The thesis focuses on optimization methods for enhancing ride comfort in road vehicles, with a focus on improving computational efficiency in suspension optimization, balancing energy savings with longitudinal comfort in eco-driving strategies, and evaluating ride comfort by using driving simulators. Ride comfort significantly impacts occupant experience, safety, and health. With the increased use of high-fidelity models that require substantial computational resources, efficient optimization algorithms are essential to achieve optimal ride comfort while minimizing computational demands. Furthermore, automated driving (AD) and advanced driver assistance systems (ADAS) have changed comfort requirements as the driver’s role shifts from vehicle controller to supervisor, or even a fully liberated passenger. Additionally, driving simulators have increasingly become essential for testing and optimizing ride comfort, allowing engineers to refine designs without physical prototypes, reducing costs and enhancing performance. From the perspective of improving computational efficiency, a multi-fidelity surrogate-based optimization framework, which combines extended kernel regression (EKR) and approximate normal constraint (ANC) method, is proposed for a suspension optimization problem. The proposed framework is compared with well-known algorithms. A linear quarter car analytical model and a nonlinear multibody model are selected as the low-fidelity and high-fidelity models, respectively. The efficiency of the proposed optimization framework is evaluated against well-known optimization algorithms and applied across various vehicle types. Results indicate that, when used for suspension optimization, the proposed method requires less high-fidelity simulations on average to obtain a single Pareto point than other algorithms, demonstrating strong potential for effectively and efficiently addressing suspension optimization problems. To investigate an eco-driving cruise strategy, so-called pulse and glide (PnG), a single objective optimization considering the ride comfort constraint on longitudinal jerk was performed to balance energy consumption and longitudinal comfort in a battery electric vehicle (BEV). A two-step optimization method was employed to obtain the optimal solution. In the first stage, a neural network (NN) is generated and trained to get an approximated optimal solution by using a genetic algorithm (GA), and in the second stage, the optimization is performed on the physical model and the optimal solution of the first stage is used as a starting point. The result shows that the optimal PnG strategy can save up to 5\% of energy compared with the constant speed (CS) strategy without the constraint of ride comfort. If comfort is considered, a reduction of about 1\% can be found. Furthermore, a subjective experiment for subjective-objective correlation is conducted by using a dynamic driving simulator at DriSMi of Politecnico di Milano. The result shows that the numerical result of the optimization has been correlated to the occupants' subjective comfort perception, demonstrating the driving simulator is a promising tool for a fast and reliable subjective evaluation of longitudinal ride comfort. To confirm the simulator’s role in evaluating comfort thresholds, a comprehensive experiment using the up-down method identified jerk thresholds, aligning well with thresholds obtained from real cars. Finally, a multi-objective optimization of the PnG strategy provided a Pareto optimal set balancing energy efficiency and comfort, with points near the subjective annoyance threshold verified through simulator testing, further confirming these thresholds. With the aim of reducing cross-directional responses to the input and ensuring that input and output directions align to be effectively transmitted to the target location in a full-spectrum simulator for ride comfort and NVH evaluation, a comprehensive optimization approach of seat bushings is presented. A flexible multibody dynamic model is first created and validated through a modal experiment. To enhance optimization efficiency, a surrogate model based on the response surface methodology (RSM) is employed, replacing the physical model during the optimization process. To solve the multi-objective optimization problem, the weighted sum method is utilized. The optimal design achieves a maximum of 57.7\% improvement in terms of crosstalk reduction. Finally, considering the significant difference in optimal stiffness between the front and rear bushings, the layout of the seat bushings has been modified, adding two extra bushings in the rear and evenly distributing the optimal stiffness. The modified layout is still effective in reducing vibration crosstalk.
La tesi si concentra sui metodi di ottimizzazione per migliorare il comfort di marcia nei veicoli stradali, con particolare attenzione all’efficienza computazionale nell'ottimizzazione delle sospensioni, al bilanciamento tra risparmio energetico e comfort longitudinale nelle strategie di eco-driving e alla valutazione del comfort di marcia mediante simulatori di guida. Il comfort di marcia ha un impatto significativo sull’esperienza, sulla sicurezza e sulla salute degli occupanti. Con l’aumento dell’uso di modelli ad alta fedeltà che richiedono risorse computazionali elevate, sono essenziali algoritmi di ottimizzazione efficienti per ottenere un comfort ottimale riducendo il costo computazionale. Inoltre, la guida autonoma (AD) e i sistemi avanzati di assistenza alla guida (ADAS) hanno modificato i requisiti di comfort, poiché il ruolo del conducente si sposta da controllore del veicolo a supervisore, o persino a passeggero completamente libero. I simulatori di guida stanno inoltre diventando strumenti fondamentali per testare e ottimizzare il comfort di marcia, consentendo agli ingegneri di affinare i progetti senza la necessità di prototipi fisici, riducendo i costi e migliorando le prestazioni. Dal punto di vista dell’efficienza computazionale, viene proposto un framework di ottimizzazione basato su un surrogato multi-fedeltà, che combina la regressione a kernel estesa (EKR) con il metodo del vincolo normale approssimato (ANC), per l'ottimizzazione delle sospensioni. Il framework proposto è confrontato con algoritmi noti. Un modello analitico lineare di un quarto di veicolo e un modello multibody non lineare sono scelti rispettivamente come modelli a bassa e alta fedeltà. L’efficienza del framework è valutata rispetto a noti algoritmi di ottimizzazione e applicata a diversi tipi di veicoli. I risultati mostrano che, nell’ottimizzazione delle sospensioni, il metodo proposto richiede in media un minor numero di simulazioni ad alta fedeltà per ottenere un singolo punto di Pareto rispetto ad altri algoritmi, dimostrando un forte potenziale nell’affrontare efficacemente ed efficientemente problemi di ottimizzazione delle sospensioni. Per studiare una strategia di eco-driving basata sulla tecnica "pulse and glide" (PnG), è stata effettuata un'ottimizzazione a singolo obiettivo considerando il vincolo sul jerk longitudinale per bilanciare il consumo energetico e il comfort longitudinale in un veicolo elettrico a batteria (BEV). È stato impiegato un metodo di ottimizzazione in due fasi per ottenere la soluzione ottimale. Nella prima fase, è stata generata e addestrata una rete neurale (NN) per ottenere una soluzione approssimata utilizzando un algoritmo genetico (GA). Nella seconda fase, l'ottimizzazione è stata eseguita sul modello fisico, utilizzando la soluzione ottimale della prima fase come punto di partenza. I risultati mostrano che la strategia PnG ottimale consente un risparmio energetico fino al 5% rispetto alla strategia a velocità costante (CS) senza considerare il comfort. Se il comfort viene considerato, il risparmio si riduce a circa l'1%. Inoltre, è stato condotto un esperimento soggettivo presso il DriSMi del Politecnico di Milano per valutare la correlazione tra i risultati numerici dell’ottimizzazione e la percezione soggettiva del comfort da parte degli occupanti, dimostrando che il simulatore di guida è uno strumento promettente per una valutazione rapida e affidabile del comfort longitudinale. Per confermare il ruolo del simulatore nella valutazione delle soglie di comfort, è stato condotto un esperimento completo utilizzando il metodo up-down per identificare le soglie di jerk, che si sono rivelate coerenti con quelle ottenute dai test su veicoli reali. Infine, un’ottimizzazione multi-obiettivo della strategia PnG ha fornito un insieme di soluzioni Pareto ottimali che bilanciano l’efficienza energetica e il comfort, con punti vicini alla soglia di fastidio soggettivo verificati tramite test in simulatore, confermando ulteriormente tali soglie. Per ridurre le risposte trasversali all’ingresso e garantire che le direzioni di ingresso e uscita siano allineate per essere efficacemente trasmesse alla posizione target in un simulatore a spettro completo per la valutazione del comfort e del NVH, viene presentato un approccio di ottimizzazione completo per i supporti del sedile. Un modello dinamico multibody flessibile è stato inizialmente creato e validato tramite un esperimento modale. Per migliorare l’efficienza dell’ottimizzazione, è stato impiegato un modello surrogato basato sulla metodologia delle superfici di risposta (RSM), sostituendo il modello fisico nel processo di ottimizzazione. Per risolvere il problema di ottimizzazione multi-obiettivo, è stato utilizzato il metodo della somma pesata. Il design ottimale raggiunge una riduzione massima del 57,7% del crosstalk. Infine, considerando la significativa differenza nella rigidità ottimale tra i supporti anteriori e posteriori, la disposizione dei supporti del sedile è stata modificata, aggiungendo due supporti extra nella parte posteriore e distribuendo uniformemente la rigidità ottimale. La nuova configurazione si è rivelata efficace nella riduzione del crosstalk delle vibrazioni.
Methods for road vehicles ride comfort optimization
Xue, Haoxiang
2024/2025
Abstract
The thesis focuses on optimization methods for enhancing ride comfort in road vehicles, with a focus on improving computational efficiency in suspension optimization, balancing energy savings with longitudinal comfort in eco-driving strategies, and evaluating ride comfort by using driving simulators. Ride comfort significantly impacts occupant experience, safety, and health. With the increased use of high-fidelity models that require substantial computational resources, efficient optimization algorithms are essential to achieve optimal ride comfort while minimizing computational demands. Furthermore, automated driving (AD) and advanced driver assistance systems (ADAS) have changed comfort requirements as the driver’s role shifts from vehicle controller to supervisor, or even a fully liberated passenger. Additionally, driving simulators have increasingly become essential for testing and optimizing ride comfort, allowing engineers to refine designs without physical prototypes, reducing costs and enhancing performance. From the perspective of improving computational efficiency, a multi-fidelity surrogate-based optimization framework, which combines extended kernel regression (EKR) and approximate normal constraint (ANC) method, is proposed for a suspension optimization problem. The proposed framework is compared with well-known algorithms. A linear quarter car analytical model and a nonlinear multibody model are selected as the low-fidelity and high-fidelity models, respectively. The efficiency of the proposed optimization framework is evaluated against well-known optimization algorithms and applied across various vehicle types. Results indicate that, when used for suspension optimization, the proposed method requires less high-fidelity simulations on average to obtain a single Pareto point than other algorithms, demonstrating strong potential for effectively and efficiently addressing suspension optimization problems. To investigate an eco-driving cruise strategy, so-called pulse and glide (PnG), a single objective optimization considering the ride comfort constraint on longitudinal jerk was performed to balance energy consumption and longitudinal comfort in a battery electric vehicle (BEV). A two-step optimization method was employed to obtain the optimal solution. In the first stage, a neural network (NN) is generated and trained to get an approximated optimal solution by using a genetic algorithm (GA), and in the second stage, the optimization is performed on the physical model and the optimal solution of the first stage is used as a starting point. The result shows that the optimal PnG strategy can save up to 5\% of energy compared with the constant speed (CS) strategy without the constraint of ride comfort. If comfort is considered, a reduction of about 1\% can be found. Furthermore, a subjective experiment for subjective-objective correlation is conducted by using a dynamic driving simulator at DriSMi of Politecnico di Milano. The result shows that the numerical result of the optimization has been correlated to the occupants' subjective comfort perception, demonstrating the driving simulator is a promising tool for a fast and reliable subjective evaluation of longitudinal ride comfort. To confirm the simulator’s role in evaluating comfort thresholds, a comprehensive experiment using the up-down method identified jerk thresholds, aligning well with thresholds obtained from real cars. Finally, a multi-objective optimization of the PnG strategy provided a Pareto optimal set balancing energy efficiency and comfort, with points near the subjective annoyance threshold verified through simulator testing, further confirming these thresholds. With the aim of reducing cross-directional responses to the input and ensuring that input and output directions align to be effectively transmitted to the target location in a full-spectrum simulator for ride comfort and NVH evaluation, a comprehensive optimization approach of seat bushings is presented. A flexible multibody dynamic model is first created and validated through a modal experiment. To enhance optimization efficiency, a surrogate model based on the response surface methodology (RSM) is employed, replacing the physical model during the optimization process. To solve the multi-objective optimization problem, the weighted sum method is utilized. The optimal design achieves a maximum of 57.7\% improvement in terms of crosstalk reduction. Finally, considering the significant difference in optimal stiffness between the front and rear bushings, the layout of the seat bushings has been modified, adding two extra bushings in the rear and evenly distributing the optimal stiffness. The modified layout is still effective in reducing vibration crosstalk.File | Dimensione | Formato | |
---|---|---|---|
PhD_Thesis_36cycle_haoxiang_xue_final.pdf
non accessibile
Dimensione
37.32 MB
Formato
Adobe PDF
|
37.32 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/10589/232912