Developing a disturbance source characterization technique for small satellite applications

Abstract

In this thesis we establish a framework with which to characterize candidate sources of disturbance for small satellite applications. By characterize we mean estimate disturbance source vibrational frequencies, and by candidate sources we mean sources previously determined with the ability to induce micro-vibrations. This framework centers on the operation of distributed sensors, and we present a set of components capable of performing a characterization effort of this nature. Our implementation of supervised learning enables us to predict actuator operational frequency values based on accelerometer readings. The standardized mean squared error (SMSE), a measure of error between the mean prediction and the true value, important for quantifying prediction performance, is shown to be a function of the Fourier transformation type used; and we conclude which considered Fourier transformation results in the lowest prediction errors. Furthermore, we analyze how different dataset sizes and sensor-actuator pairings affect the frequency predictions.