Author |
Message |
Frank Grom (franklin_g)
New member Username: franklin_g
Post Number: 1 Registered: 06-2004
| Posted on Tuesday, June 15, 2004 - 08:16 am: | |
Hi there while trying to solve a problem of electrohydrodynamics i had some difficulties. i tried to solve by penalty method described in sample/vicous. it works pretty good by setting visc=1 and dens=1, but when i change these values to realistic ones like visc=10-3 and dens=10^3 i cant get an satisfiing result. would be great to get some tips frank |
Robert G. Nelson (rgnelson)
Moderator Username: rgnelson
Post Number: 161 Registered: 06-2003
| Posted on Tuesday, June 15, 2004 - 01:38 pm: | |
Frank Grom - Your script reads table files for Temp and Efeld. What are some typical values? |
Frank Grom (franklin_g)
New member Username: franklin_g
Post Number: 2 Registered: 06-2004
| Posted on Tuesday, June 15, 2004 - 05:48 pm: | |
typical values of the Temperature field are between 20°C-20.3°C. The electrical field varies between 10^6 V/m and 10^9 V/m. |
Robert G. Nelson (rgnelson)
Moderator Username: rgnelson
Post Number: 162 Registered: 06-2003
| Posted on Wednesday, June 16, 2004 - 08:18 pm: | |
The fundamental problem is that at extremely low viscosity, the fluid equations will support many turbulent solutions, and selecting among them becomes an almost impossible job. Mathematically, this appears as an extremely ill-conditioned coupling matrix. Compounding the problem is the fact that a very low-viscosity fluid can support extremly thin shearing transitions, which cannot be represented by the finite element approximation over a coarse mesh. Your desired parameters appear to be effectively those of water in a pool 100 meters square. Since water can be stirred in a coffee cup, you can see that the possible eddy structures in the swimming pool can be enormously complex. There are a few things that can be done in such problems, though I doubt that any of them will allow you to achieve the parameter regime you want: 1) Use a staged value of viscosity, and start at a value large enough to get a stable solution, gradually lowering the viscosity and hoping to avoid turbulence in the process. (Or just use a larger viscosity and stop there.) 2) Redefine your domain to enforce known symmetries. In your case, this would appear to ba an octant of the full domain. 3) replace the no-slip boundary with a slip boundary. At tiny viscosities, the shear layer will be very thin, and will approximate a slip boundary. There are a few other comments to be made about your script: 1) You didnt say which version of FlexPDE you are running, but if it's version 4, your declarations of variable Thresholds is much too large. This destroys the meaning of the error measure and allows sloppy solutions. The Threshold ("Range" in version 3) is unnecessary in steady-state problems in any case. 2) you have specified a pressure value of 1, even though the only quantities that enter the equations and BCs are derivatives. Since your pressure variations are like 1e-4, this forces the solution process to work in the fourth decimal place of the numeric values. Specify pressure=0, so the computation is dealing only with the variations. 3) Again, since only derivatives of the pressure appear, the solution of the P equation is not unique, and can be offset by an arbitrary constant. You might do better adding an integral constraint on P (CONSTRAINTS INTEGRAL(P)=0) to make the P solution unique. There may be other formulations of the flow equations which would perform better in this parameter regime, but we are not fluid dynamicists here, and I can't tell you what it might be. |
|