Experimental nonlinear identification of a lab-scale helicopter system using MLP neural network

Artificial Neural Networks (ANN) are justified to be used as universal function approximators. That is to say, with the proper structure, ANN can approximate any nonlinear dynamics and describe the required input-output mapping. This paper introduces an experimental investigation of this concept, where a Multi-Layer Perceptron (MLP) network is used to model the entire dynamics of a nonlinear complex structure. An empirical modelling approach using MLP is established and applied to model a real-time Flexible Manoeuvring System (FMS). Various modified versions of the essential back-propagation learning algorithm are used to train the MLP model. Results are investigated and validated in both time and frequency domains; networks that are trained with the Levenberg-Marquardt (LM) algorithm conducted the best performance. Then, the validation checks indicate rapid improvements in the obtained model using the introduced MLP approach.