AAS 96-106

PHASE CANCELLATION LEARNING CONTROL USING FFT WEIGHTED FREQUENCY RESPONSE IDENTIFICATION

R. W. Longman and Y. P. Wang, Columbia University

Abstract

Previously, a phase cancellation learning control law was developed using a system model together with reidentification using OKID. Experiments on a Robotics Research Corporation robot at NASA Langley, dramatically decreased tracking error by a factor of 1000. Reidentification was necessary since high frequency modes are not identifiable from initial data, but become important once low frequency errors have been eliminated. In this paper, alternative reidentification methods are developed. Since the learning algorithm is based on frequency response, the identification is done in the frequency domain, and done using an intelligent data weighting process aimed at automatic processing. Care is exercized to give proper wieghting to initial data vs. later data, to data with large signals vs. small for each frequency component, to frequency components with large vs. small amplitude ratios to the largest amplitude component of the set, etc.