On Robustness against Measurement Noise of Iterative Learning Control Based Identification

One of the most important issues in control system design is to obtain an accurate model of the plant to be controlled. Though most of the existing identification methods are described in discrete-time, it would be more appropriate to have continuous-time models directly from the sampled I/O data. From this viewpoint, the authors have developed such a direct identification method based on ILC (Iterative Learning Control) approach. This is a new application area of ILC. The method often yields accurate models even in the presence of heavy measurement noise. The robustness against noise is achieved through (i) projection of continuous-time I/O signals onto a finite dimensional parameter space, (ii) initial models through preparatory experiment and (iii) noise tolerant learning laws. This paper examines the accuracy of the initial models and convergence property of ILC in the presence of heavy colored noise through detailed simulations, which demonstrates the robustness of the ILC based identification method.