Agricultural unmanned aerial vehicle complex load force coupling safety boundary detection method and system

Abstract

This invention discloses a method and system for detecting the safety boundary of complex load coupling in agricultural unmanned aerial vehicles (UAVs), belonging to the field of lifting and testing of agricultural UAVs. The method includes: suspending a real complex load at the end of a six-degree-of-freedom motion platform; generating a set of operating parameters according to a selected mode; acquiring time-stamped platform commands, the actual platform attitude, and six-dimensional reaction forces; inputting the time-delay-compensated six-dimensional reaction forces as an external load into a virtual UAV model to update its state and calculate rotor thrust and control torque; predicting the virtual lifting point state to generate platform commands for the next cycle, forming a continuous force coupling closed loop; judging the validity of the test based on the closed-loop response, and adjusting the operating parameters to determine the safety boundary envelope. This invention can reproduce the inverse effect of the nonlinear reaction force of the real load on the flight control under safe ground conditions, eliminate spurious oscillations caused by bench time delays, and output executable two-dimensional or three-dimensional safety boundary envelopes.

Description

Technical Field

[0001] This invention belongs to the field of agricultural unmanned aerial vehicle (UAV) hoisting and testing technology, specifically relating to a method and system for detecting the safety boundary of complex hoisting force coupling in agricultural UAVs. Background Technology

[0002] As the payload capacity of agricultural unmanned aerial vehicles (UAVs) increases, their operational targets have gradually expanded from fixed spray tanks to suspended pesticide tanks, fertilizer tanks, seedling trays, and agricultural implements. Unlike fixed payloads, suspended payloads are connected to the UAV via flexible slings, and the swinging, rotating, and tension changes of the slings continuously react upon the UAV.

[0003] For a half-full liquid pesticide tank, the liquid inside exhibits free surface sloshing, changes in added mass, liquid adhering to the tank walls, and liquid slapping. For a center-of-gravity offset load, a continuous or transient offset torque will be generated. For a two-point or multi-point suspended load, tension redistribution, local slack, and re-tensioning will occur between the various cable branches. These effects are highly nonlinear and will affect the UAV's position holding, attitude stability, and flight control output.

[0004] Currently, although controlled flight can retain the functions of real flight control, it is difficult to maintain strict consistency between different flights, and it is costly and difficult to gradually approach the failure boundary. On the other hand, whether it is a traditional ground motion test bench driven by a preset trajectory or a pure numerical simulation method, the former cannot reflect the reverse coupling of the load reaction force on the flight control, and the latter (such as an ideal pendulum or linear model) cannot accurately cover the nonlinear physical phenomena in a real pesticide tank. Neither of these methods can meet the needs of complex load verification.

[0005] When the six-dimensional reaction force generated by the real payload is applied to the virtual drone model, time delays will occur in sensor sampling, digital filtering, communication, model calculation, and motion platform execution. If the reaction force measured in the past is directly applied to the current state of the virtual drone, the six-dimensional reaction force and the virtual drone state will be out of sync and form an additional phase lag. This will cause the originally stable drone-payload system to produce unrealistic oscillations or divergence on the test bench, and may lead to the pseudo-instability caused by the test bench's own time delay being misjudged as the safety boundary of the tested drone or payload.

[0006] While general industrial control systems can use timestamps, Kalman prediction, or phase lead methods for time delay compensation, existing hoisting platforms typically do not simultaneously compare the platform command timestamp, the platform's actual pose feedback timestamp, and the six-dimensional reaction force sampling timestamp. They also do not incorporate the time when the platform will actually execute commands in the future into the virtual hoisting point state prediction, making it difficult to form a stable force coupling closed loop that can be used for safety boundary search. Summary of the Invention

[0007] Purpose of the invention: The purpose of this invention is to address the shortcomings of existing technologies by providing a method and system for detecting the safety boundary of complex load coupling in agricultural unmanned aerial vehicles. This method can retain the nonlinear reaction of real complex loads while reducing the risk of critical flight tests, and make the reaction force reverse the state of the virtual UAV, flight control output, and subsequent motion of the motion platform. At the same time, it can distinguish between real instability and bench pseudo-oscillation and form an executable safety boundary envelope.

[0008] Technical solution: This invention provides a method for detecting the safety boundary of complex load coupling in agricultural unmanned aerial vehicles, comprising the following steps:

[0009] Real, complex loads are suspended from the end of a six-degree-of-freedom motion platform using lifting devices and slings;

[0010] Set up a closed-loop reproduction mode for actual flight conditions and a parameterized safety boundary search mode. Generate the set of working condition parameters for the current test according to the selected mode, and configure the real complex load and virtual UAV model according to the set of working condition parameters. In the closed-loop reproduction mode for actual flight conditions, the hoisting point trajectory obtained by controlled actual flight is not used as an unchangeable platform command of the six-degree-of-freedom motion platform.

[0011] The platform commands with corresponding timestamps, the actual pose of the platform, and the six-dimensional reaction force transmitted by the real complex payload to the UAV side are obtained. The six-dimensional reaction force includes three-dimensional reaction force and three-dimensional reaction torque.

[0012] The platform execution delay is determined based on the time relationship between the platform command and the actual platform pose. The time correspondence between the six-dimensional reaction force and the actual platform state when the six-dimensional reaction force is generated is established based on the timestamp. The actual platform state and the six-dimensional reaction force are compensated for the time delay.

[0013] The six-dimensional reaction force that completes the time delay compensation is input as an external load into the virtual UAV model to update the virtual UAV state. The virtual UAV model then calculates the rotor thrust and control torque based on the updated virtual UAV state and the set of operating parameters.

[0014] Based on the updated virtual drone status, sling installation geometry, and platform execution delay, the virtual sling status is predicted to the moment when the six-degree-of-freedom motion platform is expected to execute the next control cycle command. The platform motion command for the next control cycle is generated and the six-degree-of-freedom motion platform is driven to move, so that the real complex load generates a new six-dimensional reaction force to form a continuous force coupling closed loop.

[0015] The validity of the current test, failure events, and hard safety thresholds are determined based on the load response, virtual UAV status, and platform operation status in the continuous force coupling closed loop. The validity determination result is output in the actual flight condition closed loop reproduction mode. In the parameterized safety boundary search mode, at least one parameter in the condition parameter set is adjusted based on the valid test results to determine the safety boundary envelope of complex loads on agricultural UAVs.

[0016] This invention also provides a safety boundary detection system for complex load-coupling of agricultural unmanned aerial vehicles, comprising:

[0017] A six-degree-of-freedom motion platform configured to suspend real and complex loads at the end via a spreader and slings.

[0018] The workstation is used to configure the closed-loop reproduction mode or parameterized safety boundary search mode for the actual flight conditions, and to generate the mode configuration package and condition parameter set for the current trial.

[0019] A unified clock and synchronous data acquisition module is used to synchronously sample the six-dimensional reaction force transmitted from the real complex load to the UAV side and generate measurement data with active timestamps.

[0020] The time delay compensation module is used to establish the time correspondence between the six-dimensional reaction force and the actual state of the platform when the six-dimensional reaction force is generated, align the actual state of the platform and the six-dimensional reaction force to a unified solution time, and predict the virtual suspension point state to the time when the six-degree-of-freedom motion platform is expected to execute the next control cycle command.

[0021] The virtual drone model is used to receive the six-dimensional reaction force after time delay compensation as an external load to update the virtual drone state, and to calculate the rotor thrust and control torque based on the updated virtual drone state and the set of operating parameters.

[0022] The real-time control module is used to generate platform motion commands for the next control cycle based on the predicted virtual lifting point status, so as to drive the real complex load to generate a new six-dimensional reaction force and form a continuous force coupling closed loop.

[0023] The workstation is also used to output the validity determination result of the reproduction in the closed-loop reproduction mode of the actual flight condition, and to adjust at least one parameter in the condition parameter set according to the valid test results in the parameterized safety boundary search mode, so as to determine the safety boundary envelope of the complex load of the agricultural unmanned aerial vehicle.

[0024] Beneficial effects: Compared with existing technologies, this invention establishes an improved path that integrates "real payload and virtual UAV." Through a six-dimensional reaction force with time correspondence and bidirectional delay compensation, the real payload directly participates in updating the virtual UAV's flight control state and generating the next cycle's motion commands. This constructs a closed-loop detection system on a ground test bench with physical realism and statistical repeatability, outputting an executable safety operation constraint envelope. Its advantages are:

[0025] This invention directly utilizes real physical loads such as half-full liquid, offset center of gravity, or multi-point suspension. Nonlinear effects such as free liquid surface sloshing, eccentric friction, and slack and re-tensioning of suspension cables naturally arise under real loads, and a six-dimensional force / torque sensor directly collects the actual transmitted resultant force and torque. This feature eliminates the assumption errors of complex physical phenomena in pure numerical simulations, allowing the transient nonlinear mechanical peak value that determines the UAV's instability boundary to be used as an objective measured input.

[0026] This invention incorporates the measured six-dimensional reaction force as an external load into the virtual drone's airframe dynamics calculation module. First, the virtual drone's position and attitude are changed. Based on this, the flight control algorithm recalculates the rotor thrust and control torque, and the real-time control module generates the platform's motion commands for the next cycle. This mechanism changes the traditional one-way open-loop mode in preset trajectory test benches where "platform motion is unaffected by load reaction," allowing liquid sloshing or sling impact to be realistically coupled in the reverse direction to the flight control output and subsequent sling movement, objectively reproducing the process of drone instability caused by load interference.

[0027] To address the inherent link and mechanical delays in hardware-in-the-loop testing, this invention first searches and establishes a correspondence between the six-dimensional reaction force and its historical platform state at the time of generation in the pose cache based on the timestamp. Then, it predicts the measured quantities to a unified solution time and further predicts the virtual lifting point state to the future time when the platform is expected to execute the next instruction. This time alignment mechanism of "compensating for the past" and "predicting the future" corrects the phase lag between the reaction torque and the UAV's motion, avoiding closed-loop divergence caused by the test bench system's own delay, thereby preventing the test bench's pseudo-instability from being mistakenly identified as a safety boundary of the tested object.

[0028] This invention divides the six-dimensional reaction force and the tension of each branch into two paths: the original signal and the finite bandwidth signal. The finite bandwidth signal is used for real-time force coupling closed-loop control to control noise and prevent divergence during differentiation calculations; the original signal, which contains high-frequency transient peaks, is directly input to an independent safety controller for hard-wired monitoring. This shunt structure ensures the stability of the high-frequency continuous force closed-loop calculation while fully preserving the actual impact peaks of liquid impact and retensioning, ensuring that overload trigger response is not missed due to filtering and smoothing.

[0029] This invention, when determining the state of dual-point or multi-point suspension, compares the current load ratio of each branch with the static baseline load ratio and extracts the dynamic tension residual after deducting the quasi-static baseline. This calculation mechanism isolates the common-mode force changes caused by the total load weight and overall vertical motion. Combined with a slack rope threshold set considering the lower limit of sensor noise, it eliminates misjudgments caused by a single noise sampling point, and can accurately distinguish between normal quasi-static load and sudden single-branch unloading or retensioning impact.

[0030] The system incorporates a closed-loop reproduction mode for real-flight conditions and a parameterized safety boundary search mode. The former inputs the mission and disturbance markers from controlled real-flight records to verify the ground test bench's ability to reproduce the nonlinear response of real-flight conditions in a closed-loop manner. The latter, based on valid verification, uses the same hardware and closed-loop solution model to progressively change wind disturbance, maneuver, and load parameters for boundary searches. This architecture eliminates the development costs of building two independent test systems and ensures that the boundary search samples are based on a mechanically validated closed-loop system obtained through real-flight testing.

[0031] This invention combines a continuous force-coupled closed loop of "real load response → bidirectional time delay compensation → virtual model update → real-time platform command generation" with a three-level parameterized progressive search program of "safety point, critical point, and failure point," producing a mutually dependent overall synergistic effect: the high-fidelity force-coupled closed loop ensures that every interaction and divergence between the UAV and the load strictly follows actual physical laws (rather than test equipment errors) during tests with continuously increasing crosswinds or maneuvering accelerations; while the automated progressive search program, relying on this stable closed-loop platform, continuously expands point-like discrete tests to the entire parameter space. The combination of these two approaches fills the testing gap where real-world flight approaching limits poses a crash risk, and pure simulation cannot capture nonlinear physical mutations. Ultimately, it transforms uncontrollable, high-risk test flights into safe, continuous, and traceable three-dimensional safety boundary envelope outputs in the laboratory (such as the maximum allowable wind speed and acceleration combination under a specific fill rate), providing directly executable operational constraints for complex loading of agricultural UAVs. Attached Figure Description

[0032] Figure 1 This is a diagram showing the overall engineering architecture of the detection system of the present invention.

[0033] Figure 2 This is the data flow and timing diagram for the k-th control cycle of this invention.

[0034] Figure 3 This is a schematic diagram of the security boundary envelope search of the present invention.

[0035] Figure 4 This is a flowchart of the detection method of the present invention. Detailed Implementation

[0036] The technical solution of the present invention will be described in detail below with reference to the accompanying drawings, but the scope of protection of the present invention is not limited to the embodiments described.

[0037] Example 1: System Overall Architecture

[0038] like Figure 1 As shown, this embodiment provides a complex load-coupling safety boundary detection system for agricultural unmanned aerial vehicles (UAVs). The system includes a flight condition acquisition component, a test configuration and safety boundary management workstation, a real-time closed-loop controller, a virtual UAV model calculation module, a unified clock and synchronous data acquisition module, a six-degree-of-freedom motion platform module, a lifting point equivalent adapter and lifting assembly, a six-dimensional reaction force measurement module, a multi-branch tension measurement module, a real complex load and load state measurement module, and an independent safety control module. The system forms the following closed data chain: real load response → six-dimensional reaction force and branch tension acquisition → time alignment and delay compensation → virtual UAV state update → next cycle platform command → actual platform motion → new real load response.

[0039] 1.1 Flight Condition Data Acquisition Component

[0040] The flight condition data acquisition component is installed during the controlled flight phase of the agricultural unmanned aerial vehicle (UAV) to collect data on the UAV's status, flight control outputs, ambient wind speed, six-dimensional reaction forces of real complex loads, multi-branch tensions, and event timestamps. This component connects to the UAV's satellite positioning device, flight inertial measurement unit, flight control data interface, wind speed measurement device, and load sensors, writing the collected data to the corresponding source timestamps to form a controlled flight condition data package. This controlled flight condition data package is imported into the test configuration and safety boundary management workstation via storage media or a communication network for configuring the closed-loop reproduction mode of the flight condition data.

[0041] 1.2 Test Configuration and Safety Boundary Management Workstation

[0042] The test configuration and safety boundary management workstation is used to select the closed-loop reproduction mode or the parameterized safety boundary search mode for the actual flight conditions, establish the test number for the current test, import the controlled actual flight data package, enter the configuration parameters of the real complex load and the sling assembly, configure the parameter search task, generate the mode configuration package for the current test, send the mode configuration package to the real-time closed-loop controller, receive the real-time status, event records and safety evaluation results of the current test, and form a two-dimensional or three-dimensional safety boundary envelope.

[0043] The mode configuration package should include at least the mode identifier, test number, payload assembly number, UAV model parameters, flight control parameter version, payload configuration parameters, mission reference, wind disturbance parameters, search step size, number of repetitions, stopping rules, failure event thresholds, calibration file version, and configuration file verification information. The configuration management link between the test configuration and safety boundary management workstation and the real-time closed-loop controller is not a high-frequency force-coupled closed loop and can use Ethernet, industrial Ethernet, or other data communication links.

[0044] 1.3 Real-time Closed-Loop Controller

[0045] The real-time closed-loop controller receives mode configuration packets, latches the mode flag for the current trial, and performs unified control cycle scheduling, real-time data caching, platform execution delay identification, measurement data time alignment, delay compensation for six-dimensional reaction forces and branch tensions, virtual lifting point state prediction, generation of the platform's next control cycle command, and output of operating and safety status. The real-time closed-loop controller receives the platform's actual pose, six-dimensional reaction forces, branch tensions, load status, wind speed feedback, and virtual UAV status; sends time-aligned measurement data to the virtual UAV model calculation module; sends position, attitude, velocity, and acceleration commands for the next control cycle to the six-degree-of-freedom motion platform; and outputs the main controller's operating status to the independent safety control module. Preferably, the real-time closed-loop controller employs a real-time industrial computer.

[0046] 1.4 Virtual Drone Model

[0047] The virtual drone model calculation module includes an airframe dynamics calculation module, a flight control algorithm module, a virtual sensor model, and an aerodynamic disturbance channel.

[0048] The airframe dynamics calculation module receives the six-dimensional reaction force after time delay compensation, the rotor combined thrust and control torque output by the flight control algorithm module, the virtual wind disturbance, and the virtual UAV state of the previous control cycle. Based on the above inputs, it updates the virtual UAV's position, velocity, attitude, and angular velocity.

[0049] The flight control algorithm module receives mission reference and virtual sensor data, and outputs control variables for each rotor, resultant rotor thrust, and control torque.

[0050] The virtual sensor model generates virtual inertial measurement, satellite positioning, altitude, attitude, and angular velocity data based on the virtual UAV state output by the body dynamics calculation module, and feeds the virtual sensor data back to the flight control algorithm module.

[0051] The aerodynamic disturbance channel receives crosswind speed, crosswind direction, gust waveform and corresponding event time stamps, and calculates the aerodynamic disturbances acting on the virtual UAV.

[0052] 1.5 Unified Clock and Synchronization Data Acquisition Module

[0053] The unified clock and synchronous data acquisition module is used to synchronously acquire platform commands and their generation time, the platform's actual position / attitude / velocity, six-dimensional reaction force, tension of each sling branch, load attitude / angular velocity / acceleration, actual wind speed, and platform servo status and fault status. This module records the corresponding generation timestamp, feedback timestamp, or source timestamp for each sampled data, and sends the timestamped data to the real-time closed-loop controller via the real-time bus.

[0054] Preferably, the platform driver and the synchronous data acquisition module are connected via EtherCAT, and a unified time reference is established using the EtherCAT distributed clock or the IEEE 1588 precision clock protocol.

[0055] In a preferred embodiment, the control and sampling period for the six-dimensional reaction force and branch tension is 1 ms, the sampling frequency of the load inertial measurement unit is not less than 200 Hz, and the operating frequency of the virtual drone model is 500 Hz to 1 kHz. These parameters are preferred implementation parameters and do not constitute a limitation on the scope of protection of this invention.

[0056] 1.6 Six-DOF Motion Platform Module

[0057] A six-degree-of-freedom (DOF) motion platform is used to generate three-way translation and three-way rotation of a virtual UAV hoisting point on the ground. A real-time closed-loop controller sends target position, target attitude, target velocity, and target acceleration commands to the six-DOF motion platform. The platform encoder and servo drivers return the platform's actual position, actual attitude, actual velocity, servo status, and fault status to the real-time closed-loop controller. The system directly uses the platform's actual pose with feedback timestamps as the actual motion excitation experienced by the real complex load.

[0058] 1.7 Equivalent adapter for suspension point and suspension assembly

[0059] An equivalent adapter for the UAV's lifting point is installed at the end of the six-DOF motion platform. Below the equivalent adapter, a six-dimensional force / torque sensor, a lifting device, a sling, and a real complex load are sequentially arranged. The equivalent adapter is used to align the geometric reference of the platform's end point during ground testing with the geometric reference of the UAV's lifting point during controlled flight, and to transfer the mechanical motion of the six-DOF motion platform to the real complex load.

[0060] Ideally, controlled flight operations and ground testing should share the same set of six-dimensional force / torque sensors, slings, slings, tension sensors, and physical assemblies for real complex loads, all located below the rapid installation reference plane. When controlled flight conditions do not allow the use of the same physical assembly, at least the same model, same installation geometry, same coordinate system definition, and traceable to the same calibration reference should be used for sensors and sling assemblies, and consistency checks should be performed on mass, center of gravity, moment of inertia, and sling mechanical properties.

[0061] 1.8 Six-Dimensional Reaction Force Measurement Module

[0062] The six-dimensional reaction force measurement module is used to measure the triaxial force and triaxial torque transmitted to the UAV by real complex loads. The six-dimensional force / torque sensor is installed between the end of the six-degree-of-freedom motion platform and the lifting device. Before the test, the installation position, installation direction, zero bias, temperature drift, and geometric relationship from the measurement point to the center of mass of the virtual UAV are calibrated for the six-dimensional force / torque sensor.

[0063] The six-dimensional reaction force measurement module outputs the original six-dimensional reaction force signal, the zero-bias compensation signal, and a finite-bandwidth signal for continuous force coupling closed-loop. The original six-dimensional reaction force signal is used for independent hard safety monitoring, while the finite-bandwidth signal is used for time delay compensation and virtual UAV dynamics calculation. Closed-loop filtering does not remove liquid impact peaks or retensioning impact peaks from the original signal.

[0064] 1.9 Multi-branch tension measurement module

[0065] When real-world complex loads are handled using a dual-point or multi-point suspension configuration, tension sensors are installed in each cable branch. A multi-branch tension measurement module independently measures the axial tension of each branch to identify tension redistribution, tension imbalance, single-branch slack, and retensioning impacts after slack. The tension signals from each branch are transmitted via a synchronous data acquisition module to both the real-time closed-loop controller and the independent safety control module.

[0066] The six-dimensional reaction force already represents the resultant force and resultant torque transmitted to the UAV by the real complex load. The tension signals of each branch are only used independently for the imbalance and impact identification of the bearing state of each branch, and do not participate in the superposition calculation of the body dynamics equation.

[0067] 1.10 Realistic Complex Load and Load State Measurement Module

[0068] Realistic complex loads include at least one of the following: half-full liquid agricultural loads, offset agricultural loads, and agricultural loads using a two-point or multi-point suspension configuration. The inertial measurement unit (IMU) of the realistic complex load is mounted on its main structure to output the load's attitude, angular velocity, and acceleration. For two-point or multi-point suspension configurations, the suspension configuration is defined according to the number of branches, the coordinates of the platform-side suspension points, the coordinates of the load-side suspension points, the effective rope length of each branch, and the static reference tension of each branch.

[0069] 1.11 Independent Security Control Module

[0070] The independent safety control module is independent of the main real-time closed-loop controller's overload, platform out-of-bounds, emergency stop, communication failure, and sensor malfunction protections. The inputs to the independent safety control module include at least the six-dimensional reaction force raw signal, branch tension hardware comparison signal, platform mechanical limit signal, platform driver fault signal, emergency stop button signal, and the main real-time closed-loop controller's operating status.

[0071] Hard safety thresholds include at least the allowable tensile force of any branch, the allowable six-dimensional reaction force of the lifting point and lifting device, the attitude limits of the virtual UAV, the lateral sweep and tilt angle boundaries of the load, the platform's workspace and kinematic limitations, and the failure of critical communication or sensors. Hard safety thresholds are stored in an independent safety control module and are not automatically increased, decreased, or masked during testing by the parameterized safety boundary search program. The allowable tensile force of a branch is the minimum of the rated load-bearing capacities of all components connected in series in that branch, divided by a specified safety factor.

[0072] Upon triggering of the independent safety control module, at least one of the following operations is performed: the platform driver is put into a safe torque off state; the platform movement is stopped; the platform mechanical braking or safety holding is initiated; the six-dimensional reaction force, branch tension, platform state, and virtual UAV state before and after triggering are frozen; and the parameterized safety boundary search program is prevented from continuing to increase crosswind speed or maneuvering acceleration.

[0073] Example 2: Dual-mode configuration location and operating rules

[0074] like Figure 1 As shown, the live flight condition closed-loop reproduction mode and the parameterized safety boundary search mode are not two independent test systems, but rather two input modes of the same ground force coupling closed-loop test system. The configuration of the two modes is divided into a mode selection layer, a real-time execution layer, a physical load layer, and a hard safety layer.

[0075] The mode selection layer is set at the test configuration and safety boundary management workstation. The workstation selects either the live flight condition closed-loop reproduction mode or the parameterized safety boundary search mode according to the test objective, establishes the test number for the current test, and generates a mode configuration package containing mode flags and corresponding operating condition parameters.

[0076] The real-time execution layer is set within the real-time closed-loop controller. The real-time closed-loop controller receives the mode configuration packet, latches the mode flag for the current trial, and generates or calls the task reference, initial state, and wind disturbance input according to the corresponding mode. Once the test enters the running state, the mode is not switched to another during platform movement. Mode switching should only occur after the six-degree-of-freedom motion platform has stopped and entered a safe holding state.

[0077] The physical load layer is set on a realistic complex load and suspension assembly. Before the start of the current trial, the liquid fill rate, center of gravity offset, lifting point position, number of sling branches, and effective rope length of each branch are actually adjusted according to the mode configuration package. Changes in liquid fill rate, center of gravity position adjustment, lifting point switching, and sling length adjustment are all completed after the six-degree-of-freedom motion platform stops and enters a safe holding state. The above physical parameters are actually realized on the realistic complex load and suspension assembly through real load filling, counterweight movement, lifting point switching, and sling length adjustment.

[0078] The hard safety layer is set within an independent safety control module, used to set and monitor the aforementioned hard safety thresholds and emergency stop conditions. When any hard safety threshold is triggered, the independent safety control module directly restricts or stops the test, and other evaluation indicators are not allowed to offset the hard safety threshold.

[0079] Example 3: Closed-loop reproduction mode of actual flight conditions

[0080] In the closed-loop reproduction mode of the actual flight condition, the test configuration and safety boundary management workstation imports the controlled actual flight data package. The controlled actual flight data package includes at least: UAV and payload configuration parameters (including mass, inertia, flight control version and physical configuration); flight mission and status data (including initial state, mission reference, satellite positioning and flight control output); environmental and event markers (including wind disturbance parameters and abrupt change timestamps such as slack rope / retensioning); and mechanical measurement data (including six-dimensional reaction forces and tension of each branch).

[0081] The initial position, velocity, attitude, and angular velocity of the virtual UAV are input into the airframe dynamics calculation module to initialize the virtual UAV state for the current trial. The mission reference obtained from controlled flight is input into the flight control algorithm module. Maneuvering events and their corresponding event timestamps are used to update the mission reference according to the event sequence of the controlled flight, enabling the flight control algorithm module to execute corresponding hovering, acceleration, deceleration, or turning tasks. Wind disturbance events are input into the aerodynamic disturbance channel according to their corresponding event timestamps to reproduce the mean wind, crosswind, or gust disturbances during actual flight.

[0082] Data such as the hoisting point trajectory obtained from controlled flight are only used for verification of the ground closed-loop reproduction results. Subsequent commands of the six-degree-of-freedom motion platform are still generated in real time by the virtual UAV model with added load response, so as to reflect the reverse coupling of the nonlinear effect of the real load on the flight control.

[0083] Example 4: Parametric Security Boundary Search Mode

[0084] In the parameterized safety boundary search mode, the test configuration and safety boundary management workstation establishes a parameter search task and divides the parameters into physical parameters, operational condition parameters, and search control parameters. Physical parameters include liquid fill rate, center of gravity offset, hoisting point configuration (e.g., number and coordinates of hoisting points), and sling length. Operational condition parameters mainly include variables such as horizontal maneuvering acceleration and crosswind speed, used to form mission references and disturbance inputs. Horizontal maneuvering acceleration, vertical acceleration, and acceleration / deceleration duration are used to form the mission reference for the flight control algorithm module. Crosswind speed and crosswind direction are input into the aerodynamic disturbance channel of the virtual UAV model. Search control parameters include parameter increment step size, number of repetitions for each parameter combination point, boundary refinement step size, and stopping rules. The search procedure starts from a safe parameter point with low crosswind and low maneuvering acceleration, gradually increasing the crosswind speed or maneuvering acceleration. When a failure event occurs at the current parameter combination point or a hard safety threshold is triggered, the increment along the dangerous direction stops, and the search returns to the last safe parameter point, narrowing the search step size between the last safe parameter point and the first failed parameter point.

[0085] Example 5: Unified Time Base and Delay Compensation

[0086] The delay compensation module stores the following time information: the timestamp of the platform command, the source timestamp of the data collected by each sensor (such as the platform encoder, six-dimensional force sensor and tension sensor), and the timestamp of the data received by the real-time closed-loop controller.

[0087] The platform execution delay is estimated by applying amplitude-limited step commands, frequency sweep commands, or pseudo-random commands to the six-DOF motion platform during the no-load calibration or low-risk operation phase, and calculating the correlation between the platform command velocity or acceleration and the platform's actual feedback velocity or acceleration within a sliding time window. The time offset corresponding to the maximum correlation is determined as the estimated platform execution delay. When there is a sudden change in the six-dimensional reaction force, branch tension, or platform drive current, the update of the platform execution delay estimate is paused, and the most recent valid estimate is continued to be used to avoid misidentifying liquid shock or sling retension as platform execution delay.

[0088] For the digitally filtered six-dimensional reaction force, let its source timestamp be... The digital filter group delay is Then the effective physical moment corresponding to the six-dimensional reaction force is: .

[0089] The latency compensation module searches for the location in the circular buffer of the platform's actual pose. The two feedback samples, one before and one after, are used to determine the actual position and attitude of the platform when the six-dimensional reaction force is generated through position interpolation and attitude interpolation. This establishes a time correspondence between the six-dimensional reaction force and the actual state of the platform, rather than directly mapping the six-dimensional reaction force to platform commands with the same data sequence number.

[0090] Under conditions of short prediction time and unsaturated signal, a first-order phase-lead prediction method with high-frequency gain limitation can be used to compensate for the six-dimensional reaction force. The basic logic is as follows: the predicted value at the unified solution time equals the original measured value at the effective physical time plus the product of the signal change rate and the prediction time difference. The change rate of the six-dimensional reaction force is filtered with finite bandwidth and its amplitude is limited to avoid noise amplification. The tensions of each branch are aligned to the unified solution time using the same phase-lead prediction method as the six-dimensional reaction force.

[0091] As an alternative to phase-lead prediction, a Discrete Kalman predictor can be used. The state variables of a Discrete Kalman predictor include at least the platform position, platform velocity, platform acceleration, six-dimensional reaction force, rate of change of the six-dimensional reaction force, tension in each branch, and rate of change of the tension in each branch. When new measurement data with source timestamps arrive, a measurement update is performed at the corresponding historical time, and the updated state is then advanced to the unified solution time, without directly treating late measurement data as data for the current time.

[0092] After completing data alignment for unified computation, compensation is also provided for future platform execution latency. The platform anticipates the execution time of the next instruction to be [time missing]. The real-time closed-loop controller predicts the virtual hoisting point state to the expected execution time based on the virtual UAV's position, velocity, attitude, and angular velocity at a unified solution moment. Based on the predicted virtual suspension point state, the system generates position, attitude, velocity, and acceleration commands for the platform's next control cycle. This simultaneously compensates for the fact that "the currently obtained six-dimensional reaction force comes from past physical moments" and "the currently generated platform commands will be actually executed in future physical moments."

[0093] When the prediction time exceeds the preset limit, the timestamp is reversed, there is continuous packet loss, or the latency jitter exceeds the allowable value, the latency compensation module stops extrapolating, determines the current trial as invalid, and puts the six-degree-of-freedom motion platform into a safe holding state.

[0094] Example 6: Virtual Unmanned Aerial Vehicle Dynamics and Flight Control Interface

[0095] The body dynamics calculation module updates the motion state of the virtual UAV's center of mass according to the following translational dynamics equations:

[0096]

[0097] in, To standardize the solution time, For virtual drone quality, Accelerating the center of mass of the virtual drone Let gravitational acceleration be the vector in the inertial coordinate system. This is the attitude mapping matrix from the body coordinate system to the inertial coordinate system. The rotor thrust generated by the flight control algorithm module, The body coordinate system reaction force is used to complete the installation direction, zero deviation, temperature drift, and time delay compensation.

[0098] The body dynamics calculation module updates the attitude motion state of the virtual UAV according to the following rotational dynamics equations:

[0099]

[0100] in, Let the rotational inertia matrix of the virtual drone relative to its center of mass be . and These are the angular velocity and angular acceleration of the virtual drone, respectively. The control torque generated for the flight control algorithm module, The reaction torque is used to complete the correction and time delay compensation of the centroid of the measurement point to the virtual UAV.

[0101] The six-dimensional reaction force is not directly superimposed onto the flight control thrust commands, attitude commands, or mission reference through the ordinary flight mission interface. Instead, it is written as an external load into the airframe dynamics calculation module. The six-dimensional reaction force first changes the virtual UAV's position, velocity, attitude, and angular velocity, and then the flight control algorithm module calculates new rotor thrust and control torque based on the updated virtual UAV state and mission reference.

[0102] When a configurable flight control algorithm is used, the flight control computing unit can run a configurable flight control program and receive virtual inertial measurement, satellite positioning, altitude, and attitude data, outputting control variables for each rotor, resultant rotor thrust, or control torque. Six-dimensional reaction forces are input to the external airframe dynamics calculation module. The airframe dynamics calculation module updates the virtual UAV state based on the flight control output, virtual wind disturbance, and real six-dimensional reaction forces, and then feeds the updated virtual sensor data back to the flight control algorithm module.

[0103] For commercial flight controllers that do not open the underlying dynamics interface, the airframe dynamics calculation module calculates sensor data such as virtual attitude and angular velocity based on the flight controller output, virtual wind disturbance and real six-dimensional reaction force, and then feeds it back to the commercial flight controller in an alternative way through the hardware-in-the-loop interface or external state interface; its original speed, attitude and other control interfaces only retain the mission reference input function.

[0104] Example 7: Multi-branch tension and failure event identification

[0105] When a complex load is applied using a dual-point or multi-point suspension configuration, the tension of each branch is aligned to a unified calculation time after zero-bias compensation, filtering, and time delay compensation. The tension imbalance of multiple branches is obtained by calculating the absolute deviation between the current load ratio of each branch and its static reference load ratio; when the deviation exceeds a set threshold, it is determined to be a tension imbalance. Using proportional comparison can reduce the influence of the total load weight and overall vertical motion on the judgment results. The dynamic tension residual of each branch is the difference between its actual tension after time delay compensation and the quasi-static branch tension baseline. Larger positive residuals are used to identify liquid impact and retensioning loads, while larger negative residuals are used to identify rapid unloading of the branch. For single-branch slack rope judgment, a slack rope threshold is set (this threshold comprehensively considers the sensor noise level of the corresponding channel and the static reference load ratio). When the actual tension of the branch is lower than this threshold and remains below it for a set time (e.g., 20–50 ms), it is determined to be slack rope. When the branch tension recovers rapidly after slack rope release and the dynamic tension residual suddenly increases, it is determined to be a retensioning impact.

[0106] Example 8: Single Control Cycle and Module Signal Chain

[0107] like Figure 2 As shown, in the first Within each control cycle, it operates according to the following process.

[0108] The real-time closed-loop controller reads the target lifting point position, attitude, velocity and acceleration generated in the previous control cycle, and latches the platform instruction generation timestamp, expected execution timestamp and current safety status.

[0109] The six-degree-of-freedom motion platform executes the platform motion commands from the previous control cycle. The platform encoder and servo driver return the platform's actual position, actual attitude, actual speed, servo status, and fault status.

[0110] The platform's actual motion induces real, complex loads such as free surface sloshing, eccentric torque, load oscillation, or multi-branch tension redistribution. Six-dimensional force / torque sensors, multi-branch tension sensors, and load inertial measurement units simultaneously sample and record corresponding source timestamps. The original six-dimensional reaction force and original branch tension enter an independent, safe channel; the finite-bandwidth signal enters a continuous force coupling closed loop.

[0111] The real-time closed-loop controller compares the platform command timestamp, the platform's actual pose feedback timestamp, the six-dimensional reaction force sampling timestamp, and the tension sampling timestamps of each branch. Through buffer interpolation and state prediction, the platform's actual state, the six-dimensional reaction force, and the tension of each branch are aligned to a unified solution time. .

[0112] The six-dimensional reaction force input to the airframe dynamics calculation module completes time delay compensation. Tension in each branch is used to determine tension imbalance, dynamic tension residual, and impact of slack rope and retensioning. The virtual UAV model updates its position, velocity, attitude, and angular velocity based on flight control output, virtual wind disturbance, and the actual six-dimensional reaction force. The flight control algorithm module recalculates the rotor thrust and control torque based on the updated virtual UAV state and mission reference.

[0113] The real-time closed-loop controller calculates the virtual hoisting point state when the platform is expected to execute the next instruction, based on the updated virtual UAV state, hoisting point installation geometry, and the identified platform execution latency, and generates the first... The platform position, attitude, velocity, and acceleration commands are generated for each control cycle. After the platform motion commands undergo safety checks on the platform workspace and kinematic boundaries (such as velocity and acceleration limits), they are sent to the platform servo driver via the real-time bus.

[0114] Before sending platform commands, the independent safety control module and safety monitoring logic check whether the current state triggers the aforementioned hard safety thresholds and failure events such as single-branch slack rope and retensioning impact. If the stopping condition is not triggered, the platform motion command for the next control cycle is sent. When the hard safety threshold is triggered, the six-dimensional reaction force, branch tension, platform state, and virtual UAV state before and after the event are frozen, and the platform enters a safe hold or restricted return-to-zero state. The above single-cycle process forms a closed loop of "real load response - measurement - time delay compensation - virtual model update - next cycle platform command - new load response".

[0115] Example 9: Experimental Validity and Safety Boundary Envelope Search

[0116] Safety evaluation first distinguishes between the validity of bench tests and the failure of the tested object. If timestamp anomalies, delay compensation failures, platform tracking errors exceeding permissible values, sensor saturation, platform workspace or kinematic limitations preventing command execution, or virtual UAV attitude divergence caused by platform execution and closed-loop timing anomalies, the current test is deemed invalid and not used as a safety boundary sample for agricultural UAVs or real complex loads. Only when the closed-loop timing relationship, sensor status, and platform execution are all valid are tension imbalance, single-branch slack rope, retensioning impact, attitude oscillation divergence, and load space exceeding limits considered critical events or failure events of the tested object.

[0117] like Figure 3 As shown, with fixed liquid fill rate, center of gravity offset, hoisting point configuration, and sling length, horizontal maneuvering acceleration and crosswind speed are used as two-dimensional coordinate axes. Starting from the safety parameter point corresponding to low crosswind and low horizontal maneuvering acceleration, the crosswind speed or horizontal maneuvering acceleration is increased step by step. Each parameter combination point is tested at least three times.

[0118] If no failure event or hard safety threshold is triggered in repeated tests, the corresponding parameter combination point is recorded as the safety point. If a recoverable short-term slack rope, tension imbalance, or non-divergent oscillation occurs, the corresponding parameter combination point is recorded as the critical point. If any hard safety threshold, attitude oscillation divergence, or load space exceedance is triggered, the corresponding parameter combination point is recorded as the failure point.

[0119] By reducing the search step size between the last safe point and the first failure point, the boundary interval between the safe region and the failure region is obtained. After changing the horizontal maneuver acceleration, the above search is repeated, and the conservative boundary points are connected to form a two-dimensional safe boundary envelope.

[0120] Further altering the liquid fill rate or center of gravity offset, a two-dimensional safety boundary search is repeated for each liquid fill rate or center of gravity offset. Multiple sets of two-dimensional safety boundary envelopes are superimposed to form a three-dimensional safety boundary envelope surface. Dominant failure modes are marked within the safety boundary envelope, and the safety domain, critical zone, failure domain, and corresponding operational constraints are output. During the safety boundary evaluation process, triggering any hard safety threshold has veto power, meaning it is directly determined as a failure point.

[0121] Example 10: Closed-loop reproduction and safety boundary search of a 20 L half-full pesticide tank

[0122] This embodiment uses a half-full pesticide box with dual-point suspension as an example to illustrate the closed-loop reproduction and safety boundary search process of the present invention under real complex suspension load conditions.

[0123] The pesticide tank is suspended at two points, with the test liquid at half full. The center of gravity is offset relative to the geometric center of the tank, resulting in different reference tensions on the left and right branches during static suspension. A six-dimensional force / torque sensor is installed between the end of the six-degree-of-freedom motion platform and the lifting device. Tension sensors are installed on each sling branch, and an inertial measurement unit is installed on the load. The allowable tensile force for each branch is set based on the rated load-bearing capacity of the slings, hooks, and mounting connectors.

[0124] Controlled flight and transient slack rope: After the test liquid was filled and allowed to stand for about 5 minutes, the UAV conducted a low-risk controlled flight over an open farmland, flying forward at low speed and then decelerating and hovering. During deceleration, the liquid first surged forward and then oscillated in the opposite direction. One branch tension decreased and fell below the slack rope threshold within about 46 ms. When this branch regained load, a retensioning impact occurred, with the peak value exceeding twice the static tension, and the load pitch angle increased significantly. The system saves the mission reference, initial state, deceleration event, measured wind speed, UAV state, six-dimensional reaction force, multi-branch tension, and load attitude. The flight suspension point trajectory is only used for verification of the ground closed-loop reproduction results and is not used as the fixed input trajectory of the six-degree-of-freedom motion platform.

[0125] Closed-loop reproduction without time delay compensation: The same payload assembly was installed on a six-DOF motion platform, and the mission reference, initial state, and wind disturbance events obtained from controlled flight were input into the virtual UAV model. In the first test, time delay compensation was not enabled, and the collected six-dimensional reaction forces were directly written into the virtual UAV model.

[0126] During the experiment, the actual platform motion was delayed relative to the platform commands, and the six-dimensional reaction force corresponded to an earlier actual platform state, causing a phase misalignment between the reaction torque and the current motion of the virtual drone. The pitch oscillation of the virtual drone continued to increase and did not decay for several consecutive cycles, at which point the safety control module terminated the experiment. Since the oscillation was caused by the bench delay, this experiment was deemed invalid for reproduction and was not included in the safety boundary.

[0127] Closed-loop reproduction with time delay compensation enabled: After enabling time delay compensation, the system aligns the platform's actual pose, six-dimensional reaction force, and branch tension to a unified calculation time, and then predicts the virtual hoisting point state to the time when the platform is expected to execute the next command. After re-executing the same working condition, the virtual UAV's pitch angle gradually decreases after the first peak. The retensioning peak after slack rope is effectively captured, and the additional pitching moment in the six-dimensional reaction force causes a corresponding change in the virtual UAV's state, preventing the closed loop from continuously diverging. This experiment verifies that the ground closed loop after time delay compensation can stably reproduce the transient slack rope and retensioning phenomena that occur in actual flight.

[0128] Parametric safety boundary search: Maintaining a liquid fill rate of approximately 50%, a center of gravity offset to the right of approximately 28 mm, a dual-point suspension configuration, and a 1.80 m sling length, a two-dimensional parametric plane was formed using crosswind speed and horizontal maneuvering acceleration. The crosswind speed started at 2.0 m / s, increasing gradually in coarse steps of 1.0 m / s, and then decreasing to a fine step of 0.25 m / s near the boundary. The horizontal maneuvering acceleration was set sequentially to 0.6 m / s², 0.9 m / s², 1.2 m / s², and 1.5 m / s². Each parameter combination point was tested three times, with liquid temperature, settling time, and static reference tension of each branch recorded before each test.

[0129] When the horizontal acceleration is 0.9 m / s²: the hard safety threshold was not triggered in the three trials with crosswinds not exceeding 5.5 m / s, and this was recorded as a safe point; when the crosswind was 6.5 m / s, the left branch experienced short-term slack rope multiple times, and the peak retensioning of the right branch was approximately 196 N, which was recorded as a critical point; when the crosswind was 7.5 m / s, the peak retensioning of the right branch was approximately 228 N, exceeding the allowable tension of 220 N for a single branch, triggering a stop and recording this as a failure point.

[0130] The system returns to the last safety parameter point, narrowing the search step size between 6.5 and 7.5 m / s to obtain the critical zones for slack rope, retensioning, and the structural allowable tensile force boundary. The search is repeated after changing the horizontal maneuvering acceleration to form a two-dimensional safety boundary envelope. The search is further repeated under conditions of 25%, 50%, and 75% liquid filling rate to form a three-dimensional safety boundary envelope of liquid filling rate, horizontal maneuvering acceleration, and crosswind velocity. The dominant failure mode is marked at the boundary locations.

[0131] In a further embodiment, the system can also output final operational limits based on the obtained safety boundaries and the reserved engineering margins. For example, when the pesticide tank filling rate is between 40% and 60%, the continuous crosswind should not exceed 5.5 m / s; when the crosswind is between 4.0 and 5.5 m / s, the horizontal maneuvering acceleration should not exceed 0.8 m / s²; when the horizontal maneuvering acceleration reaches 1.2 m / s², the continuous crosswind should not exceed 4.0 m / s; when the peak gust exceeds 6.5 m / s, the hoisting should be stopped or the system should enter a low-speed hovering recovery phase.

[0132] Example 11: Safety boundary detection method for complex loads on agricultural unmanned aerial vehicles

[0133] like Figure 4 As shown in the figure, this embodiment provides a method for detecting the safety boundary of complex load coupling in agricultural unmanned aerial vehicles, including the following steps.

[0134] Step S101: Suspend a real, complex load, including characteristics such as half-full liquid or center of gravity offset, at the end of a six-degree-of-freedom motion platform using a lifting device and slings.

[0135] Step S102: Set either the closed-loop reproduction mode for actual flight conditions or the parameterized safety boundary search mode, and generate the set of parameters for the current trial based on the selected mode. In the closed-loop reproduction mode, acquire the mission reference of the controlled actual flight, the initial state of the virtual UAV, maneuvering events, wind disturbance events, and their corresponding event timestamps. In the parameterized safety boundary search mode, configure at least one parameter from the following: liquid fill rate, center of gravity offset, sling configuration, sling length, horizontal maneuvering acceleration, and crosswind speed.

[0136] Step S103: Obtain the platform commands and their generation timestamps, the platform's actual pose and its feedback timestamps, and the six-dimensional reaction force and its source timestamps. When using a dual-point suspension configuration or a multi-point suspension configuration, also obtain the tension of each sling branch and its source timestamps.

[0137] Step S104: Compare the changes in platform commands with the actual changes in platform pose to determine the platform execution delay. Based on the source timestamps, data arrival timestamps, and digital filter group delays of the six-dimensional reaction force and branch tension, determine the effective physical time of the corresponding measurement data.

[0138] Step S105: Based on the effective physical moment of the six-dimensional reaction force, search for adjacent feedback samples in the annular buffer of the platform's actual pose. Determine the actual platform state at the time the six-dimensional reaction force is generated through position interpolation and attitude interpolation. Use the same method to determine the actual platform state corresponding to the tension of each branch.

[0139] Step S106: Using a discrete Kalman predictor or a phase lead predictor with high-frequency gain limitation, align the actual state of the platform, the six-dimensional reaction force, and the tension of each branch to a unified solution time. If the prediction time exceeds the preset upper limit, the timestamp is abnormal, there is continuous packet loss, or the delay jitter exceeds the allowable value, stop extrapolation, determine the current attempt as invalid reproduction, and put the platform into a safe holding state.

[0140] Step S107: The six-dimensional reaction force, after time delay compensation, is input as an external load into the airframe dynamics calculation module. The airframe dynamics calculation module updates the position, velocity, attitude, and angular velocity of the virtual UAV based on the flight control output, the actual six-dimensional reaction force, and the virtual wind disturbance. The flight control algorithm module calculates the rotor thrust and control torque based on the updated virtual UAV state and mission reference.

[0141] Step S108: Based on the updated virtual UAV state, sling installation geometry, and platform execution delay, predict the virtual sling state to the moment when the six-degree-of-freedom motion platform is expected to execute the next control cycle command, and generate the platform position, attitude, velocity, and acceleration commands for the next control cycle.

[0142] Step S109: Perform platform motion commands, checking the platform workspace, velocity, acceleration, jerk, and singular configuration. Drive the six-degree-of-freedom motion platform to generate new six-dimensional reaction forces and branch tensions from the real complex load. Return to step S103 to form a continuous force coupling closed loop.

[0143] Step S110: Determine the test validity, failure events, and hard safety thresholds for the current trial based on the load response, virtual UAV status, and platform operating status. In the closed-loop reproduction mode of the actual flight condition, output the reproduction validity determination result. In the parameterized safety boundary search mode, use the currently reproduced and valid trial as the safety boundary sample.

[0144] Step S111: If no failure event or hard safety threshold is triggered, increase at least one parameter among crosswind speed, horizontal maneuvering acceleration, liquid fill rate, or center of gravity offset by a preset step size. Reduce the search step size between the last safe parameter point and the first failure parameter point to obtain the boundary interval between the safe domain and the failure domain. Connect the conservative boundary points corresponding to different horizontal maneuvering accelerations and crosswind speeds to form a two-dimensional safe boundary envelope; further change the liquid fill rate or center of gravity offset to form a three-dimensional safe boundary envelope. Based on the safe domain, critical zone, failure domain, and dominant failure mode, output the allowable crosswind, allowable maneuvering acceleration, allowable load state, and allowable sling configuration for the agricultural unmanned aerial vehicle under complex sling conditions.

[0145] As described above, although the invention has been shown and described with reference to specific preferred embodiments, it should not be construed as limiting the invention itself. Various changes in form and detail may be made without departing from the spirit and scope of the invention as defined in the appended claims.

Claims

1. A method for detecting the safety boundary of complex load coupling in agricultural unmanned aerial vehicles, characterized in that, Includes the following steps: Real, complex loads are suspended at the end of a six-degree-of-freedom motion platform; Set up a closed-loop reproduction mode for actual flight conditions and a parameterized safety boundary search mode. Generate the set of operating conditions parameters for the current test based on the selected mode, and configure a real complex load and a virtual UAV model based on the set of operating conditions parameters. Obtain platform commands with corresponding timestamps, the actual pose of the platform, and the six-dimensional reaction force transmitted from the real complex payload to the UAV side; The platform execution delay is determined based on the time relationship between the platform command and the actual platform pose. The time correspondence between the six-dimensional reaction force and the actual platform state when the six-dimensional reaction force is generated is established based on the timestamp. The actual platform state and the six-dimensional reaction force are compensated for the time delay. The six-dimensional reaction force that completes the time delay compensation is input as an external load into the virtual UAV model to update the virtual UAV state. The virtual UAV model then calculates the rotor thrust and control torque based on the updated virtual UAV state and the set of operating parameters. Based on the updated virtual drone status, sling installation geometry, and platform execution delay, the virtual sling status is predicted to the moment when the six-degree-of-freedom motion platform is expected to execute the next control cycle command. The platform motion command for the next control cycle is generated and the six-degree-of-freedom motion platform is driven to move, so that the real complex load generates a new six-dimensional reaction force to form a continuous force coupling closed loop. The validity of the current test, failure events, and hard safety thresholds are determined based on the load response, virtual UAV status, and platform operation status in the continuous force coupling closed loop. The validity determination result is output in the actual flight condition closed loop reproduction mode. In the parameterized safety boundary search mode, at least one parameter in the condition parameter set is adjusted based on the valid test results to determine the safety boundary envelope of complex loads on agricultural UAVs.

2. The method for detecting the safety boundary of complex load coupling in agricultural unmanned aerial vehicles according to claim 1, characterized in that: The virtual drone model includes an airframe dynamics calculation module and a flight control algorithm module; In the closed-loop reproduction mode of the actual flight condition, the set of operating condition parameters includes the mission reference obtained from the controlled actual flight, the initial state of the virtual UAV, maneuvering events, wind disturbance events and corresponding event timestamps. The initial state of the virtual UAV is input into the airframe dynamics calculation module. The mission reference input into the flight control algorithm module is updated according to the maneuvering events and corresponding event timestamps. The wind disturbance events are input into the aerodynamic disturbance channel of the virtual UAV model according to the corresponding event timestamps. In the parameterized safety boundary search mode, the set of operating parameters includes load configuration parameters and operating parameters. At least one of the following is used as the load configuration parameter: liquid filling rate, center of gravity offset, lifting point configuration, and sling length. This is used to configure the real complex load, lifting equipment, and sling before the start of the current trial. Horizontal maneuvering acceleration is used as the operating parameter and input into the flight control algorithm module to form a mission reference. Crosswind speed is used as the operating parameter and input into the aerodynamic disturbance channel. The six-dimensional reaction force that has completed time delay compensation is input as an external load into the body dynamics calculation module to update the virtual UAV state. The flight control algorithm module then calculates the rotor thrust and control torque based on the updated virtual UAV state and the task reference corresponding to the operating condition parameter set.

3. The method for detecting the safety boundary of complex load-coupling of agricultural unmanned aerial vehicles according to claim 1, characterized in that, The real complex loads include at least one of the following: half-full liquid agricultural loads, offset agricultural loads, and agricultural loads using a two-point suspension configuration or a multi-point suspension configuration.

4. The method for detecting the safety boundary of complex load coupling in agricultural unmanned aerial vehicles according to claim 3, characterized in that, The real complex load adopts a dual-point suspension configuration or a multi-point suspension configuration. Each cable branch is equipped with a tension sensor to acquire the branch tension with an active timestamp. The tension of each branch is aligned to a unified calculation time using the same time correspondence and time delay compensation method as the six-dimensional reaction force. Based on the deviation between the current load ratio of each branch and the static reference load ratio, the difference between the measured branch tension and the quasi-static branch tension baseline, the comparison results of the branch tension and the slack rope threshold, and the rapid rise of the branch tension or six-dimensional reaction force after slack rope, tension imbalance, single branch slack rope, and retensioning impact are identified.

5. The method for detecting the safety boundary of complex load-coupling of agricultural unmanned aerial vehicles according to claim 1, characterized in that, The platform execution delay is obtained by applying commands to the six-degree-of-freedom motion platform during the no-load calibration phase and determining the time offset when the correlation between the platform command change and the actual pose change is greatest within a sliding time window. The effective physical moment of the six-dimensional reaction force is determined based on the source timestamp, data arrival timestamp, and digital filter group delay. Adjacent feedback samples are searched in the annular buffer of the platform's actual pose based on the effective physical moment, and the actual state of the platform when the six-dimensional reaction force is generated is determined by position interpolation and attitude interpolation. The delay compensation adopts a discrete Kalman predictor or a phase lead compensation method with high-frequency gain limitation. When the prediction time exceeds a preset upper limit, the timestamp is abnormal, or there is continuous packet loss, the delay compensation is stopped and the platform enters a safe state.

6. The method for detecting the safety boundary of complex load coupling in agricultural unmanned aerial vehicles according to claim 2, characterized in that, The body dynamics calculation module updates the virtual UAV state according to the following translational and rotational dynamics equations: ； ； in, For the unified solution time, For virtual drone quality, Accelerating the center of mass of the virtual drone The vector of gravitational acceleration in the inertial coordinate system. This is the attitude mapping matrix from the body coordinate system to the inertial coordinate system. The rotor thrust generated by the flight control algorithm. To complete the translational force component in the six-dimensional reaction force for time delay compensation, Let the rotational inertia matrix of the virtual drone relative to its center of mass be . For the virtual drone angular velocity, For virtual drone angular acceleration, The control torque generated by the flight control algorithm. The torque component in the six-dimensional reaction force used to complete time delay compensation.

7. The method for detecting the safety boundary of complex load coupling in agricultural unmanned aerial vehicles according to claim 1, characterized in that, The platform commands, the actual pose of the platform, the six-dimensional reaction force, and the tension of each branch when acquiring the tension of multiple branches are recorded with corresponding generation timestamps, feedback timestamps, or source timestamps under the established unified time reference. The hard safety threshold includes at least the allowable tension of any branch when multiple sling branches exist, the allowable six-dimensional reaction force of the lifting point and the lifting device, the virtual UAV attitude limit, the load lateral sweep and tilt angle boundaries, the platform workspace and kinematic limitations, and the failure of critical communication or sensors; the six-dimensional reaction force and the tension of each branch when acquiring the tension of multiple branches respectively retain the original signal for hard safety threshold monitoring and the finite bandwidth signal for the continuous force coupling closed loop, the original signal is input to an independent safety controller or a hard-wired safety loop, and the filtering processing of the finite bandwidth signal does not delete the impact peak value in the original signal.

8. The method for detecting the safety boundary of complex load-coupling forces in agricultural unmanned aerial vehicles according to claim 1, characterized in that, In the parameterized safety boundary search mode, at least one parameter in the set of working condition parameters includes horizontal maneuvering acceleration and crosswind speed. When the liquid fill rate, center of gravity offset, lifting point configuration, and sling length are fixed, horizontal maneuvering acceleration and crosswind speed are used as search parameters. Each parameter combination point is tested repeatedly at least three times. If no failure event or hard safety threshold is triggered in the repeated tests, the corresponding parameter combination point is determined as a safe point. If any hard safety threshold is triggered, the corresponding parameter combination point is determined as a failure point. If a recoverable short-term slack rope, tension imbalance, or non-divergent oscillation occurs, the corresponding parameter combination point is determined as a critical point. The search step size is reduced between the last safe point and the first failure point, and the obtained conservative boundary points are connected to form a two-dimensional safety boundary envelope. Multiple sets of two-dimensional safety boundary envelopes are superimposed to form a three-dimensional safety boundary envelope by changing the liquid fill rate or center of gravity offset, and the dominant failure mode is marked in the safety boundary envelope.

9. A safety boundary detection system for complex load coupling of agricultural unmanned aerial vehicles, characterized in that, include: The workstation is used to configure the closed-loop reproduction mode or parameterized safety boundary search mode for the actual flight conditions, and to generate the mode configuration package and condition parameter set for the current trial. A six-degree-of-freedom motion platform configured to suspend real and complex loads at the end; A unified clock and synchronous data acquisition module is used to synchronously sample the six-dimensional reaction force transmitted from the real complex load to the UAV side and generate measurement data with active timestamps. The time delay compensation module is used to establish the time correspondence between the six-dimensional reaction force and the actual state of the platform when the six-dimensional reaction force is generated, align the actual state of the platform and the six-dimensional reaction force to a unified solution time, and predict the virtual suspension point state to the time when the six-degree-of-freedom motion platform is expected to execute the next control cycle command. The virtual drone model is used to receive the six-dimensional reaction force after time delay compensation as an external load to update the virtual drone state, and to calculate the rotor thrust and control torque based on the updated virtual drone state and the set of operating parameters. The real-time control module is used to generate platform motion commands for the next control cycle based on the predicted virtual lifting point status, so as to drive the real complex load to generate a new six-dimensional reaction force and form a continuous force coupling closed loop. The workstation is also used to output the validity determination result of the reproduction in the closed-loop reproduction mode of the actual flight condition, and to adjust at least one parameter in the condition parameter set according to the valid test results in the parameterized safety boundary search mode, so as to determine the safety boundary envelope of the complex load of the agricultural unmanned aerial vehicle.

10. The complex load-coupling safety boundary detection system for agricultural unmanned aerial vehicles according to claim 9, characterized in that, The virtual drone model includes a body dynamics calculation module and a flight control algorithm module; The real complex loads include at least one of the following: half-full liquid agricultural loads, offset center of gravity agricultural loads, and agricultural loads using a two-point hanging configuration or a multi-point hanging configuration. In the closed-loop reproduction mode of the actual flight condition, the airframe dynamics calculation module receives the initial state of the virtual UAV obtained from the controlled actual flight, the flight control algorithm module receives the mission reference obtained from the controlled actual flight and the mission reference updated according to the maneuvering event and the event time stamp, and the aerodynamic disturbance channel of the virtual UAV model receives the wind disturbance event obtained from the controlled actual flight. In the parameterized safety boundary search mode, the real complex load receives at least one configuration parameter among liquid fill rate and center of gravity offset, the lifting gear and sling receive at least one configuration parameter among lifting point configuration and sling length, the flight control algorithm module receives horizontal maneuvering acceleration for forming mission reference, and the aerodynamic disturbance channel receives crosswind speed.

Images