Author:

Keywords:

Iterative learning control, optimization, optimal control, nonlinear, control, nonlinear programming

Abstract:

An important evolution of the past decades in industry is the mechanization and automation of production processes. Industrial robots and other mechatronic systems play a crucial role in increasing production speed, and in the continued miniaturization of components, for example in the electronics industry. One of the major challenges for such systems is to improve the positioning accuracy by applying modern control strategies. This can be achieved by including a mathematical model into the control design, and increased complexity of the system therefore requires the use of more advanced models, such as nonlinear models. Furthermore, time and cost requirements increase the need to operate mechatronic systems at the maximum of their potential, so modern control approaches must be able to include system constraints into their design.The repetitive nature of industrial processes has lead to the development of iterative learning control (ILC) approaches, for which the controls are calculated beforehand and updated after each repetition, or trial. Such a control strategy can achieve very high accuracy, while being robust with respect to small variations or changes to the controlled system. Traditional ILC methods have been developed for linear models, and without dealing with system constraints. The goal of the current research is to develop ILC methods which can deal with nonlinear models, and which take constraints into account directly.In this thesis, modern dynamic optimization techniques are used to design efficient and general calculation strategies for the inversion of nonlinear models, which is essential in the design of nonlinear ILC. This approach is based on the solution of inequality constrained nonlinear least squares problems. Using this calculation strategy, an existing ILC method has been improved by dramatically reducing the required calculation time. Another contribution of this thesis is the development of a novel ILC approach, based on the concept of explicit model correction, and consisting of two steps: in the first step the nominal model is corrected based on the previous trial’s output, and in the second step the corrected model is inverted to track a given reference. The equivalence of the proposed approach with existing methods, under certain conditions, is discussed. An important benefit of the method is an increased applicability to other control problems, such as point-to-point motion control, iterative model identification, constrained feedback control, and the control of systems which experience sudden changes in dynamics. Each of these applications is discussed in the thesis. In order to facilitate the application of the proposed ILC approach, an open source software package is developed, which integrates state-of-the-art third party packages for automatic differentiation and dynamic optimization using an interior point method.