Control and Estimation of Automotive Powertrains with Backlash
In automotive powertrains, backlash causes problems with vehicle driveability, specifically at so called tip-in and tip-out maneuvers. These maneuvers may trigger phenomena known in the automotive industry as shunt and shuffle, which are partially caused by the backlash. These phenomena are considered difficult to cope with, and until recently, only relatively simple controllers have been used to reduce the discomfort. Also, the tuning of these controllers is made quite subjectively.
A number of approaches to the control of systems with backlash are reported in the literature. Most of these approaches assume another system structure around the backlash than what is the case in automotive powertrains. Very few approaches are therefore directly applicable to the control of automotive powertrains.
The first part of this thesis gives an overview of available control strategies for backlash control. The strategies can be divided into active and passive strategies, depending on the way the controller handles the backlash. An active controller compensates the backlash nonlinearity by a more active control signal, while the passive controller becomes more cautious when the backlash gap is entered. Some of the strategies, e.g. switched linear controllers and model predictive controllers, are evaluated in the powertrain application by means of simulation. The results show that active nonlinear controllers have a potential for improved backlash control. However, the robustness of these controllers needs further investigation. Open-loop optimal control is used in this thesis as a way to find theoretical limits on backlash compensation performance.
High-performance controllers for backlash compensation require high-quality measurements of the current state of the powertrain. Information about the size of the backlash is also needed. These problems are addressed in the second part of this thesis. Two nonlinear estimators based on Kalman filtering theory have been developed, one for state estimation and one for the estimation of backlash size. Simulation and experimental results show that the resulting estimates are of high quality.
model predictive control