Why is my 2-way FSI going to take so long to complete? Can I change something to make it go faster?

simonhackettsimonhackett Member
edited November 21 in Fluids

So I still have work to do for my simulation but my report is due in a week so I need to get something done to hand in. I ran this simulation before without knowing too much about the Fluent setup and so it was overly simplified, but not much has changed.

Last night I started the coupled analysis with the changes I made (some minor mesh refinements, enabling the energy equation, and changing a few physics settings) and it's been running for about 9.5 hours now at only 48% complete. Why is it taking so much longer than before? I'm doing the same amount of iterations but now it's going to take nearly 10x as long (it took just under an hour and a half before).

Is there a single specific setting that would do this? When I was running just Fluent on its own it would solve within a few minutes at the very most for even up to about 100 iterations. I'm only running 250 iterations of system coupling right now.

Unless someone has a solution, I'm just going to let this finish and lose any productivity I was going to have today because I don't want to make ANSYS crash by overstressing my CPU or something, but I am still going to be working on this after my formal project is due, and I would like to solve for a longer total simulation time (or at the very least, change some things and try again) without needing to forfeit my computer for nearly a whole day.

Answers

  • RobRob UKForum Coordinator

    The obvious one is parallel for Fluent, but you also want to look at update frequency and what's going on in the refined Fluent model. On the latter, if the solution is mesh dependent small changes in the mesh or settings can have a more significant effect on the solution, this then effects convergence and solver speed.

  • SteveSteve Forum Coordinator

    I agree with @Rob, running the refined Fluent only case should give more insight into what's happening.

    Some further information:

    It is very important to build up the FSI simulation in stages as opposed to setting up the 2-way FSI right at the start. This document "Best Practices for Coupled Fluid-Structure Interaction (FSI)" describes this process and is available here: https://ansyshelp.ansys.com/account/secured?returnurl=/Views/Secured/corp/v202/en/sysc_ug/sysc_bestpractices_fsi.html, System Coupling User's Guide \\ Best Practices for System Coupling \\ Best Practices for Coupled Fluid-Structure Interaction (FSI)

    FSI simulations with very soft materials or membranes are prone to numerical instabilities. In 2020R1 we have introduced a stabilization method in System Coupling called the Quasi-Newton Stabilization Algorithm. Note that this has to be used with the new System Coupling GUI or Command Line Interface that is run outside of Workbench. More information here: https://ansyshelp.ansys.com/account/secured?returnurl=/Views/Secured/corp/v202/en/sysc_ug/sysc_gen_scservice_dt_supplemental_iqnils.html, System Coupling User's Guide \\ System Coupling Data Transfers \\ Supplemental Processing Algorithms \\ Quasi-Newton Stabilization Algorithm

    You can also look at the following tutorial run with the System Coupling GUI or Command Line Interface that is run outside of Workbench: https://ansyshelp.ansys.com/account/secured?returnurl=/Views/Secured/corp/v202/en/sysc_tut/sysc_tut_oscplate_cli_fluent.html, System Coupling Tutorials \\ Tutorials with Command-Line Interface (CLI) Workflows \\ Oscillating Plate FSI with Fluent and Mechanical

  • @Rob what do you mean by "parallel for Fluent" and update frequency? About the mesh dependency, I do have a dynamic mesh enabled at the fluid-solid interface but that was enabled before when I did not encounter this problem with the run-time (is this what you mean?).

    @Steve, prior to running this coupled system I had done a lot of work with just the Fluent system alone in a separate project and it was giving me no such problems in regards to how long it took to solve. I essentially replicated all the things I did in that Fluent system for this FSI simulation, and only now is it taking so long. I had previously ran a very similar FSI but with some different Fluent settings and it took an hour and a half to solve.

  • RobRob UKForum Coordinator

    Fluent runs on a single core by default, to accelerate simulations we run in parallel, ie on many cores. I think you can use about 16 for a standard academic seat, but please check. Similarly, if you update the mech model every step it takes time, I'll leave @Steve to comment, but it may be possible to update Mechanical less frequently. If deformation is low you may find the FSI bit isn't too tightly coupled.

    As an aside, have a look at the intrinsic FSI in Fluent in the 2020 release, that avoids Mechanical entirely but doesn't have all the features.

  • @Rob, I run using 6 cores (I'm pretty sure my CPU only has 6). I haven't changed any of those settings and it was running just fine before tweaking those things in Fluent that were in my other discussion thread.

    I noticed something in a tutorial about FSI just using Fluent, but it was very different from what I'm used to dealing with so I figured it would be best to try and stick with making it as similar as the one-way case (that uses CFX) as possible.

    Can you think of any other settings I should check? Maybe I accidentally enabled (or disabled) something I shouldn't have? I don't see why adding viscous heating and those other things would be an issue only with the coupled system, my standalone Fluent model would solve in less than a few minutes.

  • RobRob UKForum Coordinator

    Not sure then, there shouldn't be any change in Fluent behaviour when hooked up to FSI but check your RAM & cpu load incase you're holding Mech or system coupling in the background. @Steve knows a lot more about this, I seem to cover multiphase & weird stuff in Fluent and have very little knowledge about how structures bend (I'm a Chemical Engineer so didn't do much solids at Uni).

  • SteveSteve Forum Coordinator

    Hi Simon,

    So to recap, there are three cases that are being discussed.

    1. Fluent only case with "some minor mesh refinements, enabling the energy equation, and changing a few physics settings"
    2. FSI case with Fluent setup from 1.
    3. Initial FSI case that runs in 1.5 hours.

    It's important to note that these cases are all very different and comparing the run times for them can be misleading without comparing the all the setup differences. If you attach some images of your geometry and list all changes between cases, it might be more clear why there are run time differences. There are many settings that can affect runtime so the investigation needs to start with comparing the differences between cases. Also at the end of the System Coupling log file there is a timing summary. This should give more information.

    +============================================================================+

    |               Timing Summary [s]               |

    +============================================================================+

    | Coupling Engine Time :                     2.8753E+00 |

    | Coupling Participant Time                         |

    |  Fluid Flow (CFX) :                      1.7293E+01 |

    |  Transient Structural :                    4.2373E+00 |

    | Total Time :                          2.4406E+01 |

    +============================================================================+

  • @Steve, I think I understand what you mean. I probably don’t have time to get to it today but I’ll do my best to go through it once I do have the time.

    (The FSI is being done partially academic project and also for a student engineering team. The academic project report has been finished and I’m just at the end of my term which is concentrated with midterms, assignments, and group project deadlines.)

    I have a preliminary project report which predates the updates I made in Fluent and describes the setup. I think I glossed over a few things but going off of that and my memory, I should be able to list everything that’s different. I may even have an archive of the FSI project when it ran in 1.5 hours too.

Sign In or Register to comment.