Propulsion vibration analysis using neural network inverse modeling

Neural networks are employed to predict the amount and location of propulsion system rotor unbalance. Vibration data used to train and test inverse system models are generated via a high-order structural dynamic finite element model. Several neural network methods, including feed forward neural network using back propagation, node-decoupled Kalman filter (NDEKF) and support vector machines (SVMs) are investigated. Training results and performance among the various methods are compared. Original applications to nonlinear structural models and damaged structure models are shown