Solver blows up when applying inflation layers

Provides a system for patient-specific cardiovascular modeling and simulation.
POST REPLY
User avatar
Dylan van der Schoot
Posts: 10
Joined: Wed Jan 16, 2019 7:40 am

Solver blows up when applying inflation layers

Post by Dylan van der Schoot » Fri Mar 15, 2019 4:18 am

Hi all,

I'm encountering a problem when trying to solve flows on a mesh with inflation layers on the wall. When I try to solve the flow in a simple straight tube with, for example, 2 inflation (boundary) layers on the wall, the solver blows up. This is not the case when I'm not using inflation layers. I tried temporal refinement, but whatever I try, the solution will blow up. Does anybody have an idea what causes this problem and how to solve it?

Dylan

User avatar
David Parker
Posts: 1718
Joined: Tue Aug 23, 2005 2:43 pm

Re: Solver blows up when applying inflation layers

Post by David Parker » Fri Mar 15, 2019 10:38 am

Hi Dylan,

What version of SimVascular / svSolver and which OS are you using?

It might be that there are bad elements (Jacobian <= 0) in your mesh. You can check using ParaView.

If you want you can put your project on DropBox and I'll take a look.

Cheers,
Dave

User avatar
Dylan van der Schoot
Posts: 10
Joined: Wed Jan 16, 2019 7:40 am

Re: Solver blows up when applying inflation layers

Post by Dylan van der Schoot » Wed Mar 27, 2019 6:59 am

Hi David,

I've updated my version of Simvascular, and the meshes with inflation layers are now generated without bad elements. So that's great!

However, I am now encountering another problem, that is that I can't seem to reach grid convergence.
For testing purposes, I am simulating with a simple straight tube with D=1.4 and L=10cm. I gave an aortic flow curve on the inlet (scaled version of this: https://ars.els-cdn.com/content/image/1 ... 16-gr1.gif, with max speed at peak systole v = 30cm/s) with a parabolic velocity profile. I am doing three cycles of 1 sec per cycle, time steps are chosen such that CFL is always below 1.

So I've used the same geometry and I am only changing the global max edge size and, if required for stability, the time step size.
I am up to 8.000.000 elements right now, and the solution is still changing by lot. There seems to be no convergence in any variable (Parameters at probe points, velocity profiles, pressures on volume, wss, etc.). Any idea which might be the cause for this? I have the feeling that the backflow stabilization introduces an error relative to the gridsize, but that's a wild guess. For the sake of checking this I am currently running a refinement study on a flow without bulk backflow. To proceed with my research I need to prove that I am able to get grid convergence, else the results will be very questionable.

Cheers,

Dylan

User avatar
Weiguang Yang
Posts: 110
Joined: Mon Apr 07, 2008 2:17 pm

Re: Solver blows up when applying inflation layers

Post by Weiguang Yang » Thu Mar 28, 2019 8:58 am

Can you tell me your mesh size and upload a plot/figure that shows the changes in your quantities of interest vs your mesh size? What are the residuals printed in histor.dat? Do they look ok? Usually it is a very difficult quantity for WSS to reach satisfactory convergence with pulsatile flow with a complex geometry. Steady flow is easier. Pressure should not vary too much if the result is converged.

User avatar
Dylan van der Schoot
Posts: 10
Joined: Wed Jan 16, 2019 7:40 am

Re: Solver blows up when applying inflation layers

Post by Dylan van der Schoot » Mon Apr 01, 2019 6:51 am

Hi Weiguang Yang,

Thanks for your response! I've gathered some results and combined them in the excel sheet provided here:
https://we.tl/t-NtjtI08dsk
For all the simulations the residuals were below 10^-3, so no blowup is occuring in the simulations. Pressure is for instance also varying quite a lot, and is not converging.

User avatar
Weiguang Yang
Posts: 110
Joined: Mon Apr 07, 2008 2:17 pm

Re: Solver blows up when applying inflation layers

Post by Weiguang Yang » Wed Apr 03, 2019 10:29 am

Thanks for your plots. Could you please clarify the "velocity in x-dir over centerline of tube"? Is it a time-averaged or instantaneous plot? How did you define the error for pressure. If the flow is fully developed, the velocity profile should be independent of the longitudinal coordinate assuming a rigid wall. Thus the velocity profile in the middle of your tube should be the same as the inlet profile. If I were you, I would try 1. Test a steady case 2. Extend the length to >10D say 20-25 cm.

POST REPLY