Generating a custom geometry and setting up the simulation

[Sanketvy1999]

Hello,
I am trying to benchmark a paper (An J, Wang H, Liu B, et al (2020) A Deep Learning Framework for Hydrogen-fueled Turbulent Combustion Simulation). The geometry is similar to backward step flame but in addition it has a ramp in the right. For the custom geometry I made a UserDefined function and took an intersection between a plane at 45 degrees and a box. The geometry created is shown below,

The ramp has small steps and thus is not smooth. I think this is not visualization error. I can attached the UserDefined function as well.
May you please help me create a smooth ramp?

[baperry2]

This is a visualization artifact. With the EB method and AMReX plt files, visualization software will render cut cells as if they are fully fluid. If you plot the volFrac variable you should see which cells are cut cells. For visualization, you can approximate the EB surface by taking an isosurface of volFrac, although bare in mind that this won't exactly correspond to the exact EB surface.

[Sanketvy1999]

Thankyou very much for your response! I will look into it. Also one more question. Will the boxes in EBs defined extends slightly beyond the domain? As my entire domain goes from x=0.0 to 0.1. So will the EB box on the left have x_lo = -0.01 and EB box on right have x_hi = 0.11?

[baperry2]

In general defining the EB surface to extend slightly beyond the domain boundaries as you have described is a good idea.

[Sanketvy1999]

In the same problem, I am not able to set the jet inflow properly. Before going to H2, I am trying to input CH4. I have set the massfrac[CH4_ID] = 1 in bcnormal_eb at the jet inlet. The plotfile after 10 iterations is shown below

Here you can see the range of Y(CH4) is not going to 1. I don't understand why it is happening. Moreover I try print statements and found out that bctype_eb function is never called. bctype_eb is used to mention where is the inflow and assign rest of the walls adiabatic.

Also while using H2 as the fuel instead of CH4, what changes do I have to make? Currently the mechanism file is drm19. How to change the derived controls for H2?
May you please help me set up the problem correctly.

Thank you.

[Sanketvy1999]

After 400 timesteps, I get the following profile,

Now the range of Y(CH4) went till 1. But after 500 iterations, the CVODE fails to integrate giving the following error,
[ERROR][rank 0][/usr/local/usrapps/techekki/SANKET_WORKSPACE/PeleLMeX/Submodules/PelePhysics/Submodules/sundials/src/cvode/cvode.c:3682][cvHandleFailure] At t = 0 and h = 1.60395e-16, the corrector convergence test failed repeatedly or with |h| = hmin.
giving Nans. I had set AMR level = 2, refinement ratio 2 2 2 2 and regrid after 5. The grid is 320 \times 128. What shall I do? The simulation should go until the flame develops completely.

[Sanketvy1999]

I am getting the following error.
Est. time step - Conv: 1.483716371e-06, divu: 9.603212958e-06
STEP [1059] - Time: 0.001466566592, dt 1.483716371e-06
SDC iter [1]
[WARNING][rank 0][/usr/local/usrapps/techekki/SANKET_WORKSPACE/PeleLMeX/Submodules/PelePhysics/Submodules/sundials/src/cvode/cvode.c:1457][CVode] Internal t = 6.64128e-07 and h = 5.28196e-23 are such that t + h = t on the next step. The solver will continue anyway.
[WARNING][rank 0][/usr/local/usrapps/techekki/SANKET_WORKSPACE/PeleLMeX/Submodules/PelePhysics/Submodules/sundials/src/cvode/cvode.c:1462][CVode] The above warning has been issued mxhnil times and will not be issued again for this problem.
[ERROR][rank 0][/usr/local/usrapps/techekki/SANKET_WORKSPACE/PeleLMeX/Submodules/PelePhysics/Submodules/sundials/src/cvode/cvode.c:1428][CVode] At t = 6.64128e-07, mxstep steps taken before reaching tout.
SDC iter [2]
amrex::Abort::11::Error: divU dtcell larger than actual dt ! !!!
SIGABRT
amrex::Abort::7::Error: divU dtcell larger than actual dt ! !!!
SIGABRT
amrex::Abort::0::Error: divU dtcell larger than actual dt ! !!!
SIGABRT
See Backtrace.11 file for details
See Backtrace.7 file for details
See Backtrace.0 file for details

It appears to me that due to stiffness, the timestep is so small that t + h = t .i.e. addition of h to current time has no numerical effect due to floating point limits. I can also attach the input file if needed. Please let me know how to resolve this.

Thank you.

[Sanketvy1999]

Hello,
Your suggestions on this will be helpful. Thanks.