System Identification For An Unmanned Aerial Vehicle Using The Maximum Likelihood Method

Unmanned Aerial Vehicles(UAVs)can be defined as vehicles that are remotely piloted or autonomously controlled to execute a predefined flight task,which have been widely used in military and civil fields.In view of its characteristics of strong nonlinearity and aerodynamic derivatives that are difficult to obtain,the system identification approach is introduced.The mathematical models of UAVs are determined by system identification based on input and output data,besides,the premise of traditional control theory is the need for correct mathematical models of systems.To a certain extent,the accuracy of system identification largely affects the controlling quality.Meanwhile,accurate mathematical models of dynamic process gain more and more importance for analysis and simulation purpose in fields like control design and optimization as system identification has a good applicability for getting dynamic models efficiently.In this paper at hand,according to the structure characteristics and modeling purposes of small UAVs,it is vital to determine a suitable system identification method.Compared with the frequency domain and several classical algorithms,the maximum likelihood method in time domain is chosen for system identification.First of all,aimed at a small UAV,the mathematical model is built for small unmanned aerial vehicles using the rigid motion equations based on the geometry and aerodynamic information,and reference coordinates as well as the conversion matrices are determined.According to the inherent characteristics of the aircraft,besides parameters to be identified,variables of state,control and output of the models also need to be explicit.Secondly,the output response is obtained by simulating standard signals to the model.And according to the mathematical model,the output error equations are determined.To test the model and the algorithm,this paper applies the maximum likelihood method for system identification in simulated data,where model parameters are adjusted iteratively to minimize the error between the measured variables(system output)and the estimated(model predicted)responses.The simulation results are excellent,proving that both the model and the algorithm are suitable for the study of the aircraft.Finally,the flight test is done artificially.After obtaining the recorded flight data,a step is undertaken to reconstruct the aircraft’s actual states during the test flight,from which erroneous measurements can be determined.The final task is to use the maximum likelihood method to obtain aerodynamic data from real flight test results,in which process the error source and sensitivities of the model to the change of parameters are analyzed.The method and strategy are proved feasibly and acceptable according to the results.