This research addresses the modeling, analysis, and control of flutter, in a high order nonminimum phase aeroservoelastic UAV system with Piezoelectric single servo input Flutter suppression requires output feedback supplied by four accelerometer sensor readings. Linear time invariant plant models vary with the exogenous parameter, velocity. The f w gain scheduling control strategy, io conjunction with Linear Quadratic Gaussian regulation, with prescribed degree of stability, at nominal operating conditions, produces interpolated controller and observer gains over a uniform grid of operating velocities. "Wont case" highest velocity eigenvalues are used for stabilization. Simulations of augmented closed-loop systems show settling times ranging between 0.2 s and 0.4 s. Control efTort is uniform and converges to 0 rad within 0.1 s. The end result is a robust global controller that suppresses flutter over the entire velocity envelope.
I. ~NTRoDUCllONOver the past several years, flutter suppression techniques have been actively investigated, on the Benchmark Active Control Technology (BACT) wind-tunnel model, developed by NASA Langley Research Center [1,2,3,4]. The BACT statespace design model consists of ten states and is parameterized by dynamic pressure. Transonic flutter occurs at approximately Mach number 0.77. One flutter mechanism causes classical flutter instability [5]. This model has provided a testbed for the development and testing of passivity based robust control, LPV gain scheduling control, H-infinity, p-synthesis generalized predictive control, fuU order and reduced order LQG (Linear Quadratic Gaussian), and others.Recently, altemative methods, for achieving flutter suppression, in high-order nonminimum phase systems, with three flutter mechanisms, have been proposed. These control strategies include 1) gain scheduling techniques for interpolating between linear time invariant (LTI), parameter dependent controllers [6,3,7], 2) the construction of a single LQG regulator satisfying a prescribed degree of stability QDS) criterion[S], and 3) recurrent neural networks for adaptive predictive control [9].In the case of gain scheduling, LMI (Linear Matrix Inequalities) techniques have generally been applied for gain construction. For the Piezoelectric system under study, feasibility gains could not be constructed, using LMI, for both the controller and observer modules. This was due, in large part, to the ill-conditioning properties of high-order state space matrices. In the case of the single LQG-PDS controller, robust performance could not be achieved for velocities below a certain value. The thiid alternative, recurrent neural networks, require extensive training times and extensive line-tuning of parameters for configuration of appropriate network topologies. Such networks achieve l i i t e d robustness over the flutter envelope.The current study introduces fuzzy gain scheduling as a technique for robust flutter suppression in a high-order n o n " phase UAV system that operates at various speeds at const...