Research On Adaptive Control Of Loaded Quadrotor UAV Based On Neural Network

Quadrotor UAV can be used to complete some dangerous and complex tasks instead of manual work.When performing flight mission for quadrotor UAV,it is usually inevitably to carry payloads,and the suspended loads will swing within a certain range.In this case,the swing angle is hard to be measured generally,which makes it impossible to accurately obtain the impact acting on UAV,leading to modeling and control design more complex.Furthermore,considering the safety and physical characteristics,the position and attitude of quadrotor UAV need to be limited within a certain range to avoid potential safety hazards.Based on the above difficulties,this paper comprehensively considers the influencing factors such as model uncertainty,actuator saturation and failure as well as output saturation to conduct dynamic modeling and adaptive fault-tolerant control research for the loaded quadrotor UAV.The main tasks are showed as follows:(1)Aiming at the dynamic modeling problem for quadrotor UAV,this paper extends the classic ontology model of quadrotor UAV to the integrated dynamic model with suspended load.When analyzing the force of the suspended load,the inconsistency between the suspension point and the UAV geometric center,the change of COG with time and the unknown non-diagonal moment of inertia are considered,so as to make the model more accurate and facilitate the subsequent control design.(2)For the unknown load swing angles,this paper uses the method of matrix decomposition to solve it.According to the loaded quadrotor UAV model established before,the gain matrix of the rotating subsystem model is unknown and time-varying due to the unknown swing angle.Based on this situation,this article decomposes the gain matrix into three matrices with known structures,which can be used in the controller design derivation process.Finally,the controller will no longer have unknown variables appear,so the design have higher universality.(3)In view of output constraint problem,this paper introduces the BLF function to convert the system output constraint problem into an output error problem,so that the position and attitude of UAV do not exceed the constraint range.Next,this paper approximates the saturation function by a well-defined smooth function,and combines the backstepping method to solve the input saturation and actuator failure problems.In addition,as for the uncertainty of the system parameters,the number of online update parameters is significantly reduced and the amount of calculation is greatly reduced in the neural network approximation by using the indirect processing method of virtual parameters.(4)The efficacy of control scheme has been fully confirmed via theoretical derivation and simulation in this paper,which renders all signals in the closed-loop system are bounded while keeping the state output within a certain constraint range.Finally,the practical feasibility of design scheme is further verified through the experimental platform.