Multioutput Autoregressive Method for Aeroelastic System Identification and Flutter Prediction

The paper presents a numerical study of aeroelastic system identification and flutter prediction based on large structural response data, such as strains recorded in fiber-optic sensors or accelerations from multiple IMUs. As in existing flutter prediction methods, the aeroelastic frequencies and damping values are estimated from the time histories of structural responses at stable, pre-flutter conditions. The modes relevant to flutter are tracked over the airspeeds and used to compute a Flutter Margin (FM) parameter, which points to the predicted flutter onset speed. As the accuracy of the estimated frequencies and damping values affects the flutter prediction by the FM, the current study proposes two approaches that take advantage of the multioutput data available for accurate estimation of the dynamics, FM, and flutter onset. First approach is to model the aeroelastic responses as Vector Autoregressive Moving Average (VARMA) processes, accounting for the interdependencies of the responses on each other’s time history. This approach is shown to work well and produce more accurate estimations than an ARMA model approach, in which each response is only dependent on its own history. However, the VARMA modeling approach is limited to only handling responses from a few sensors. A second approach is tested, in which a poly-reference Least Square Complex Frequency-domain (LSCF) estimator is used to identify the two aeroelastic modes that participate in flutter and determine the FM and onset speed from their properties. The method yielded somewhat less accurate modal parameters and flutter prediction, but could be favorable in cases of very large response data sets, where the VARAMA approach hits its limits.