Learning Forum

[fuyu-m]

Hello,

I investigated "the resultant force in the X direction" applied to the sphere in the flow analysis around the sphere(see temp_p.1).

Model and machine used are the same, graphic boards are P5000, P4000 and K2000. The accuracy is Max Fiderity.

 

In R19.0 and R19.1, there was a big difference in "resultant in X direction" (see temp_p.2-4).

Please tell us why there was a big difference between R19.0 and R19.1 and whether the result of R19.0 or R19.1 is more correct.
Also, I would like to know if there is an effective countermeasure to eliminate this difference.
(Note that R 19.0 is closer to the theoretical value (0.046507) when "the resultant force in the X direction" is regarded as drag)

 

Thank you.

🛈 This post originally contained file attachments which have been removed in compliance with the updated Ansys Learning Forum Terms & Conditions

[Brian Bueno]

Wintercat

Discovery Live is intended to function primarily as a tool for design exploration, as opposed to design validation.  The idea is to provide real time feedback to changes in the design in order to give the user some indication of the predicted behavior of the model while avoiding the need to run a full simulation.  The trade off for this speed is accuracy.

Additionally, Discovery Live performs all the calculations for the simulation on the GPU of the graphics card.  Different graphics cards with different specifications (or more/less CUDA cores) may provide slightly different results.  

I believe some changes were made to improve the fluid solver for R19.1.  These are most evident when displaying the results for Pressure as a surface, composite, or iso-surface.  

Lastly, since the fluid models are transient, you may want to try comparing the time averaged results to see if they align more closely (if you have not done so already).

[fuyu-m]

BBueno_ANSYS 

Thanks for the advice.

Of course I know that Doscovery Live is not software to pursue accuracy.
I also understand that the results will change depending on the graphics card.

What I am currently addressing is that despite using the same graphics card, the result varies greatly depending on the version of Discovery Live.

For example, the result of "X direction resultant force" when using P5000 for graphics card was 0.044275 [N] in R19.0 and 0.03557 [N] in R19.1.
When the "resultant force in the X direction" is regarded as drag, the theoretical value of the drag of this model is 0.046507 [N], so the error is -3.14% for R19.0 and 23.52% for R19.1, The error is larger when calculated with R19.1.

This result is due to improvement of solver at R19.1, is it a useless result?

 

Thank you.

[Brian Bueno]

Wintercat

I assume the difference in the results is clear regardless of the diameter of the sphere, but can you please tell me what diameter you are using for that test for consistency? 

I can test this to see if similar results are produced locally and report this to the development team.

[fuyu-m]

BBueno_ANSYS 

Thank you for considering the test.

The diameter of the sphere is 10 cm and the flow velocity is 5 m / s.

I expect it from the result of the test.
 

Thank you.

[fuyu-m]

BBueno_ANSYS 

How is the progress of the test?
Please tell me if anything has been found.

Thank you.

[Brian Bueno]

Wintercat 

Sorry for the delay.  I'm awaiting a response from R&D and will update you as soon as I have the information.

[fuyu-m]

BBueno_ANSYS 

Thank you for your reply.
I am looking forward to the test results.

Thank you.

[Brian Bueno]

Wintercat

If you run the same problem, on the same hardware with the same fidelity setting, using the same release, you will get exactly the same results. If you change the fidelity setting, you may get different results (just like changing a mesh). The higher the fidelity, the more accurate the results should be. Note that the fidelity limits are hardware dependent. This means that running a problem at max fidelity on different hardware is not running the same simulation as the maximums are different.

This second question about the sphere results, we are looking into. The short answer is that we improved the fluids solver between 19.0 and 19.1. Whether the results on this model are better, is yet to be determined.

[fuyu-m]

BBueno_ANSYS 

Thank you for your attentive confirmation.

This time we use the same fidelity setting on the same problem on the same hardware (fidelity is set to maximize accuracy).
Only the release differs between R19.0 and R19.1.

I also understand that there are some differences in results as the release differs. However, I do not think the difference between the results this time is somewhat (for example in the result of P 5000 the result of R 19.1 is about 80% of R 19.0).
I expect that further verification will be done.

Thank you.

[fuyu-m]

BBueno_ANSYS 

How is the progress of the test?
Please tell me if anything has been found.

Thank you.