Last week, last post…
It is the end of August and sadly the Summer comes to an end. It is the most awaited time of the year for most of people, that time where we can forget of our usual routines and obligations to focus on our hobbies and passions.
As I explained you in my first blog post which now seems to be a hundredth of years ago, my passion since a really young age are racing cars and its technical aspect, especially aerodynamics. Therefore, one of my main hobbies is learning about aerodynamics and this Summer of HPC project has been a brilliant opportunity to do so.
First of all, starting the summer full of illusion with the HPC training, a great formation to anticipate this nice program and then meeting my project mate, Paolo Scuderi, and my project mentor from the University of Luxembourg, Ezhilmathi Krishnasamy. The three of us together we have formed a great team in order to go ahead with this project.
All Computational Fluid Dynamics projects are formed by three main steps:
- Pre-processing: dividing the domain in the control volumes where the velocity and pressure values will be computed. This step is also known as creating the mesh and it is crucial to obtain accurate results. The smaller the control volumes, the more accurate will be the results but the longer the computation will be. Hence it is essential to choose where the elements have to be fine and where they can be coarse.
- Solver setup: this step is based on setting the adequate boundary conditions and numerical configuration in the used solver. Any problem or mistake in this step will lead to a divergent solution or will slow the whole process. As part of this step there is also the choice of the adequate turbulence and wall models, as well as its set up.
- Postprocessing: it is the turn of checking the results and analyse them. Usually, it is good to compare them with experimental results to validate the whole process and know if the mesh and solver configuration are valid.
Personally, along the development of the project I focused first in the pre-processing side as I had a vast previous experience on professional software to prepare meshes. Then, I started learning together with my project mate how to use OpenFOAM (the used solver). This was definitely the most challenging part in the summer as I had to learn from scratch all the needed knowledge to perform this type of simulations.
It has been extremely challenging but rewarding all the battles against the OpenFOAM error messages in the Iris cluster of the University of Luxembourg, they have led me to gain a lot of CFD experience but also about how to manage jobs in HPC clusters. A part from that, I must say that all the performed studies were not only focused on aerodynamics and CFD, there were some about parallelization. Precisely about how to divide the elements of the mesh among the processors in order to get the most of the HPC benefits.
To finish this blog post I would also like to explain you a bit more of the obtained results. I will not give you much details now as I do not want to spoil our presentation which I hope that you all enjoy next Tuesday. Together with Paolo we have been analysing the three different Drivaer geometries without considering the effect of the wheels at three different Reynolds numbers (equivalent to velocities of 10, 20 and 30 m/s) trying different paralelization strategies and turbulence models. We have validated our model with experimental data from well known authors and thanks to that we managed to get to the conclusions that you will all see on Tuesday!
I hope that you all enjoyed this Summer and this program as much as I did and let’s do the final sprint during this week!