A flight control method of a bird-like flapping-wing aircraft

Abstract

The application discloses a flight control method of a bird flapping aircraft, and belongs to the technical field of aircraft control. Unmodeled nonlinearities and unknown disturbances in attitude dynamics are summarized as total disturbances of the system, a self-disturbance control device composed of a tracking differentiator, an extended state observer and an error feedback controller is used for attitude control in the roll channel and the pitch channel, the yaw channel is open-loop controlled, and the deflection angles of two independent rudders of a tail fin are solved through a control distribution matrix; a cascade nonlinear dynamic inverse combined with L1 adaptive compensation is used for horizontal position and speed, a sliding mode variable structure control is used for height, and a RBF neural network is used for approaching unknown parts of the system. The application can realize attitude stabilization of the bird flapping aircraft in pure flapping, pure gliding and flapping / gliding conversion modes, improve disturbance resistance, and realize three-dimensional trajectory tracking.

Description

Technical Field

[0001] This invention belongs to the field of aircraft control technology, specifically relating to a flight control method for a bird-inspired flapping-wing aircraft, which is used to achieve attitude stability control and three-dimensional trajectory tracking control of the bird-inspired flapping-wing aircraft in various flight modes such as flapping, gliding, and flapping / gliding transition. Background Technology

[0002] A bird-inspired flapping-wing aircraft is a biomimetic aircraft that mimics the flight patterns of birds in nature. Its flapping wings primarily function through reciprocating flapping motions, generating lift and thrust for flight, while the tail fins deflect to generate control torque. Compared to fixed-wing and rotary-wing aircraft, bird-inspired flapping-wing aircraft possess a biomimetic shape and superior aerodynamic and maneuverability, making them suitable for close-range reconnaissance and information gathering missions within confined spaces such as buildings and forests.

[0003] In existing technologies, attitude control for bird-inspired flapping-wing aircraft often employs proportional-integral-derivative (PID) or proportional-derivative (PD) methods. For example, the Robo Raven flapping-wing aircraft in the United States uses the ArduPilot hardware platform and PID attitude control algorithms to achieve automatic hovering and automatic flight path. However, due to factors such as external environmental wind interference, sensor noise, and tail control errors, the trajectory tracking error is relatively large. The attitude control of the DelFly series of flapping-wing aircraft from Delft University of Technology in the Netherlands is also based on a PD / P controller. In terms of modeling, existing studies mostly use quasi-steady-state assumptions to establish aerodynamic models of flapping wings. However, when the airfoil camber, stiffness-flexibility distribution, and other characteristics of the flapping wing change and the forward speed is high, the empirical models under the quasi-steady-state assumptions are no longer applicable.

[0004] Bird-inspired flapping-wing aircraft face several challenges during flight. First, in pure flapping mode, aerodynamic forces and torques exhibit periodic and drastic changes, displaying strong nonlinear characteristics. In pure gliding mode, however, these forces and torques change approximately linearly within a certain range. The flapping frequency varies significantly during transitions between these modes, leading to substantial changes in system dynamics. Attitude controllers designed for single flight modes cannot simultaneously guarantee attitude stability across multiple modes, potentially resulting in excessive control errors or even system divergence. Second, the bird-inspired flapping-wing aircraft has relatively small mass and inertia, making it susceptible to interference from external environmental gusts and internal electrical interference. This interference is time-varying and unpredictable, making it difficult to establish accurate interference models. Third, the flapping wing wake's aerodynamic impact on the tail is complex and time-varying, making it difficult to accurately model the tail aerodynamics. Fourth, model-based position control methods are sensitive to model accuracy; inaccurate model information is difficult to obtain. Nonlinear dynamic inverse control methods suffer performance degradation under unmodeled errors and disturbances. Sliding mode variable structure control requires large switching gains to resist disturbances, causing high-frequency chattering in the control input. The aforementioned problems severely restrict the autonomous flight capability and practical application of bird-inspired flapping-wing aircraft. Summary of the Invention

[0005] To address the problems of existing bird-inspired flapping-wing aircraft, such as difficulty in achieving stable attitude control in multiple flight modes, insufficient resistance to internal and external interference, and trajectory tracking accuracy being affected by model uncertainty, this invention provides a flight control method for bird-inspired flapping-wing aircraft.

[0006] The present invention adopts the following technical solution.

[0007] In terms of attitude control, the unmodeled nonlinear components and unknown disturbances in the attitude dynamics of the bird-like flapping-wing aircraft are summarized as the total system disturbance. Active disturbance rejection controllers (ADRCs) are used for attitude control in the roll and pitch channels, while open-loop control is used in the yaw channel. Each ADRC consists of a tracking differentiator, an extended state observer, and an error feedback controller. The tracking differentiator smooths the desired attitude angle and generates the target signal and its differential estimate, avoiding system overshoot caused by drastic changes in the target signal and protecting actuators such as servos. The extended state observer expands the total system disturbance into a new system state for estimation, unifying the dynamic changes during flapping / gliding transitions, the interference of flapping wing wake on the tail wing, and gust disturbances as a unified total system disturbance for real-time estimation. The error feedback controller constructs intermediate control quantities based on the error between the target signal and the estimated system state, and uses the estimated total system disturbance to pre-compensate for the intermediate control quantities in the control law. The error feedback controller adopts either a linear feedback form or a nonlinear feedback form based on the non-smooth function fal. The coupling effect between the roll and pitch channels is estimated and compensated for by the extended state observer as an external disturbance. The outputs of the two channel controllers are used to calculate the deflection angles of the two independent control surfaces of the tail fin through the control allocation matrix. The two control surfaces deflect in the same direction to generate pitch control torque, and the differential deflection to generate roll control torque.

[0008] During flapping / gliding intermittent flight, the flapping frequency of the flapping wing is controlled by changing the throttle signal, allowing the aircraft to switch between pure flapping mode, pure gliding mode, and flapping / gliding transition mode. The active disturbance rejection controller uses the same controller parameters in all three modes.

[0009] In trajectory tracking control, under the action of the attitude controller, the aircraft position control model is decomposed into a horizontal position and velocity control model and an altitude control model. For the horizontal position and velocity control model, a cascaded outer-loop nonlinear dynamic inverse controller and an inner-loop nonlinear dynamic inverse controller are used as baseline controllers. A dynamic inverse system is constructed using known system information to eliminate nonlinearity, and an L1 adaptive controller is introduced to compensate for unmodeled errors and disturbances. The L1 adaptive controller consists of a state predictor, an adaptive law, and a control law, outputting the desired velocity command and the desired roll angle command. For the altitude control model, a sliding mode variable structure controller is used to generate the desired altitude command. The sliding surface is constructed based on the altitude tracking error and its derivative, and an RBF neural network is used to approximate the unknown nonlinear functions in the model. A sliding mode control law is constructed based on the output of the RBF neural network to reduce high-frequency chattering of the control quantity caused by switching gain. The desired roll angle command, desired velocity command, and desired altitude command serve as the desired inputs to the inner-loop attitude controller. The inner-loop attitude controller employs the roll channel and pitch channel active disturbance rejection controllers from the aforementioned attitude control steps. Attitude control and trajectory tracking control together constitute the complete flight control scheme for the bird-like flapping-wing aircraft.

[0010] Compared with existing technologies, this invention has the following advantages. First, by summarizing unmodeled nonlinearities and unknown disturbances into a total system disturbance and using an extended state observer for real-time estimation and compensation, the attitude controller can adapt to a wide range of changes in system dynamics during flapping / gliding transitions, while ensuring attitude stability in pure flapping, pure gliding, and transition modes. Flight tests show that, compared with the PD controller, the active disturbance rejection controller has smaller tracking errors in flapping mode, and the deviation between the peak pitch angle fluctuation and the desired angle in gliding mode is reduced from 0.2 rad to 0.1 rad, with faster convergence speed. Second, the tracking differentiator smooths the desired attitude signal and generates a differential signal, avoiding system overshoot caused by drastic changes in the target signal and protecting the actuators from damage. Third, the L1 adaptive controller quickly compensates for the unmodeled errors and disturbances of the cascaded nonlinear dynamic inverse controller. Simulation results show that, under superimposed disturbance conditions, the dynamic inverse controller combined with L1 adaptive control can converge the position deviation and yaw angle tracking errors to near zero, while the X-axis position deviation cannot converge when the dynamic inverse controller is used alone. Fourth, the RBF neural network approximates the unknown parts in the altitude control model. Simulation results show that the sliding mode controller combined with the RBF neural network has a faster convergence speed, smaller overshoot, and significantly reduces high-frequency chattering of the control quantity. Fifth, flight tests verify that the proposed method can achieve intermittent flapping / gliding flight, hovering descent flight, and autonomous flight over a distance of two kilometers. The altitude error of the trajectory tracking is about 2 meters, and the horizontal error is about 3 meters, which is within an acceptable range. Attached Figure Description

[0011] Figure 1 This is a schematic diagram of the attitude control system architecture for a bird-inspired flapping-wing aircraft. Figure 1 : A schematic diagram of the attitude control system architecture for a bird-inspired flapping-wing aircraft. It includes a roll channel active disturbance rejection controller, a pitch channel active disturbance rejection controller, a torque distribution unit, tail control surfaces, and the aircraft itself. The coupling effect between the roll and pitch channels is estimated by an extended state observer and compensated in the controller output. The yaw channel is an open-loop control, and the motion law of the flapping wing determines the resultant torque of the wing.

[0012] Figure 2 This is a flowchart of a flight control method for a bird-inspired flapping-wing aircraft. Figure 2 : A flowchart of a flight control method for a bird-inspired flapping-wing aircraft. It includes the following steps: S101 Acquiring flight state data and fusing it to estimate the flight state; S102 Generating desired attitude commands based on the desired position and speed; S103 Performing active disturbance rejection attitude control on the roll and pitch channels; S104 Calculating the tail fin deflection angle and driving the control surfaces. The steps are connected in a top-to-bottom order.

[0013] Figure 3 This is a schematic diagram of the flight control system of a bird-inspired flapping-wing aircraft. Figure 3 A schematic diagram of a flight control system for a bird-inspired flapping-wing aircraft. It includes a position and velocity controller, an attitude controller, a control distributor, actuators, a bird-inspired flapping-wing aircraft, sensors, and a data fusion unit. The output of the position and velocity controller serves as the desired input to the attitude controller. The output of the attitude controller drives the actuators via the control distributor. The sensors measure the flight state, and the data is fused and fed back to the controller, forming a closed-loop control system. Detailed Implementation

[0014] The invention will now be described in further detail with reference to the accompanying drawings. This embodiment uses the developed biomimetic flapping-wing aircraft as the control object. The aircraft has a total weight of 230 grams, a wingspan of 0.6 meters, a flapping wing mass of 11 grams, and a flapping frequency range of 0 to 12 Hz. The main motion of the flapping wing is reciprocating up and down flapping accompanied by passive structural deformation. The flight attitude and trajectory are controlled by two independently deflecting control surfaces on the tail and horizontal stabilizer. The two control surfaces deflect in the same direction to generate pitch control torque, and deflect differentially to generate roll control torque. The yaw channel is stabilized by the vertical stabilizer.

[0015] I. Establishment of Mathematical Model With the aim of control algorithm simulation and design, a mathematical model of a bird-inspired flapping-wing aircraft was established by combining wind tunnel experiments with aerodynamic estimation. The aerodynamic model of the flapping wing was established through wind tunnel experiments: in a low-turbulence wind tunnel, the incoming wind speed, angle of attack, and flapping frequency were used as experimental factors. An experimental state table was designed using a combination of orthogonal and full-scale experiments. In the experiment, the incoming wind speed was set to 6 m / s, 8 m / s, and 10 m / s, the angle of attack was set to 0° to 18° in 3° intervals, and the flapping frequency was set to 4 Hz, 6 Hz, and 8 Hz. The average aerodynamic surrogate model of the flapping wing was established using a third-order RSM model with average lift, average thrust, and average pitching moment as outputs. The inputs of the model were the incoming wind speed, angle of attack, and flapping frequency. The aerodynamic model of the tail fin was established using Tornado aerodynamic estimation software. Both the horizontal and vertical stabilizers are flat airfoils, with the horizontal stabilizer having a half-span of 0.105m and a control surface width of 0.03m, while the vertical stabilizer has a half-span of 0.08m. The aerodynamic influence of the flapping wing wake on the tail fin was treated as a control input disturbance. The forces and moments generated by the tail fin were expressed as the sum of the aerodynamic components under free flow and the nonlinear disturbance of the flapping wing wake. The aircraft was treated as a single rigid body, and a dynamic model was established using the resultant external force equation and the resultant moment equation.

[0016] II. Attitude Control Attitude control is implemented as follows: The attitude dynamics equations are transformed into a second-order system described by Euler angles. The unmodeled nonlinear components, the interference of the flapping wing wake on the tail wing, external gust disturbances, sensor measurement uncertainties, and nonlinearities caused by the actuator's own physical limitations are all uniformly summarized into a total system disturbance f(t). Taking the pitch channel as an example, let y(t) = θ(t), the longitudinal dynamics model is rewritten as y''(t) = f(t) + b0·u1(t), where b0 is the input signal gain, and the value of b0 is adjusted according to the controlled object in engineering applications. The roll channel is handled in the same way, while the yaw channel uses open-loop control.

[0017] The tracking differentiator is constructed using a nonlinear function fhan, in the form v1'(t) = v2(t), v2'(t) = fhan(v1(t) - θd(t), v2(t), r0, h0), where θd(t) is the desired attitude angle, v1(t) is the smoothed estimate of the desired attitude angle, v2(t) is the estimate of the rate of change of the desired attitude angle, r0 is the tracking speed parameter (a larger value of r0 results in faster convergence until the physical limit is reached), and h0 is the filtering parameter, adjusted according to the desired signal smoothness, typically chosen as an integer multiple of the control time step. The target signal output by the tracking differentiator and its differential estimate are used to construct the error feedback control law.

[0018] The extended state observer employs a linear extended state observer. The total system disturbance f(t) is extended to a new system state x3(t). The observer form is z1'=z2+β1e1, z2'=z3+β2e1+U(t), z3'=β3e1, where e1=y-z1 is the output estimation error, and z1, z2, and z3 are the estimated values ​​of the system state and the total system disturbance, respectively. The observer gain is tuned according to the observer bandwidth, letting β1=3w0, β2=3w0², and β3=w0³, where w0 is the observer bandwidth and is positive. When the total system disturbance is continuously differentiable and bounded, the estimation error of the extended state observer is bounded, and the upper bound of the estimation error decreases as the observer bandwidth w0 increases.

[0019] The error feedback controller employs either linear or nonlinear feedback. The linear feedback form is: intermediate control quantity U(t) = k1(v1-z1) + k2(v2-z2) + v3-z3, where v3 is an estimate of the second-order derivative of the target signal. To avoid higher-order derivative calculations, the higher-order derivative values ​​of the target signal are omitted. The control quantity u1(t) = [k1(v1-z1) + k2(v2-z2)-z3] / b0, where k1 and k2 are controller gains, adjusted according to the actual flight performance. The nonlinear feedback is constructed using the nonsmooth function fal, which gives the control law u(t) = [α1·fal(ζ1,λ1,δ0) + α2·fal(ζ2,λ2,δ0) - z3] / b0', where ζ1 = z1 - v1, ζ2 = z2 - v2, α1 and α2 are controller gains, λ1 and λ2 are pre-given constants satisfying 0 < λ1 < 1 < λ2, and δ0 is a pre-given positive number. The fal function outputs the product of the independent variable raised to the power of λ and the sign function when the absolute value of the independent variable is greater than δ0, and outputs the ratio of the independent variable to the power of (1-λ) of δ0 when the absolute value of the independent variable is not greater than δ0. Nonlinear feedback features large gain with small errors and small gain with large errors, achieving good control performance during wide-range fluctuations in the flapping frequency. The coupling effect between the roll and pitch channels is estimated as an external disturbance by their respective extended state observers and compensated in advance in the controller output.

[0020] Control allocation is implemented as follows. Let δl and δr represent the deflection angles of the left and right control surfaces, respectively, with downward deflection being the positive direction. The relationship between the roll channel controller output u1 and the pitch channel controller output u2 and the control surface deflection angle is [u1;u2]=η·Λ·[δl;δr], where Λ is the allocation matrix, Λ=[[-1,-1],[1,-1]], and η is the scaling factor. The control surface deflection angle [δl;δr]=Λ⁻¹[u1;u2] / η is calculated based on the controller output and the inverse solution of the allocation matrix.

[0021] During flapping / gliding intermittent flight, the flapping frequency of the flapping wing is controlled by changing the throttle signal r: r=1 represents full-throttle flapping flight, and r=0 represents gliding flight with the throttle completely closed. During the transition, the throttle signal is gradually decreased or increased according to a preset pattern, allowing the flapping frequency to vary widely between flapping and gliding modes. Throughout the entire process, the active disturbance rejection controller maintains the same controller parameters. The control effect is verified through flapping / gliding intermittent flight tests and hovering descent flight tests: In the two-stage landing route test, the aircraft descends from an altitude of 502 meters to 462 meters in the first stage, with a target descent rate set to 1 m / s. The lateral flight follows circular trajectories with diameters of 55 meters and 35 meters respectively. After reaching the lowest altitude, the aircraft briefly levels off. In the second stage, the throttle signal is reduced to r=0 within 3 seconds, and the target roll angle is changed to 0 rad and the target pitch angle is changed to -0.09 rad, achieving the transition from flapping to gliding and gliding to the ground.

[0022] III. Trajectory Tracking Control Trajectory tracking control is implemented as follows. Under the action of the attitude controller, the kinematic model simplifies to X'=Vcosψ, Y'=Vsinψ, ψ'=g·tanφ / V, V'=(Vr-V) / αV, φ'=(φr-φ) / αφ, h''=-h' / αh'+(hr+h) / αh, where X and Y are the horizontal positions in the inertial frame, V is the flight speed, ψ is the yaw angle, φ is the roll angle, h is the flight altitude, Vr, φr, and hr are the speed, roll angle, and altitude commands, respectively, g is the gravitational acceleration, and αV, αφ, αh', and αh are system model constants related to flight performance and low-level attitude control. In engineering, αV≈1 / 2 and αφ≈1 / 0.55 are taken. The horizontal position and speed control models are described by the first three equations, while the altitude control model is a second-order system.

[0023] For the horizontal position and velocity control model, a cascaded outer loop and inner loop nonlinear dynamic inverse controller are designed. The outer loop uses the position tracking error as the state, with the dynamic inverse control law ζ_out=g1⁻¹(ν1-f1) and the pseudo-inverse control quantity ν1=A1(ζd-ζ), where A1 is the Hurwitz matrix representing the desired response speed of the outer loop. The inner loop uses the vector composed of velocity V and roll angle φ as the state, and the desired input of the inner loop is calculated from the output of the outer loop, with the dynamic inverse control law ξ_out=g2⁻¹(ν2-f2) and ν2=A2(υin-υ), where A2 is the Hurwitz matrix. In the simulation, A1=diag([1,1,1]) and A2=diag([50,50]). Since the nonlinear dynamic inverse control is sensitive to model accuracy and unmodeled errors, an L1 adaptive controller is introduced in the inner loop for compensation. The output of the inner loop controller consists of the output of the dynamic inverse controller and the output of the L1 adaptive controller. The L1 adaptive controller consists of a state predictor, an adaptive law, and a control law. The state predictor constructs a state output with the same dynamic characteristics as the controlled object. The adaptive law uses a projection operator to estimate the adaptive estimation parameters. In the simulation, the adaptive gain Γ = 5 × 10⁻⁶. 5 Q = diag([1,1]); the control law is ξ_L1(t) = -Kd·D(s)(η̂(s) - Kg(s)·υin(s)), where Kd = diag([4,4]) is the feedback gain matrix, D(s) = diag([1 / s,1 / s]) is the regular transformation matrix, and Kd and D(s) together constitute the strictly positive real low-pass filter transfer function matrix C(s) = Kd·D(s) / (I + Kd·D(s)) and satisfy the L1 norm condition. In the simulation verification, a sinusoidal disturbance with an amplitude of 5 rad / s and a frequency of 5 Hz is superimposed on the inner loop roll rate. The results show that the dynamic inverse controller combined with L1 adaptive can converge the position deviation and yaw angle tracking error to near zero, while the X-direction position deviation cannot converge when the dynamic inverse controller is used alone.

[0024] For the height control model, a sliding mode variable structure controller is designed. Let the height tracking errors be h_e = h_d - h and h_e' = h_d' - h', the sliding surface s = ch_e + h_e', and c be the sliding surface parameters. When the model information is known, the sliding mode control law is h_r1 = [ch_e' + h_d'' - f3(h,h',t) + κ·sgn(s)] / g3(t), where κ is the switching gain, satisfying κ ≥ D, and D is the boundary of the disturbance d3(t). When the disturbance boundary is difficult to determine, a larger switching gain is used to ensure system stability. When the model information is unknown, an RBF neural network is used to approximate the unknown nonlinear functions f3 and g3. The RBF network has a three-layer structure. The input layer has an input vector Z=[h,h'], and the activation function of the hidden layer neurons is the Gaussian radial basis function s_j=exp(-||x-c_j||² / (2b_j²)). The network output is ĝ3=V̂ᵀSg(Z) and f̂3=ŴᵀSf(Z). The sliding mode control law is h_r2=[ch_e'+h_d''-f̂3+κ·sgn(s)] / ĝ3(t). The weights of the neural network are updated according to the adaptive laws V̂'=-(1 / λ1)·s·Sg(Z) and Ŵ'=-(1 / λ2)·s·Sf(Z), where λ1 and λ2 are adaptive gains. In the simulation verification, the sliding surface parameter c=5, the target height h_d=8+5sin(t), the disturbance d3(t)=sin(10t), and the model parameters αh=2 and αh'=0.5. The results show that the sliding mode controller combined with the RBF neural network has a faster convergence speed, smaller overshoot, and significantly reduced high-frequency chattering of the control quantity.

[0025] IV. Overall Work Process Combination Figure 2 The overall process of the flight control method is as follows: S101, flight status data is collected by airborne sensors and processed by a multi-sensor data fusion algorithm to obtain flight status estimates; S102, the position and velocity controller generates the desired attitude command according to the desired position, desired velocity, and flight status estimates, following the aforementioned horizontal position and velocity control method and altitude control method; S103, the attitude controller calculates the desired control torque using the active disturbance rejection controllers in the roll and pitch channels based on the desired attitude and the actual attitude in the flight status estimates; S104, the deflection angles of the two tail surfaces are calculated according to the control allocation matrix, and the servo drives the control surfaces to deflect, generating the desired control torque to complete flight control. The above process is executed cyclically at a certain period, forming a complete closed-loop control. The reliability and stability of the flight control system were verified through a two-kilometer long-distance automatic flight test. During the flight, the flight path altitude was set at 80 meters, and the trajectory tracking effect was good.

Claims

1. A flight control method for a bird-inspired flapping-wing aircraft, characterized in that, include: The attitude control step involves summarizing the unmodeled nonlinear components and unknown disturbances in the attitude dynamics of the bird-like flapping-wing aircraft into a total system disturbance. Active disturbance rejection controllers (ADRCs) are used for attitude control in the roll and pitch channels, while open-loop control is employed in the yaw channel. Each ADRC includes a tracking differentiator, an extended state observer, and an error feedback controller. The tracking differentiator smooths the desired attitude angle and generates a target signal and its differential estimate. The extended state observer estimates the system state and the total system disturbance. The error feedback controller constructs an intermediate control quantity based on the error between the target signal and the estimated system state, and compensates for this intermediate control quantity using the estimated total system disturbance. The control allocation step involves calculating the deflection angles of the two independent tail surfaces using a control allocation matrix based on the outputs of the ADRCs for the roll and pitch channels. Co-directional deflection of the two tail surfaces generates a pitch control torque, while differential deflection generates a roll control torque.

2. The flight control method for the bird-like flapping-wing aircraft according to claim 1, characterized in that, The tracking differentiator is constructed using a nonlinear function fhan, in the form v1'(t)=v2(t), v2'(t)=fhan(v1(t)-θd(t),v2(t),r0,h0), where v1(t) is the estimated value of the target signal, v2(t) is the estimated value of the derivative of the target signal, θd(t) is the desired attitude angle, r0 is the tracking speed parameter, and h0 is the filtering parameter.

3. The flight control method for the bird-like flapping-wing aircraft according to claim 1, characterized in that, The extended state observer is a linear extended state observer, which estimates the total system disturbance by expanding it into a new system state. The gain of the linear extended state observer is tuned according to the observer bandwidth. When the total system disturbance is bounded, the estimation error of the linear extended state observer is bounded, and the upper bound of the estimation error decreases as the observer bandwidth increases.

4. The flight control method for the bird-like flapping-wing aircraft according to claim 1, characterized in that, The error feedback controller is a linear feedback controller. The intermediate control quantity U(t) = k1(v1-z1) + k2(v2-z2) + v3-z3, where v1 and v2 are the target signal and the first-order differential estimate of the target signal, v3 is the second-order differential estimate of the target signal, z1, z2, and z3 are the system state estimate and the total system disturbance estimate output by the extended state observer, k1 and k2 are the controller gain, and after omitting the higher-order differential value of the target signal, the control quantity u1(t) = [k1(v1-z1) + k2(v2-z2) - z3] / b0, where b0 is the input signal gain.

5. The flight control method for the bird-like flapping-wing aircraft according to claim 1, characterized in that, The error feedback controller is a nonlinear feedback controller constructed using a non-smooth function fal(ζ,λ,δ). The control law is u(t)=[α1·fal(ζ1,λ1,δ0)+α2·fal(ζ2,λ2,δ0)-z3] / b0', where ζ1=z1-v1, ζ2=z2-v2, α1 and α2 are controller gains, λ1 and λ2 are pre-given constants satisfying 0<λ1<1<λ2, and δ0 is a pre-given positive number. The non-smooth function fal outputs the product of the independent variable raised to the power of λ and the sign function when the absolute value of the independent variable is greater than δ0, and outputs the ratio of the independent variable to the power of (1-λ) of δ0 when the absolute value of the independent variable is not greater than δ0.

6. The flight control method for the bird-like flapping-wing aircraft according to claim 1, characterized in that, It also includes flapping / gliding mode control steps: by changing the throttle signal to control the flapping frequency of the flapping wing, the aircraft can switch between pure flapping mode, pure gliding mode and flapping / gliding transition mode. The active disturbance rejection controller uses the same controller parameters in the pure flapping mode, the pure gliding mode and the flapping / gliding transition mode.

7. A trajectory tracking control method for a bird-inspired flapping-wing aircraft, characterized in that, include: The model decomposition step decomposes the aircraft position control model into a horizontal position and velocity control model and an altitude control model. The horizontal position and velocity control model is described by flight speed, yaw angle, and roll angle, and the altitude control model is a second-order system. The horizontal position and velocity control step uses a cascaded outer-loop nonlinear dynamic inverse controller and an inner-loop nonlinear dynamic inverse controller as baseline controllers for the horizontal position and velocity control model, and introduces an L1 adaptive controller to compensate for unmodeled errors and disturbances, outputting desired velocity and desired roll angle commands. The altitude control step uses a sliding mode variable structure controller to generate desired altitude commands for the altitude control model, and uses an RBF neural network to approximate the unknown nonlinear functions in the altitude control model, constructing a sliding mode control law based on the output of the RBF neural network. The command output step uses the desired roll angle command, the desired velocity command, and the desired altitude command as desired inputs to the inner-loop attitude controller, which performs attitude control using the attitude control steps of the flight control method described in claim 1.

8. The trajectory tracking control method for a bird-like flapping-wing aircraft according to claim 7, characterized in that, The L1 adaptive controller includes a state predictor, an adaptive law, and a control law. The state predictor constructs a state output with consistent dynamic characteristics with the controlled object. The adaptive law uses a projection operator to estimate the adaptive estimation parameters. The control law generates a control signal through a low-pass filter transfer function matrix C(s) = Kd·D(s) / (I+Kd·D(s)), where Kd is the feedback gain matrix, D(s) is the regular transformation matrix, and I is the identity matrix. C(s) is strictly positive real, and Kd and D(s) satisfy the L1 norm condition.

9. The trajectory tracking control method for a bird-like flapping-wing aircraft according to claim 7, characterized in that, The sliding surface of the sliding mode variable structure controller is s = ch_e + h_e', where h_e is the height tracking error and c is the sliding surface parameter. The sliding mode control law is h_r2 = [ch_e' + h_d'' - f̂3(h,h',t) + κ·sgn(s)] / ĝ3(t), where f̂3 is the approximation output of the RBF neural network for the unknown function f3 of the height control model, ĝ3 is the approximation output of the RBF neural network for the unknown function g3 of the height control model, κ is the switching gain, and sgn is the sign function.

10. The trajectory tracking control method for a bird-like flapping-wing aircraft according to claim 7, characterized in that, The weights of the RBF neural network are updated according to the adaptive laws V'=-(1 / λ1)·s·Sg(Z) and Ŵ'=-(1 / λ2)·s·Sf(Z), where s is the sliding surface function value, Sg(Z) and Sf(Z) are the radial basis function outputs of the approximation functions g3 and f3, respectively, Z is the input vector of the RBF neural network, and λ1 and λ2 are the adaptive gains.

Images