Convergence and discretization

karananand · August 5, 2010, 00:09

I have a model where the turbine blades are cooled internally. The discretization to NS equation was first performed with 1st order upwind for momentum, energy and viscous model chosen (k-w SST) and standard for pressure. The solution converged with residual set to 1e-4 for all except 1e-6 for energy. I also have mass avrged pressure, velocity and temperature monitors at outlet.

After the solution converged (just with residual criteria, surface monitors are still varying), to get more accurate results, I switched to Second order upwind for all except pressure which was left as standard and reducing the relaxation factors. this time monitoring both residuals and surface monitors. I ran the simulation for a while with 'none' as convergence criteria.

I have a couple of questions:
Firstly, I plotted the residuals and surface monitors. In the velocity monitor, the velocity has almost stabilized, but there exists a few wiggles (extremely minor oscillations). Can I consider the solution to be converged (all other monitors are flat, and converged) ? Is this expected for the discretization method chosen? Also, the residual for energy did not satisfy the 1e-6 mark, can i still say the solution is converged? (I have added a few pics for the grid and convergence history)

Secondly, Could doing this help? ->
Instead of changing the discretization to 2nd order upwind, I first change it to Power law from 1st order upwind. As far as i remember , power law is 1st order accurate but it is way less diffusive and it seems quite similar to hybrid differencing scheme. Is the solution obtained here good enough?
After I switch to power law, I come back to 2nd order, which is a higher order scheme. Could this reduce the wiggles, or help me approach a converged solution? I thought of going for QUICK, but i have a lot of tet meshes.

Other info:
The solver is Steady state -SIMPLE scheme for the staggered grid. The model has no significant curvature , nor high values of natural convection.
Grid is Hybrid with hex core and tetra around. Have a prism boundary layer at the airfoil surface. Entire meshed domain is fluid. Grid is conformal.

Images for grid: (coarse mesh)
http://img85.imageshack.us/gal.php?g...meshstrctu.jpg

Images for convergence (a bit refined mesh and better transition between mesh regions-i didn't add images for them, but you get the basic idea from above)
http://img651.imageshack.us/gal.php?g=residual.png

karananand · August 28, 2010, 23:31

Could anyone please help?

xrs333 · August 29, 2010, 02:07

More often than not physical flow is inherently transient, even though steady solver is used.

descendents1 · August 29, 2010, 07:07

I would say that you can call your solution "converged" after the 6000th iteration. It's probably still going to do the wiggles for the next 4000 iterations if you let it run on. You may have an issue with your mesh. There are sharp transitions in your mesh density that I don't like. See if you can improve those transitions and run again to 6000 iterations and compare. What's your current mesh size? Basically you might want to do a grid convergence study.

August 5, 2010, 00:09	Convergence and discretization	#1
karananand New Member Karan Anand Join Date: Feb 2010 Posts: 23 Rep Power: 16	I have a model where the turbine blades are cooled internally. The discretization to NS equation was first performed with 1st order upwind for momentum, energy and viscous model chosen (k-w SST) and standard for pressure. The solution converged with residual set to 1e-4 for all except 1e-6 for energy. I also have mass avrged pressure, velocity and temperature monitors at outlet. After the solution converged (just with residual criteria, surface monitors are still varying), to get more accurate results, I switched to Second order upwind for all except pressure which was left as standard and reducing the relaxation factors. this time monitoring both residuals and surface monitors. I ran the simulation for a while with 'none' as convergence criteria. I have a couple of questions: Firstly, I plotted the residuals and surface monitors. In the velocity monitor, the velocity has almost stabilized, but there exists a few wiggles (extremely minor oscillations). Can I consider the solution to be converged (all other monitors are flat, and converged) ? Is this expected for the discretization method chosen? Also, the residual for energy did not satisfy the 1e-6 mark, can i still say the solution is converged? (I have added a few pics for the grid and convergence history) Secondly, Could doing this help? -> Instead of changing the discretization to 2nd order upwind, I first change it to Power law from 1st order upwind. As far as i remember , power law is 1st order accurate but it is way less diffusive and it seems quite similar to hybrid differencing scheme. Is the solution obtained here good enough? After I switch to power law, I come back to 2nd order, which is a higher order scheme. Could this reduce the wiggles, or help me approach a converged solution? I thought of going for QUICK, but i have a lot of tet meshes. Other info: The solver is Steady state -SIMPLE scheme for the staggered grid. The model has no significant curvature , nor high values of natural convection. Grid is Hybrid with hex core and tetra around. Have a prism boundary layer at the airfoil surface. Entire meshed domain is fluid. Grid is conformal. Images for grid: (coarse mesh) http://img85.imageshack.us/gal.php?g...meshstrctu.jpg Images for convergence (a bit refined mesh and better transition between mesh regions-i didn't add images for them, but you get the basic idea from above) http://img651.imageshack.us/gal.php?g=residual.png

Similar Threads
Thread	Thread Starter	Forum	Replies	Last Post
Discretization and convergence	karananand	Main CFD Forum	0	August 4, 2010 17:19
Convergence question with regards to discretization	karananand	FLUENT	8	July 16, 2010 15:49
Convergence question with regards to discretization	karananand	Main CFD Forum	0	July 16, 2010 01:27
Convergence problems	Chetan	FLUENT	3	April 15, 2004 20:13
CAVITATION (convergence criteria & discretization)	ROOZBEH	FLUENT	1	October 6, 2003 10:32

August 28, 2010, 23:31		#2
karananand New Member Karan Anand Join Date: Feb 2010 Posts: 23 Rep Power: 16	Could anyone please help?

August 29, 2010, 02:07		#3
xrs333 Senior Member xrs333 Join Date: Aug 2010 Posts: 125 Rep Power: 17	More often than not physical flow is inherently transient, even though steady solver is used.

August 29, 2010, 07:07		#4
descendents1 New Member Ben Jimenez Join Date: Aug 2010 Posts: 5 Rep Power: 16	I would say that you can call your solution "converged" after the 6000th iteration. It's probably still going to do the wiggles for the next 4000 iterations if you let it run on. You may have an issue with your mesh. There are sharp transitions in your mesh density that I don't like. See if you can improve those transitions and run again to 6000 iterations and compare. What's your current mesh size? Basically you might want to do a grid convergence study.