The work aims at showing the importance of flutter predictions in the preliminary design of Propfan-Open Rotors. Both single rotating propellers and contra rotating open rotors are investigated. The related structural sub-systems are modelled through finite element analyses, the aerodynamic sub-systems exploit a finite volume full potential formulation suitable for unsteady transonic flows. An ad hoc technique is developed for simulating the flow field around a contra rotating open rotor configuration. The effectiveness of the proposed aeroelastic analysis is successfully assessed for single propellers through comparisons with reference numerical and experimental data available in the literature, as well as against Euler flow based solutions. Results for contra rotating open rotors cannot be validated because of the lack of corresponding open literature data.

The effectiveness of using neural networks to predict rotor loads on the AW609 tilt-rotor is proven in this work. The main objective is to find a viable architecture for a neural network simple enough to be implemented in real time, with the aim to have a reliable prediction of rotor loads during telemetry monitoring sessions of flight test operations. The real time comparison of the loads predicted by the neural network with those measured by the aircraft instrumentation can provide immediate hints of incipient anomalies. A simple Feed Forward neural network has been tested, analyzing briefly the pros and cons of such a choice versus other possible architectures. The proposed neural network will estimate the bending loads (beam and chord) and the pitch link axial load, given the parameters that describe the aircraft trim point and how it is maneuvering. Instead of trying to estimate directly the time history of the loads, with all its associated dynamics, an approach based on a harmonic decomposition is here proposed. In particular, the signal is first decomposed in its harmonic components and various neural networks are trained efficiently to predict a single harmonic at a time. The complete time history is then reconstructed a-posteriori by combining all the signals predicted by the different neural networks.