Neural network-based modelling of a two-degrees-of-freedom twin rotor multiple input, multiple output system using conjugate gradient learning algorithms

Abstract

This paper presents a neural network (NN)-based non-linear dynamic modelling approach for a twin rotor multiple input, multiple output system (TRMS), in terms of its two-degrees-of-freedom dynamics. The TRMS is a highly non-linear system with significant cross-coupling between its horizontal and vertical axes. It is perceived as an aerodynamic test rig, representing the control challenges of modern air vehicles. Accurate dynamic modelling is a prerequisite to address such challenges satisfactorily. A feedforward NN has been trained using the Powell—Beale version of conjugate gradient and scaled conjugate gradient learning algorithms. The data for training comprises sine and square waves with various frequencies and amplitudes, pseudo random binary sequence (PRBS), and composite PRBS signals with different amplitudes. The trained NN-based models have been tested with a set of data that are different from those used for training purposes. For more validation, the power spectral density of the model is compared with that of the real TRMS and also the correlation validations of the test results are presented in order to show the effectiveness of the proposed model. The results show that the developed model can adequately represent the highly non-linear features of the system and can be used for sophisticated controller development.