Research On Methods Of Identification Modeling Of Unmanned Helicopter’s Flight Dynamics Near Steady Condition

Abstract:Accurate parametric model of the UAV dynamics are difficult to be identified, because of complex operation and its characteristics of multiple coupling, instability, high order and easy disturbance. The current domestic methods on system identification modeling are inadequate to meet the requirements of the UAV’S autonomous flight of high quality. The simple dynamic model of helicopter is established firstly. The identification algorithms and model which is versatile、easy to be used by control system, and describes the dynamics of the helicopter accurately is researched based on manual flight test data.(1) the off-line time domain approach of identification based on state space model which is white box is researched. The "single data channel" method is proposed to estimate initial parameters of multi-channel model effectively and efficiently.PEM-GA is used to identify the parameters of a4-DOF UAV state space model to improve the precision. The predict outputs of the final model match the mesures outputs varing slowly well.(2) Efforts to enhance the accuracy of the angular velocity of the channel model are made, and the off-line time domain approach of identification based on state space model which is black box is researched Expanding subspace method is proposed based on the subspace-PEM method to guarantee the the precision of model and robustness of the algorithms. The6channels model fits well with the data sections in the identification and verification both, and dynamic characteristics of the SVU can be analyzed based on the model. But the model order is high and highly sensitive to data so that it is not conducive to be handled by the control system.(3) The off-line identification in frequency domain which is more robust and can limit the model order based on transform function model is carried out. It is the first time that the FRF(frequency response function) of the unmanned helicopter have been identified by the local polynomial method:"fast" method. The simulation results show that the construction method can effectively suppress the color noise estimated and leakage errors of DFT, while maintaining a high frequency resolution, better than the traditional method of CIFER. The transfer function models of longitudinal and lateral motion of SVU200are identified through least squares-maximum likelihood method based on the nonparametric FRF. Test results show that models owe the excellent ability to adapt to the rapid change of the data. With the high overall accuracy and the low order, the model identified reached the expected requirements.(4) The results of the off-line identification are verified further via recursive algorithms. It is found that Hammerstein-Wiener model accuracy is better than NARX model, through study on the nonlinear model identification of dual-channel angular velocity. There may improve the accuracy of unmanned helicopter simulator model using this model.