Urban Geometry External Flow: Convergence and Validation
 

As a first time poster, I would like to preface this post by stating that I am relatively new to CFD, so I apologise in advance if my questions are trivial.

I've built an external flow model in Star-CCM+ to simulate the flow around twin building configurations. Using the literature provided, I managed to get some nice initial values and boundary conditions but I'm concerned about their validity, given my inexperience. I've attached screenshots of the residuals and XY, XZ vector plots to accompany my queries.

Firstly, boundary conditions: Wall for the ground+buildings, velocity inlet for the inlet, pressure outlet for the outlet and symmetry for the rest. Reasonable? The explanation of the function of some of the boundaries are a little hard to grasp, in terms of their applicability.

I had trouble with convergence at first but then I changed the turbulent viscosity and turbulent intensity (I'm using a k-epsilon model) to more reasonable values and it made a huge difference. However, I'm still troubled by the "flatlining" of the residuals, even though they've reduced by a considerable amount. Is this an acceptable convergence? I've even read in some places that a converged solution doesn't necessarily imply a correct one. How can I be confident that a simulation result is reasonable?

I've read the documentation, but it's no substitute for a forum of well-informed individuals...

Cheers.

P.S. Any reference to literature which acts as a "CFD for dummies" would also be appreciated...

I would say your solution domain is a little small, but acceptable.

Are my results valid? is a good question, which begs another question - what makes a solution accurate? You need to compare your analysis with some quantities of interest - drag, lift, pressures, something.

Your residuals look fine, but you need to also know if your engineering quantities are converging.

Thanks for the response. I've since changed the domain size to 1000x517x300m for a pair of 50m high 17m square buildings in the centre of the domain.

My only problem with an independent engineering quantity in this case is that I'm not sure what would be appropriate. I've been monitoring scalar and vector plots during the simulation to see if they match up with other plots in the literature, flow separation, vortex generation etc. However, I don't think there is any quantity that can be numerically monitored in this case...

What is the goal? What is the purpose of running this study? Are you just trying to learn cfd?

I know I look at the residuals, but that is not what I use for convergence. I use whatever report/reports I am looking for in the simulation. For my internal flow problems, I look at my outlet mass flow rate most of the time for convergence. For external aerodynamic problems, I look at coefficient of drag and lift.

Well, there's a few things you can do.

Try and search your literature for some kind of experimental data. Perhaps drag from scaled models, or pressure on the sides of a building? There has to be something.

Furthermore, you can always monitor values just to see if they stop changing. Calculate a drag report or something similar, and see if it converges to a value.

I'm trying to generate a 3D array of numerical velocity values to include in an analysis of the dynamic response of a small UAV. I'm also trying to learn CFD at the same time.

For different building configurations, there are specific flow characteristics which I am simply looking out for... Horseshoe vortices, rolled up vortices at building base with increased vorticity in downwind building etc... These are well defined in the literature...

As I don't have anything numerical to monitor, my plan was just to let it run for a considerable number of iterations until the flow characteristics match that of previous studies. I keep the scalar and vector plots open and stop when: the residuals have converged and I'm satisfied that the flowfield is representative of what I'm expecting...

I might just do that. As I explained in reply to another poster, I am determining convergence "visually", as it were, until I'm satisfied that the relevant scalar and vector plots match results found in the literature. A prerequisite for this is the convergence of the residuals, I figure.

Not very scientific, I know. I may look for numerical pressure data. That's a good suggestion.