Attitude Control System Design of UAV Guo Li1, a, Xiaoliang Lv2, b, Yongqing Zeng3, c

Transcription

1 4th National Conference on Electrical, Electronics an Computer Engineering (NCEECE 205) Attitue Control ystem Design of UAV Guo Li, a, Xiaoliang Lv2, b, Yongqing eng3, c Automation chool, University of cience an Technology Beijing, Beijing, 00083, China 2 Automation chool, University of cience an Technology Beijing, Beijing, 00083, China 3 Automation chool, University of cience an Technology Beijing, Beijing, 00083, China a guo6393@sina.com, b lv_maxwell@sina.com, c @qq.com Keywors: Nonlinear; ANNs; Preict Moel; Attitue Control; imulation. Abstract. Base on the characteristics of the attitue control system, this paper use the neural network preictive moel controller which can ientify accurately attitue system an obtain real-time control sequence applie to the controlle object through the moel preiction, feeback correction an rolling optimization each link to complete motion attitue control of unmanne aerial vehicle. This control system has goo control effect, meanwhile, the fast, accuracy an anti-isturbance capability tracking a given process can be effectively improve.. Introuction There is a very significant piece of legislation that attitue control is extremely important for performance of the unmanne aerial vehicle (UAV) flight system. However, Movement of UAV attitue in the actual process contains a lot of uncertain factors. To get a higher control precision, we only use more avance control theory. In this paper, first of all, posture moel was establishe through simplifying scientifically. econly, combining with the expecte control goal an the actual conitions, we aopt the preiction algorithm base on BP neural network in the esign an expect to achieve better control effect. Then in orer to better show the avance esign performance of the controller, we will compare an analyze with the classic PID control performance. Finally, the system performances of robust an aaptability were stuie an analyze. 2. ystem Moel In orer to stuy conveniently, when establish UVA s moel [], we have the following two assumptions: aerial robot is rigi an quality is constant; ignore the groun curvature an the influence of the gravitational acceleration in moel. P =V () m V = C eb fb + [0 0 mg ]T (2) R = Hωb ϕtθ Insie, H = 0 Cϕ 0 ϕ Cθ (3) CϕTθ ϕ, o attitue angle motion equation is following: Cϕ Cθ θ t = q cos ϕ r sin ϕ ϕ = p + tan θ (q sin θ + r cos ϕ ) t ψ = (q sin ϕ + r cos ϕ ) cos θ t 206. The authors - Publishe by Atlantis Press (4) 735

2 Attitue kinetic equation by the moment of momentum theorem is: ω b + ωb ωb = τb. (5) Ixx 0 Iyy Insie, inertia matrix is: = 0 Iyy 0. Ixx 0 Izz Instea of blae element theory, we use straight the way of ieal moment research to simplify the system moel, ignore coupling effects between main rotor an tail rotor. Theory suggests that this kin of approximation is still meaningful within a certain range in stuy [2]. By the motion equation Eq.4, Eq.5, selecte [ w ] T x wy wz φ θ y as state variables, the initial state is R= [0; 0; 0], the state space expression can be obtaine. x = Ax + Bu y = Cx + Du (6) Insie, = A , B = C = , D = Controller Design 3. Neural Network Preiction Moel Controller Compare with the traitional PID controller, the moel preiction control not only use eviation information both the past an present, but also can get the eviation value of the process via the preictive moel at future [3]. Therefore, the control effect is better [4]. Preictive Moel: Due to the characteristics of the nonlinear system, we aopt neural network to make the preictive moel. Moel output is: y m = O (7) O is the output of neural network, BP neural network weights between each neurons parameters was obtaine by the training sample. Reference Trajectory: et y ( ) T = ϕ, θ, y, reference trajectory is: y t + k = -α y + αy t (8) r ( ) ( ) ( ) T s T k Insie, α = ( e ), T s is the sampling perio an T is a time constant of reference trajectory. Feeback Correction: Use error between measure an preicte quantity to correct moel output. With correcting by feeback of historical error, preictive moel output is: y t + k = y t + k + β y t y t (9) ( ) ( ) [ ( ) ( )] p m - m Rolling Optimization: Optimization is base on real-time manipulation at each sampling time, an get optimal control rule since starting sampling. Actual optimal control law applie to 736

3 controlle obtain is only the current item, the item of next sampling moment is calculate again. Accoring to the optimization algorithm an the optimal control: N 2 N e ( t k) u 2 = + + µ u ( t + k) N (0) 0 e t + k = y t + k y t k, u ( t + k) = u( t + k) u( t + k ).The resulting control Insie, ( ) r ( ) p ( + ) sequence is ( t k) u +, this is our controller output. 3.2 A Cascae Control tructure Block There form a ual preictive control metho we use neural network preictive moel to control respectively the ynamics an kinematics moule [5]. et an output reference angle value given for the secon part of kinematics moule, the input of the ynamics moule in the first part refer from the actual control sequence that is occurre by neural network preiction of the kinematics moule in the secon part, because this sequence is constantly changing, so the ynamics moule reference is ynamic, then apply Dynamics moule control input to the real moel of the controlle object, establish a cascae controller As shown in the Fig.. Insie, NN Preictive Controller is shown in Fig. 2 an Fig. 3. Fig. A cascae control structure block iagram of neural network preictive moel Fig. 2 NN preictive control parameter Fig. 3 Plant ientification parameter 3.3 A Three Dimensional PID Controller A single imensional PID controller is use to generate a control effect on the output of the roll angle, pitch angle an yaw angle. The control torque of the three irections are the same, ue to the axis of rotation inertia,, 2, 3, are not equal to each other, an changes in't get negative feeback control in the irection of the other two output. Therefore, the integrator input to an illegal value, not only system cannot achieve feeback attitue angle control, but also the tren of feeback attitue angle cannot stabilize the set value, the reason is the coupling effect between the axis. Uner the ranom isturbance signal, the steay state fluctuation in the roll Angle is opposite bigger, shaft coupling also affects the stability of the other axis an persistent oscillation process. 737

4 In orer to overcome these rawbacks, we aopt a three-imensional PID controller, which is the three attitue angles were fe back to the three PID controller, an then the control output value of the each controller is mixe as the total control quantity. When the three attitue angles are equal to the initial value, we can achieve the attitue control perfectly. Control moel an control results are shown in figure an iagram. In the process of solving the PID parameters of the ieal 4: control output curve, we use the critical ratio metho, an we get the parameters: Kp=2, K=0.375, Ki= imulation an Analysis A signal in Fig. 4 was input to control system with PID, so we acquire output of Roll as shown in Fig. 5, an error curve in Fig. 6. Then interference signal in Fig. 7 was ae to control system too, system output is shown in Fig. 8 uner isturbance. When controlle quantity shown in Fig. 9 is input to control system with neural network preictive moel, we obtain control curve as shown in Fig. 0, Fig., Fig. 2. While research the output curve of the neural network preiction moel, we have the following features: (). Rise time, ajust time shorter, system response is well; (2). Overshoot less, an even can achieve no overshoot state; (3). Output curve is smooth, there is no steay-state error; (4). There is more stable state uner unpreictable isturbances. But the multistage neural network preiction control increase greatly account of caculation, an there will have big error between the output of the neural network preiction moel an the actual in some situation an appear peaks an troughs similar to the shock signal in some cases. Its range of attitue angle controlle stably is less than PID control. Fig. 4 Control quantity with PID Fig. 5 Roll curve with PID Fig. 6 Error curve with PID Fig. 7 Interference signal Fig. 8 PID output Fig. 9 Controlle quantity Fig. 0 Pitch curve Fig. Roll curve Fig. 2 Yaw curve 738

5 5. Conclusion The attitue control is an important part of UAV control system, it is interisciplinary science fiel incluing air ynamics, the boy ynamics, kinematics, transmission signal etection, an many other key fiels of science [6]. How to achieve the process of flight attitue control stability an high precision nee to consier the coupling nonlinear an unpreictable isturbance factors. This research has been carrie out on the basis of the existing research theories an esign cases, aiming at the technical ifficulties of UAV attitue control. Accoring to theoretical analysis an simulation show that, compare with the classical PID controller, using the neural network preiction moel of cascae control system has obvious avantages at the rapiity an optimization of controlle quantity, controlle object still can ensure the stability to some extent uner the unknown time-varying torque isturbance. Acknowlegements In this paper, the research was sponsore by the Funamental Research Funs for the Central Universities (Project No ). References [] Mun-Taek Choi an Henryk Flashner. Neural-network-base pacecarft Attitue Control an Momentum Management. AIAA, 2000, [2] NWE T T, HTIKE T, MON K M. Application of an inertial navigation system to the qua-rotor UAV using MEM sensors. Proceeings of worl acaemy of science, 2008, 32(8). [3] Calise A, Rysyk R T. Nonlinear Aaptive Flight Control Using Neural Networks []. IEEE Control ystems Magazine, 998, 8(6): [4] iang Ping, Rolf Unbehauen. Iterative Learning Neural Network Control for Nonlinear ystem Trajectory Tracking[]. Neurocomputing, 2002, 25(48): [5] CATILLO P, DUL A, LOANO R. Real-time stabilization an tracking of a four rotor mini rotorcraft. IEEE transactions on control systems technology, 2004, 2(7). [6] iao Xiaohong, Guan Xinping. Nonlinear ystem Analysis an Design [M]. Beijing: Publishing House of Electronics Inustry,