ILC Initial Input Selection with Experimental Verification

Abstract

Error convergence in Iterative Learning Control (ILC) is generally highly dependent on the selection of the initial input signal applied to the system. Techniques for generation of an initial choice of input are therefore considered in this paper, based on i) a frequency-domain model-based approach, ii) a time-domain model-free method involving use of previously stored tasks and their associated convergent input demands, and iii) a combination of these approaches. Each is shown to significantly decrease the error over subsequent trials using a common form of linear ILC algorithm compared with a more arbitrary initial input selection. Experimental results are then presented using a gantry robot test facility in order to establish the efficacy and practical applicability of each technique.