Identification of Helicopter Dynamics based on Flight Data using Recurrent Neural Networks: A Comparative Study

In this paper, we present a comparative analysis of different artificial neural networks (ANN) for identification of longitudinal and lateral dynamics of helicopter using flight data. These methods have an advantage over the traditional methods for identification because the model structure need not be defined apriori. In case of helicopter dynamics, defining apriori model is difficult because of highly nonlinear aeromechanical characteristics due to the interplay between various subsystems like rotor, fuselage, power plant, transmission, empennage, and tail rotor. For a given input-output data set, ANN’s are capable of fully capturing the underlying relationship. Three different ANN architectures namely, Non-linear Auto Regressive eXogenous input (NARX) model, ANN with internal memory known as Memory Neuron Networks (MNN) and Recurrent MultiLayer Perceptron (RMLP) networks have been used to identify longitudinal and lateral dynamics of the helicopter at various speeds. Actual flight data are used for simulation studies that are carried out using various ANN architectures and their performances are compared. Based on the identification results, the practical utility, advantages and limitations of the three models are critically appraised.