Neural Network for Positioning Space Station Solar Arrays

Abstract As a Shuttle approaches the Space Station Freedom for a rendezvous, the Shuttle's reaction control jet firinga pose a risk of excessive plume impingement load of Freedom solar arrays. The current solution to this problem, in which the array. are locked in a feathered position prior to the approach, may be neither accurate nor robust, and is also expensive. An alternative solution it proposed here: the active control of Freedom's. beta gimbals during the approach, positioning the arrays dynamically in such a way that they remain feathered relative to the Shuttle jet molt likely to cause an impingement load. An artificial neural network is proposed as a means to determining the gimbal angles that would drive plume angle of attack to zero. Such a network would be both accurate and robust, and could be less expensive to implement than the current solution. A network was trained via backpropagation, and results, which compare favorably to the current solution as well as to some other alternative, are presented. Other training options are currently being evaluated.