Non-linear Impulse Methods for Aeroelastic Simulations

We present the results of extending the methodology of system identification techniques as proposed by Won et al. to study three-dimensional multimodal structures. Critical to the success of the system identification technique are data representation and the type of training data involved. Once the appropriate model has been identified and trained, subsequent predictions would be efficient and fast. In this work, an algebraic and neural network model were used in lieu of computationally intensive CFD solvers to expediate aeroelastic simulations. The transonic flutter of the AGARD 445.6 wing was used as a test case. To generate the training data necessary for the creation of the system identification models, two approaches were investigated. In the first approach, non-linear filtered impulse signals were applied mode by mode to the dynamical system to obtain the system’s responses. In the other approach, a staggered sequence of filtered impulses was used to elicit the responses in a single CFD run. Results show that generally the non-linear neural network model of radial basis function trained with the staggered filtered impulse signals performed much better than the algebraic autoregressive model.