AAS 97-154

EXPERIMENTS IN THE USE OF LEARNING CONTROL CONCEPTS TO ELIMINATE ERROR IN CONSTANT VELOCITY BELT DRIVES

Y. Hsin and R. W. Longman, Columbia University; E. J. Solcz and J. de Jong, Xerox Ccorporation

Abstract

Learning and repetitive control use previous experience performing specific commands to improve the performance in future repetitions of the commands. In learning control the system returns to the same initial condition before each repetition, and the computations can be made in a batch mode between repetitions. In repetitive control the command or disturbance is periodic, but the system keeps running without resetting the initial conditions. Here we develop conditions for convergence when the repetitive control problem is addressed in a batch mode by applying learning action, waiting for steady state behavior, learning from this data, adjusting the command, and repeating.