Estimating rolling resistance for autonomous dumper

Sammanfattning

The vehicle industry is moving towards electrification and autonomous driving. Improvements in actuators and sensor technology have enabled greater degree of control for safe operation of such autonomous vehicles. These vehicles use a lot of information about the surrounding environment to manage the vehicle power demands. The external resistive forces to be overcome by the vehicle contributes to most of the demands. Measurement of all these external forces is sometimes impractical and inaccurate. Rolling resistance is one such resistive force that cannot be measured directly using sensors. However, using advanced control methods like parameter estimation, it is possible to combine a system model and available sensor information to estimate such unknown parameters. With this information, it is viable to model the vehicle system accurately and manage the energy demands while the vehicle is running. Such online estimation and control strategies are vital for highly autonomous systems like the electric dumper TA 15, developed by Volvo Construction Equipment, that navigates through challenging terrains like quarries and mines. This thesis work aims to recursively estimate the rolling resistance coefficient for TA 15 using two estimation techniques, namely, Recursive Least Square (RLS) and Extended Kalman Filter (EKF). By using available sensor information on vehicle velocity, motor torque, inclination and vehicle mass, it is possible to accurately estimate the rolling resistance coefficient from a longitudinal vehicle model. A vehicle model is setup virtually to test different input scenarios and the estimators are tuned accordingly. Finally, the estimators are run offline using real data from the vehicle to extract the unknown rolling resistance coefficient. It can be concluded that it is difficult to estimate rolling resistance separately without any information about vehicle mass and road slope. However, when accurate mass and road slope information are available, the rolling resistance estimate tends to fall within acceptable range for the vehicle under consideration. Although the RLS estimator is straightforward and efficient for linear systems, its usefulness in complex real-world situations is limited by its inability to handle non-linear dynamics and vulnerability to initial assumptions and noise. The EKF is the recommended option for estimating the rolling resistance coefficient, out of the two investigated methods, due to its greater sensor fusion capabilities, robustness, and adaptability. However, based on the results, there may be other better estimation methods that should be explored.