Perform LBM simulations on iRonCub-Mk1 and compare the results with FVM simulations

[antonellopaolino]

Following #139 we would like to start testing the Lattice Boltzmann Method (LBM) based open-source code FluidX3D with the results obtained on iRonCub whole-body geometry using a Finite Volume Method (FVM) as the one implemented in the commercial code ANSYS Fluent.
This activity can be the launching point leading to GPU massively parallelized CFD simulations.

cc @fabiodinatale @gabrielenava @DanielePucci

[antonellopaolino]

LBM simulations cold start with FluidX3D

Reproducing the same commands used in a couple of tutorials I made a first LBM simulation with FluidX3D using iRonCub .stl generated via SpaceClaim. I did this first test to see if there are any critical points in importing faceted surface stl files generated from Ansys SC, but it seems the procedure is pretty straightforward.

The result has no coherent physical parameters with the actual flow around the robot since I didn't change the $Re$ or the flow velocity $U$ but I just used a pre-set simulation. I report it just because it's pretty cool :)

iRonCub_FluidX3D_1080p.mp4

Next step is to investigate the correct way to setup simulations which are consistent with the physics in the Fluent simulations (i.e. boundary conditions, resolution, flow properties, ...), and then run everything on the iitbmp014srv001

cc @gabrielenava @fabiodinatale @DanielePucci

[antonellopaolino]

Perform a first simulation with the moving robot

One of the biggest advantages of this methodology is the meshing time, in FVM method-based softwares (as Fluent) it can take some minutes even with a relatively coarse geometry, while the procedure in LBM it's much faster, since just the boundaries are meshed (voxelized), therefore we can make the robot move in the simulation in a very simple and straightforward way.
In this test I made a simulation (reducing the resolution to have faster result, since it's for testing purpose) where the robot it's still for 0.5 sec, then it rotates forward from 90 deg to 0 deg in 1 sec (it's super fast but it's just to understand how it works) and then it stays still for 0.5 sec.
The final result is shown in the following slowmo vide. I used the laptop GPU, with the super powerful ws it can be much better, also there are some vortices (e.g. between the arms) which appear just for the reduced resolution, while they should not appear with a very refined one.
Another progress is that the video in the previous comment was a capture of the laptop screen, while now it's automatically generated from the software itself saving the .png captures and making a video with a basic python script.

iRonCub_moving_FluidX3D_720p.mp4

cc @gabrielenava @DanielePucci @fabiodinatale

[DanielePucci]

This is simply fantastic @antonellopaolino.

[antonellopaolino]

Hovering simulation LBM vs RANS forces and torques comparison

I made a low resolution test (using 1 GB of GPU VRAM) to understand how to extract values and compare them to the results already obtained from classic CFD using RANS on the robot reduced model.

Plots

Video

iRonCub_hovering_720p.mp4

Data comparison

At the same flow conditions the obtained forces are:

	RANS	LBM
$C_DA$	0.240	0.264
$C_LA$	0.0152	-0.00861
$C_SA$	-0.0161	-0.000863

In terms of average the results seem not to differ too much, further investigations will follow, considering a more refined lattice resolution, probably using the most powerful GPUs.
On the torques I didn't compare the values since for now I'm not sure of the center w.r.t. they are computed.

[S-Dafarra]

Awesome videos! I would have two curiosities:

• at what velocity is the wind flowing?
• does it consider the air flowing out from the turbines?

[antonellopaolino]

• at what velocity is the wind flowing?

The wind is flowing at 17 m/s but the video is slowed down of 10x.

• does it consider the air flowing out from the turbines?

As all the simulations done until this moment the flow from the turbines is not considered since it won't have any comparison with the wind tunnel data and the FVM simulations (we couldn't turn on the turbines in the wind tunnel, so we validated the calculations with incompressible flow equations, which wouldn't be valid with the turbines outflow).

[akhilsathuluri]

There is this group from TU Munich that works on physics based fluid simulations: https://ge.in.tum.de/research/
They have some open source libraries too. Might not be what you are looking for now, yet relevant.

[antonellopaolino]

Thanks @akhilsathuluri! I had a quick look and their work seems quite interesting, even if more oriented to learning-based simulation approaches, but could be a useful reference for future developments!

[antonellopaolino]

Code for LBM simulations

I added a commit (d9f3b2e) to store in this repo the main() function in the setup.cpp file used to run LBM simulations on iRonCub robot geometry as reported in the previous comments using the FluidX3D software.
With this file there are also some utilities to generate the videos and force/torque plots of the simulations.
These scripts are written in python language but I added an environment.yml file to install the required packages from conda to be able to run the related code.

cc @fabiodinatale @gabrielenava @DanielePucci