Dynamic Neural Units for Adaptive Magnetic Attitude Control of Spacecraft

A neural network based on dynamic neural units has the capability to handle any type of nonlinearity. In addition, it can adapt itself to parameter changes in real time. In this paper, such a dynamic neural network is used to design a controller through inverse modeling to address the attitude control of an Earth-pointing magnetically actuated spacecraft. Furthermore, normalization of weights of the dynamic neural units is proposed to ensure their convergence for proper learning. The dynamic neural controller developed in this paper, being adaptive, not only takes care of any unknown disturbance torques, but it is also robust and can adapt itself if there are any large changes in the parameters in the plant, such as the moments of inertia. It is shown, that the stabilization accuracy of the plant is better under the proposed neural controller as compared with a proportional-derivate controller. Proof of stability, for the dynamic neural units and the system as a whole, is also presented.