In this article, we devise a nonlinear model predictive control framework for the energy management of nonhybrid hydrostatic drive transmissions. The controller determines the optimal control commands of the actuators by minimising a cost function over a receding horizon. With our approach, the velocity-tracking error is minimised while keeping the fuel economy of the system high. The hydrostatic drive transmission system studied in this article is a typical commercial work machine, that is, there is no energy storage or alternative power source in the system (a nonhybrid hydrostatic drive transmission). We evaluate success with a validated simulation model of the hydrostatic drive transmission of a municipal tractor. In our experiments, a detailed system model is used both in the system simulation and in the prediction phase of the nonlinear model predictive control. The use of a detailed model in the nonlinear model predictive control framework places our design as a benchmark for controlling nonhybrid hydrostatic drive transmissions, when compared to solutions using simplified models or computationally less intensive control methods as in earlier work by the authors. Our nonlinear model predictive control approach enables numerically robust optimisation convergence with the utilised complex nonlinear model. Above all, this is accomplished with stabilising terminal constraints and distinctive terminal cost, both based on an optimal steady-state solution. In addition, a simple method to generate initial guesses for optimisation is introduced. When compared with the performance of a controller based on quasi-static models, our results show notable improvement in velocity tracking while maintaining high fuel economy. Furthermore, our experiments demonstrate that framing energy management as a nonlinear model predictive control provides a flexible and rigorous framework for fast velocity tracking and high energy efficiency. We also compare the results with those of an industrial baseline controller.
Implementation and validation of an event-based real-time nonlinear model predictive control framework with ROS interface for single and multi-robot systems
