Linear parameter-varying model identification for flutter prediction

Abstract

Flutter-induced vibrations can easily cause large structures like airplanes and bridges to fail, which even today makes flutter prediction an important topic of research. To predict flutter, recursive system identification can be used to capture the time-varying behavior using time-invariant techniques combined with the forgetting of old data. However, forgetting older data takes time, which introduces a delay in the predictions. In order to circumvent this prediction delay the use of Linear Parameter-Varying (LPV) model identification from pre-flutter data is proposed, as this directly identifies a time-varying model. This thesis will compare recently published global, local, and glocal LPV identification algorithms and assess their flutter prediction capabilities both in simulation and using experiments. The simulation results show that using LPV model identification the flutter speed could be predicted to within 10 % even in the presence of significant noise. Furthermore, guidelines are given on how these methods can be used and what their limitations are. Regarding the experiments, flutter-induced vibrations could not be reproduced due to fundamental problems in the experimental setup for which recommendations are made.