Industrial processes are, in many cases, difficult to model and control due to variable loads and nonlinearities. When the tasks of the processes are repetitive as can be seen in the pick-and-place, welding, etc. Robot operation, however, some of the difficulties due to uncertain dynamic models may be overcome if the controller is smartly designed as a control method called iterative learning control(ILC). In a way similar to a human being learning a desired motion pattern through repeated trials, the ILC system is able to acquire dynamically real-time data of the controlled system during each trial and make changes accordingly to the control input signal for each successive repetitive operation.
However, the conventional ILC has several deficiencies such as lack of robustness, lack of the knowledge representation and lack of proper model which describes learning process as well as the system dynamics. To solve part of these deficiencies, three ILC algorithms are proposed in this thesis.
I. A learning control algorithm with feedback based on the Fourier series approximation of the system input/output(I/O) signals is proposed for the continuous time linear systems, knowing that the inverse model of the system is one of the most appropriate functions to construct an iterative learning controller. The convergence condition of the proposed algorithm is provided and the existence and uniqueness of the desired control input is discussed. The effectiveness of the proposed algorithm is illustrated by computer simulation for a robot trajectory tracking. It is shown that, by adding a feedback term in learning control algorithm, robustness and convergence speed can be improved.
II. A learning control algorithm based on Fourier series approximation of the system signals is extended to the continuous time nonlinear systems, knowing that any function in $L_2$ [O,T] space can be represented as Fourier series. The convergence condition of the proposed algorithm is provided and it is shown that robot manipulators with a proper feedback satisfy the given assumptions. The effectiveness of the proposed algorithm is illustrated by computer simulation for a robot trajectory tracking.
III. An iterative learning control algorithm based on the 2-D system theory with feedback term is proposed for a class of unknown discrete-time linear systems and a sufficient condition for convergency is provided on the approximated space. The proposed algorithm has been successfully applied for the periodic disturbance rejection which occurs due to the imperfect mechanical assembly process or magnet inhomogeneity of motors for a Video Cassette Recorder (VCR) servo system.
