MENUBlog by Yves-Marie Lefebvre, Tecplot Chief Technical Officer
It’s that time of the year again—Formula 1 enthusiasts are beginning to dust off their F1 team apparel and get ready for the races. New car liveries have been presented, pre-season tests deliver hints of what teams have been working on during the winter, the first grand prix is approaching, and Netflix released a new season of Formula 1: Drive to Survive.
Personally, all of the buzz brings me back to an experience I had several years ago: I lived this pre-season excitement from within one of the top teams, where I worked for a couple of months as a CFD methods consultant. I remember detailed pictures from spy photographers being streamed live from the track, highlighting new design features from competitive teams, and engineers gathering behind a monitor as everybody tried to understand what aerodynamic effect their opponents were counting on to unlock a few tenths of a second per lap.
How did I end up in a Formula 1 team, you may ask?
At the time, this team was using a competing CFD post-processor. This specific tool had been identified as the biggest bottleneck in their workflow, which is not a good situation to be in. It’s important to note that over the years, all Formula 1 teams have developed very similar CFD workflows. All operations are fully automated, from CAD surface preparation all the way to post-processing, including a work of art mesh and a massive parallel solver run. The pressure to perform is high. If, for any reason, your part of the process is holding up the fresh images from being ready for the morning meeting, you’ll be in the hot seat.
This team approached us with a very simple mission: produce the couple thousands of images expected for each solver run (see examples below), using their current hardware, but achieve this feat ten times faster than our competition, to allow for the post-processing of 1,000 runs per week. Oh, and I forgot to mention we only had 3 months to achieve this (you may say Formula 1 is fast paced).
Typical images used by Formula 1 teams to compare car design iterations through CFD.
Why would any team want to run 1,000 simulations per week? First, each design must be evaluated under several conditions: high-speed in a straight, lower speed in a curve, etc. Secondly, during the racing season, continuous and incremental design evolution propels the aerodynamic development. Formula 1 teams use CFD as a filter for design ideas. RANS provides sufficient accuracy and fidelity for picking up the correct trends from sequences of small design changes. Only the best designs will be tested in the wind tunnel, which requires building a very expensive model. And further down the pipeline, when a part is placed on the car at the track, there is reasonable certainty that it will bring the expected improvement in aerodynamic efficiency (The amount of downforce you can produce over the drag it costs you – see image below). If you miss your target, the competition will cruise away.
Aerodynamic Efficiency is defined as Fz / Fx.
It’s funny to think that there used to be a debate on whether CFD would eventually replace testing, particularly wind tunnel testing. It feels very similar to the discussion agitating our CFD community today about whether AI models are going to replace CFD simulations. But at the end of the day, good engineers know how to use the best tool for each task.
So, what were the results of our mission? I was given an account on their HPC system and started documenting all automation scripts used for post-processing, with their inputs and outputs. I ported these scripts to our scripting language (FVX). Thanks to a combination of better raw performance from FieldView and an innovative workflow which better utilized the team’s existing hardware, the goal we were given was exceeded and we outperformed the competition.
To this date, when I watch this team’s cars compete in Formula 1 GPs, I feel like my work back from that time still contributes to propelling them to the podium.
