University of Freiburg
Tuesday, December 20, 2016, 11:00
Room 02-016, Georges-Koehler-Allee 101, Freiburg 79110, Germany
This thesis deals with new approaches for model predictive control (MPC) of redundant permanent magnet synchronous machines (PMSM), where each subsystem is connected to a two-level inverter. Traditionally, an ordinary PMSM is controlled by a PI-controller calculating the continuous input voltages and a modulator transforming the voltages into discrete switching states of the inverter.
The control methods developed in this thesis combine these two tasks. This control approach, known as finite control set model predictive control (FCS-MPC) has gained popularity in the past decade and results in one of the two major challenges of this work. The nature of the optimization problem, which has to be solved in every iteration of the control scheme, is discrete. This class of optimization problems is provably NP-hard. Despite the computational burden of FCS-MPC, it could be shown that it is superior compared to conventional approaches in terms of reduction of switching losses in the inverter.
The second aspect of this work is the exploitation of the redundancy in order to handle faults in the inverter, which is of particular interest in the field of autonomous driving.
First, the control of the currents in the machine is addressed. Formulating the cost function in different ways results in two different methods for the online solution of the optimization problem. One optimization algorithm is a branch & bound type method, while the other approach leads to a linear program (LP).
In the next step, these methods are extended in order to control the torque of the machine. This quantity is a nonlinear function in the currents and will therefore be linearized. The fault case can be incorporated easily in the control structure. Both methods will be compared in terms of control performance and computational efficiency.