Technical Support Request: Poor Performance of FLUENT‑EDEM Coupling Simulation and Coupling‑step Que

yanyanchen

Dear Altair and ANSYS Technical Support Team,

I am conducting submerged sediment‑pile scour two‑way coupling simulation on an AMD EPYC workstation with Intel‑MPI. My case contains 170 000 EDEM particles and 970 000 FLUENT fluid mesh elements, using time‑step sizes of 5e‑5 s for FLUENT and 5e‑7 s for EDEM. The simulation only reaches 0.2 physical seconds after one‑week continuous computation, which severely delays my research. Could you please provide official optimization suggestions for accelerating the coupled calculation, tell me where to set the coupling data‑exchange interval (how many FLUENT steps trigger data exchange with EDEM) in GUI, and share recommended time‑step ratio as well as MPI/parallel tips for AMD EPYC platform?

Looking forward to your reply.

Best regards,
[yanyan CHEN]

Stephen Cole

Hi,

It does seem very slow, I'd check if you can run each solution uncoupled just to check the time. I'd estimate that 0.2 s for EDEM alone should be taking ~30 minutes to run. If you run each case for a short time uncoupled and find one or the other is taking excessively long this helps check where the gains are, or if its purely related to coupling the two together.

Can you tell where the time is been spent in the simulation, its not always so easy to measure but the programs run sequentially, so one pauses while the other runs, and then there is a data transfer time. Based on what you see can you estimate how much time (percentage) is spent on each?

Also are you running CPU or GPU for EDEM? EDEM is significantly faster if you use a GPU solver rather than CPU.

What shape of particle is used in EDEM?

The data exchange is based on the time-steps of each, so if Fluent's time-step is 100x larger than EDEM then EDEM will take 100 steps to 'catch up' to Fluent's one. This 100:1 ratio is about the upper limit of what is normally recommended (10:1 to 100:1 is a usual range). However this is only a guide for stability, in the e-learning 'time based errors' shows what happens if the ratio is too big https://learn.altair.com/learn/course/cfd-dem-coupling-edem-and-fluent-elearning

Typically though you can increase the EDEM time-step by reducing the particle shear modulus, within reason. https://community.sw.siemens.com/s/article/33314-particle-shear-modulus-it-can-save-you-time

Another area to check is if the EDEM and Fluents domains match, the EDEM particles should always be inside the fluent mesh. If you have the EDEM domain larger and particles existing outside the domain this can cause the simulations to run longer as the program tries to find a mesh cell to contain the particle.

Regards

Stephen