MSCA-COFUND-CLEAR-Doc-PhD Position#CD22-63: A machine learning based rolling resistance prediction model for electric vehicles

Worldwide governmental bodies and authorities are adopting policies and allocating resource to promote the development of cleaner and more efficient vehicle technologies with the objective of replacing the internal combustion engines that consume large amounts of fossil fuels and emit huge quantities of air pollutants and greenhouse gases (GHG). As a result, the increasing penetration of electric vehicles (EV) in the transportation market witnessed in the last years offers a promising outlook for the diversification of the sector’s fuel sources.

One of the factors affecting vehicle’s fuel economy is rolling resistance. Rolling resistance is a physical phenomenon related to the dissipation of energy that occurs during the passage of a tire on a road pavement. This loss of energy generates forces opposed to the vehicle movement, which in turn increase fuel consumption. Three physical phenomena can be identified that explain rolling resistance: 1) deformation of the tire in the tire/road contact area, 2) aerodynamic drag of the rotating tire, and 3) slip between the tire tread and the pavement surface. Rolling resistance can represent until 30% of the resistive forces depending on the vehicles’ characteristics and the driving conditions (rural or urban roads, motorways). According to several studies, rolling resistance is responsible for 5 to 20% of the fuel consumption of a passenger car and 15 to 40% of trucks’ fuel consumption. It depends on multiple factors related to the vehicle type (load, suspension), tyre properties (rubber, inflation pressure, rubber temperature), vehicle operating conditions (speed), atmospheric conditions (wind, temperature) and road pavement characteristics (roughness, macrotexture). From state of the art, it is clear that all the factors described previously can have opposite effects on rolling resistance. This makes it harder to understand and decouple their individual contribution. Moreover, other factors related to vehicle dynamics can also affect the generation of rolling resistance (e.g., tire camber, torque, etc.) but are less analysed in literature.

In recent decades several approaches for estimating rolling resistance have been studied and developed. They proposed models suitable for diesel and gasoline vehicles based on numerical approaches (dynamical equations) or statistical approaches (correlation). The overwhelming majority of these approaches requires prior knowledge on the road pavement characteristics, usually roughness and macrotexture. Additionally, they exhibit accuracy problems, robustness issues, require sometimes long calibration of the parameters and are limited on their field of application. However, so far, few research studies have been performed on the quantification of the effect of rolling resistance on the energy consumption of EV, let alone while considering the dynamic states of vehicle, tyre and based on the real time perception of pavement characteristics.

Thus, by harnessing the potentialities of data analytics and machine learning this thesis aims to develop a data-driven rolling resistance prediction model based on the acquisition of data related to the dynamic states of the vehicle, tyre and pavement characteristics. To achieve this purpose, an instrumented EV will be used both on test tracks exhibiting a wide range of pavement mixes and on real road sections. The EV is equipped with multiple sensors allowing to measure continuously rolling resistance forces in the tire/road contact area, temperature (rubber and pavement), dynamical behaviour of the vehicle (speeds, accelerations), tire deformation and pavement characteristics. These sensors are controlled periodically to check their accuracy and robustness and the data is acquired at a frequency of 100 Hz.

Lastly, the PhD thesis includes both wide experimental work to collect data and theoretical mathematical development to clean dataset, select relevant parameters and propose accurate and robust model based on machine learning technics. The candidate will have access to instrumented EV, to test tracks and a team of technicians to implement experimental campaigns. He will be hosted in the EASE laboratory (Environmental Assessment, Planning, Safety, Ecodesign)