Dynamic system model identification of inertial sensors by means of neural networks

The thesis presents the design of a feedforward neural network for the model identification of a dynamical system of a general inertial sensor. According to the universal approximation theorem, feedforward network with a linear output layer and at least one hidden layer with any activation function can approximate any measurable function with any desired non-zero amount of error provided that the network is given enough hidden units. This theorem simply states that no matter what function we are trying to learn there is always a neural network which will be able to represent the function. So, strong of this result, the work consisted to simulate a black box approach in which the data of input and output of the dynamical system have been used first to train the neural network to predict the future behaviour of the system itself, and then to validate the model.