# Transient Simulations - Effect of number of grid points

 Register Blogs Members List Search Today's Posts Mark Forums Read

 April 16, 2018, 06:03 Transient Simulations - Effect of number of grid points #1 Member     Thamilmani M Join Date: Sep 2017 Location: IIT Bombay, Mumbai Posts: 52 Rep Power: 8 Hello everyone, I am working on a transient 2D laminar flow over two tandem cylinders in ANSYS. I used finer mesh by using different grid points on each side of cylinder. Let that be a. I used values of 40, 60 and 80. so that the total number of cells also keep increasing as we increase a. Now, I got the a 80 converges to a steady solution after around 5 lakh iterations. But a 40 or a 60 has run even more than that many iterations, but didn't converge to a steady solution, but the Cl plots are just growing. We know that, increasing number of grid points increase the fineness of the mesh, which improves the accuracy of the solution, however also increase the computational time greatly. But, Here we see that decreasing the number of grid points didn't actualy reduce the computational time as expected. Though a 80 would still be at higher accuracy. Why? Following case conditions are same of all those meshes: Upstream distance is 10D, and Downstream distance is 20D and the walls are at 10D each. Re is 100. With PISO solver. All discretizations are Second order. Can anyone reason out this Why? Or is there something fundamentally wrong in me? Thanks for all. Always Thedal... __________________ Always Thedal

 April 16, 2018, 07:00 #2 Senior Member   Filippo Maria Denaro Join Date: Jul 2010 Posts: 6,810 Rep Power: 72 What is wrong is that at Re=100 the solution is physically unsteady. Are you using second order upwind? I can suggest to try the NITA setting the second order central discretization. Then, be careful about the grid resolution around the cylinders

April 16, 2018, 07:23
#3
Member

Thamilmani M
Join Date: Sep 2017
Location: IIT Bombay, Mumbai
Posts: 52
Rep Power: 8
Quote:
 Originally Posted by FMDenaro What is wrong is that at Re=100 the solution is physically unsteady. Are you using second order upwind? I can suggest to try the NITA setting the second order central discretization. Then, be careful about the grid resolution around the cylinders
Dear Sir,

Thanks for the reply. I am using Second Order Implicit scheme for Transient formulations sir. Yes, sir Solution is physically unsteady, But why does a lesser number of cells take more time to converge than a higher one is my primary doubt?
__________________
Always
Thedal

April 16, 2018, 08:06
#4
Senior Member

Filippo Maria Denaro
Join Date: Jul 2010
Posts: 6,810
Rep Power: 72
Quote:
 Originally Posted by thedal Dear Sir, Thanks for the reply. I am using Second Order Implicit scheme for Transient formulations sir. Yes, sir Solution is physically unsteady, But why does a lesser number of cells take more time to converge than a higher one is my primary doubt?

As the case is unsteady, I assume that for "convergence" you do not mean the steady state but the convergence of the iterative methods at each time step. The slower convergence can be due to high local truncation error on the coarse grid as well as to the guess solution that is far from the finaln one.

April 16, 2018, 08:54
#5
Member

Thamilmani M
Join Date: Sep 2017
Location: IIT Bombay, Mumbai
Posts: 52
Rep Power: 8
Quote:
 Originally Posted by FMDenaro As the case is unsteady, I assume that for "convergence" you do not mean the steady state but the convergence of the iterative methods at each time step. The slower convergence can be due to high local truncation error on the coarse grid as well as to the guess solution that is far from the finaln one.
Sorry for my misconception and hence the wrong answer, I am actually not asking of the convergence then. Within each timestep, However the convergence is faster than 80, But, to attain steady state, 40 takes more number of timesteps than 80. Why would that be? And Hence more computational time for 40 as well.
__________________
Always
Thedal

April 17, 2018, 07:33
#6
Member

Thamilmani M
Join Date: Sep 2017
Location: IIT Bombay, Mumbai
Posts: 52
Rep Power: 8
Quote:
 Originally Posted by FMDenaro As the case is unsteady, I assume that for "convergence" you do not mean the steady state but the convergence of the iterative methods at each time step. The slower convergence can be due to high local truncation error on the coarse grid as well as to the guess solution that is far from the finaln one.
So, There is no relation between the No. of timesteps taken for the flow to attain steady (oscillations are steady) and refinement of mesh?
__________________
Always
Thedal

April 17, 2018, 07:36
#7
Senior Member

Filippo Maria Denaro
Join Date: Jul 2010
Posts: 6,810
Rep Power: 72
Quote:
 Originally Posted by thedal So, There is no relation between the No. of timesteps taken for the flow to attain steady (oscillations are steady) and refinement of mesh?

The number of time steps, or better the final time to reach a developed flow, generally depends on the initial condition and on the Reynolds number.

The time step is a parameter depending on the mesh size when the numerical stability constraint must be fulfilled.

 Tags ansys, computational time, cylinders, meshing, unsteady