior worse as well:

https://www.dropbox.com/s/l9du7y3cy2o2ehd/P1_new.png?dl=0
https://www.dropbox.com/s/j1r3fnb1yqvx2a7/P5_new.png?dl=0

I also decreased dp again to 0.005, which (at least at P5) increased the oscillations and also introduced new peaks (especially negative ones), see here:

https://www.dropbox.com/s/hzm99ck6w37ezw4/P1_fine.png?dl=0
https://www.dropbox.com/s/fkvsfsqkjoa7urh/P5_fine.png?dl=0

moh

Did you ALSO use the delta-SPH scheme (DeltaSPH=0.1)?
If yes, then the target is achieved by changing the "coefh".

JuliaB

Yes, I did also use DeltaSPH=0.1 (turning it off results in very strong fluctuations, of course) and I tried coef=0.8 and 1.2 (before it was 1.0) and both produced stronger fluctuations than before. Thank you for your effort, though!

alexmees

I'm working on a stepped spillway model which, admittedly, is quite different from yours but has similar dimensions, velocity range and geometry detail to that of Kleefman (2005). Here is what I can share with you:
1) I could only get results that were close to reality when I reduced dp to 0.0015. Larger than that you could tell that the results, even visually, were wrong.
2) The best pressure results I got were with coefh=0.9. Assuming that larger values allow for better wave propagation this seems like it is a coherent proposition since the structure by itself "breaks" wave continuity.
3) Changing the viscosity did change results but less that altering other parameters mentioned.
4) I actually got the "worst" results on pressure values if comparing with velocity profiles and wave height.

Alex

Dear all

Remember that when computing pressures, the pressure gauge location should be move 1.5*h far from the boundaries particles, since the boundary layer is effective 1.5*h far due to the gap between fluid and boundaries created by DBC. You can see results of applying this shift in CaseDambreak and CaseSloshing. This will avoid the need of drastically reducing "dp"

Regards

AldoP

Hi,

I start working with SPH for 2D slamming problems. I'm writing in this tread because I also found some issues with pressures fluctuations.

My simulation is a simple 2D wedge falling at prescribed velocity. I have used dp=0.002, CFL=0.2, Visco=0.05,coefh=1,coefsound=20,deltaSPH=0.1. The domain sizes have been set to avoid reflections at least during the impact. I have used the periodic BC at the sides of the domain (XPeriodicIncZ).

Pressures are taken using the measure tool according to what I found in the case for the sloshing. I have tried both a distance of 1.5h and also 2.0h from the bottom surface of the wedge. The probes moves with the same motion law of the wedge, so the distance (wedge-probe) is kept all over the simulation.

The problem is that repeating 3 times the same simulation I found 3 different time series of the pressures, using the same probe location. Since it seems to me that all the formulation is deterministic, is there something that I'm missing or is it possible that this happens?

Thanks a lot

Here the results. The experimental data are taken from Yettou et al. (2006). 'Experimental study on the water impact of a symmetrical wedge'. In case I can also send the xml file.

https://www.dropbox.com/s/u9ej97haf5rppg9/SIMN_storyP3.png?dl=0

moh

The problem is that you get different answers by using the same parameters.
This problem is absolutely related to your CPU!
The voltage of the CPU core has some abnormal fluctuations which infects the CPU.

Alex

That is very strange, you should obtain always the same results!
You can try double precision but I do not think this is the issue.
parameter key="PosDouble" value="2"

AldoP

Hi,

thanks for the replies.

Alex: I see in another discussion in the forum that there is an issue with measuretool using double precision. With v4.0 I obtain the error 'pos not defined...'. I have downloaded again the complete windows package but there is still v4.0 of the tool. So at the moment I'm not able to check your suggestion. BTW when do you think it will be released? Thanks

moh: I'm not sure if I understood your answer but I expect that this kind of issues related to the hardware settings will affect the resuls in the range of the truncation error, while I obtain substantial differences. If you anyway think this could be the problem do you have any practical suggestions to fix it?

Thanks again