Iterative Identification of Feedforward Controllers for Iterative Learning Control

Abstract

One the the main disadvantages to the standard iterative learning control (ILC) problem is that the learning filter is static and does not update based on the current iteration information. This means that while the learning control is updating the feedforward control input, the learning controller itself is not updating. In this paper, we present an iteration varying learning filter which is based on identification techniques. Every iteration, this learning filter is computed to be the closed loop system inverse based on a least squares approximation. We show that the iteration varying learning filter successfully reduces the 2-norm error as fast as a static, model-based learning filter and much better than a P-Type learning filter. This results suggests the benefit of extensive preliminary system identification efforts for iterative systems is limited at best.