کنترل پیشرفته و هوش مصنوعی مبتنی بر داده *

This paper addresses the error convergence rate of data-driven iterative learning control (ILC) for single-input-single-output (SISO) systems. Since the error convergence rate depends on the learning filter, which ideally should invert the plant dynamics, the challenge lies in creating the ILC learning filters that approximate the plant inverse without having the plant model. Zero-phase or time-reversal filtering ILC is applied to track smoothened impulse, where the learning filter is progressively updated while trajectory learning proceeds. The approach drastically accelerates the error convergence rate of the time-reversal based ILC. The progression of the ILC learning filter brings an additional degree of freedom for the learning filter design with proven stability properties. Simulation results for tracking a chirp reference on a linear motor positioning system demonstrate the effectiveness of the approach.