Strange grid senstivity behaviour

  • Last Post 15 October 2018
ahmadzaki posted this 11 October 2018


I've been conducting a grid sensitivity analysis for an opened building inside a domain. I took five different grids, all structured, the meshing order was the same for all of them (meshing steps), and the reduction percentage was the same. However, the analysis didn't give a decreasing/ increasing behaviour it looked as shown; the mass flow rate percentage difference across the building relative to the inlet mass flow rate was compared for each case. which are the accurate ones? and why it is behaving like this? 

 Thank you


Order By: Standard | Newest | Votes
klu posted this 12 October 2018

Hi Ahmad,

Can you please replace the X axis with the cell numbers? Also would you please show us the convergence for each mesh case?

ahmadzaki posted this 12 October 2018

I'm using FLUENT, so my iterations were stopped when the residuals levelled up, however, I unified the number of iterations in all of them (the convergence was set to 10^-6). This is the curve with the number of grid elements where M stands for million


rwoolhou posted this 12 October 2018

Please can you post images of the domain and residuals. Whole building models rarely show convergence at 10^-6 as there is nearly always flow separation somewhere in the domain and/or interacting jets because of the scale. 

ahmadzaki posted this 12 October 2018


This is the picture of the domain.


and this is a screenshot of two different grid (coarse and fine) residuals (all of them have similar  behaviour)


thank you

Kremella posted this 12 October 2018


Couple of thoughts:

  • What kind of model are you running - K-epsilon: enhanced wall treatment or wall functions, and what are your y+ values?
  • Do you have any monitors and have they become steady? Generally, it is always a good practice to create as many monitors as possible when performing grid independence studies. If it likely that your residuals are flat, but your monitors are oscillating and therefore, you are seeing different results. Please make sure you have reached true steady state solution by using appropriate monitors.

I hope this helps.

Best Regards,


ahmadzaki posted this 13 October 2018

Hello Karthik,


I'm using K_epsilon, with default wall functions.

and yes, I had velocity monitors and they became steady (levelled up into a straight line).


thank you

rwoolhou posted this 15 October 2018

I think the finer mesh is picking up some transient eddy shedding on the building. Can you plot contours of pressure on the sides of the building and take an image every 10 iterations for a few hundred iterations?  I'd use the execute commands feature to do this: doing it manually will be tedious! 

For a built environment model I usually reckon on a high imbalance due to these effects, and switching to transient isn't practical. 

  • Liked by
  • ahmadzaki