Research On System Identification Of Small-Scale Unmanned Helicopter

Compared with the fixed-wing UAV,small-scale unmanned helicopters have many good flight features,such as vertical take-off and landing,hovering,more flexible attitudes and so on.Moreover,because of many advantages,such as low application cost,convenient deployment,small-scale unmanned helicopters have been proven to have broad application prospects in military and civilian fields.The small-scale unmanned helicopter industry has developed rapidly in recent ten years,and some national defense research institutions,enterprises,and universities regard it as a very attractive research topic.Our laboratory has worked in this field for a long period of time.On the basis of the achievements of our laboratory,this thesis further explores and studies the related issues of system identification of small-scale unmanned helicopters.Firstly,based on the first-principles modeling and linearization theory,an aerodynamic model of the helicopter in hovering condition is established,and an efficient and reliable frequency domain identification procedure is proposed to obtain the values of parameters in that model.This helicopter aerodynamic model has been rigorously validated by flight experiences and has been applied in the design of the flight control system by other research groups in our laboratory.Next,in order to solve the problems that the frequency domain identification method often involves complex computation and numerous parameters,an orthogonal basis identification method based on reinforcement learning technique is designed.The designed method can avoid the complex calculation of frequency domain transformation,effectively solve the problem of parameter selection involved in the calculation process,and has the ability to accomplish the system identification task flexibly and accurately.At last,in order to obtain models which can predict the real system more accurately,two online identification methods are proposed in the framework of online identification algorithm and time-varying system.By fixing the aerodynamic model of the small-scale unmanned helicopter online,the prediction accuracy of the aerodynamic model is improved.It should be emphasized that the above three main contents have been verified by the actual flight data.The main work and innovations of this thesis are summarized below:(1)We introduce the research background and significance of small-scale unmanned helicopter,and briefly review the research status in both domestic and aboard.The related system identification theory and its application are described and analyzed in detail.(2)We introduce the basic knowledge for a small-scale unmanned helicopter's modeling and present the whole aerodynamic model to be identified in this thesis.Based on the frequency domain identification theory and CIFER software,an efficient and reliable system identification process is designed,and the identification and validation of the above model are also completed.(3)In order to solve the problem that the frequency domain characteristics of the coupling coefficients between the lateral channel and the longitudinal channel of our small-scale unmanned helicopter are not obvious and the identification results are inaccurate,we propose an iteration method between the two channel which descend the cost functions of lateral and longitudinal channels respectively,and the problem of identification for coupling parameters is solved.The actual flight test shows that the accuracy of the system model considering the coupling terms is better than the system model without considering the coupling terms.(4)The orthogonal basis theory is applied to the system identification of small-scale unmanned helicopter,which reduces the number of parameters to be identified and improves the numerical conditions in the identification process.Aiming at the key point of orthogonal basis system identification — pole selection problem,a pole iteration method based on reinforcement learning algorithm is designed.In order to overcome the dimension curse problem in reinforcement learning,we modify the algorithm and solve the dimension curse problem,thus ensuring the effectiveness of our algorithm.This method can make full use of the user's prior knowledge and solve the problem of system identification under orthogonal basis function intuitively and effectively.(5)In order to improve the prediction accuracy of the model,realize the online identification procedure of small-scale unmanned helicopter,after fully considering the timevarying characteristics of helicopter system,we apply the recursive least squares algorithm with a damping term to the online identification experiment of small UAVs based on the framework of classical least squares method.The result shows that the algorithm can restrain the jitter of parameter estimation and make the identification result easier to read.This method has a positive engineering significance and can be considered as a "negative" scheme against the influence caused by the time-varying parameters.In order to overcome the influence of timevarying parameters on the estimation results,we propose a new information-weighted algorithm.By estimating and compensating the influence of time-varying parameters,the algorithm can achieve an asymptotic convergence result assuming there is no measurement noise,a rigorous proof of convergence is also provided.Both numerical simulation and system identification flight experiment of the small-scale unmanned helicopter verify the validity of the algorithm.At last,some problems are proposed for further research and exploration after the summary of this thesis.