THE APPLICATION OF ARTIFICIAL NEURAL NETWORKS FOR TELESCOPE GUIDANCE: A FEASIBILITY STUDY FOR LYMAN FUSE

Abstract

ABSTRACT Since the fine error sensor camera of the Lyman Far Ultraviolet Spectrographic Explorer (FUSE) has significant residual field curvature aberration, the star images over the field of view have a wide variety of shapes. A search for appropriate backup guiding techniques led to the investigation of artificial neural networks (ANN). Such a technique is shown to be capable of learning the image shapes of stars if they are sufficiently differnet. This study investigates the feasibility of using image patterns as positional references for telescope guidance to satisfy redundancy requirements for the mission. For this initial simulation, the ANN was trained to categorize images according to how far they were from the center of the field of view (radius). We found that a non-linear, single hidden layer ANN learned 90 percent of the training patterns, then correctly classified 89 percent of a set of patterns randomly spread over the field of view. This indicates that the network interpolates between training images. Half of the misclassifications are attributed to the image pattern degradation caused by the secondary support structure spider.