Fluid Flow over a bus -- Convergence Problem

Roberto J. · May 11, 2016, 13:52

Hi everyone.

I am am relatively new to CFD and I am hoping to get some advice with my current project. I have model air flow around a passenger bus and determine separation points. Essentially a rectangular box with wheels. the bus is 110mm wide, 126mm high and 487mm long.

I have modeled it using the following models:
All y + Wall treatment
SST K-Omega
K-Omega Turbulence
RANS
Turbulent
Constant Density
Segregated flow
Gas
Steady
3D

I used a polyhedral mesh with 12 layer 3mm prism layer which gives y+ values of between 0-1. The number of cells is around 470 000 with a base size of 5 mm.

However when I run it after 3 hours and 600 iterations it doesn't seem to converge. The velocity inlet condition was set at 120kph. Is there anyway I can improve my convergence? I have attached my residual results and Cd results

marmot · May 11, 2016, 15:52

If you are going for a y+ <1 you should use the low Re approach not the all Y+ approach.

Looking at your residuals, its not great but one could live with that convergence, you always need to look at other variables like you did with drag, I would say drag is your main indicator for convergence,

fluid23 · May 12, 2016, 13:05

This is almost certainly an unsteady flow (due to separation) which is why you are seeing such poor convergence. Switch your physics models from steady to implicit unsteady and set your inner iteration stopping criteria to 100, at least to start with. You may eventually need to play around with different time steps (keeping your largest CFL number below 5-10 if time and resources allow). Assuming you don't have mesh issues then you will see your residuals drop, then spike and drop off again at each new time step.

Your turbulence model selection is appropriate. K-W SST is good for separated flows. The use of a low-Re model doesn’t make sense with K-W since there is no such thing as a low-Re k-w model. There are low-Re damping parameters already built into the model which you can adjust, but I do not recommend doing so. The use of the all wall y+ approach simply gives you flexibility to leave some wall boundary meshes coarse while refining others that may be more important to you while maintaining some level of accuracy. In general, k-w sst works best with y+ around 1.

Similar Threads
Thread	Thread Starter	Forum	Replies	Last Post
convergence problem with compressible flow	sarah l	FLUENT	0	July 1, 2015 15:47
Waterwheel shaped turbine inside a pipe simulation problem	mshahed91	CFX	3	January 10, 2015 11:19
convergence problem in 2D steady supersonic flow	sehs15	FLUENT	4	November 6, 2014 07:06
Problem with Convergence (Over flow)	armlic	CFX	4	July 14, 2014 04:48
Convergence Problem in Axisymmetric Periodic Flow	atheresia	FLUENT	3	February 10, 2014 03:00

May 11, 2016, 15:52		#2
marmot Senior Member kevin alun Join Date: Sep 2011 Location: Germany Posts: 106 Rep Power: 14	If you are going for a y+ <1 you should use the low Re approach not the all Y+ approach. Looking at your residuals, its not great but one could live with that convergence, you always need to look at other variables like you did with drag, I would say drag is your main indicator for convergence,

May 12, 2016, 13:05		#3
fluid23 Senior Member Matt Join Date: Aug 2014 Posts: 947 Rep Power: 17	This is almost certainly an unsteady flow (due to separation) which is why you are seeing such poor convergence. Switch your physics models from steady to implicit unsteady and set your inner iteration stopping criteria to 100, at least to start with. You may eventually need to play around with different time steps (keeping your largest CFL number below 5-10 if time and resources allow). Assuming you don't have mesh issues then you will see your residuals drop, then spike and drop off again at each new time step. Your turbulence model selection is appropriate. K-W SST is good for separated flows. The use of a low-Re model doesn’t make sense with K-W since there is no such thing as a low-Re k-w model. There are low-Re damping parameters already built into the model which you can adjust, but I do not recommend doing so. The use of the all wall y+ approach simply gives you flexibility to leave some wall boundary meshes coarse while refining others that may be more important to you while maintaining some level of accuracy. In general, k-w sst works best with y+ around 1.