In this thesis, we tackle the problem of the design of wind tunnel experiments for an F1 car. Wind tunnel experiments assess the goodness of new aerodynamic configurations, providing a method to estimate their lap time without running expensive track tests. Improving the set of states assumed by the car during these tests provides two main advantages: a lower uncertainty on the estimated lap time, and faster and shorter wind tunnel experiments. Historically, F1 was dominated by the best engine manufacturers, but, nowadays, the focus has turned more and more towards the aerodynamics of cars as the key aspect to improve performance. This is a joint work with Scuderia Ferrari F1, who decided the study of the design of wind tunnel tests to become more and more crucial in the last years to identify the best aerodynamic configurations that are worth trying on track. Various limitations are imposed by FIA on both wind tunnel runs and track tests, so it is crucial to select the best configurations providing the lowest possible uncertainty on the lap time estimates. The core point that will be discussed in this thesis is the choice of sequences of states assumed by the car during wind tunnel experiments. The goal is to develop data-driven methodologies, based on machine learning techniques, to optimize the design of the experiments to guarantee low uncertainty on the measures retrieved in the process. In particular, the design of the experiments has to generate sequences of positions that could be useful to Ferrari aerodynamicists and that satisfy particular constraints to allow high customization. Genetic Algorithms have been chosen as the most suitable solution to this problem. These algorithms were studied and modified, in this thesis, to better suit the specific problem of aerodynamic map design. Finally, experiments with different constraints were tested and compared, analyzing and proposing the best set of parameters for this kind of application.
In questa tesi viene affrontato il problema dell'ottimizzazione dei test in galleria del vento per minimizzare l'incertezza della stima dei tempi su giro per vetture di Formula Uno. I test in galleria del vento valutano la qualità di nuove componenti aerodinamiche dell'auto e migliorandoli si hanno due principali vantaggi: una riduzione dell' incertezza della stima dei tempi su giro e una riduzione della durata degli esperimenti in galleria del vento Il progetto è un lavoro in collaborazione con Scuderia Ferrari F1, he negli ultimi anni ha visto lo studio del design dei test in galleria del vento diventare sempre più importante per valutare quali sono i migliori componenti sviluppati e che andranno quindi testati in pista. La FIA ha imposto diverse limitazioni sia sui test in galleria del vento che sui test in pista ed è quindi fondamentale avere delle valutazioni delle componenti aerodinamiche con la più grande accuratezza possibile. Il scopo di questa tesi è quello di sviluppare metodologie basate su tecniche di Machine Learning in grado di ottimizzare il design degli esperimenti in galleria del vento, garantendo di ottenere misurazioni con bassa incertezza. In particolare, quello che si vuole ottenere è una sequenza di posizioni che il veicolo assumerà durante il test in galleria del vento. Questa sequenza deve soddisfare diversi vincoli fissati dagli aerodinamici di Ferrari, i quali devono poter personalizzare la soluzione ottenuta. La soluzione proposta si basa sull'uso di Algoritmi Genetici, che sono stati adattati a questo problema specifico. Esperimenti sui vari vincoli sono stati effettuati e sono stati proposti i valori dei parametri che migliorano le prestazioni dell'applicazione.
Optimizing wind tunnel experiments to minimize the uncertainty of laptime estimation for F1 cars
GREGORI, GIACOMO
2018/2019
Abstract
In this thesis, we tackle the problem of the design of wind tunnel experiments for an F1 car. Wind tunnel experiments assess the goodness of new aerodynamic configurations, providing a method to estimate their lap time without running expensive track tests. Improving the set of states assumed by the car during these tests provides two main advantages: a lower uncertainty on the estimated lap time, and faster and shorter wind tunnel experiments. Historically, F1 was dominated by the best engine manufacturers, but, nowadays, the focus has turned more and more towards the aerodynamics of cars as the key aspect to improve performance. This is a joint work with Scuderia Ferrari F1, who decided the study of the design of wind tunnel tests to become more and more crucial in the last years to identify the best aerodynamic configurations that are worth trying on track. Various limitations are imposed by FIA on both wind tunnel runs and track tests, so it is crucial to select the best configurations providing the lowest possible uncertainty on the lap time estimates. The core point that will be discussed in this thesis is the choice of sequences of states assumed by the car during wind tunnel experiments. The goal is to develop data-driven methodologies, based on machine learning techniques, to optimize the design of the experiments to guarantee low uncertainty on the measures retrieved in the process. In particular, the design of the experiments has to generate sequences of positions that could be useful to Ferrari aerodynamicists and that satisfy particular constraints to allow high customization. Genetic Algorithms have been chosen as the most suitable solution to this problem. These algorithms were studied and modified, in this thesis, to better suit the specific problem of aerodynamic map design. Finally, experiments with different constraints were tested and compared, analyzing and proposing the best set of parameters for this kind of application.File | Dimensione | Formato | |
---|---|---|---|
Thesis_Giacomo_Gregori.pdf
non accessibile
Descrizione: Testo della tesi
Dimensione
2.91 MB
Formato
Adobe PDF
|
2.91 MB | Adobe PDF | Visualizza/Apri |
I documenti in POLITesi sono protetti da copyright e tutti i diritti sono riservati, salvo diversa indicazione.
https://hdl.handle.net/10589/148517