On-orbit nonlinear structural parameters realization via artificial neural network

Abstract

Structural parameters realization is formulated as a pattern recognition problem. Candidate mathematical models are designated as with which computer simulations are conducted to generate simulated system responses. Patterns are organized into pattern classes in a topdown dichotomous manner based on the variation of the simulated system responses such that the coherence property of patterns within any pattern class is embedded. An adaptive neural network serves as a pattern classifier. The actual response of the real world system is classified as the pattern class of the most similar system response to determine the most probable mathematical descriptors of structural parameters. The proposed methodology was successfully applied to the realization of the disturbance damping torques at the alpha gimbals of the Phase I Space Station Freedom model. Our experimental data were obtained analytically by simulation with additive Gaussian noise. The results are encouraging, showing a high percentage of correct classification in a noisy environment.