Unknown input recovery for closed loop systems with saturations

An inverse method is developed to recover for unknown stochastic inputs. The proposed approach relies on an approximate finite impulse response model of the system which is seen as a linearization around the current input estimation. The inversion itself is done within the frequency domain in order to lower the computation cost. This classical approach requires a regularization technique to be used in order to limit sensitivity to noise. Several tools are discussed here, including shaping filters. An original procedure is then introduced for the automatic tuning of the regularization parameters. It is justified by a SVD analysis. The second main contribution is the saturation effects processing. These effects are taken into account for by separating inversion of the linear and nonlinear parts of the output error. This original method relies on the projection of signals on a dedicated basis. The overall procedure is applied to the recovery of the roll rate of an aircraft that is induced by a spatially non uniform turbulent wind along the wing. Such an input is not measured. Its estimation allows to minimize the difference between the recorder lateral displacement of the plane with respect to the runway axis and the simulated one.