Fuel-Efficient Look-Ahead Control for Heavy-Duty Vehicles with Varying Velocity Demands
Time: Fri 2020-06-12 14.00
Location: Disputation in Zoom: https://kth-se.zoom.us/webinar/register/WN_Atk97qAwS9OiEW8dTC0Fyg (English)
Subject area: Electrical Engineering
Doctoral student: Manne Held , Reglerteknik
Opponent: Professor Ardalan Vahidi, Clemson University
Supervisor: Jonas Mårtensson, Reglerteknik, Integrated Transport Research Lab, ITRL; Oscar Flärdh, Scania CV AB
The fuel consumption of heavy-duty vehicles can be reduced by using information about the upcoming road section when controlling the vehicles. Most manufacturers of heavy-duty vehicles today offer such look-ahead controllers for highway driving, where the information consists of the road grade and the velocity only has small variations. This thesis considers look-ahead control for applications where the velocity of the vehicle has large variations, such as distribution vehicles or vehicles in mining applications. In such conditions, other look-ahead information is important, for instance legal speed limits and curvature. Fuel-efficient control is found by formulating and solving the driving missions as optimal control problems.
First, it is shown how look-ahead information can be used to set constraints in the optimal control problems. A velocity reference from a driving cycle is modified to create an upper and a lower bound for the allowed velocity, denoted the velocity corridor. In order to prevent the solution of the optimal control problem from deviating too much from a normal way of the driving, statistics derived from data collected during live truck operation are used when formulating the constraints. It is also shown how curvature and speed limits can be used together with actuator limitations and driveability considerations to create the velocity corridor.
Second, a vehicle model based on forces is used to find energy-efficient velocity control. The problem is first solved using Pontryagin's maximum principle to find the energy savings for different settings of the velocity corridor. The problem is then solved in a receding horizon fashion using a model predictive controller to investigate the influence of the control horizon on the energy consumption. The phasing and timing of traffic lights are then added to the available information to derive optimal control when driving in the presence of traffic lights.
Third, the vehicle model is extended to include powertrain components in two different approaches. In a first approach, a Boolean variable is added to represent open or closed powertrain. This enables the vehicle to freewheel, in order to save fuel by reducing the losses due to engine drag. The problem is formulated as a mixed integer quadratic program. In a second approach, the full powertrain is modeled including a fuel map and a model of the gearbox losses, both based on measurements on real components. The problem is solved using dynamic programming, with transitions between states including gear shifts, freewheeling, and coasting in gear.
Forth, the optimal control framework is used to implement an optimal control-based powertrain controller in a real Scania truck. The problem is first solved offline resulting in trajectories for velocity and freewheeling. These are used online in the vehicle as references to the existing controllers for torque and gear demands. Experiments are performed with fuel measurements, resulting in 16% fuel savings, compared to 18% savings by solving the optimal control problem.