We consider a two degree of freedom airfoil
oscillating in pitch and plunge, with either polynomial or freeplay structural
non-linearities. Using only a limited amount of transient noisy data, we
present a new approach based on the expectation maximization algorithm, for
parameter estimation, filtering and prediction. For systems with polynomial
nonlinearities the expectation maximization algorithm is constructed using the
unscented filter, and is illustrated for experimental data. For the freeplay
case we use switching linear Kalman filters and results with simulated data are
reported.