講演情報
[t3FRZuAM-09]Machine learning-based modeling of pavement-induced rolling resistance in a battery electric vehicle
Gerardo Gravante1, Joao Santos2, *Veronique Cerezo1 (1. Université Gustave Eiffel (France), 2. University of Twente (Netherlands))
キーワード:
rolling resistance、pavement texture、pavement roughness、machine learning、battery electric vehicle
Pavement-induced rolling resistance (RR) is one of the resistive forces opposing the motion of a vehicle. Representing the energy dissipated in the pavement-vehicle interaction (PVI), it results from a complex and dynamic interplay of variables related to the characteristics and operating conditions of the tire, road surface, and vehicle. In the case of battery electric vehicles (BEVs), this phenomenon is expected to be more pronounced than in comparable-sized internal combustion engine vehicles (ICEVs), owing to specific features, such as increased tire load and different torque performance. Nevertheless, most existing RR models for BEVs are based on simplified formulations that neglect the contribution of pavement texture and roughness. This study investigates the potential of machine learning (ML) techniques to predict pavement surface-induced RR in a passenger BEV. To achieve this, multiple experimental campaigns were performed in testing conditions close to real driving conditions. These involved an instrumented BEV driven on a test track with asphalt pavement sections featuring diverse texture and roughness characteristics. Standardized pavement descriptors, including mean profile depth (MPD) and international roughness index (IRI), along with vehicle speed, served as input features, while the corresponding RR coefficients (RRCs), derived from the collected data, were used as the target variable for model development. Several supervised ML algorithms were trained and evaluated, and their predictive performance was compared with that of conventional regression-based models. The findings show that ML-based approaches are more effective at capturing the nonlinear and joint effects of pavement surface characteristics and operating conditions on RR, resulting in superior predictive accuracy. However, the findings also reveal practical challenges related to model generalization and interpretability. Ultimately, this work presents a promising tool for enhancing pavement-related RR modeling and may support more informed pavement management and environmental impact assessments.
