DETECTION OF HELICOPTER ROTOR SYSTEM SIMULATED FAULTS USING NEURAL NETWORKS

Simulated fault data from a mathematical model of a damaged rotor syst em are used to develop a neural network based approach for rotor syste m damage detection. The mathematical model of the damaged rotor is a c omprehensive rotorcraft aeroelastic analysis based on a finite element approach in space and time, Selected helicopter rotor faults are simu lated through changes in inertial, damping, stiffness and aerodynamic properties of the damaged blade, Noise is added to the numerical simul ation to account for sensor noise and inherent uncertainty in the real system, A feedforward neural network with backpropagation learning is trained using both ''ideal'' and ''noisy'' simulated data. Testing of the trained neural network shows that it can detect and identify dama ge in the rotor system from simulated and noise contaminated blade res ponse and vibratory hub loads data, For accurate estimation of the typ e and extent of damages, it is important to train the neural network w ith noise contaminated response data (Ref, 1).