Neural Network Based System Identification of 3-Dof Quadcopter Dynamics

Quadcopters are used frequently in various applications around the world. However, the complexity of the quadcopter's dynamic behavior requires advanced stabilizing control for its movement. In addition, precise mathematical modeling for quadcopter is highly challenging task because of its highly non-linear behaviour and complex structure. In this paper, we identify the dynamical system of the quadcopter using two system Identification techniques. The dynamics includes the 3-movement angles pitch, roll and yaw. The first approach includes the direct identification of the transfer function using TF estimation from the input output data with the help of system identification toolbox in MATLAB. The second technique involves the machine learning, where we use feedforward neural network to first train the MIMO model that will later correctly identifies the model from data generated by the same model. In all techniques data is generated from Simulink using Pseudo Random Binary Signal (PRBS) and step signal in some dynamics depend upon the complexity of dynamic model.