Autonomous Neural Control of Space Platforms

ORINCON proposes to develop an Adaptive Neural Controller (ANC) based on the extended Kalman filter (EKF) principle, which requires neither the controlled process (or plant) nor the controll to have fixed, parameterized structures, as a superior method of performing identification of flexible zero-gravity structure dynamics. To demonstrate than an ANC can perform unsupervised identification of a dynamic model, a software-based ANC will produce a plant identification for a model of a simple beam structure. Model response will be calculated from a finite number of bending moments to generate training data. The control scheme will use an EKF on the first five beam moments plus the neural network. The neural network will training to reduce the error between the model's response and the EKF output. A final report will describe the system configuration, analyze test results, and make recommendations for future investigation. We anticipate that ANCs can provide real-time stabilization of large structures more accurately than traditional control of space structures with a wider range of unknown parameters than are other conventional controllers, and is generally applicalbe to mechanical structure control and robotics.