Examenspresentation av Oskar Hurtig och Marcus Lindohf

Examiner: Torsten Wik

Abstrakt:
We propose two versatile and optimized model predictive controllers suitable for em-
bedded implementation in heavy-duty trucks, one based on the open-source mixed-
integer nonlinear solver OpEn and one based on dynamic programming. The con-
trollers are set to optimize the gear selections fed to an automatic gearbox during
pedal driving. By utilizing road data in terms of the slope ahead, vehicle information
such as the current engine speed, and the driver demanded output shaft torque, the
gear selections can be optimized with regard to energy efficiency, or power, depend-
ing on the tuning of the penalty coefficients within the controller. The hardware
requirements are evaluated in terms of the number of floating point operations per
second needed per controller for embedded suitability. On an ARM Cortex A53 pro-
cessor, the OpEn-based solver only requires on average a tenth of the computational
time compared to the DP-based controller. However, implementing the DP-based
solver in C/C++ will probably improve the computational efficiency of the solver,
decreasing the difference in solve time between the two solvers. Also, the DP-based
solver is more flexible in terms of altering the optimization algorithms and more ro-
bust in the sense of behaving according to the tuning parameters. It also returns a
more optimized solution compered to the OpEn-based solver given the problem for-
mulation at hand. Hence, the controller aligning most with the objectives presented
in section 1.1 is the DP-based MPC

Welcome!
Oskar, Marcus and Torsten