(607c) Integration of Iterative Learning Control and Model Predictive Control for Point-to-Point Tracking Problem

Lee, J. M. - Presenter, Seoul National University
Oh, S. K., Seoul National University
Iterative learning control (ILC) is an effective control technique for handling repetitive, cyclic, iterative or batch processes which repeat a same task on a finite time interval. ILC was originally introduced for manipulating robots and has been applied to many industrial processes. However, conventional ILC cannot reject real-time disturbances because control actions are calculated based on the tracking error of the previous batch.

Much research has been conducted to incorporate feedback controller into ILC. Among them, the ILC method combined with model predictive control (MPC), referred to as iterative learning model predictive control (ILMPC), has attracted much attention due to their applicability to constrained multivariable control problems in the process industry. The main purpose of the ILMPC method is to track an entire reference trajectory at every time step of the operation. However, the entire reference trajectory is not always relevant for many practical applications such as the robotic “pick and place” task, crane control, rapid thermal process, and chemical batch reactor. The errors at certain time points are only relevant, and regulating the system at the time points of interest is the point-to-point (PTP) tracking problem.

For real-time feedback control of PTP tracking problem, we propose a PTP ILC method combined with MPC, called point-to-point iterative learning model predictive control (PTP ILMPC). Unlike general tracking problems, PTP tracking problems have output errors at desired time points only. Thus, conventional feedback controller cannot be used for PTP tracking problems because it requires output errors at every time step as the real-time feedback. The proposed PTP ILMPC uses state variables instead of output errors of each time step as feedback signals and minimizes the output errors at the specified time points in the future time window. Thus, it can be applied using the desired reference points without an arbitrary reference trajectory that passes through the desired reference points. The proposed method can seamlessly handle tracking, quality control, and constraints handling problems with a single controller. This study also presents the nominal and robust stability analyses of the proposed method.