Stable flight control method for heavy-payload UAVs with compound wing configuration

Abstract

This invention discloses a stable flight control method for heavy-load unmanned aerial vehicles (UAVs) with compound wing configurations, relating to the field of UAV flight control technology. This invention utilizes a third-order nonlinear ESO to unify all internal and external disturbances, such as rotor-fixed-wing coupling disturbances, wind field disturbances, and model uncertainties caused by load changes, into a total disturbance for real-time estimation and feedforward compensation. This eliminates the need for precise modeling of complex coupled aerodynamic models and load change models, significantly reducing modeling difficulty. Simultaneously, it achieves full compensation for both matched and mismatched disturbances, significantly improving the anti-disturbance capability and robustness of heavy-load compound wing UAVs under strong disturbance scenarios, and solving the problems of attitude instability under sudden load changes, coupling disturbances, and complex wind fields.

Description

Technical Field

[0001] This invention relates to the field of unmanned aerial vehicle (UAV) flight control technology, specifically to a stable flight control method for heavy-load UAVs with a compound wing configuration. Background Technology

[0002] The compound wing layout UAV combines the vertical take-off and landing and hovering capabilities of multi-rotor UAVs with the long range, high speed and large payload of fixed-wing UAVs, and has been widely used in heavy-load application scenarios such as logistics transportation, emergency rescue and material delivery.

[0003] However, in heavy-load scenarios, the flight control of compound-wing UAVs faces three major technical challenges: First, compound-wing UAVs have both rotor and fixed-wing actuator systems. During the transition from vertical takeoff and landing to level flight, there is a strong coupling disturbance between the rotor wash and the fixed-wing aerodynamic field. Under heavy-load conditions, this coupling effect is further amplified, which can easily lead to attitude oscillations. Second, during material delivery and hoisting transportation, the payload weight and center of gravity of heavy-load UAVs can change abruptly. At the same time, wind field disturbances such as gusts and turbulence in complex flight environments can bring strong uncertainties to the system. Traditional control methods are insufficient to suppress large disturbances, which can easily lead to trajectory deviation, attitude instability, or even the risk of crash. Third, heavy-load compound-wing UAVs are redundant actuator systems. The control characteristics of multiple rotors and multiple control surfaces vary greatly. Traditional linear pseudo-inverse allocation methods cannot handle the nonlinear characteristics and saturation constraints of the actuators, making it difficult to achieve coordinated optimal control of multiple actuators. Under heavy loads, the control efficiency is low and the actuators are prone to exceeding limits.

[0004] In existing technologies, flight control for compound-wing UAVs mostly employs PID cascade control or linear model predictive control (LMPC). PID control relies on manual parameter tuning and has poor robustness to highly nonlinear systems and large disturbances. LMPC can only achieve effective control near the system equilibrium point, and its control accuracy drops significantly under heavy loads and large state deviations or disturbances. Some existing technologies introduce extended state observers (ESOs) for disturbance estimation, but they do not combine predictive control to compensate for trajectory deviations in advance, nor do they perform specific optimizations for heavy load changes and coupled disturbances. Other technologies have improved the control allocation for compound wings, but they mostly use linear allocation strategies, which cannot adapt to the nonlinear characteristics of actuators under heavy loads and the priority requirements throughout the flight phase, resulting in poor multi-actuator cooperative control performance.

[0005] In summary, existing technologies cannot simultaneously solve the control challenges posed by strong coupling, large disturbances, and strong nonlinearity in heavy-load, repetitive-wing UAVs. Therefore, it is necessary to develop a stable flight control method that is adaptable to all flight phases, has strong anti-disturbance capabilities, high trajectory tracking accuracy, and excellent multi-actuator coordination. Summary of the Invention

[0006] The purpose of this invention is to address the aforementioned deficiencies in existing technologies by providing a stable flight control method for heavy-load unmanned aerial vehicles (UAVs) with a compound wing configuration. This method utilizes a three-stage cascaded control architecture—"real-time ESO total disturbance estimation, NMPC trajectory prediction control with feedforward compensation, and nonlinear OCA optimal allocation"—to solve the problems of poor flight stability, low trajectory tracking accuracy, and insufficient multi-actuator coordination in strongly coupled, large-disturbance, and highly nonlinear scenarios for heavy-load compound wing UAVs. This achieves stable, accurate, and reliable control throughout the entire flight phase.

[0007] To achieve the above objectives, the present invention provides the following technical solution: a stable flight control method for a heavy-load unmanned aerial vehicle (UAV) with a compound wing configuration, comprising the following:

[0008] S1: Multi-source data acquisition, real-time acquisition of multi-source data during the flight of a heavy-load repetitive wing-shaped UAV, and output of the multi-source data after preprocessing;

[0009] S2: Real-time estimation and compensation of total disturbance. Based on preprocessed multi-source data, the total disturbance caused by rotor-fixed wing coupling disturbance, wind field disturbance and load change during UAV flight is estimated in real time through the preset extended state observer (ESO), and the corresponding disturbance compensation amount is output.

[0010] S3: Trajectory prediction and compensation command generation. The disturbance compensation amount and the preprocessed multi-source data are input into the preset nonlinear model predictive control model NMPC. The NMPC predicts the flight trajectory deviation in the preset prediction time domain and outputs the corresponding virtual compensation control command through rolling optimization solution.

[0011] S4: Multi-actuator collaborative control allocation. Based on the virtual compensation control command, the virtual compensation control command is optimally allocated to each actuator of the UAV through the preset nonlinear optimal control allocation strategy OCA, generating the actual control command for each actuator, and realizing multi-actuator collaborative stable control.

[0012] Furthermore, in S1, the multi-source data includes UAV body state data, actuator state data, environmental state data, and payload state data; wherein, the UAV body state data includes the UAV's spatial position, flight speed, attitude angle, attitude angular velocity, and linear acceleration data; the actuator state data includes the real-time rotational speed of the rotor motor, the actual deflection angle and deflection rate of the control surface data; the environmental state data includes real-time wind speed, wind direction, atmospheric density, and atmospheric pressure data; and the payload state data includes the real-time weight of the payload, center of gravity offset, and payload swing amplitude and frequency data.

[0013] Furthermore, in S1, preprocessing includes timestamp synchronization of multi-source data, outlier removal, and filtering and fusion processing. Among these, the 3σ criterion is used for outlier removal, and extended Kalman filtering is used to perform multi-sensor fusion processing on the UAV's body state data. All data is synchronized to the same control time base, with a time synchronization accuracy of not less than 1ms.

[0014] Furthermore, S2 specifically includes:

[0015] S2.1: Establish a six-degree-of-freedom nonlinear dynamic model for a heavy-load repeating wing UAV, and unify the model uncertainties and unmodeled dynamics caused by rotor-fixed-wing aerodynamic coupling disturbances, wind field disturbances, and load changes into the total disturbance state variables of the system, and construct the system state equations containing the extended states;

[0016] S2.2: Based on the system state equation containing the extended state, a third-order nonlinear extended state observer (ESO) is designed. The UAV body state data in the preprocessed multi-source data is used as the input of the ESO to observe the estimated values ​​of each state variable and the total disturbance state variable of the system in real time.

[0017] The ESO expression for the third-order nonlinear extended state observer is:

[0018]

[0019] in, This is the system output, i.e., the preprocessed state variables of the UAV itself; These are the observed values ​​of the system state variables, the derivatives of the state variables, and the total disturbance, respectively. For the gain parameter of ESO; It is a nonlinear power function, used to improve the convergence speed and anti-interference capability of the observer; For system control gain;

[0020] S2.3: Based on the estimated value of the total disturbance state variable, generate the corresponding feedforward disturbance compensation amount, and superimpose the disturbance compensation amount onto the input of the NMPC prediction model to achieve feedforward cancellation of the total disturbance.

[0021] Furthermore, in S21, the six-degree-of-freedom nonlinear dynamic model incorporates the real-time changes in the UAV mass and moment of inertia caused by load variations, as well as the coupling effect parameters of the rotor wash on the aerodynamic characteristics of the fixed wing; the total disturbance is the sum of matched and unmatched disturbances, eliminating the need to model each sub-disturbance separately, thus significantly reducing the modeling difficulty.

[0022] Furthermore, S3 specifically includes:

[0023] S3.1: Construct an NMPC prediction model with disturbance feedforward compensation. The prediction model is a nonlinear dynamic model of a large-load repeating winged UAV that incorporates disturbance compensation. The current state of the UAV in the preprocessed multi-source data is used as the initial value for prediction, and the preset reference flight trajectory is used as the tracking target.

[0024] S3.2: Construct the rolling optimization objective function of NMPC, wherein the objective function takes the minimum trajectory tracking error, the minimum rate of change of control quantity, and the minimum actuator constraint violation as optimization objectives; at the same time, system state constraints, control quantity constraints, and actuator physical constraints are set.

[0025] The objective function expression for NMPC's rolling optimization is:

[0026]

[0027] in, To predict the time domain, To control the time domain; for Time prediction Step future state quantity, For reference trajectory state variables; To control the rate of change of quantity; Here is the state error weight matrix. The weight matrix represents the changes in the control quantity. For relaxation factor weights, To constrain violations of the relaxation factor, used to handle soft constraints;

[0028] S3.3: Based on the current flight phase of the UAV, adaptively adjust the prediction time domain and control time domain, and perform rolling optimization to solve the optimal control problem in the finite time domain within each control cycle, and output the corresponding virtual compensation control commands; the virtual compensation control commands include the UAV's total lift command, roll moment command, pitch moment command, and yaw moment command.

[0029] Furthermore, the flight phase includes a vertical takeoff and landing phase, a transition phase, and a fixed-wing cruise phase; wherein, the vertical takeoff and landing phase and the transition phase adopt short prediction time domain and short control time domain to improve the dynamic response speed of the system; the fixed-wing cruise phase adopts long prediction time domain and long control time domain to improve flight stability and trajectory tracking accuracy.

[0030] Furthermore, in S4, the UAV's actuators include a vertical takeoff and landing rotor, a tilt rotor, ailerons, elevators, rudders, and flaps, forming a redundant actuator system; the nonlinear optimal control allocation strategy (OCA) aims to minimize actuator control efficiency loss and control quantity variation, while incorporating the following constraints:

[0031] Actuator physical hard constraints: including the maximum / minimum speed constraints and speed change rate constraints of the rotor motor, and the maximum / minimum deflection angle constraints and deflection rate constraints of the control surfaces;

[0032] Nonlinear constraints on aerodynamic efficiency: Efficiency constraints are constructed based on the nonlinear model of actuator aerodynamic efficiency under different flight conditions;

[0033] Flight phase priority constraints: During the vertical takeoff and landing phase, the rotor actuators have the highest priority, and the control surface actuators do not participate in control; during the transition phase, the rotor and control surface actuators work together, and the priority is adaptively adjusted; during the fixed-wing cruise phase, the control surface actuators have the highest priority, and the rotor only participates in auxiliary attitude correction.

[0034] Furthermore, in S4, the nonlinear optimization problem of OCA is solved by sequential quadratic programming, and the virtual compensation control command is converted into the actual control command of each actuator, including the target speed of each rotor motor and the target deflection angle of each control surface, and sent to the flight control execution unit in real time to realize the coordinated stable control of multiple actuators.

[0035] Furthermore, S1-S4 are executed cyclically within each control cycle, with a control cycle of 5-20ms, and the entire process is adapted to all flight phases of the UAV, including vertical take-off and landing, transition, cruise, go-around, and emergency landing.

[0036] Compared with the prior art, the beneficial effects of the present invention are:

[0037] 1. This invention uses a third-order nonlinear ESO to unify all internal and external disturbances, such as rotor-fixed wing coupling disturbances, wind field disturbances, and model uncertainties caused by load changes, into a total disturbance for real-time estimation and feedforward compensation. This eliminates the need for precise modeling of complex coupled aerodynamic models and load change models, significantly reducing modeling difficulty. At the same time, it achieves full compensation for matched and mismatched disturbances, significantly improving the anti-disturbance capability and robustness of heavy-load repetitive wing UAVs in strong disturbance scenarios, and solving the problems of attitude instability under load abrupt changes, coupling disturbances, and complex wind fields.

[0038] 2. This invention uses NMPC with disturbance feedforward compensation as the upper-level control for trajectory and attitude, which can fully adapt to the strong nonlinear characteristics of the compound airfoil system. At the same time, by predicting trajectory deviation in the time domain and optimizing rolling, it realizes advance compensation of control commands. Compared with traditional PID and linear MPC, it significantly improves trajectory tracking accuracy and dynamic response performance under heavy load conditions. In particular, it solves the problem of smooth control in the transition phase and avoids attitude oscillation and stall risk in the transition phase.

[0039] 3. This invention adopts a nonlinear OCA strategy as the bottom actuator allocation stage, which fully considers the nonlinear aerodynamic characteristics, physical hard constraints and priority requirements of the actuators throughout the flight phase, and optimally allocates the upper-level virtual control commands to each redundant actuator. This solves the problem that the traditional linear allocation method cannot handle nonlinearity and is prone to actuator saturation, and realizes the cooperative optimal control of multiple actuators, which greatly improves the control efficiency and system reliability in heavy load scenarios.

[0040] 4. The three-level control architecture of this invention is fully adaptable to all flight stages of compound wing UAVs. It is highly adaptable to heavy loads, large load changes, and complex flight environments. It has a short control cycle, high real-time performance, does not rely on high-precision aerodynamic modeling, and is easy to implement in engineering. It has extremely high application value in scenarios such as heavy-load logistics transportation and emergency material delivery. Attached Figure Description

[0041] To more clearly illustrate the technical solutions of the embodiments of the present invention, the accompanying drawings used in the description of the embodiments will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0042] Figure 1 This is a schematic diagram of the process of the present invention. Detailed Implementation

[0043] The technical solutions of the present invention will be clearly and completely described below with reference to the accompanying drawings of the embodiments of the present invention. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments.

[0044] This embodiment takes a heavy-duty vertical take-off and landing compound wing UAV as the implementation object. The UAV adopts a compound wing layout of "4 vertical take-off rotors and 2 tilt rotors". The fixed wing section is equipped with ailerons, elevators, rudders and flaps. The maximum take-off weight is 200kg and the rated payload is 50kg. It has the capability of vertical take-off and landing, transition and fixed-wing cruise in the entire flight phase. The flight control system control cycle is 10ms.

[0045] See Figure 1 A stable flight control method for heavy-payload UAVs with a compound wing configuration, comprising the following steps:

[0046] S1: Multi-source data acquisition and preprocessing

[0047] The drone uses a multi-sensor system to collect multi-source data in real time during flight, specifically including:

[0048] UAV body status data: Spatial position and ground speed data are collected through differential GPS module, attitude angle (roll angle, pitch angle, yaw angle), attitude angular velocity and linear acceleration data are collected through IMU inertial measurement unit, and altitude and airspeed data are collected through barometric altimeter and airspeed tube, with a sampling frequency of 100Hz.

[0049] Actuator status data: Real-time speed data of 6 rotor motors are collected through ESC, and actual deflection angle and deflection rate data of each control surface are collected through servo feedback, with a sampling frequency of 200Hz;

[0050] Environmental status data: Real-time wind speed, wind direction, atmospheric density, and atmospheric pressure data are collected through wind vanes and atmospheric sensors at a sampling frequency of 50Hz;

[0051] Load status data: The real-time weight of the load is collected by the weighing sensor on the machine body, the center of gravity offset is collected by the center of gravity monitoring module, and the load swing amplitude and frequency data are collected by the hoisting attitude sensor, with a sampling frequency of 50Hz.

[0052] The collected multi-source data is preprocessed as follows: First, all data is synchronized to the 10ms control time base of the flight control system using hard synchronization, with a time synchronization accuracy of no less than 1ms. Then, the 3σ criterion is used to remove outliers from each data to avoid abnormal data interfering with the control effect. Finally, extended Kalman filtering is used to perform multi-sensor fusion of the body state data from GPS, IMU, altimeter, and pitot tube, outputting high-frequency, high-precision UAV state variables to provide reliable input for subsequent control links.

[0053] S2: Real-time estimation and compensation of total disturbance based on ESO

[0054] First, a six-degree-of-freedom nonlinear dynamic model of a heavy-load, repetitive-wing UAV is established. The model incorporates parameters related to the real-time changes in UAV mass and moment of inertia caused by load variations, as well as parameters related to the coupling effect of rotor wash on the aerodynamic characteristics of the fixed wing. The rigid body dynamic equations are expressed as follows:

[0055]

[0056] in, This is the system state vector, which includes the UAV's position, velocity, attitude angle, and angular velocity. This is the system control input vector; The nominal nonlinear dynamic function of the system; The total system disturbance includes rotor-fixed-wing coupling disturbances, wind field disturbances, model uncertainties caused by load variations, and unmodeled dynamics, encompassing both matched and mismatched disturbances. The total disturbance... Expanding into another new state variable of the system A third-order system state equation containing an extended state is constructed, and a third-order nonlinear extended state observer (ESO) is designed based on this state equation. Its expression is:

[0057]

[0058] in, This is the system output, i.e., the preprocessed state variables of the UAV itself; These are the observed values ​​of the system state variables, the derivatives of the state variables, and the total disturbance, respectively. For the gain parameter of ESO; It is a nonlinear power function, used to improve the convergence speed and anti-interference capability of the observer; This is the system control gain.

[0059] The preprocessed UAV body state data is used as input to the ESO. Within each 10ms control cycle, the estimated values ​​of each state variable and the total disturbance of the system are observed in real time. The corresponding feedforward disturbance compensation is generated based on the total disturbance estimate. The disturbance compensation is then superimposed on the input of the subsequent NMPC prediction model to achieve feedforward cancellation of the total disturbance and eliminate the impact of various disturbances on the system control effect.

[0060] S3: Trajectory Prediction and Compensation Command Generation Based on NMPC

[0061] A predictive model for NMPC with perturbation feedforward compensation is constructed. This predictive model is a nonlinear dynamic model of a composite wing that incorporates the aforementioned perturbation compensation, and its expression is as follows:

[0062]

[0063] The preprocessed current state of the UAV is used as the initial value for prediction, and the reference flight trajectory and reference attitude preset by the ground station are used as the tracking target.

[0064] The objective function for NMPC's rolling optimization is expressed as follows:

[0065]

[0066] in, To predict the time domain, To control the time domain; for Time prediction Step future state quantity, For reference trajectory state variables; To control the rate of change of quantity; Here is the state error weight matrix. The weight matrix represents the changes in the control quantity. For relaxation factor weights, It is used to handle soft constraints by constraining violations of the relaxation factor.

[0067] At the same time, system constraints are set, including: state constraints corresponding to the UAV flight envelope, control quantity amplitude constraints, and actuator physical hard constraints, to prevent control commands from exceeding the actuator limits.

[0068] Based on the current flight phase of the UAV, adaptive adjustments are made to the prediction time domain and control time domain:

[0069] Vertical Takeoff and Landing Phase: Prediction in the Time Domain Control time domain This enhances the system's ability to respond quickly to hovering disturbances in the short time domain.

[0070] Transition Phase: Prediction Time Domain Control time domain It balances response speed and transition smoothness;

[0071] Fixed-wing cruise phase: Prediction time domain Control time domain This improves the accuracy of cruise trajectory tracking and flight stability over a longer time domain.

[0072] In each During the control cycle, the ACADO optimization solver is used to solve the optimal control problem in the finite time domain, and the corresponding virtual compensation control commands are output, including the UAV's total lift command, roll torque command, pitch torque command, and yaw torque command, so as to realize the early prediction and compensation of trajectory deviation.

[0073] S4: OCA-based multi-actuator cooperative control allocation

[0074] The UAV actuator in this embodiment includes four vertical take-off and landing rotors, two tilt rotors, ailerons, elevators, rudders, and flaps, totaling 10 control channels, forming a redundant actuator system. A nonlinear optimal control allocation strategy (OCA) is constructed, with the optimization objectives being to minimize actuator control efficiency loss and control variable variation. Its objective function expression is:

[0075]

[0076] in, These are virtual compensation control commands (total lift, triaxial torque) output by the upper-level NMPC. These are the actual control quantities for each actuator; The nonlinear control efficiency matrix of the actuator, and the flight state Relatedly, the nonlinear characteristics of actuator aerodynamic efficiency are incorporated; This is the actuator control quantity from the previous cycle; The weight matrix is ​​used to track virtual control commands; This is the weight matrix for controlling the changes in the variable.

[0077] Simultaneously include three types of constraints:

[0078] Actuator physical hard constraints: rotor motor speed range is 0-6000rpm, speed change rate ≤1000rpm / 0.1s; control surface deflection angle range is ±30°, deflection rate ≤60° / s;

[0079] Nonlinear constraints on aerodynamic efficiency: Based on the nonlinear models of actuator aerodynamic efficiency at different airspeeds and angles of attack calibrated by wind tunnel and flight tests, efficiency constraints are constructed to prevent the actuator from operating in the inefficient range.

[0080] Flight phase priority constraints:

[0081] During vertical takeoff and landing: the rotor actuators have the highest weight, while the control surface actuators have their weight reset to 0 and do not participate in control.

[0082] Transition phase: The weights of the rotor and control surface actuators are adaptively adjusted with the tilt angle. During the tilt angle from 0 to 90°, the rotor weight decreases linearly and the control surface weight increases linearly, achieving smooth coordination.

[0083] During the fixed-wing cruise phase: the control surface actuators have the highest weight, while the rotor actuators have a low weight and only participate in minor attitude corrections to reduce energy consumption.

[0084] The above nonlinear optimization problem is solved in each control cycle using the Sequential Quadratic Programming (SQP) method. The virtual compensation control command is converted into the actual control command of each actuator, including the target speed of the 6 rotor motors and the target deflection angle of the 4 control surfaces. The command is sent to the flight control execution unit in real time to drive the actuators to work together, and finally achieve stable flight control of the heavy-load repetitive wing-shaped UAV throughout the entire flight phase.

[0085] The core innovation of this invention is a three-level cascaded control architecture consisting of "ESO total disturbance feedforward compensation, nonlinear model predictive control (NMPC) with disturbance compensation, and nonlinear optimal control allocation (OCA)". Three sets of comparative examples (hereinafter referred to as comparative examples) are set up to provide a precise comparison with the embodiments of this invention, targeting three mainstream schemes in the prior art.

[0086] General test basic conditions

[0087] All examples / comparative examples used the same test platform: a 4-vertical-rotor + 2-tilt-rotor compound wing UAV with a maximum takeoff weight of 200 kg, a rated payload of 50 kg, a flight control system control cycle of 10 ms, and completely identical actuator hardware parameters, aerodynamic models, and sensor configurations. All tests were conducted under the same environmental model, disturbance input, and reference trajectory to ensure the uniqueness and credibility of the comparison results.

[0088] Comparative Example 1: Traditional PID cascade control + linear pseudo-inverse distribution

[0089] Control scheme: The most common four-loop PID cascade control architecture in the industry for compound wing UAVs is the "position loop-velocity loop-attitude angle loop-angular velocity loop". The PID parameters of each loop are tuned to the optimal state through engineering experience. The bottom layer uses the linear pseudo-inverse method for control allocation, without distinguishing the priority of actuators in the flight phase, and only satisfies the physical hard constraints of the actuators.

[0090] Key differences: No ESO disturbance observation and compensation, no model-predicted trajectory prediction, no nonlinear optimal allocation, relying entirely on linear feedback control, corresponding to the baseline scheme of existing technologies.

[0091] Comparative Example 2: Linear Model Predictive Control (LMPC) + ESO + Linear Pseudo-Inverse Assignment

[0092] Control scheme: The total disturbance estimation and feedforward compensation are performed using a third-order ESO with the same parameters as the present invention; the upper layer adopts linear model predictive control (LMPC), and the model is linearized at the two equilibrium points of hovering and cruising. The prediction time domain and control time domain parameters are completely consistent with the present invention; the lower layer adopts the same linear pseudo-inverse method as Comparative Example 1 for control allocation.

[0093] Key differences: The linear MPC is used instead of the nonlinear NMPC of this invention, and the linear pseudo-inverse allocation is used instead of the nonlinear OCA of this invention. Only the ESO disturbance compensation module is retained, corresponding to the "ESO + linear MPC" combination scheme in the prior art.

[0094] Comparative Example 3: Nonlinear Model Predictive Control (NMPC) + Nonlinear OCA

[0095] Control scheme: It adopts the same nonlinear NMPC architecture, prediction time-domain / control time-domain parameters, and constraints as this invention; the underlying layer adopts the same nonlinear OCA strategy, optimization objective, and constraints as this invention; there is no ESO total disturbance observation and feedforward compensation module, and the system disturbance is processed only by the feedback correction of NMPC.

[0096] The key difference is that the ESO total disturbance feedforward compensation core module of this invention is removed, which corresponds to the "NMPC+OCA" non-feedforward disturbance compensation scheme in the prior art.

[0097] Simulation test proves

[0098] This experiment is divided into two parts: a semi-physical simulation test and a flight verification test. The simulation test covers extreme conditions throughout the entire flight phase and quantitatively compares various control performance indicators. The flight verification test verifies the engineering applicability and repeatability of the simulation results, fully meeting the requirements of "full disclosure" and "verifiable technical effects" for patent examination.

[0099] Experimental platform construction

[0100] (1) Semi-physical simulation platform

[0101] A hardware-in-the-loop (HIL) semi-physical simulation platform was built using MATLAB / Simulink, Speedgoat real-time target machine, and FlightGear visualization system.

[0102] Dynamics Model: A six-degree-of-freedom nonlinear dynamics model was built, incorporating rotor-fixed-wing aerodynamic coupling, real-time changes in load mass / moment of inertia, and wind field disturbances. The model parameters were determined through wind tunnel testing and flight calibration, and the model's characteristics match those of real UAVs by ≥95%.

[0103] Hardware environment: The flight controller main control board, sensors, and actuator ESCs / servos all use real flight hardware to achieve hardware-in-the-loop simulation and eliminate the idealization errors of pure digital simulation;

[0104] Disturbance Model: Built-in Dryden turbulence model, discrete gust model, and load mutation model, which can accurately simulate various disturbance inputs under heavy load scenarios.

[0105] (2) Flight verification platform

[0106] It adopts a 200kg-class heavy-load, repetitive, double-wing UAV that is completely consistent with the simulation model. It is equipped with differential GPS, high-precision IMU, wind vane sensor, and load weighing sensor, and is equipped with a ground station data recording system with a sampling frequency of 100Hz to completely record the status data, control commands and actuator feedback data of the entire flight process.

[0107] 2.2 Test Condition Design

[0108] To address the core application scenarios and control challenges of high-payload, repetitive, wing-shaped unmanned aerial vehicles (UAVs), four typical test conditions were set up, covering the entire flight phase. All conditions included rated load and disturbance inputs to simulate real-world operational scenarios.

[0109] Operating condition number Operating condition name Core test content Disturbance settings Operating Condition 1 Hovering and Disturbance Resistant Operation (Vertical Take-off and Landing Phase) Fixed-point hovering control under rated load verifies the disturbance rejection capability and attitude stability during vertical takeoff and landing. 1. A lateral gust of wind at 1.8 m / s, lasting for 2 seconds; 2. Sudden load change: from 50 kg to 35 kg (30% of rated load delivered). Operating Condition 2 Transitional operating conditions The entire transition process from hovering to level flight was controlled, verifying the stability and oscillation-free control capability during the strongly coupled phase. The tilting mechanism rotates uniformly from 0° to 90° within 90 seconds, with the air velocity increasing linearly from 0 to 25 m / s, and a longitudinal turbulence of 4 m / s superimposed throughout the process. Operating Condition 3 Cruise trajectory tracking conditions Long-endurance cruise control for fixed-wing aircraft verifies trajectory tracking accuracy and wind resistance over long distances. Cruise at a constant speed of 25 m / s, tracking a circular reference route with a radius of 5 km, with continuous turbulence of 10 m / s throughout, and random wind speed fluctuations of ±3 m / s. Operating Condition 4 Extreme load change conditions Load change control under heavy loads: verifying robustness and instability prevention under extreme conditions. Takeoff weight 200kg, payload 60kg, 20kg payload jettisoned instantaneously during flight (payload drop of 33%), simultaneously superimposed with a 6m / s side gust.

[0110] Quantitative evaluation index system

[0111] To address the core control requirements of high-payload, repeatable-wing unmanned aerial vehicles (UAVs), four categories of quantitative evaluation indicators were established. All indicators were calculated using statistical methods to ensure data objectivity.

[0112] Tracking performance: Root mean square error of horizontal position (RMSE), altitude RMSE, airspeed RMSE;

[0113] Attitude control performance: roll / pitch RMSE, maximum attitude angle overshoot, attitude stabilization time;

[0114] Disturbance immunity performance: position recovery time after disturbance, disturbance rejection ratio (ratio of peak error to steady-state error after disturbance);

[0115] Actuator performance: The number of times the actuator is saturated under all operating conditions and the peak value of the rate of change of the control quantity are used to evaluate the actuator's coordination and service life.

[0116] Experimental Results and Comparative Analysis

[0117] (1) Quantitative results of semi-physical simulation experiments

[0118] The table below shows the quantitative comparison results of the core indicators of the embodiments of the present invention and the three comparative examples under four working conditions. All data are the average values ​​of 10 repeated experiments:

[0119] Comparison of core indicators for working condition 1 (hovering and anti-disturbance condition)

[0120] Evaluation indicators Embodiments of the present invention Comparative Example 1 Comparative Example 2 Comparative Example 3 Horizontal position RMSE (m) 0.32 2.18 1.07 1.26 Height RMSE (m) 0.18 1.25 0.62 0.74 Pitch angle RMSE (°) 0.21 1.86 0.89 0.97 Attitude stabilization time after perturbation (s) 0.42 3.27 1.58 1.83 Actuator saturation count (times) 0 12 5 4

[0121] Comparison of core indicators for operating condition 2 (transitional operating condition)

[0122] Evaluation indicators Embodiments of the present invention Comparative Example 1 Comparative Example 2 Comparative Example 3 Airspeed RMSE (m / s) 0.28 1.92 0.95 1.03 Maximum roll overshoot (°) 0.75 5.82 2.64 2.91 Pitch angle RMSE (°) 0.26 2.05 0.98 1.12 Number of attitude oscillations during the transition phase (times) 0 6 2 2 Actuator saturation count (times) 0 17 7 6

[0123] Comparison of core indicators for Condition 3 (cruise trajectory tracking condition)

[0124] Evaluation indicators Embodiments of the present invention Comparative Example 1 Comparative Example 2 Comparative Example 3 Horizontal position RMSE (m) 0.47 2.36 1.15 1.32 Height RMSE (m) 0.22 1.41 0.68 0.79 Yaw angle RMSE (°) 0.18 1.53 0.72 0.81 Perturbation suppression ratio 92.3% 38.6% 67.2% 62.5% Actuator saturation count (times) 0 9 3 3

[0125] Comparison of core indicators for operating condition 4 (extreme load change condition)

[0126] Evaluation indicators Embodiments of the present invention Comparative Example 1 Comparative Example 2 Comparative Example 3 Maximum overshoot of attitude angle (°) 1.02 8.74 3.95 4.28 Position recovery time after disturbance (s) 0.68 4.82 2.16 2.47 Steady-state error (m) after sudden load change 0.38 2.79 1.24 1.46 Actuator saturation count (times) 1 26 11 9

[0127] Core analysis of experimental results:

[0128] The performance of this invention is superior to that of existing technologies across all operating conditions. Comparing data from all operating conditions, the embodiments of this invention achieve orders-of-magnitude improvements in trajectory tracking accuracy, attitude control stability, and disturbance resistance: compared to baseline Comparative Example 1, trajectory tracking accuracy is improved by 78%~85%, attitude control error is reduced by 85%~89%, and disturbance stabilization time is shortened by 87%~90%; compared to improved Comparative Examples 2 and 3, core performance indicators are also improved by more than 45%, fully demonstrating the synergistic effect of the three-level cascaded architecture of this invention, rather than a simple superposition of a single module.

[0129] The ESO module addresses the robustness challenge in scenarios with large disturbances.

[0130] Compared with Comparative Example 3, under the conditions of sudden load change and sudden wind interference, Comparative Example 3 has no ESO feedforward compensation and relies only on NMPC feedback correction. The overshoot after the disturbance is increased by more than 3 times and the stabilization time is extended by more than 2 times. This fully demonstrates that the ESO module of the present invention can realize real-time estimation and feedforward cancellation of the total disturbance, fundamentally solve the model uncertainty problem caused by sudden load change, coupled disturbance and wind disturbance of heavy-load UAV, and greatly improve the robustness of the system.

[0131] Nonlinear NMPC+OCA solves control challenges in strongly coupled and highly nonlinear scenarios.

[0132] Compared with Comparative Example 2, under transition and cruise turbulence conditions, the linear MPC of Comparative Example 2 cannot adapt to the nonlinear characteristics of the system under large state deviations, and the linear pseudo-inverse allocation cannot handle the nonlinear aerodynamic characteristics of the actuators, resulting in a trajectory tracking error increase of more than 140%, obvious attitude oscillations in the transition section, and frequent actuator saturation. In contrast, the nonlinear NMPC of this invention can accurately predict the trajectory deviation in the entire state domain, and the nonlinear OCA can achieve the optimal cooperative allocation of actuators in the entire flight phase, completely solving the industry pain points of strong coupling and poor coordination of redundant actuators in the transition section of the compound wing.

[0133] Engineering practicality verification

[0134] Under all operating conditions, the actuator saturation times in this embodiment are almost zero, and the rate of change of control quantity is much lower than that in the comparative embodiment, which can effectively reduce the mechanical wear of the actuator and extend the service life of the UAV. At the same time, the control cycle of this invention is 10ms, and the optimization solution time is ≤2ms, which fully meets the real-time requirements of embedded flight control and has strong engineering feasibility.

[0135] (3) Flight verification results

[0136] The live flight test was conducted under conditions of a rated payload of 50kg and a force 6 wind, completing the entire flight process from vertical takeoff and landing, transition, cruise, payload delivery, and vertical landing. The deviation between the live flight data and the simulation results was ≤8%, fully replicating the excellent performance of the simulation.

[0137] The attitude remained stable throughout the entire flight, with no significant oscillations during the transition phase, and the attitude angle control error was ≤0.35°.

[0138] After the load is delivered (30% of the rated load), the attitude stabilization time is ≤0.5s, with no risk of instability;

[0139] The tracking error of the 5km cruise route is ≤0.6m, which fully meets the operational accuracy requirements for heavy-duty logistics transportation and material delivery.

[0140] It will be apparent to those skilled in the art that the present invention is not limited to the details of the exemplary embodiments described above, and that the invention can be implemented in other specific forms without departing from its spirit or essential characteristics. Therefore, the embodiments should be considered in all respects as exemplary and non-limiting, and the scope of the invention is defined by the appended claims rather than the foregoing description. Thus, all variations falling within the meaning and scope of equivalents of the claims are intended to be included within the present invention. No reference numerals in the claims should be construed as limiting the scope of the claims.

Claims

1. A stable flight control method for heavy-load unmanned aerial vehicles with a compound wing configuration, characterized in that: Includes the following steps: S1: Real-time acquisition of multi-source data during the flight of a high-payload, repetitive wing-jointing UAV, and output of the multi-source data after preprocessing; S2: Based on the preprocessed multi-source data, the total disturbance caused by rotor-fixed wing coupling disturbance, wind field disturbance and load change during the UAV flight is estimated in real time through the preset extended state observer (ESO), and the corresponding disturbance compensation amount is output. S3: Input the disturbance compensation amount and the preprocessed multi-source data into the preset nonlinear model predictive control model NMPC, predict the flight trajectory deviation in the preset prediction time domain through NMPC, and output the corresponding virtual compensation control command by rolling optimization solution. S4: Based on the virtual compensation control command, the virtual compensation control command is optimally allocated to each actuator of the UAV through the preset nonlinear optimal control allocation strategy OCA, generating the actual control command for each actuator, and realizing multi-actuator coordinated stable control.

2. The stable flight control method for heavy-load UAVs with a compound wing configuration according to claim 1, characterized in that: The multi-source data includes UAV body status data, actuator status data, environmental status data, and payload status data; The UAV's physical status data includes the UAV's spatial position, flight speed, attitude angle, attitude angular velocity, and linear acceleration data; The actuator status data includes the real-time rotational speed of the rotor motor, the actual deflection angle of the control surface, and the deflection rate data. The environmental status data includes real-time wind speed, wind direction, atmospheric density, and atmospheric pressure data; The load status data includes real-time load weight, center of gravity offset, load swing amplitude and frequency data.

3. The stable flight control method for heavy-load UAVs with a compound wing configuration according to claim 2, characterized in that: The preprocessing includes timestamp synchronization of multi-source data, outlier removal, and filtering and fusion processing. Outlier removal is performed using the 3σ criterion, and extended Kalman filtering is used to perform multi-sensor fusion processing on the UAV's body state data. All data is synchronized to the same control time base with a time synchronization accuracy of no less than 1ms.

4. The stable flight control method for heavy-load UAVs with a compound wing configuration according to claim 1, characterized in that: S2 includes at least the following steps: S2.1: Establish a six-degree-of-freedom nonlinear dynamic model for a heavy-load repeating wing UAV, and unify the model uncertainties and unmodeled dynamics caused by rotor-fixed-wing aerodynamic coupling disturbances, wind field disturbances, and load changes into the total disturbance state variables of the system, and construct the system state equations containing the extended states; S2.2: Based on the system state equation containing the extended state, a third-order nonlinear extended state observer (ESO) is designed. The UAV body state data in the preprocessed multi-source data is used as the input of the ESO to observe the estimated values ​​of each state variable and the total disturbance state variable of the system in real time. The ESO expression for the third-order nonlinear extended state observer is: in, This is the system output, i.e., the preprocessed state variables of the UAV itself; These are the observed values ​​of the system state variables, the derivatives of the state variables, and the total disturbance, respectively. For the gain parameter of ESO; It is a nonlinear power function, used to improve the convergence speed and anti-interference capability of the observer; For system control gain; S2.3: Based on the estimated value of the total disturbance state variable, generate the corresponding feedforward disturbance compensation amount, and superimpose the disturbance compensation amount onto the input of the NMPC prediction model to achieve feedforward cancellation of the total disturbance.

5. The stable flight control method for heavy-load UAVs with a compound wing configuration according to claim 4, characterized in that: The six-degree-of-freedom nonlinear dynamics model incorporates parameters on the real-time changes in UAV mass and moment of inertia caused by load variations, as well as parameters on the coupling effect of rotor wash on the aerodynamic characteristics of the fixed wing. The total disturbance is the sum of matched and unmatched disturbances, and there is no need to model each individual disturbance separately.

6. The stable flight control method for heavy-load UAVs with compound wing configuration according to claim 1, characterized in that: The S3 includes at least the following steps: S3.1: Construct an NMPC prediction model with disturbance feedforward compensation. The prediction model is a nonlinear dynamic model of a large-load repeating winged UAV that incorporates disturbance compensation. The current state of the UAV in the preprocessed multi-source data is used as the initial value for prediction, and the preset reference flight trajectory is used as the tracking target. S3.2: Construct the rolling optimization objective function of NMPC, wherein the objective function takes the minimum trajectory tracking error, the minimum rate of change of control quantity, and the minimum actuator constraint violation as optimization objectives; at the same time, system state constraints, control quantity constraints, and actuator physical constraints are set. The objective function expression for NMPC's rolling optimization is: in, To predict the time domain, To control the time domain; for Time prediction Step future state quantity, For reference trajectory state variables; To control the rate of change of quantity; Here is the state error weight matrix. The weight matrix represents the changes in the control quantity. For relaxation factor weights, To constrain violations of the relaxation factor, used to handle soft constraints; S3.3: Based on the current flight phase of the UAV, adaptively adjust the prediction time domain and control time domain, and perform rolling optimization to solve the optimal control problem in the finite time domain within each control cycle, and output the corresponding virtual compensation control commands; the virtual compensation control commands include the UAV's total lift command, roll moment command, pitch moment command, and yaw moment command.

7. The stable flight control method for heavy-load UAVs with a compound wing configuration according to claim 6, characterized in that: The flight phases include the vertical takeoff and landing phase, the transition phase, and the fixed-wing cruise phase. The vertical takeoff and landing phase and the transition phase adopt a short prediction time domain and a short control time domain. The fixed-wing cruise phase employs a long prediction time domain and a long control time domain.

8. The stable flight control method for heavy-load UAVs with a compound wing configuration according to claim 1, characterized in that: The actuators in S4 include a vertical takeoff and landing rotor, a tilt rotor, ailerons, elevator, rudder, and flaps, forming a redundant actuator system; The nonlinear optimal control allocation strategy (OCA) aims to minimize actuator control efficiency loss and control quantity change, while incorporating the following constraints: Actuator physical hard constraints: including the maximum / minimum speed constraints and speed change rate constraints of the rotor motor, and the maximum / minimum deflection angle constraints and deflection rate constraints of the control surfaces; Nonlinear constraints on aerodynamic efficiency: Efficiency constraints are constructed based on the nonlinear model of actuator aerodynamic efficiency under different flight conditions; Flight phase priority constraints: During the vertical takeoff and landing phase, the rotor actuators have the highest priority, and the control surface actuators do not participate in control; during the transition phase, the rotor and control surface actuators work together, and the priority is adaptively adjusted; during the fixed-wing cruise phase, the control surface actuators have the highest priority, and the rotor only participates in auxiliary attitude correction.

Images