How can I reduce the SPH analysis time?

Ovunc2

Hello,

I am working on sloshing motion of liquid in a tank. I'm using the SPH method for this problem in Radioss.

 

1. Is there a way to increase the time step and shorten the solution time without reducing the number of particles? What would you suggest me in this regard?
2. Also, I have SPH analysis of 9000 particles. The solution of this analysis takes 4 hours using 8 cores. For example, when I use 20 cores, the solution time takes much longer. What is the reason of this? Could it be that the model is too small, etc.? So could it take too long to collect data from cores?

 

Thanks 

 

Best Regards, 

Ovunc 

 

Rodrigo Merigo_22161

Hi!

There are a few points that you can review:

- Check the contact parameters: Make sure that the gap size is physical and the contact stiffness is adequate.

- Add the card /SPHGLO with alpha=0.25.

- Check the smoothing length (h) in the SPH property, if it is different from the default value, make sure it is compatible with the inter-cell distance.

Please check the RD-E: 4900 Bird Strike on Windshield in the Example Guide for more details. The SPH section in the User Guide can be very helpful too. 

Thank you.

Ovunc2

Rodrigo Merigo_22161 said:

Hi!

There are a few points that you can review:

- Check the contact parameters: Make sure that the gap size is physical and the contact stiffness is adequate.

- Add the card /SPHGLO with alpha=0.25.

- Check the smoothing length (h) in the SPH property, if it is different from the default value, make sure it is compatible with the inter-cell distance.

Please check the RD-E: 4900 Bird Strike on Windshield in the Example Guide for more details. The SPH section in the User Guide can be very helpful too. 

Thank you.

Hi, 

Firstly, thanks for your answer @Rodrigo Merigo.

 

• I added the card / SPHGLO with alpha= 0,25. But I did not observe the shortening of the solution time.

• Can you explain more about the item below? Because, I create SPH particle with 1D panel. (1D panel- SPH ) . For a liquid of a constant volume and filling ratio, I create a particle by entering the values ​​of h and density. As a result, the SPH property is formed by itself and and I don't change the h value in this property. 

"Check the smoothing length (h) in the SPH property, if it is different from the default value, make sure it is compatible with the inter-cell distance."

I think the part you mean is the part of the user guide that I sent in the attachment. Is it correct? Could you please  explain these sections too? 

Thank you . 

Best Regards, 

Ovunc

 

Ovunc2

Hi, 

Can you comment on why the solution time is not shortened despite the increase in the number of cores ?

Thank you. 

Ovunc

Rodrigo Merigo_22161

Ovunc2 said:

Hi, 

Can you comment on why the solution time is not shortened despite the increase in the number of cores ?

Thank you. 

Ovunc

Yes you are right, it is possible that you find that some number of cores is too much for the problem. In this case the estimated run time increases as the communication between cores is also increased. You can try running with different number of cores to find out the most suitable for the problem.

Rodrigo Merigo_22161

Ovunc2 said:

Hi, 

Firstly, thanks for your answer @Rodrigo Merigo.

 

• I added the card / SPHGLO with alpha= 0,25. But I did not observe the shortening of the solution time.

• Can you explain more about the item below? Because, I create SPH particle with 1D panel. (1D panel- SPH ) . For a liquid of a constant volume and filling ratio, I create a particle by entering the values ​​of h and density. As a result, the SPH property is formed by itself and and I don't change the h value in this property. 

"Check the smoothing length (h) in the SPH property, if it is different from the default value, make sure it is compatible with the inter-cell distance."

I think the part you mean is the part of the user guide that I sent in the attachment. Is it correct? Could you please  explain these sections too? 

Thank you . 

Best Regards, 

Ovunc

 

Yes, these texts are related to the smoothing length. You dont need to calculate it because the property will be filled automatically (default).

In order to decrease simulation time you can modify the time step. You can check in the starter out file (_0000.out) the time steps calculated for the model and also the imposed time steps you can use with added mass. Maybe adding a small percentage of mass with /DT/NODA/CST can reduce run time without changing the results significantly.

Ovunc2

Rodrigo Merigo_22161 said:

Yes you are right, it is possible that you find that some number of cores is too much for the problem. In this case the estimated run time increases as the communication between cores is also increased. You can try running with different number of cores to find out the most suitable for the problem.

Hi @Rodrigo Merigo , 

Thank you very much for the informations. 

Best Regards, 

Ovunc