|M.Sc Student||Matan Argaman|
|Subject||Multi-Output Autoregressive Aeroelastic System|
Identification and Flutter Prediction
|Department||Department of Aerospace Engineering||Supervisor||Professor Raveh Daniella|
The research presents a methodology for reduced-order modeling of aeroelastic systems and flutter prediction. The aeroelastic system is modeled as a Multi-Output Autoregressive process and the model parameters are identified, using a least-squares estimator, based on aeroelastic responses to initial conditions, simulated in a CFD code. The aeroelastic system is identified at a few sub-critical (pre-flutter) dynamic pressure values. System identification is performed using all of the flutter-participating modal response output channels. Flutter onset is determined from a stability parameter that is computed for the two dominant aeroelastic modes, picked out of the overall flutter participating modes that are used for system identification.
The methodology is demonstrated on three test cases.
The first test case, a linear subsonic 2D airfoil, is used to demonstrate the methodology on a case where linear analysis can be used for accurate flutter prediction.
The second test case, a transonic 2D wing, is a test case for the nonlinear flight regimes, with comparison to wind tunnel flutter tests.
The third test case, a subsonic generic transport aircraft, is used particularly as a multi-mode test case.
The research compares the Multi-Output Autoregressive to Single-Output Autoregressive model and shows that the former is more straightforward to implement, robust, and requires shorter identification data. Furthermore, in some of the nonlinear and multi-mode test cases Single-Output Autoregressive models proved to be inadequate for aeroelastic system identification. Close to the flutter onset these models might predict an unstable system.
The research includes sensitivity analyses for several key parameters of the system identification and flutter prediction process. A sensitivity analysis for the required data length shows that only two cycles of the lowest frequency mode are required for a well converged
Multi-Output Autoregressive model. In terms of the required number of output channels it is shown that only the two dominant flutter modes are needed for system identification. The Multi-Output Autoregressive model is also shown to be insensitive to the magnitude of the initial condition used to excite the system in the CFD simulation, even in nonlinear cases.
Overall the methodology is simple to implement as it only requires an aeroelastic simulation capability and basic processing. It is computationally efficient and therefore suitable for CFD-based flutter prediction. Other than predicting the flutter onset, the method can be used to determine whether the aeroelastic system is stable at a specific dynamic pressure value, without resorting to a lengthy CFD aeroelastic simulation.