In recent years, the automotive industry has undergone rapid evolution, with increasing interest in electric vehicles (EVs) as a response to environmental challenges and rising fossil fuel costs. In this context, improving driving comfort and vehicle safety has become a crucial factor for the widespread adoption of EVs. In particular, longitudinal vehicle vibrations, although often overlooked compared to vertical dynamics, can have a significant impact on driving comfort. This thesis presents the development of a controller based on the Deep Deterministic Policy Gradient (DDPG) approach, a Reinforcement Learning (RL) algorithm, for torque correction aimed at reducing longitudinal vibrations in electric vehicles. The proposed controller has been implemented for both the front and rear axles, modulating the torque requested by the driver in response to road irregularities or obstacles. The agent was trained on various road scenarios and tested in simulations that included obstacles such as steps and bumps, as well as real road profiles sampled experimentally. The results show that the proposed RL controller offers performance comparable to that of a Nonlinear Model Predictive Controller (NMPC) in terms of vibration reduction, while maintaining a lower computational load, making it suitable for real-time applications in road vehicles. Further robustness tests were conducted to assess the controller's response to various road conditions and vehicle parameter variations, confirming the validity of the proposed approach. Finally, the controller was implemented on a dSPACE MicroAutoBox II device to test its computational time.
Negli ultimi anni, l'industria automobilistica ha subito un'evoluzione rapida, con un crescente interesse per i veicoli elettrici (EV) come risposta alle sfide ambientali e all'aumento dei costi dei combustibili fossili. In questo contesto, il miglioramento del comfort di guida e della sicurezza dei veicoli è diventato un fattore cruciale per la diffusione di massa degli EV. In particolare, le vibrazioni longitudinali del veicolo, sebbene spesso trascurate rispetto alle dinamiche verticali, possono avere un impatto significativo sul comfort di guida. Questo articolo presenta lo sviluppo di un controllore basato sull'approccio Deep Deterministic Policy Gradient (DDPG), un algoritmo di Reinforcement Learning (RL), per la correzione della coppia motrice al fine di ridurre le vibrazioni longitudinali nei veicoli elettrici. Il controllore proposto è stato implementato sia per l'asse anteriore che posteriore, modulando la coppia richiesta dal guidatore in risposta alle irregolarità della strada o agli ostacoli. L'agente è stato addestrato su diversi scenari stradali e testato in simulazioni che includono ostacoli come gradini e dossi, infine su profili stradali reali campionati. I risultati mostrano che il controllore RL proposto offre prestazioni comparabili rispetto a un controllore NMPC (Nonlinear Model Predictive Controller) in termini di riduzione delle vibrazioni, mantenendo tuttavia un impegno computazionale inferiore, rendendolo adatto per applicazioni in tempo reale nei veicoli stradali. Ulteriori test di robustezza sono stati condotti per valutare la risposta del controllore a diverse condizioni stradali e variazioni dei parametri del veicolo, confermando la validità dell'approccio proposto. Infine il controllore è stato implementato su un dispositivo dSPACE MicroAutoBox II al fine di testarne il tempo computazionale.
Miglioramento del comfort del veicolo utilizzando un approccio IA
ADAMI, MARCO
2023/2024
Abstract
In recent years, the automotive industry has undergone rapid evolution, with increasing interest in electric vehicles (EVs) as a response to environmental challenges and rising fossil fuel costs. In this context, improving driving comfort and vehicle safety has become a crucial factor for the widespread adoption of EVs. In particular, longitudinal vehicle vibrations, although often overlooked compared to vertical dynamics, can have a significant impact on driving comfort. This thesis presents the development of a controller based on the Deep Deterministic Policy Gradient (DDPG) approach, a Reinforcement Learning (RL) algorithm, for torque correction aimed at reducing longitudinal vibrations in electric vehicles. The proposed controller has been implemented for both the front and rear axles, modulating the torque requested by the driver in response to road irregularities or obstacles. The agent was trained on various road scenarios and tested in simulations that included obstacles such as steps and bumps, as well as real road profiles sampled experimentally. The results show that the proposed RL controller offers performance comparable to that of a Nonlinear Model Predictive Controller (NMPC) in terms of vibration reduction, while maintaining a lower computational load, making it suitable for real-time applications in road vehicles. Further robustness tests were conducted to assess the controller's response to various road conditions and vehicle parameter variations, confirming the validity of the proposed approach. Finally, the controller was implemented on a dSPACE MicroAutoBox II device to test its computational time.File | Dimensione | Formato | |
---|---|---|---|
Adami_Marco.pdf
embargo fino al 14/06/2026
Dimensione
27.38 MB
Formato
Adobe PDF
|
27.38 MB | Adobe PDF |
The text of this website © Università degli studi di Padova. Full Text are published under a non-exclusive license. Metadata are under a CC0 License
https://hdl.handle.net/20.500.12608/80325