Aircraft System Identification using Pseudospectral Parameter Optimization with Adaptive Nodes

Abstract

The US Air Force (USAF) Flight Test Center has made recent advancements in its system identification capabilities. The most recent project from the USAF Test Pilot School (TPS) was to implement a near real-time parameter identification curriculum sortie. This sortie development entailed automatic parsing of input doublets and then analyzing these doublets using equation error. The Global Vigilance Combined Test Force would like to take the next step toward more rapid parameter identification, specifically by extending beyond the previous equation error techniques and advancing towards trajectory reconstruction, i.e. data compatibility and output error. The paper herein demonstrates how to solve for aerodynamic stability and control derivatives via dynamic optimization using pseudospectral methods. Future research may include time delay which can be added as another optimization parameter, thus resolving issues with heavily filtered instrumentation data. The inherent implicit integration within the pseudospectral method has already shown improved stability over traditional explicit integration within shooting methods (1) . Also, by optimizing the solution using a smaller finite number of nodes, the convergence time may drastically improve over explicit or sequential integration. Lastly, by using segmented adaptive node distribution, vice a single segment, a more accurate solution may be obtained using the same number of nodes. The work herein introduces a new method for system identification algorithms. This is intended to be an extension, or improvement, upon the existing accomplishments within the System Identification Programs for Aircraft (SIDPAC) software documented in Aircraft System Identification: Theory and Practice by V. Klein and E.A. Morelli. The pseudospectral method and the General Pseudospectral Optimal Control Software (GPOPS) to solve dynamic optimization problems are already established. This paper extends these established techniques towards this new application. These results are modeled after textbook examples for output error. This technique is motivated by the desire to develop a practical real time parameter estimation capability.