Neural networks for residual motion identification of a suspended mirror

In this work we present a Neural Networks based approach to identify the residual mechanical motion of a suspended mass after that a digital control system has acted to damp the oscillations. The control system is based on optical levers that read the position of the mass. The digital control system is based on both standard linear filtering of signal. The Neural Network, used to realize a non linear model of the residual motion, is the Multilayer Perceptron, with the Bayesian learning scheme. In this work we show that, after a standard linear control, a Neural Network is able to identify the residual motion so it can be used to implement a "inverse system" to model predictive controlling.