By Carlo Platella
Artificial intelligence is now perceived as the next great IT revolutionwith calculation potential superior to traditional algorithms. This is a technology with implications for numerous sectors, including Formula 1, where it in turn already finds application in various fields. Among the most interesting, aerodynamic analysis stands out, with the prospect of reducing the use of wind tunnels and traditional instruments in the future.
Existing technologies
Virtual simulation CFD programs they start from the equations that describe physical phenomena to predict the behavior of an aerodynamic component, calculating the evolution of speed, pressure and forces released by the air when interacting with the surface in question. However, CFD models are always simplified compared to reality, as a complete physical analysis would be too expensive in terms of time and computing resources. Since it is not possible to take every single aspect into consideration, the task of the design teams is to select only those that are actually relevant. It is therefore a question of choosing what type of phenomena to analyse, for example whether only aerodynamic or also thermal, whether of a stationary or transitory nature, and above all on what scale of turbulence, with the vortices ranging from being microscopic to having several meters of diameter.
Also the wind tunnel has its limitationsfailing to faithfully replicate the car’s behavior in the real world. For example, it is possible to analyze the car at various heights from the ground, but not simulate its movements on the suspension, without therefore being able to collect results in more dynamic conditions. The possibility of investigating a physical phenomenon also depends on the availability of sensors suitable for measuring its indicative parameters, not to mention the difficulties in controlling the temperature of the air flow, to which modern single-seaters are quite sensitive. CFD and wind tunnel are therefore useful tools for building an approximate idea of a certain solution, but only the track can offer the definitive verdict.
Reverse route
Contrary to what one might think, artificial intelligence is not a recent innovation. The first theoretical foundations date back to a few decades ago, but it is only in recent times that processor technology has reached a level that offers the computing power necessary to apply them. Already in the nineties for example Benetton experimented with a neural network project to study how to improve the drivability and vehicle dynamics of their cars.
If with CFD we start from physical laws to study the behavior of an aerodynamic solution, with artificial intelligence algorithms the opposite happens. The starting point is already existing examplesidentifying trends in the available data that are useful for predicting the behavior of new solutions or of the same object in different conditions. The ‘reasoning’ of artificial intelligence is therefore deductive, starting from information already known. A prerogative is the construction of a database, a large sample of data which in the case of aerodynamics can come from the track, wind tunnel or CFD analysis, necessary to train the model to recognize trends that can predict scenarios different from the starting ones .
The examples
Formula 1 is no stranger to this technology, having it already used for example for the drafting of the 2022 regulations. The interest at the time was entirely in the aerodynamic disturbance for the following car, with the need to study the turbulent wake of the single-seaters, extremely complex to model and analyze with CFD. However, the FIA and Formula 1 have also exploited the predictive power of Machine Learning, with the algorithms then matured to the point that now, starting from the study of the turbulent wake one meter away, it is possible to predict its behavior 40 meters further back.
Even before that there was the case of Porsche, when between 2017 and 2018 it developed the EVO version of the 919 Hybrid LMP1 prototype. The German company designed the rear wing started from an existing database of 1600 airfoilstraining a calculation model to identify the most representative parameters, such as chord, geometry incidence and curvature. It was therefore a question of mapping the ‘genes’ that described the DNA of each profile, identifying 5 key parameters. However, with a variability of 5 values each, the overlap of two flaps in the same wing and other variables such as relative incidence and distance, the possible combinations were over 30 billion.
Porsche then took advantage another Machine Learning algorithm to solve the optimization problem. The model returned between hundreds and a few thousand of the most worthy configurations, then virtually simulated through traditional CFD analyzes using a cluster. The sample became smaller and smaller until the final configuration of the wing was reached, subject to a final test in the wind tunnel.
Cohabitation possible
The last example highlights how in its various forms Artificial Intelligence lends itself to cooperating with classic simulation tools, without necessarily replacing them. As seen, these are models that apply a deductive logic, but still strongly limited in replicating the intuitiveness of the human mind. If to predict changes with respect to an already known solution or context, AI must start from a pre-existing case study, to explore completely new forms, concepts or situations it is necessary to return to the CFD or the wind tunnel, starting from the underlying physics. Added to this is the possibility of using traditional tools to verify the predictions of Artificial Intelligence, exactly as happened in the case of the Porsche 919 Hybrid EVO.
This does not mean that it is not a technology with enormous potentialgoing beyond the limits of CFD and wind tunnel. In recent years there has been a lot of talk about the teams’ investments in infrastructure, with the modernization of production departments and the construction of new wind tunnels. However, IT research and development in the field of Artificial Intelligence could make an even greater difference in the medium-term future, exactly as happened with the advent of CFD between the late 1990s and the early 2000s.
#Formula #revolution #wind #tunnels #artificial #intelligence