Very good thesis from the University of Bologna. Not directly related to F1, but tons of useful and applicable work.
The aim of this thesis is to investigate a method that, by using Artificial Intelligence technics, it is able to predict the overall downforce coefficients (Clf and Clr) of a racing car and the discrete pressure distribution over the car floor for a given particular car attitude.
The main result of this study is the design and development of computer models for each single pressure tap placed on the car floor and, thanks to that, to generate the pressure distribution over the whole car floor even in absence of pressure tap sensors.
Three types of models have been implemented using three different Machine Learning algorithms:
• Support Vector Machine: Support Vector Regression – SVR
• Neural Networks: Multilayer Perceptron - MLP
• Neural Networks: Recurrent Neural Network – RNN
and the respective results are presented and discussed in detail.
The advantage of having a model for determining the pressure distribution over the
car floor with respect of having one single model to predict the single values of downforce (Clf, Clr) consists in the ability of obtaining a 3D description of the car aerodynamics, without reducing it to the sole definition of the downforce coefficients. A continuous distribution over the car floor will not be considered in this thesis since it would imply extrapolation and the consequent introduction of an error.
Models would need to be built for each of the car chassis used in the IndyCar competition according to the type of circuit where the racing event takes place:
• Road course RC configuration.
• Short oval SO configuration.
• Super Speedway SSW configuration.
However, this thesis is focused on the creation of downforce and Cp models from track data for only the RC car since only the track mapping data for this chassis were accessible to me for this study; in any case, the exact same procedure described in this thesis for the RC car is valid in general for all the car chassis, and so it can be applied to generate the models for also the SO and SSW cars, provided good enough mapping data.
The advantages of having generated the models of this thesis using Track data with respect to Wind Tunnel ones are also discussed.
i
https://morethesis.unimore.it/theses/av ... Milani.pdf