Model Predictive Trajectory Set Control with Adaptive Input Domain Discretization

From a theoretical point of view, model predictive control (MPC) promises a high control quality since the future system performance is predicted and evaluated in every sampling interval. Additionally, desired optimization objectives can be realized while explicitly adhering to state and control input constraints. This control concept has the potential to set a new standard in industrial applications. In practical terms, functioning and properties of online optimization algorithms must be familiar to the developer to adjust the control performance of a mechatronic system. Moreover, it is challenging to realize a computationally expensive control concept for mechatronic systems with fast dynamics due to the limited computing power. It is worth mentioning that industrial systems usually exhibit hardware with low computing power. This contribution presents the model predictive trajectory set control (MPTSC) that constitutes a sub-optimal MPC with a sparse discretization of the control input domain. The optimal control input is determined without the use of iterative optimization techniques. To mimic quasi-continuous control input values similar to MPC, an adaptive input domain discretization is developed. MPTSC includes most advantages of MPC and is still computationally efficient. The implementation is less complex and error-prone and thus addresses especially industrial applications. The performance and the computation time are evaluated in comparison to MPC with nonlinear benchmark systems. Furthermore, the approach is tested experimentally on an industrial plant emulator with a sample rate of 100 Hz.