On-Orbit Multiple-Input and Multiple-Output System Realization for Space Structures

This work seeks to develop a systematic approach to closed-loop MIMO structural realization that incorporates the Eigensystem Realization Algorithm developed by Juang and Pappa [Juang and Pappa, 1985]. Measurement data is transformed into the frequency domain for multiple time vector lengths and then used in a Bayesian classification algorithm in order to optimize data selection for the realization algorithm. Resulting scores from the realization algorithm, in the form of modal amplitude coherence values, are also included in the classification algorithm to help determine the best combination of sensor data for various bandwidths of modes. The block-Hankel matrix characteristics are determined with the aid of a neural network. The neural network provides the nonlinear mapping between the block-Hankel matrix size characteristics and resulting modal amplitude coherence values. This research builds a strategy to provide dependable system realization for a space structure that is operating in closed-loop. Two steps assure sufficient richness and quality in the data used by the realization algorithm. First, sensor data is pruned through classification, as alluded to above; second, an excitation strategy is enforced