[aestas]

In my case, i solved two UDS and flow.
Density method was used, the solution control is like this:
QQ??20151018232814.jpg
2ddp solver was used
when i set the discretization method of flow as first order upwind, the solution could converged quickly.
But when i change it to second order upwind(at about 921 step), after some iteration error quick occured like this:
2222.jpg
================================================== ==========
time step reduced in 2 cells
1115 2.0238e+00 2.5777e-01 1.9400e+00 1.9950e+00 3.0582e-05 1.9039e-06 0:00:05 805

time step reduced in 2 cells

absolute pressure limited to 1.000000e-002 in 2 cells on zone 2
1116 5.6257e+00 7.1932e-01 5.4224e+00 5.5522e+00 1.2272e+00 7.9264e+09 0:00:04 804

time step reduced in 1145 cells

absolute pressure limited to 1.000000e-002 in 7 cells on zone 2

absolute pressure limited to 5.000000e+012 in 30 cells on zone 2

temperature limited to 5.000000e+012 in 86 cells on zone 2

Error: Floating point error: overflow

Error Object: ()
================================================== =========

i don't know how it happens and how to avoid this,could someone help me?

P.S. how much difference can the two method cause to the results?

[skewness abyss]

I have had this happen to me before when switching between 1st order and 2nd order upwind using the density solver but without UDFs. I solved the problem by reducing the Courant number prior to the switch and letting the simulation run for some time before ramping up the Courant number slowly. Try this to see if it works.

[aestas]

Originally Posted by skewness abyss

I have had this happen to me before when switching between 1st order and 2nd order upwind using the density solver but without UDFs. I solved the problem by reducing the Courant number prior to the switch and letting the simulation run for some time before ramping up the Courant number slowly. Try this to see if it works.

Thanks for your advice, my friend.
But I reduced the Courant number from 1 to 0.1 or even 0.01 prior to the switch, but the error still occurred....

Sometimes when i switch 1st order to 2nd order error like this occurred:
============================
862 1.9177e+00 3.1486e-01 1.7713e+00 1.8499e+00 3.7691e-05 3.2944e-06 0:00:01 99732
863 3.5531e+00 5.7955e-01 3.2954e+00 3.4302e+00 9.5398e-05 4.0079e-06 0:00:01 99731
864 1.1012e+01 1.8027e+00 1.0253e+01 1.0635e+01 4.3546e-03 1.2504e-04 0:00:01 99730

time step reduced in 2 cells
865 6.5079e-01 1.5483e-01 6.5359e-01 6.3682e-01 3.7083e-03 2.9342e-02 0:00:01 99729
866 2.9763e+00 -1.#INDe+00 2.4637e+00 -1.#INDe+00 1.#QNBe+00 1.#QNBe+00 0:00:01 99728

reversed flow in 18 faces on pressure-outlet 5.
! 867 solution is converged
867 -1.#INDe+00 -1.#INDe+00 -1.#INDe+00 -1.#INDe+00 1.#QNBe+00 1.#QNBe+00 0:00:00 99727
============================

[aestas]

Originally Posted by skewness abyss

I have had this happen to me before when switching between 1st order and 2nd order upwind using the density solver but without UDFs. I solved the problem by reducing the Courant number prior to the switch and letting the simulation run for some time before ramping up the Courant number slowly. Try this to see if it works.

And i tried to solve the problem first without any UDFs(only solve the flow) with Courant number=1 , the result could converge quickly within 50 iterations.

However, i changed Courant number to 0.01 before switching discretization of Flow to 2nd order upwind, and the error still exists
======================
432 2.4575e-01 2.2115e+00 2.5251e-01 2.5380e-01 0:00:48 99620
433 2.4643e-01 2.2096e+00 2.5320e-01 2.5450e-01 0:00:39 99619
! 434 solution is converged
434 -1.#INDe+00 -1.#INDe+00 -1.#INDe+00 -1.#INDe+00 0:00:31 99618
==========================
3333333333333.jpg

[skewness abyss]

When I had this happen to me in the past, it was caused by a meshing issue. I had several skewed cells that prevented convergence using second order upwind. I switched to a structured mesh which gave me high quality cells and the solution converged just fine. Can you report the minimum quality of the mesh that you have and post a picture of your mesh so that we can diagnose the problem better?

[aestas]

Originally Posted by skewness abyss

When I had this happen to me in the past, it was caused by a meshing issue. I had several skewed cells that prevented convergence using second order upwind. I switched to a structured mesh which gave me high quality cells and the solution converged just fine. Can you report the minimum quality of the mesh that you have and post a picture of your mesh so that we can diagnose the problem better?

Ok,here they are：
=========================
Grid Check

Domain Extents:
x-coordinate: min (m) = -5.000000e-003, max (m) = 5.000000e-003
y-coordinate: min (m) = 0.000000e+000, max (m) = 1.000000e-002
Volume statistics:
minimum volume (m3): 2.506945e-007
maximum volume (m3): 2.770834e-007
total volume (m3): 9.500001e-005
Face area statistics:
minimum face area (m2): 5.000000e-004
maximum face area (m2): 5.555556e-004

=========================


> grid

/grid> quality

Grid Quality:
Applying quality criteria for quadrilateral cells.
Maximum cell squish = 1.22896e-003
Maximum 'aspect_ratio' = 1.53168e+000


i know little about mesh, in my case, first i meshed all the edges by interval size, then created the mesh of face
4444444444.jpg

[skewness abyss]

What is it exactly that you are trying to model? I don't see any geometry in this picture. Is this flow over a flat plate?

[aestas]

Originally Posted by skewness abyss

What is it exactly that you are trying to model? I don't see any geometry in this picture. Is this flow over a flat plate?

No， i am trying to solve a plasma problem in vacuum condition, the lower edge is cathode(pressure-inlet), upper is anode (pressure outlet), and the sides are set as walls or far field to simulate free diffusion.

[aestas]

Originally Posted by aestas

No， i am trying to solve a plasma problem in vacuum condition, the lower edge is cathode(pressure-inlet), upper is anode (pressure outlet), and the sides are set as walls or far field to simulate free diffusion.

Today i used a previous model which could successful converge with both 1st and 2nd order discretization of Flow to compare the difference of the results. And i meshed the grid more thick.
1.png

I found two problems:
1. The previous model with thin grid mesh density, could be directly converged with 2nd order discretization of Flow after initialization. But this new model would become divergence.
The way to use 2nd order discretization is, use 1st order first until convergence, then switch it to 2nd order(without any other change including Cournt number)
2.png

2. I compared the pressure and other parameters, the two discretization method greatly affected the results.

1st order

3.png

2nd order
4.png


This means i have to figure out how to use 2nd order without any error in the model i posted...

[devesh.baghel]

Hi,

1. Is it the result of stabilized flow field ?
2. what is residuals limit looking for convergence i.e. 10^-3 ?
If yes, are you switching over to 2nd order descritizatio once you reached 1st order convergence.
3. Results shown are till 1800 iterations or more ?

since 2nd order descritization impose tight calculation to stabilze the flow field, also depends on good guess of initialization.

Pressure contour reflects that, pressure loop is travelling towards outlet, I think you should run for more iterations by monitoring some scalar\ flow parameter for better convergence.

[aestas]

Originally Posted by devesh.baghel

Hi,

1. Is it the result of stabilized flow field ?
2. what is residuals limit looking for convergence i.e. 10^-3 ?
If yes, are you switching over to 2nd order descritizatio once you reached 1st order convergence.
3. Results shown are till 1800 iterations or more ?

since 2nd order descritization impose tight calculation to stabilze the flow field, also depends on good guess of initialization.

Pressure contour reflects that, pressure loop is travelling towards outlet, I think you should run for more iterations by monitoring some scalar\ flow parameter for better convergence.

Hi,my friend, thanks for your reply.
1. Yes, it is a steady case of the flow.
44444444444444444444444.jpg
2. Yes, except i set the residuals limit of UDS0 to 1e-6
123123123.jpg
And it is after the solution reached 1st order convergence that i switching over to 2nd order descritization
3.Result stopped at about1300 iterations, after i switch to 2nd order, the residuals quickly reached the criteria again.

The expected result is closer to 2nd result (the maximum pressure center is not at the upper outlet boundry but near the center of the geometry.)

And my main problem is, 2nd order may not converge regarding different cases (like the case i post in #1 with declining edge of the right and left sides). (Since 1st order can always converge) I don't know why.

[aestas]

My email address is hellofdy@gmail.com, i am looking forward to your reply and help, my friends.