Hierarchical Coordination Control Method for Unmanned Aerial Vehicles

Abstract

This invention discloses a hierarchical coordinated control method for unmanned aerial vehicles (UAVs), relating to the field of UAV control technology. The method includes real-time monitoring of flight state parameters, dividing the flight mission into multiple flight modes based on a finite state machine, and introducing a forward prediction window to output advance warning signals. An improved power-following strategy is employed, dynamically allocating engine output power and battery charging / discharging power according to the relationship between total power demand and the engine's optimal operating range, while limiting the engine power change rate to suppress power surges. In transitional flight modes, a gradual tilt control law is used, causing the propulsion vector angle to change with airspeed according to a Sigmoid curve. Attitude angle error feedback correction is introduced, and the gain coefficient of the feedback correction is dynamically adjusted based on the ratio of the current airspeed to the wing stall speed. This achieves smooth mode switching, efficient energy management, and safe tilt transitions for an internal combustion engine-extended-range electric vertical takeoff and landing (EVVT-i) fixed-wing UAV throughout its entire flight envelope.

Description

Technical Field

[0001] This invention relates to the field of unmanned aerial vehicle (UAV) control technology, and more particularly to a hierarchical coordinated control method for UAVs. Background Technology

[0002] Internal combustion engine-powered electric vertical takeoff and landing (VTOL) fixed-wing drones combine the advantages of vertical takeoff and landing with long endurance flight, attracting widespread attention in the industrial drone field. The power system of these drones consists of an internal combustion engine, generator, battery, and multiple electric thrusters, with control involving multiple aspects such as flight mode switching, energy management, and thrust vectoring.

[0003] The existing control methods for this type of drone mainly have the following problems:

[0004] Flight mode switching is usually based on simple state threshold judgment. After the switching command is issued, there is a significant time lag in the power response of the engine and motor, which leads to a mismatch between power output and demand during the mode transition, and can easily cause voltage fluctuations or thrust fluctuations.

[0005] The power distribution strategy is rather crude and fails to effectively coordinate the power flow between the engine and the battery. During flight phases with large fluctuations in power demand, the engine often deviates from its efficient operating range, resulting in increased fuel consumption. At the same time, the drastic power changes also impact the electrical system.

[0006] Tilting control during the transitional flight phase mostly adopts preset linear or piecewise rules, lacking the ability to adaptively adjust to changes in parameters such as airspeed and angle of attack during actual flight, making it difficult to guarantee attitude stability and stall safety under different flight conditions.

[0007] Furthermore, in special operating conditions such as sensor malfunctions or near stall, existing control methods often lack effective fault tolerance and protection mechanisms, resulting in insufficient safety margins for drones. Summary of the Invention

[0008] Based on the above problems, the purpose of this invention is to provide a hierarchical coordinated control method for unmanned aerial vehicles (UAVs). Through hierarchical coordinated control and multimodal adaptive strategies, it achieves smooth mode switching, efficient energy management, and safe tilt transition of internal combustion engine extended-range electric vertical take-off and landing fixed-wing UAVs throughout the entire flight envelope.

[0009] This invention provides a hierarchical coordinated control method for unmanned aerial vehicles (UAVs), characterized in that it is applied to internal combustion engine-powered extended-range electric vertical takeoff and landing (VTOL) fixed-wing UAVs. The method is executed under a hierarchical control architecture and includes:

[0010] Real-time monitoring of flight status parameters; division of flight missions into multiple flight modes based on finite state machines; introduction of forward prediction windows; prediction of mode switching requirements based on flight mission profile and system response delay; and output of advance warning signals.

[0011] An improved power follower strategy is adopted to dynamically allocate engine output power and battery charging and discharging power according to the relationship between total power demand and the engine's optimal operating range, and to limit the rate of change of engine power to suppress power surges.

[0012] In the transitional flight mode, a gradual tilt control law is adopted, which makes the propulsion vector angle change with the airspeed according to the Sigmoid curve, introduces attitude angle error feedback correction, and dynamically adjusts the gain coefficient of the feedback correction according to the ratio of the current airspeed to the wing stall speed.

[0013] Compared with existing technologies, the beneficial effects of this invention include at least the following: by combining a finite state machine with a forward prediction window, accurate division and early prediction of flight modes are achieved, effectively overcoming the mode switching impact problem caused by the response delay of the internal combustion engine range extender system; by dynamically allocating engine and battery power and limiting the rate of power change through an improved power following strategy, mechanical wear and grid fluctuations caused by sudden increases and decreases in range extender power are avoided, improving the stability of the energy system and fuel economy; in the transition flight mode, a stepwise tilt control law based on the Sigmoid curve is adopted, and an airspeed-related attitude error feedback gain adjustment is introduced, which makes the propulsion vector angle change smooth and has strong anti-disturbance capability, especially in the low-speed transition phase, it can significantly suppress attitude oscillation, improving the flight quality of the UAV and the safety within the stall boundary. Attached Figure Description

[0014] Figure 1 This is a schematic diagram of a hierarchical coordinated control method for unmanned aerial vehicles (UAVs) according to an embodiment of the present invention.

[0015] Figure 2 This is a schematic diagram of the layered architecture of an embodiment of the present invention;

[0016] Figure 3 This is a schematic diagram of flight mode conversion and operating mode mapping according to an embodiment of the present invention;

[0017] Figure 4 This is a schematic diagram of the power follower control logic according to an embodiment of the present invention;

[0018] Figure 5 This is a schematic diagram of the relationship between the propulsion vector angle and airspeed during the transition flight phase of an embodiment of the present invention. Detailed Implementation

[0019] Exemplary embodiments will now be described more fully with reference to the accompanying drawings. However, these exemplary embodiments can be implemented in many forms and should not be construed as limited to the embodiments set forth herein; rather, they are provided to make the invention more comprehensive and complete, and to fully convey the concept of the exemplary embodiments to those skilled in the art. The same reference numerals in the drawings denote the same or similar structures, and therefore repeated descriptions of them will be omitted.

[0020] The terms used to express position and direction in this invention are illustrated with reference to the accompanying drawings, but changes can be made as needed, and all such changes are included within the scope of protection of this invention.

[0021] Example 1, referring to Figure 1 and attached Figure 2 This embodiment provides a hierarchical coordinated control method for unmanned aerial vehicles (UAVs), applied to an internal combustion engine-powered extended-range electric vertical takeoff and landing (VTOL) fixed-wing UAV. The method includes:

[0022] Real-time monitoring of flight status parameters; division of flight missions into multiple flight modes based on finite state machines; introduction of forward prediction windows; prediction of mode switching requirements based on flight mission profile and system response delay; and output of advance warning signals.

[0023] An improved power follower strategy is adopted to dynamically allocate engine output power and battery charging and discharging power according to the relationship between total power demand and the engine's optimal operating range, and to limit the rate of change of engine power to suppress power surges.

[0024] In the transitional flight mode, a gradual tilt control law is adopted, which makes the propulsion vector angle change with the airspeed according to the Sigmoid curve, introduces attitude angle error feedback correction, and dynamically adjusts the gain coefficient of the feedback correction according to the ratio of the current airspeed to the wing stall speed.

[0025] The method is executed under a hierarchical control architecture, including a task decision layer, an energy management layer, and a bottom execution layer;

[0026] The task decision layer is used to determine the system's operating mode; the energy management layer is used to allocate power and formulate range extender commands; and the bottom execution layer performs closed-loop control of motor speed / torque at a frequency of 100-500Hz to compensate for the internal combustion engine's response delay.

[0027] The flight modes include vertical takeoff mode, transitional flight mode, cruise mode, vertical landing mode, emergency return mode, and nearest landing mode;

[0028] The specific method for dividing flight modes based on finite state machines is as follows:

[0029] The entry conditions for vertical takeoff mode are defined as follows: the airspeed is less than the takeoff airspeed threshold and the vertical speed is greater than the ascent vertical speed threshold. For example, the takeoff airspeed threshold is 5 m / s and the ascent vertical speed threshold is 2 m / s.

[0030] The entry conditions for the transition mode are defined as follows: the airspeed is between the transition low-speed threshold and the transition high-speed threshold, and the thrust vector angle is within the range of 0 to 90 degrees. In this invention, the thrust vector angle refers to the angle between the thrust direction of the propulsion device (including rotors or propellers) driven by the tilt mechanism and the longitudinal axis of the UAV fuselage (from tail to nose). When the thrust vector angle is 0°, the thrust direction is parallel to the longitudinal axis of the fuselage and forward, suitable for fixed-wing cruise flight; when the thrust vector angle is 90°, the thrust direction is perpendicular to the longitudinal axis of the fuselage and upward, suitable for vertical takeoff and landing hovering. In the transition flight mode, the thrust vector angle continuously varies between 0° and 90°.

[0031] The entry conditions for cruise mode are defined as follows: the airspeed is greater than the cruise airspeed threshold and the propulsion vector angle is equal to 0 degrees. For example, the cruise airspeed threshold is 15 m / s.

[0032] Hysteresis intervals are dynamically set between each mode based on the flight status. The hysteresis intervals are related to the sensor measurement accuracy to suppress mode jitter; for example, the hysteresis amount is ≥ twice the state measurement noise.

[0033] The method further includes:

[0034] In vertical flight mode or transitional flight mode, a tilt mechanism or distributed rotor is used to provide lift and attitude control. The thrust response bandwidth of the electric motor is configured to be no less than 20Hz to meet the hovering stability requirements.

[0035] In fixed-wing mode, cruise propulsion is undertaken by the main thruster driven by the range extender system, and the distributed electric rotor is configured to be foldable or stationary to reduce cruise drag.

[0036] For example, taking a vertical takeoff and landing (VTOL) fixed-wing UAV with a maximum takeoff weight of 35 kg as an example, this section specifically illustrates how the hierarchical coordinated control method is applied in the vertical takeoff mode. The UAV's power system includes: a 120cc two-stroke gasoline engine (rated power 8kW, peak power 10kW, dry weight 2.89kg), a 6kW permanent magnet synchronous generator (efficiency 92%, weight 3kg), a set of 6S Li-ion 21700 power batteries (22.2V / 32000mAh, 2 parallel 6 series, rated discharge capacity 10C, instantaneous 15C), a 3kW (peak power 10kW) main propulsion brushless DC motor, and four 1.5kW VTOL motors distributed at the front and rear ends of the wing linkages. In addition, the system is equipped with an electromagnetic clutch to achieve power decoupling and engagement between the engine and the main propulsion motor.

[0037] The control architecture is divided into three layers: the mission decision layer is responsible for deciding the system's operating mode based on flight mode, battery state of charge (SOC), and engine health status; the energy management layer is responsible for allocating the output power of the internal combustion engine-generator set and the charging and discharging power of the battery, and formulating commands for range extender start-stop and speed regulation; the bottom execution layer executes closed-loop control of motor speed / torque at a frequency of 100–500Hz to compensate for the delay in internal combustion engine response.

[0038] The specific steps of the vertical takeoff mode control process are as follows:

[0039] The mission decision layer determines that takeoff conditions are met based on the current flight mode (ground standby) and battery SOC (above 30%), and sets the system operating mode to vertical takeoff preparation. The energy management layer controls the electromagnetic clutch to be disengaged, the engines to shut down, and the power battery to supply power solely to the avionics and vertical takeoff and landing motors. The bottom execution layer maintains zero torque output from each motor, awaiting takeoff command.

[0040] Upon receiving the takeoff command, the mission decision layer switches the flight mode from ground standby to vertical takeoff mode and sends a mode switching signal to the energy management layer and the bottom execution layer. The bottom execution layer immediately activates the four vertical takeoff and landing motors, outputting rated power with a closed-loop control system having a thrust response bandwidth of no less than 20Hz to drive the UAV vertical climb. Simultaneously, the energy management layer issues an engine start command: the engine is driven to 2500rpm idle speed via an electric starter and maintained for 30 seconds to reduce friction and stabilize lubrication. During the warm-up period, the battery continuously discharges to meet the instantaneous high power requirements of vertical climb (up to 6kW). The climb phase ends after the UAV stably climbs to an altitude of 15m.

[0041] Once the mission decision layer detects that the altitude has reached the threshold and the engine warm-up is complete, it transitions to the transitional flight preparation sub-phase. The energy management layer increases the engine speed command to 5000 rpm and engages the electromagnetic clutch, enabling the generator to output DC power. At this point, the generator begins supplying power to the DC bus, forming a parallel power supply network with the battery. The energy management layer employs an improved power-following strategy: based on the relationship between the total power demand in the vertical takeoff mode (approximately 4kW for vertical motor hovering, plus other loads) and the engine's optimal operating range (5000–6500 rpm corresponding to approximately 4–6kW high-efficiency range), it dynamically allocates the generator output power (approximately 3kW) and the battery supplemental power (approximately 1kW), while limiting the generator power change rate to no more than 500W / s to suppress power surges to the bus.

[0042] The mission decision layer introduces a forward prediction window (e.g., predicting the flight state for the next 2 seconds in 0.5-second increments). As the airspeed gradually increases from 0 to near the transition low-speed threshold (e.g., 5 m / s), the system outputs an advance warning signal based on the flight mission profile (vertical climb to 15 m followed by horizontal acceleration) and the range extender response delay (approximately 1–1.5 seconds for the engine to accelerate from 5000 rpm to cruise speed), indicating that it is about to enter the transition flight mode. The energy management layer then pre-increases the engine speed and main propulsion motor power accordingly.

[0043] Upon receiving the warning signal, the underlying execution layer gradually increases the power of the main propulsion motor to 5kW. Simultaneously, the VTOL motors begin to tilt forward according to a gradual tilt control law: the thrust vector angle changes with airspeed according to a Sigmoid curve, and attitude angle error feedback correction is introduced. The feedback gain coefficient is dynamically adjusted based on the ratio of the current airspeed to the wing stall speed (assumed to be 10m / s). When the airspeed accelerates to 15m / s (exceeding the cruise airspeed threshold) and the thrust vector angle returns to zero, the mission decision layer determines that the vertical takeoff mode has ended and officially switches to cruise mode. At this time, the energy management layer adjusts the generator output power to the cruise requirement (approximately 2–3kW), the battery switches to low-current charging or float charging, and the VTOL motors automatically fold or stop to reduce cruise drag.

[0044] The effects of the above technical solution are as follows:

[0045] The task decision-making layer rationally plans the timing of engine start-up and warm-up to avoid the battery bearing high power output for a long time alone; the energy management layer adopts an improved power following strategy to ensure that the engine always operates in the optimal fuel consumption range, and uses the generator and battery to supply power in parallel, which reduces the battery discharge pressure (the peak discharge current is less than 10C and does not exceed the rated capacity), while limiting the power change rate and avoiding the impact of current peaks on the generator and battery.

[0046] By combining a finite state machine with a forward prediction window, precise division and early prediction of flight modes are achieved, effectively overcoming the mode switching impact problem caused by the response delay of the internal combustion engine range extender system. An improved power-following strategy dynamically allocates engine and battery power and limits the rate of power change, avoiding mechanical wear and grid fluctuations caused by sudden increases and decreases in range extender power, thus improving the stability of the energy system and fuel economy. In transitional flight modes, a gradual tilt control law based on the Sigmoid curve is adopted, and airspeed-related attitude error feedback gain adjustment is introduced, resulting in smooth propulsion vector angle changes and strong anti-disturbance capabilities. Especially during the low-to-medium speed transition phase, it significantly suppresses attitude oscillations, improving the UAV's flight quality and safety within the stall boundary. These three features work synergistically, enabling the internal combustion engine-range-extended electric vertical takeoff and landing fixed-wing UAV to achieve a comprehensive control effect of smooth mode switching, efficient energy management, and safe tilt transition throughout the entire flight envelope.

[0047] Throughout the vertical takeoff process, the bottom-level execution layer uses a high-frequency closed-loop control of the motor speed / torque at 100–500Hz to quickly compensate for fluctuations in internal combustion engine power (such as airspeed changes caused by wind disturbances), ensuring hovering stability (thrust response bandwidth ≥20Hz). Simultaneously, a hysteresis interval (correlated with sensor measurement accuracy) is set based on finite state machine mode partitioning, effectively suppressing repeated mode jumps caused by measurement noise and improving the robustness of the control system.

[0048] By rationally allocating the power levels of the range extender, generator, battery, and motor, this 35kg-class UAV achieves a combination of vertical takeoff and landing (VTOL) and long-endurance cruise. The range extender intervenes in a timely manner during VTOL takeoff, eliminating the need for an excessively large battery capacity (only 32,000mAh is sufficient for mission requirements). Furthermore, the folding rotor design further reduces cruise drag, demonstrating the supporting role of layered control in the overall design.

[0049] In one possible implementation, the method further includes the step of controlling the power flow to smoothly transition between multiple operating modes based on flight mode and battery state of charge:

[0050] The various operating modes include pure electric mode, range-extended mode, combined power supply mode, charging mode, and energy recovery mode;

[0051] See attached document Figure 3 Specifically: in vertical takeoff and vertical landing modes, pure electric mode or energy recovery mode is preferred; in transitional flight mode, combined power supply mode is preferred; in cruise mode, range-extended mode or charging mode is preferred; in emergency return mode, emergency return mode is preferred, which is a combination of range-extended mode and pure electric mode, and the power output limit is locked to ensure range.

[0052] A power-weighted transition filter is used during mode switching, where the first derivative of the transition function is continuous.

[0053] When exiting the combined power supply mode or entering the charging mode, active bus voltage balancing is first performed: the difference between the DC bus voltage output by the generator after rectification and the battery pack voltage is detected. If the difference exceeds a preset first voltage threshold, the voltage difference is gradually reduced to below a preset second voltage threshold by adjusting the generator excitation current or the DC converter, and then the power switch is closed or opened, wherein the second voltage threshold is less than or equal to the first voltage threshold. Before the voltage difference is reduced to within the second voltage threshold, the mode switching is in a waiting state, and the power switch is prohibited from operating.

[0054] For example, based on the aforementioned hierarchical coordinated control method, a specific implementation method for achieving smooth power flow transition between multiple operating modes according to flight mode and battery state of charge (SOC) is further provided. The UAV's power system is as described above: a 120cc two-stroke gasoline engine, a 6kW permanent magnet synchronous generator, a 22.2V / 32000mAh power battery, a 3kW main propulsion motor, and four 1.5kW vertical takeoff and landing motors, equipped with an electromagnetic clutch and a bidirectional DC-DC converter (for battery charging and discharging voltage matching). In the control system, the mission decision layer continuously monitors the current flight mode, battery SOC, and generator status, while the energy management layer determines the system's operating mode according to preset rules and executes mode switching operations.

[0055] The system defines five working modes:

[0056] Pure electric mode: The engine is off, the generator is not working, the electromagnetic clutch is disengaged, and the power battery provides power only. This mode is used for the initial stage of vertical takeoff, the later stage of vertical landing, or low-power cruise.

[0057] Range extender mode: The engine drives the generator to generate electricity and supplies power to the DC bus independently. The battery is neither charged nor discharged (or float charged). It is suitable for cruising when the engine is in its optimal fuel consumption range and the battery SOC is moderate.

[0058] Combined power supply mode: The engine / generator set and the power battery supply power to the bus at the same time to facilitate the transition of high power demand phases such as flight mode or climb.

[0059] Charging mode: When the generator output power is greater than the load demand, the excess power is charged to the battery in a constant current / constant voltage manner through a bidirectional DC-DC converter. This mode is suitable for when the battery SOC is low during cruise mode.

[0060] Energy recovery mode: During vertical landing or deceleration, the vertical take-off and landing motor or the main propulsion motor acts as a generator to convert mechanical energy into electrical energy and feed it back to the battery. This mode is suitable for vertical landing.

[0061] The mission decision-making layer prioritizes flight modes and battery SOC according to the following rules:

[0062] Vertical takeoff mode (airspeed < 5 m / s, vertical speed > 2 m / s): Pure electric mode is preferred; if the battery SOC is below 30% and the takeoff command has been issued, the combined power supply mode is entered in advance (but the engine power climb rate must be limited).

[0063] Vertical landing mode: Energy recovery mode is used first, and pure electric mode is used as a supplement according to landing power requirements; when the altitude is lower than the safe hovering height (e.g., 3m), it is forced to switch to pure electric mode to ensure response speed.

[0064] Transitional flight mode (airspeed 5–15 m / s, thrust vector angle 0–90°): The combined power supply mode is preferred, with the generator providing the base power (approximately 3–5 kW) and the battery supplementing the peak power (approximately 2–3 kW) to ensure smooth forward acceleration.

[0065] Cruise mode (airspeed ≥ 15 m / s, thrust vector angle 0°): Range extender mode (when SOC is between 40% and 80%) or charging mode (when SOC < 40%) is preferred. If SOC > 85%, switch to pure electric mode and shut down the engine to save fuel.

[0066] Emergency Return Mode: This mode is triggered by the mission decision-making layer when an engine failure, extremely low battery SOC (<15%), or critical sensor failure is detected. It forces the use of a combination of range-extended mode and pure electric mode, where the engine continuously generates electricity at maximum safe power (not exceeding 6kW), while the battery replenishes it with a small current not exceeding 0.5C, and the total power output limit is locked (e.g., ≤7kW), prioritizing return range over maneuverability.

[0067] To prevent drastic bus voltage fluctuations or mechanical shocks caused by sudden power command changes during mode switching, the energy management layer employs a power-weighted transition filter during each mode switch. Specifically, this is implemented as follows:

[0068] Set target power value Actual output power Update according to the following transition function:

[0069]

[0070] Where t is a time variable, and its value ranges from arrive + ; This is a time-weighted function with a continuous first derivative. In this embodiment, a cosine S-shaped curve is used.

[0071]

[0072] Transition time constant Dynamically set according to mode differences: for example, the switching time between pure electric and combined power supply is 0.5 seconds, the switching time between range-extended and charging modes is 1 second, and the emergency return mode switching time is 0.3 seconds (fast response but continuous derivative).

[0073] The target power value refers to the desired power command determined by the energy management layer based on the current operating mode, flight mode, battery state of charge (SOC), and total power demand. This command can be for generator output power (e.g., requiring the generator to output 3kW), total DC bus power, or battery charging / discharging power. In practice, it can be uniformly converted to the DC bus power target value to facilitate coordinated operation of all execution units. Actual output power represents the power command value actually applied to the execution layer (such as the generator controller or motor driver) after a smooth transition.

[0074] This filter allows the power command to smoothly transition from the old value to the new value, avoiding engine speed overshoot or motor current spikes caused by abrupt changes.

[0075] When mode switching involves the parallel connection or disconnection of the generator and battery (e.g., exiting from combined power supply mode to charging mode, or switching back from charging mode to range extender mode), the operation of the electromagnetic clutch or power switching transistor (MOSFET) may generate a large current surge due to the mismatch between the bus voltage and the battery voltage. This embodiment designs an active bus voltage balancing process. Taking an actual flight mission as an example: the UAV uses charging mode in cruise mode (battery SOC=35%), and the generator outputs 6kW, of which 3kW is used for main propulsion and 3kW is charged into the battery in a constant current manner. When the SOC rises to 70%, the mission decision layer decides to switch to range extender mode (generating power only to meet the load, and stopping charging). At this time, the energy management layer performs the following steps:

[0076] Voltage difference detection: The DC bus voltage after generator rectification is acquired through a voltage sensor. With battery pack voltage (Measured via the front end of the bidirectional converter). Assume at this time... =28.5V, =25.2V (due to the increase in battery polarization voltage caused by charging current), difference ΔV=3.3V. The first voltage threshold is preset to 1.5V.

[0077] Voltage adjustment: Because the voltage difference ΔV > 1.5V, disconnecting the charging circuit is prohibited. The energy management layer first gradually reduces the generator excitation current to lower the generator's output power, while simultaneously switching the bidirectional DC-DC converter from charging mode to voltage follower mode. The voltage drops slowly. Simultaneously, the battery management system allows for a short, low-current discharge to balance the voltage. After approximately 0.8 seconds of adjustment, The voltage dropped to 25.8V and stabilized at 25.6V, with a difference of ΔV=0.2V, which is lower than the second voltage threshold (preset to 0.5V).

[0078] Switching Execution: The second voltage threshold is 0.5V. Once this condition is met, the energy management layer issues a command to disconnect the generator-side power switch, simultaneously switching the bidirectional converter to standby mode. The electromagnetic clutch remains engaged (the engine still drives the generator, but the load is 0 or minimal). At this time, the battery has stopped charging, and the bus voltage is maintained at approximately 25.6V by the generator, allowing the system to smoothly enter range-extending mode. Throughout the standby period, the power switch is disabled, and the enable signal of the switch drive circuit is locked by hardware logic and the voltage comparator output to ensure safety.

[0079] Conversely, when exiting the combined power supply mode and entering the pure electric mode, the engine speed and generator excitation are reduced first to lower the bus voltage to near the battery open-circuit voltage (difference <0.5V) before disengaging the clutch to prevent overvoltage damage caused by a sudden rise in generator side voltage.

[0080] In emergency return mode, if it is necessary to switch from combined power supply mode to emergency combined mode, since it does not involve disconnecting generators but limiting the total power limit, there is no need to perform voltage balancing. The power limit can be reduced linearly directly through the power weighted transition filter.

[0081] The effects of the above technical solution are as follows:

[0082] Based on the different requirements for power response speed and range capability in different flight phases, the system automatically selects between pure electric, range-extended, and combined power supply modes. During vertical takeoff and landing, the high power density of the battery ensures hovering stability; during cruise, the high efficiency range of the range extender reduces fuel consumption; and the energy recovery mode recovers some energy during landing to extend the overall flight time.

[0083] A power-weighted transition filter with continuous first-order derivatives is used to eliminate engine speed oscillations and motor torque pulsations caused by power command jumps. The active bus voltage balancing mechanism detects voltage differences and adjusts generator excitation or converters to ensure that the voltage difference is reduced to within a safe threshold before the power switching transistors are opened or closed, thus avoiding arcing or surge currents caused by relays or MOSFETs operating under voltage differences.

[0084] In emergency return mode, the power output limit is locked and a combination of range extender and pure electric power supply is used to avoid exhausting the battery or overloading the engine due to blindly pursuing return speed.

[0085] In summary, this embodiment achieves safe, efficient, and smooth power flow control across the entire flight envelope of an internal combustion engine-powered electric vertical takeoff and landing fixed-wing UAV through precise operating mode division, smooth transition filters, and active voltage balance interlocking.

[0086] In one possible implementation, the forward prediction window is implemented in the following ways:

[0087] The system records the time of issuance of the mode switching command and the actual time when the corresponding power response reaches the target in real time, calculates the time delay between the two, and uses the recursive least squares method to identify the time delay sequence online, so as to obtain the equivalent delay time constant between the issuance of the mode switching command and the actual power response.

[0088] Set a lead time for the prediction, which is equal to the equivalent delay time constant plus a safety margin.

[0089] Read the pre-loaded flight mission profile and extract the target airspeed sequence and target altitude sequence within a time period equal to at least the predicted time lead.

[0090] Based on the current flight status parameters and the extracted target sequence, the trend of state change within the predicted time lead time is predicted; when it is predicted that from the current moment, within the predicted time lead time, any flight status parameter will first reach or exceed the preset mode switching threshold, a mode switching warning signal is immediately output; the mode switching warning signal includes at least the predicted target mode type and the estimated remaining time.

[0091] The forward prediction window also introduces a confidence verification mechanism for the uncertainty interval of prediction: when the root mean square error of the prediction parameter exceeds a preset threshold within three consecutive prediction periods, the prediction switching function is temporarily disabled, and the system returns to the normal state monitoring of the finite state machine until the error returns to within the threshold before the prediction window is enabled.

[0092] Taking a typical mission performed by a 35kg-class vertical takeoff and landing fixed-wing UAV as an example: the mission profile includes vertical takeoff (climbing to 15 meters), transition flight (accelerating to 15 m / s), cruise (15 m / s, altitude 50 meters), transition deceleration, and vertical landing. The UAV collects flight status parameters (airspeed, vertical velocity, thrust vector angle, power, etc.) every 10 milliseconds, and the underlying execution layer controls the motors in a closed loop at a frequency of 200 Hz.

[0093] The task decision layer records in real time the moment each mode switching command is issued and the moment the corresponding power response actually reaches the target. Power response reaching the target is defined as: the generator output power or motor thrust reaching more than 90% of the target value and remaining stable for at least 20 milliseconds. The difference between the two is the actual delay of this switch. For example, when switching from pure electric mode to combined power supply mode, the engine speed increases from 2500 rpm to 5000 rpm, driving the generator to output rated power. The measured delay is typically between 0.6 and 1.2 seconds, affected by factors such as engine temperature and battery state of charge.

[0094] The task decision layer maintains a fixed-length delay time series of the most recent 20 switching events. The equivalent first-order inertia constant of this series is identified online using recursive least squares. Specifically, a model is built based on the linear relationship between the delay value of the current sampling period and the delay value of the previous period. Each time a new delay sample is obtained, the covariance matrix and the estimated model parameters are updated. A forgetting factor of 0.95 is used to gradually decay the weight of older data to adapt to time-varying characteristics. The identified dominant time constant is called the equivalent delay time constant, which represents the equivalent lag between command issuance and actual power response.

[0095] The prediction lead is set as the equivalent time delay constant plus a safety margin. The safety margin is 0.3 seconds, used to cover model errors and random disturbances. For example, when the equivalent time delay constant identified online is 0.8 seconds, the prediction lead is 1.1 seconds. This lead is dynamically updated to ensure that the prediction window length is always slightly larger than the actual delay.

[0096] The mission decision layer pre-loads flight mission profile files (e.g., sequence points indexed by timestamps or range, including target airspeed, target altitude, target thrust vector angle, etc.). Based on the current flight time, it reads forward a sequence of targets with a duration at least equal to the predicted time lead. Since the data sampling period is 0.1 seconds, if the predicted time lead is 1.1 seconds, then 11 future target points need to be extracted.

[0097] Subsequently, based on the currently measured flight state parameters (airspeed, vertical velocity, thrust vector angle) and the extracted target sequence, the trend of state change within the future prediction time lead is predicted using linear extrapolation or a Kalman filter. Example of a prediction rule: If the target airspeed sequence indicates that it will increase from 8 m / s to 12 m / s in 0.8 seconds, and the currently measured airspeed is 8.5 m / s with an acceleration of approximately 0.5 m / s², then it can be predicted that the airspeed after 0.8 seconds will be approximately 8.9 m / s, which has not yet reached the transitional high-speed threshold of 15 m / s; however, it is predicted that the airspeed will exceed 15 m / s in 1.0 seconds. Therefore, the mission decision-making layer can determine at the current moment that, starting from the current moment, the airspeed will reach the transitional high-speed threshold for the first time within 1.1 seconds (i.e., at the 1.0 second).

[0098] Once it is predicted that any flight status parameter will first reach or exceed a preset mode switching threshold (e.g., airspeed greater than or equal to 15 m / s, or vertical velocity less than or equal to -2 m / s) within the predicted lead time, the mission decision layer will immediately output a mode switching warning signal. This warning signal contains at least two aspects: first, the predicted target mode type; and second, the estimated remaining time, i.e., from the current moment to the predicted moment.

[0099] The warning signal is sent to the energy management layer and the bottom execution layer. Based on this, the energy management layer adjusts the engine speed in advance (e.g., starts increasing the throttle 0.5 seconds in advance) or the generator excitation to make the power response as synchronized as possible with the mode switching; the bottom execution layer adjusts the control parameters in advance (e.g., the gain coefficient of the tilt control law).

[0100] To prevent frequent false alarms due to sensor noise or flight mission profile deviations, this method introduces a confidence verification mechanism based on root mean square error. The specific implementation is as follows:

[0101] In each prediction period (e.g., every 0.5 seconds), the root mean square error (RMSE) between the predicted and actual values ​​for the three most recent consecutive prediction periods is calculated. Taking airspeed as an example: for each prediction, the predicted airspeed value 0.5 seconds in the future is subtracted from the measured airspeed value 0.5 seconds later. The error values ​​for three consecutive periods are stored, and the square root of the average of the sum of the squares of these three error values ​​is calculated as the RMS error.

[0102] A preset error threshold is set to 1.0 m / s for airspeed (other parameters are set proportionally). When the root mean square error exceeds this threshold for three consecutive prediction cycles (i.e., within a 1.5-second period), the mission decision layer determines that the forward prediction is unreliable and temporarily disables the prediction switching function. At this time, the mission decision layer reverts to the conventional state monitoring mode of the finite state machine, that is, it only switches modes based on whether the currently measured parameters have reached the threshold, and no longer outputs a warning signal.

[0103] During this period, the recursive least squares algorithm continues to identify the delay time online, but the prediction function is locked. Once the root mean square error of the subsequent three consecutive prediction cycles falls below the threshold (e.g., the error drops to 0.7 m / s, 0.6 m / s, and 0.8 m / s respectively), the task decision layer automatically re-enables the prediction window and clears the disabled flag.

[0104] During flight testing, when encountering sudden gusts of wind causing severe fluctuations in airspeed, the root mean square error of the prediction briefly increased to 1.2 m / s, and the prediction function was automatically disabled for approximately 2 seconds. After the disturbance decayed, the error recovered to 0.6 m / s, and the prediction function immediately resumed. This mechanism effectively avoids false warning signals caused by transient disturbances while ensuring high sensitivity under normal flight conditions.

[0105] The effects of the above technical solution are as follows:

[0106] A recursive least squares method is employed to identify the equivalent delay time constant online and dynamically adjust the prediction lead to ensure a precise match between the forecast time and the actual power response time. The forecast signal includes the target mode type and the estimated remaining time, providing ample preparation window for the energy management layer and the underlying execution layer. A root mean square error confidence verification mechanism is introduced to automatically downgrade to routine state monitoring when the prediction error exceeds the limit, avoiding false forecasts caused by sensor noise or flight profile deviations.

[0107] See attached document Figure 4 In one possible implementation, the improved power follower strategy includes:

[0108] Set the total power demand, battery state of charge, and optimal operating range for the engine, which includes a lower limit, an economic value, and an upper limit.

[0109] When the total power demand is within the optimal operating range, the engine output power is adjusted to be equal to the total power demand plus the charging compensation amount. This charging compensation amount is used to maintain the battery state of charge within a preset normal range. The charging compensation amount is calculated by a proportional-integral controller based on the deviation between the battery state of charge and the target midpoint value of the normal range, and its maximum amplitude is limited.

[0110] When the total power demand is lower than the lower limit of the optimal operating range, if the battery state of charge is higher than the preset high charge threshold, the engine will be shut off and the battery will supply power alone; otherwise, the engine will operate at the lower limit of the optimal operating range, and the excess power will be used to charge the battery.

[0111] When the total power demand exceeds the upper limit of the optimal operating range, the engine operates at the upper limit of the optimal operating range, and the difference in power is instantly replenished by the battery; at the same time, the rate of change of engine power is limited to not exceeding the preset rate limit.

[0112] This embodiment, based on the aforementioned hierarchical coordinated control method, further provides a specific implementation of an improved power-following strategy, using a 35kg-class vertical takeoff and landing fixed-wing UAV as an example. The UAV's internal combustion engine range extension system includes a 120cc two-stroke gasoline engine, a permanent magnet synchronous generator, a set of lithium-ion power batteries, a main propulsion brushless DC motor, and four vertical takeoff and landing motors. The energy management layer is responsible for executing the power-following strategy, while the bottom execution layer performs closed-loop control of the motors at a frequency of 200 Hz.

[0113] The energy management system pre-defines the engine's optimal operating range, which includes three key power values:

[0114] The lower limit value is denoted as This represents the lowest power point at which the engine can maintain stable combustion and has acceptable thermal efficiency (e.g., approximately 4 kW for the engine in this embodiment).

[0115] Economic value, denoted as This represents the power point at which the engine has the lowest fuel consumption rate (e.g., approximately 5.5 kilowatts).

[0116] Upper limit, denoted as This represents the highest power point at which the engine can operate sustainably without exceeding mechanical and thermal load limits (e.g., approximately 7 kilowatts).

[0117] Simultaneously, a normal range for the battery state of charge (SOC) is set, with the lower limit denoted as... The upper limit is denoted as For example, set them to 0.4 and 0.8 respectively; the target midpoint value of this interval is denoted as... For example, set it to 0.6. A proportional-integral controller is introduced for calculating the charging compensation amount, and the proportional coefficient is denoted as... The integral coefficient is denoted as The specific values ​​are pre-tuned based on the dynamic characteristics of the engine and battery (e.g., 0.2 and 0.05 respectively). The maximum limit of the charging compensation amount is denoted as... For example, set at 1 kilowatt (or no more than 15% of the engine's rated power). The engine power variation rate limit is denoted as... For example, set it to 1 kilowatt per second (or no more than 15% of the engine's rated power per second).

[0118] The energy management layer receives total power demand from the underlying execution layer in real time. (Including the sum of the power of all loads such as the main propulsion motor, vertical takeoff and landing motor, and avionics equipment) and the current battery SOC; according to With the lower limit of the engine's optimal operating range and upper limit The target engine power is determined according to the following rules based on the relationship between the engine and the target power. :

[0119] Scenario 1: The total power demand is within the optimal operating range, i.e. ≤ ≤ ;

[0120] The engine target power is set as the total required power plus the charging compensation amount. :

[0121]

[0122] The cumulative deviation between the integral term's accumulated SOC and the target midpoint value over time. Compensation amount. The amplitude is limited to - arrive Between, meaning that the load may be reduced or increased by no more than [a certain amount]. The power; through this compensation, when the SOC is lower than At this time, the engine outputs additional power to charge the battery; when the SOC is higher than 100%, the engine outputs more power to charge the battery. At this time, the engine outputs less power, and the battery provides auxiliary power, thereby realizing the SOC to... return.

[0123] Scenario 2: Total power demand is lower than the lower limit of the optimal operating range, i.e. Less than :

[0124] At this point, the energy management layer determines the battery's State of Charge (SOC): If the SOC is greater than a preset high-capacity threshold... This threshold can be equal to If the value is slightly higher than the target midpoint, for example, set to 0.6, the engine will shut down, the electromagnetic clutch will disengage, the generator will stop working, and the battery will provide power alone. For example, at the end of the vertical descent, when the total power demand is low and the battery is fully charged, the engine will stop, relying entirely on battery power to save fuel and reduce noise.

[0125] If SOC≤ The engine operates at the lower limit of its optimal operating range. Running, excess power - Used to charge the battery. For example: during the low-power gliding phase of a drone, the total power demand is less than... If the current SOC is at a moderate level, the engine will still operate at... The output is used to supply part of the load and the remainder to charge the battery in a constant current manner through a bidirectional DC-DC converter.

[0126] Scenario 3: The total power demand exceeds the upper limit of the optimal operating range, i.e. > :

[0127] The engine operates at the upper limit of its optimal operating range. Operation, differential power - Instantaneous battery replenishment. For example, during transitional flight modes, the drone needs to accelerate, and the total power demand exceeds [a certain threshold]. Then the engine uses A constant output is maintained, with any excess power replenished by battery discharge. Simultaneously, the energy management system monitors the rate of change of the engine's actual power in real time, ensuring that the change per second does not exceed a preset rate limit. If the difference between the engine's target power calculated by the energy management system and the current actual power is too large, adjustments are made gradually in multiple steps; for example, only a small step size is allowed per control cycle (step size = ...). × Control cycle duration), until the target is achieved; The preset upper limit of the engine's power change rate per second.

[0128] The above three cases can be unified into a single proportional-integral control expression with a limit. The energy management layer first calculates an intermediate target power. :

[0129]

[0130] Then, this value is limited to ensure it is not lower than... Not higher than ;

[0131] when Calculated value is lower than At that time, according to the logic of scenario two above: if SOC is greater than ,

[0132] The engine will then actually shut down (target power is 0); otherwise, it will be forced to shut down. ;when The calculated value is higher than At that time, forced to The difference is supplemented by the battery. This unified formula simplifies programming implementation while ensuring smooth power tracking.

[0133] The energy management layer calculates the engine target power in each control cycle (e.g., every 50 milliseconds). Then, compared with the current actual output power of the engine. (Based on feedback from the generator controller) A comparison is made. If the absolute value of the difference between the two exceeds the allowable single-step change... Then this instruction will be restricted to It also records the limiting indicator. This mechanism prevents drastic changes in engine speed due to sudden changes in power demand, and avoids combustion instability caused by an overly rich or lean air-fuel mixture.

[0134] The mission decision-making layer determines the basic parameters of the power-following strategy based on the current flight mode (cruise, transition, vertical, etc.) and battery SOC. For example, in emergency return-to-home mode, it can... Temporarily increase to a higher value that the engine is safe to allow (such as near peak power), while also... The power output is reduced, sacrificing some economy for power margin and return range. The energy management layer executes the aforementioned power-following algorithm, outputting engine speed commands (via electronic throttle) and generator excitation current commands, while simultaneously sending power allocation commands for the main propulsion motor and vertical takeoff and landing motor to the lower execution layer. The lower execution layer performs closed-loop control of the actual power of each motor at a higher frequency (e.g., 100 to 500 Hz), providing real-time feedback on the current total power demand and actual engine output power to the energy management layer, thus forming a closed loop.

[0135] Through clear segmented control rules, proportional-integral charging compensation, and rate-of-change limits, economical, stable, and rapid energy management of internal combustion engine-extended electric vertical take-off and landing UAVs is achieved across the entire power range.

[0136] In one possible implementation, the gradual tilt control law includes:

[0137] The relationship between the thrust vector angle and airspeed is determined using a Sigmoid curve; the corresponding base thrust vector angle is obtained by mapping the current airspeed from this Sigmoid curve.

[0138] The pitch angle θ_pitch and pitch rate ω_pitch of the UAV are acquired in real time, as well as the current pitch command. When the deviation of the pitch angle from the pitch command exceeds a preset deviation threshold, or the pitch rate exceeds a preset rate threshold, a correction increment is superimposed on the base thrust vector angle. The correction increment is linearly related to the pitch angle deviation and the pitch rate. The correction increment is obtained by multiplying the pitch angle deviation and the pitch rate by a first gain coefficient and a second gain coefficient, respectively, summing the results, and then taking the negative of the sum.

[0139] The first gain coefficient and the second gain coefficient are dynamically adjusted according to the ratio of the current airspeed to the wing stall speed: as the ratio decreases, the first gain coefficient gradually decreases and the second gain coefficient gradually increases, so that when the airspeed approaches the stall boundary, the control law mainly suppresses the pitch rate; at the same time, anti-integral saturation logic is introduced into the control loop of the tilt mechanism to prevent the integral depth caused by the saturation of the actuator.

[0140] The dynamic adjustment method of the first gain coefficient and the second gain coefficient is as follows: the first gain coefficient is positively correlated with the ratio of the current airspeed to the stall speed, and the second gain coefficient is negatively correlated with the ratio, so that as the airspeed changes from much higher than the stall speed to close to the stall speed, the control law smoothly transitions from being dominated by pitch angle deviation feedback to being dominated by pitch angle rate feedback.

[0141] During the transitional flight mode, the mission decision-making layer maps the corresponding basic thrust vector angle from the pre-stored Sigmoid curve based on the current airspeed.

[0142] like Figure 5 As shown, the relationship between the thrust vector angle and airspeed is illustrated using a Sigmoid curve. This curve is divided into three characteristic regions based on the airspeed range: the low-speed range corresponds to the vertical takeoff and landing (VTOL) zone, where the thrust vector angle tends to 90°, corresponding to VTOL and VTOL modes; the high-speed range corresponds to the fixed-wing cruise zone, where the thrust vector angle tends to 0°, corresponding to cruise mode; and the intermediate-speed range is the transition zone, where the thrust vector angle smoothly decreases from 90° to 0° as airspeed increases, corresponding to transitional flight modes. By consulting this curve, the corresponding basic thrust vector angle can be obtained based on the current measured airspeed. This Sigmoid curve exhibits a smooth, S-shaped monotonically decreasing characteristic: the change is gradual at low airspeeds, the rate of change is greatest at intermediate airspeeds, and it tends to level off again at high airspeeds. This curve shape avoids attitude disturbances caused by abrupt changes in the thrust vector angle.

[0143] During the tilting process, the pitch attitude of the UAV is prone to deviation due to changes in aerodynamic forces. To suppress this phenomenon, a correction increment is superimposed on the basic thrust vector angle in the control law. The calculation method for this correction increment is as follows:

[0144] The system acquires two pitch-related parameters of the UAV in real time: pitch angle (the angle between the aircraft's longitudinal axis and the horizontal plane) and pitch rate (the rate of change of the pitch angle over time). When the absolute value of the pitch angle exceeds a preset deviation threshold (e.g., 5 degrees), or the absolute value of the pitch rate exceeds a preset rate threshold (e.g., 15 degrees per second), attitude feedback correction is initiated. The correction increment equals the sum of the pitch angle deviation and pitch rate multiplied by their respective gain coefficients, then the negative of the sum. In other words, if the nose pitches up (positive pitch angle) or the pitch rate is too fast, the correction increment will adjust the thrust vector angle in the direction that reduces the vertical component, generating a nose-down torque; conversely, the opposite occurs.

[0145] The preset deviation threshold is set based on the UAV's static stability margin and flight quality requirements. For UAVs with a maximum takeoff weight of 35 kg, the typical range is 3° to 10°. A value that is too small will cause the attitude feedback to be overly sensitive and susceptible to turbulence noise interference; a value that is too large will fail to suppress attitude divergence in a timely manner. It is usually determined through linearized model simulation: the minimum angle that ensures the pitch attitude closed-loop system has a phase margin of not less than 45° at typical transition airspeeds is selected as the threshold. In actual flight testing, it can be fine-tuned based on the UAV's response to gusts, and finally fixed in the flight control parameters. Similar thresholds for other flight modes can be set independently with reference to the logic in this example.

[0146] For example, the preset deviation threshold is 5 degrees; the preset rate threshold is 15 degrees / second;

[0147] when At that time, a correction increment is superimposed on the base propulsion vector angle:

[0148] ;

[0149] Where Δθ is the correction increment; Z1 and Z2 are the first gain coefficient and the second gain coefficient, respectively.

[0150] When both pitch angle deviation and pitch rate are less than the threshold, the correction increment is zero, and only the basic Sigmoid mapping is used to ensure the optimal tilt trajectory under undisturbed conditions.

[0151] The two gain coefficients, pitch angle deviation gain and pitch angle rate gain, are not fixed values, but are dynamically adjusted according to the ratio of the current airspeed to the wing stall speed. The specific rules are as follows:

[0152] The pitch angle deviation gain is positively correlated with the ratio of airspeed to stall speed: the higher the airspeed (the larger the ratio), the greater the gain, and the control law focuses more on correcting pitch attitude based on the angle deviation.

[0153] The pitch rate gain is negatively correlated with this ratio: the lower the airspeed (the closer the ratio is to 1, i.e. the closer to the stall boundary), the greater the gain, and the control law focuses more on suppressing the pitch rate to prevent attitude divergence caused by pre-stall chatter.

[0154] The reason for adopting this complementary adjustment is that at high speeds, the wing aerodynamic efficiency is high, and pitch angle deviations will accumulate significant altitude errors, requiring rapid correction; at low speeds (close to stall), the aerodynamic control surface effect deteriorates, and at this time, feedback mainly based on rate damping can effectively suppress pitch oscillations caused by gusts or roll, avoiding stall induced by overcorrection.

[0155] By having the two gain coefficients change continuously and smoothly according to the above rules as the airspeed changes from much higher than the stall speed to close to the stall speed, the control law naturally transitions from being dominated by pitch angle deviation feedback to being dominated by pitch angle rate feedback, without the need for manual parameter switching.

[0156] Because tilt actuators (such as servo motors or electric actuators) have physical limits (0 to 90 degrees), integral depth saturation is prone to occur in integral controllers. This means that the actuator has reached its mechanical limit, but the error integral term continues to accumulate, requiring a long time to exit saturation when the error reverses. Therefore, this control law introduces anti-integral saturation logic into the integral stage of the tilt loop:

[0157] When the propulsion vector angle command exceeds the limit (less than 0 degrees or greater than 90 degrees) and the error signal direction is consistent with the limit direction (e.g., the command is greater than 90 degrees but the angle still needs to be increased), the accumulation of the integral term is frozen, and only the proportional and derivative terms are output.

[0158] When the instruction exits the limit or the error direction reverses, the integral accumulation immediately resumes.

[0159] Meanwhile, during the control cycle, the output of the actuator is monitored in real time to see if it is in a saturated state (i.e., whether the actual propulsion vector angle has reached the mechanical limit and the command direction still points to the limit). If it is saturated, the integral term is forcibly cleared to zero or set to a value that matches the limit to avoid subsequent response delays caused by the depth of integration.

[0160] The mission decision layer provides the current flight mode (transitional flight) and stall speed (calculated or looked up from a table based on current weight, angle of attack, etc.). The energy management layer is responsible for coordinating the overall requirements of engine power and tilt control. The bottom execution layer performs dual closed-loop control of thrust vector angle and motor speed at a frequency of no less than 200 Hz. When the airspeed approaches the stall speed, the energy management layer prioritizes ensuring that the rotor provides sufficient vertical force, while the tilt control law automatically increases the pitch rate gain. The two work together to prevent stall.

[0161] The effects of the above technical solution are as follows:

[0162] The Sigmoid curve allows the propulsion vector angle to change continuously with airspeed, avoiding the abrupt changes in angular velocity at inflection points that occur in traditional piecewise linear tilting methods, significantly reducing the amplitude of UAV attitude fluctuations. Attitude angle error feedback correction can suppress pitch deviations caused by crosswinds or turbulence in real time. When encountering a gust of wind that causes a sudden drop in airspeed, the pitch rate gain automatically increases, and the control law immediately generates a damping torque, enabling the UAV to recover stability within 0.5 seconds and avoiding divergent oscillations. The dynamic adjustment of the gain coefficient with the airspeed-to-stall rate ratio allows the control law to automatically switch to rate damping dominance when approaching the stall boundary, avoiding the risk of excessive pitching or plunging due to blindly chasing deviations under traditional fixed-gain control, which can induce stall. Anti-integral saturation logic effectively prevents the tilt servo from being in the saturation region for a long time and the integral term from accumulating infinitely, avoiding repeated "bumps" or delayed responses of the servo at the limit position. Since the adjustment of the gain coefficient depends only on the airspeed-to-stall rate ratio (a dimensionless quantity), there is no need to repeatedly calibrate the absolute value for different takeoff weights or different aircraft models. The same set of control laws can automatically adapt to different load configurations, greatly reducing the amount of debugging work.

[0163] In one possible implementation, the transitional flight mode further includes attitude correction and asymmetric compensation:

[0164] Real-time monitoring of the angle of attack sensor difference between the two wings and the actual position feedback of the tilting mechanisms on both sides;

[0165] When the difference in angle of attack between the two sides exceeds the preset angle difference threshold, or the difference in angle between the tilting mechanisms on both sides exceeds the preset mechanical angle difference threshold, the asymmetrical torque generated is compensated by adjusting the differential rotation speed of the two rotors.

[0166] The specific method is as follows:

[0167] While keeping the total lift / thrust requirement constant, the rotational speed of one rotor is increased while the rotational speed of the opposite rotor is decreased by the same amount. For example, if the angle of attack on the left side is higher than that on the right side (due to greater lift on the left wing causing a tendency to roll to the right), the rotational speed of the right rotor is increased and the rotational speed of the left rotor is decreased to generate a leftward rolling torque to balance the load.

[0168] The magnitude of the compensation torque is proportional to the square of the speed difference between the two sides. The control law calculates the required target speed difference based on the absolute value of the angle of attack difference or the propulsion vector angle difference through a proportional-integral-derivative controller; the differentiated speed adjustment acts simultaneously on both the front and rear rotors (ensuring that the pitching torque does not change significantly due to left and right adjustments).

[0169] If the required compensation torque exceeds a predetermined proportion (e.g., 80%) of the maximum compensation torque that the motor's rated thrust can generate, and the duration of this state exceeds a preset duration threshold (e.g., 2 seconds), it is determined to be a hardware failure of the mechanism (e.g., one side of the tilt servo is stuck, the rotor motor partially fails, or the sensor is severely biased). It will be forced to enter the safety mode, lock the tilt mechanism at the current angle or recover it to the vertical position to ensure the center of gravity balance, and perform an emergency landing at the nearest point in time.

[0170] The specific operations include:

[0171] Tilting mechanism locking or retraction: If the current thrust vector angle is between 0 and 90 degrees and remains stable, it is locked at the current angle; if the current thrust vector angle is unstable or the airspeed is close to the stall speed, all tilting mechanisms are retracted to the vertical position (90 degrees), and the attitude is maintained by the lift of the rotor.

[0172] Balance center of gravity: After recovering to the vertical position, the aircraft switches to a vertical flight state dominated by rotor force. At this time, the center of gravity position is automatically compensated by the flight control (through the battery mounting position or movable counterweight, if any); if there is no counterweight mechanism, the vertical hovering attitude is maintained first.

[0173] Emergency landing at the nearest safe point: Based on the current altitude, remaining battery power, and terrain data below, the mission decision-making level automatically plans the nearest safe landing point and executes a vertical descent. During the descent, the engines are shut down or maintained at minimum power, relying solely on battery power to ensure a safe landing.

[0174] The effects of the above technical solution are as follows:

[0175] In actual flight, due to manufacturing and assembly errors or crosswinds, the left and right tilt mechanisms often exhibit an angle difference of 1 to 2 degrees. Through differentiated speed compensation, the roll angle can be controlled within 3 degrees without increasing mechanical complexity, far superior to the more than 10 degrees without compensation. When one servo is stuck or the motor response is abnormal, traditional control methods often fail to maintain attitude. This embodiment quickly diagnoses hardware faults by monitoring whether the compensation torque continuously exceeds 80% of the motor's capacity and switches to a safe mode within 2 seconds, reducing the risk of loss of control. Asymmetric compensation relies on existing rotor speed differential control, eliminating the need for additional ailerons, spoilers, or other aerodynamic control surfaces, as well as high-precision mechanical synchronization mechanisms, simplifying the tilt mechanism design and reducing overall aircraft weight. Retrieving the tilt mechanism to a vertical position and prioritizing landing at the nearest available location avoids stall or spin that might occur during transitional flight under asymmetric thrust conditions.

[0176] In one possible implementation, during the transitional flight mode, the method further includes a fault-tolerant handling step for airspeed anomalies:

[0177] When the airspeed signal mutation rate is detected to exceed the preset mutation rate threshold, or when the airspeed signal is invalid, the tilt control law automatically enters the hold mode: the current thrust vector angle is locked and unchanged, while the thrust response gain of the electric rotor is increased, and the rotor directly generates the anti-disturbance torque.

[0178] If the airspeed signal is invalid, the virtual airspeed estimator is activated to calculate the current airspeed estimate based on the ground speed and wind field estimation model. If the difference between the estimated value and the valid airspeed value at the previous moment does not exceed the preset difference threshold, the Sigmoid tilt is continued based on the estimated value, and the rate of change of the propulsion vector angle is reduced to below the preset rate of change upper limit. Otherwise, the hold mode is maintained and an emergency landing is triggered.

[0179] The virtual airspeed estimator has a built-in fusion estimation module based on extended Kalman filtering. It inputs the acceleration signal from the inertial measurement unit, the ground speed measured by the global positioning system, the altitude change rate obtained by the barometric altimeter, and the fuselage attitude angular rate into the filtering module to generate an airspeed estimate and its estimation error covariance matrix. When the trace of the estimation error covariance is less than a set threshold, tilt control is allowed to resume; otherwise, an alarm is continuously issued and the hold mode is maintained.

[0180] During the transitional flight modes, airspeed is the core parameter determining the propulsion vector angle. The mission decision-making layer continuously monitors the signal quality from the pitot tube airspeed indicator to identify the following two anomalies:

[0181] Excessive airspeed signal abrupt change rate: Calculate the first derivative of the airspeed measurement (i.e., the change per second). When the absolute value of this rate of change exceeds a preset abrupt change rate threshold (e.g., ±10 meters per second), it is determined that the airspeed signal has been severely disturbed (such as a sudden clearing of a blocked pitot tube or instantaneous pseudo-velocity caused by strong turbulence). This threshold is much larger than the airspeed change rate corresponding to the maximum acceleration that the UAV may reach during normal transitional flight (usually not exceeding ±5 meters per second), thus effectively distinguishing normal maneuvers from signal anomalies.

[0182] Invalid airspeed signal: When the pitot tube heating fails, the sensor output exceeds the reasonable range (e.g., negative value or greater than 1.2 times the maximum flight speed), or the communication verification fails for more than a preset time (e.g., 0.5 seconds), the airspeed signal is determined to be completely invalid.

[0183] Once any of the above conditions are detected, the task decision layer immediately triggers the fault tolerance process.

[0184] Once an airspeed anomaly is confirmed, the underlying execution layer automatically enters the tilt control law hold mode, as follows:

[0185] Locking the current thrust vector angle: The tilt mechanism servo motor maintains the angle value at the instant before the airspeed anomaly occurs, and no longer performs Sigmoid curve mapping with subsequent airspeed changes. This avoids violent fluctuations in the thrust vector angle caused by sudden changes in airspeed.

[0186] Increase the thrust response gain of the electric rotor: Temporarily increase the proportional gain of the thrust controller of the vertical / horizontal motor to 1.5 to 2 times the normal value. At the same time, reduce the integral time constant of the control loop, enabling the rotor to respond to attitude errors more quickly. In this way, when the aircraft experiences a sudden drop or increase in aerodynamic forces due to abnormal airspeed, the rotor can instantaneously generate disturbance-resistant torque to maintain pitch and roll stability.

[0187] In the hold mode, the propulsion vector angle is locked, and the UAV flies in a hybrid state of quasi-fixed wing or quasi-rotor, reducing its reliance on airspeed information.

[0188] If the airspeed signal is invalid (rather than just an excessively high abrupt change rate), a virtual airspeed estimator is activated. The core of this estimator is a fusion estimation module based on extended Kalman filtering, which integrates information from multiple sensors to calculate an estimate of the current airspeed. Specific inputs include:

[0189] The acceleration signal (specific force) measured by the inertial measurement unit: that is, the raw specific force data output by the triaxial accelerometer (after deducting the gravity component), is used as the control input of the filter state equation to predict velocity changes.

[0190] Ground speed measured by the Global Positioning System: GPS directly provides the drone's horizontal velocity vector (ground speed) relative to the ground, including its magnitude and direction.

[0191] The rate of change of altitude obtained by the barometric altimeter: vertical velocity is obtained by differentiating altitude with respect to time.

[0192] Aircraft attitude angular rates: Pitch rate, roll rate, and yaw rate measured by gyroscopes, used to convert acceleration signals from the body coordinate system to the navigation coordinate system.

[0193] Ground Speed ​​and Wind Field Estimation Model: The virtual airspeed estimator utilizes the fundamental relationship that airspeed = ground speed - wind speed. Ground speed is directly provided by GPS; wind speed is not directly measured but is estimated online as a state variable of the Kalman filter. The wind field estimation model assumes that wind speed can be considered a slowly changing quantity over a short timescale (e.g., seconds to tens of seconds) and is modeled using a first-order Markov process. Specifically, the filter's state vector includes: the UAV's three-dimensional velocity relative to the air (i.e., the airspeed component in the navigation frame) and the three-dimensional wind speed component. In the process equations, the change in the wind speed component is modeled as a white noise-driven random walk. The measurement equations integrate the velocity prediction obtained by integrating the acceleration signal internally (corresponding to the sum of airspeed and wind speed), the GPS ground speed (equal to airspeed plus wind speed), and the altitude change rate (related to the sum of vertical airspeed and vertical wind speed). By extending the prediction and update steps of the Kalman filter, the filter outputs the optimal estimate of airspeed in real time, along with the covariance matrix of the estimation error.

[0194] When the airspeed signal is invalid, the mission decision layer calls the estimator, using the current filter state as the initial value, and recursively obtains the airspeed estimate based on the latest IMU and GPS data.

[0195] After obtaining the virtual airspeed estimate, the system compares it with the previous time step (i.e., the last valid airspeed value before airspeed failure):

[0196] If the absolute value of the difference does not exceed a preset difference threshold (e.g., 3 m / s), the estimated value is considered reliable. In this case, the mission decision layer allows continued Sigmoid curve mapping based on the estimated value, but for safety, the rate of change of the propulsion vector angle is simultaneously reduced to below a predetermined rate of change (e.g., 50% of the normal rate of change). This ensures that even if the estimated value has some error, the adjustment of the propulsion vector angle is slow and reversible.

[0197] If the absolute value of the difference exceeds the preset difference threshold, it indicates that the estimated value may deviate too much from the actual airspeed. At this time, the hold mode is maintained (the propulsion vector angle is locked and no adjustments are made), and the emergency landing procedure is triggered: the mission decision layer switches the flight mode to the emergency return mode, plans the nearest safe landing point, and controls the UAV to descend vertically at a high descent rate.

[0198] The virtual airspeed estimator outputs an estimate along with an estimation error covariance matrix in each control cycle. The trace of the covariance matrix represents the total uncertainty of the estimate. A recovery threshold is set (e.g., the covariance trace value corresponding to an airspeed standard deviation of approximately 0.5 m / s). When the following two conditions are met, the mission decision layer determines that the airspeed signal has returned to normal and allows the exit from fault tolerance processing:

[0199] The pitot tube airspeed indicator signal remained stable for more than 1 second, and the mutation rate was below the threshold.

[0200] The trace of the estimation error covariance output by the virtual airspeed estimator is less than the set recovery threshold.

[0201] At this point, switch back to normal mode: the tilt control law resumes using measured airspeed for Sigmoid curve mapping, and the rotor gain is restored to its nominal value. If airspeed anomalies recur during subsequent flights, and the estimated error covariance trace remains greater than the recovery threshold for more than 5 seconds, a serious airspeed system malfunction is detected, maintaining hold mode and guiding the landing.

[0202] By employing three layers of measures—maintaining mode lock, virtual airspeed estimation, and covariance-based recovery judgment—the internal combustion engine extended-range electric vertical take-off and landing fixed-wing UAV can still fly safely and perform controlled landings even if it loses airspeed signals or encounters severe signal disturbances during transitional flight modes.

[0203] The effects of the above technical solution are as follows:

[0204] The virtual airspeed estimator fully reuses the existing IMU, GPS, and barometer in the flight controller, obtaining airspeed estimates independent of the Pitot tube principle without adding any sensors or extra hardware. When the physical airspeed meter fails due to icing, blockage, or sensor malfunction, the estimator provides crucial backup airspeed information, preventing mission failure due to a single point of failure. The fusion estimation module based on extended Kalman filtering fully utilizes the high-frequency dynamic response of the acceleration signal and the low-frequency steady-state accuracy of GPS ground speed, outputting the airspeed estimate and its estimation error covariance matrix in real time. The trace of the covariance matrix intuitively reflects the uncertainty of the current estimate, providing a quantifiable safety criterion for recovering tilt control (recovery is only allowed if the trace is less than a threshold), avoiding the risks of blindly trusting the estimated value. For airspeed mutation rates exceeding the limit (the signal is still valid but disturbed), a hold mode is adopted (locking the thrust vector angle and increasing rotor gain), responding within milliseconds to maintain attitude stability.

[0205] If the airspeed is completely invalid, a virtual estimator is activated, and the estimated value is compared with the historical valid value: if the deviation is small, the aircraft continues to tilt according to the descent rate sigmoid curve to ensure uninterrupted transitional flight; if the deviation is large, the aircraft maintains hold mode and triggers an emergency landing. This tiered approach avoids unnecessary emergency landings while ensuring safety under extreme failure conditions.

[0206] The recovery conditions also require a stable physical airspeed indicator signal and an estimation error covariance trace less than a threshold, providing dual protection against repeated switching of control modes due to intermittent sensor failures. When the estimation uncertainty remains excessively high, continuous warnings are issued and landing guidance is provided, eliminating the risk of continuing flight with unreliable data.

[0207] In one possible implementation, the control logic under the transitional flight mode employs an adaptive gain scheduling strategy based on the lift allocation factor:

[0208] The lift distribution factor is defined as the ratio of estimated lift to UAV weight, where the estimated lift is obtained through real-time airspeed, angle of attack, and pre-stored aerodynamic data;

[0209] The propulsion vector angle change rate command is equal to the reference change rate multiplied by a coefficient, which is equal to the value obtained by mapping the lift distribution factor through a monotonically increasing Sigmoid function, so that the propulsion vector angle change rate increases with the increase of the lift distribution factor; it is slower in the early stage of lift transfer and faster in the later stage.

[0210] The attitude feedback control quantity is reconstructed by weighted averaging of the pre-stored hovering mode attitude control law and cruise mode attitude control law, where the weight of the hovering mode attitude control law is equal to 1-λ and the weight of the cruise mode attitude control law is equal to λ, where λ is the lift distribution factor.

[0211] Based on the aforementioned hierarchical coordinated control method, an adaptive gain scheduling strategy for transitional flight modes is further proposed. The core idea of ​​this strategy is to dynamically adjust the propulsion vector angle change rate and attitude control law according to the current proportion of wing aerodynamic lift to the total weight of the UAV, so as to achieve a smooth and efficient transition from vertical take-off and landing mode to fixed-wing cruise mode.

[0212] Taking a vertical takeoff and landing fixed-wing UAV with a maximum takeoff weight of 35 kg as an example, during the transitional flight mode, the lift source gradually shifts from the vertical thrust of the four distributed rotors to the aerodynamic lift of the wings. To quantify this transfer process, a key parameter, the lift distribution factor denoted as λ, is introduced. It is defined as the ratio of the current wing aerodynamic lift to the total weight of the UAV. The current wing aerodynamic lift is obtained as follows: based on real-time measured airspeed and angle of attack, combined with lift coefficient curves (i.e., wing lift coefficients at different angles of attack) pre-stored in the aerodynamic database for this aircraft type, the aerodynamic lift generated by the current wing is calculated. This aerodynamic lift is then divided by the current weight of the UAV (weight after considering fuel consumption) to obtain the lift distribution factor λ. When the drone is in vertical hovering, the lift of the wing is negligible, and λ is close to 0. As the airspeed increases, the lift of the wing gradually increases, and λ increases from 0 to 1. When the drone fully enters cruise mode, the wing generates all the required lift, and λ reaches 1 (or slightly greater than 1 indicates excess lift).

[0213] During transitional flight, a uniform change in the thrust vector angle from 90 degrees (vertical) to 0 degrees (horizontal) can lead to the following problems: Initially (just leaving hover), the wings have not yet generated sufficient lift; if the tilt is too rapid, the aircraft may lose altitude due to insufficient lift. Later (approaching cruise), lift is mainly provided by the wings; if the tilt is too slow, the drag component of the rotor will persist, reducing forward acceleration efficiency. Therefore, this embodiment employs adaptive thrust vector angle change rate scheduling:

[0214] A baseline rate of change (e.g., 15 degrees per second) is set, and then a real-time adjustment coefficient is generated based on the lift distribution factor λ. This coefficient is equal to f(λ), where f is a monotonically increasing sigmoid function whose output monotonically increases from near 0 to near 1 as the input varies between 0 and 1. Specifically, when λ is small (i.e., the wing's lift contribution is very low, and the rotor still provides the main lift), f(λ) is close to 0, and the thrust vector angle changes slowly, preventing altitude loss. As λ gradually increases, f(λ) increases smoothly accordingly, causing the thrust vector angle change rate to gradually accelerate. This automatically accelerates in the later stages of lift transfer, allowing the aircraft to quickly complete thrust vector steering and reduce drag loss.

[0215] The thrust vector angle change rate command = baseline change rate × f(λ). In the initial transition phase, the wing lift is insufficient, and the tilt proceeds slowly to ensure a smooth lift transfer. In the later transition phase, the wing bears most of the lift, and the tilt accelerates to avoid persistent rotor drag affecting acceleration efficiency. Simultaneously, to ensure that the thrust vector angle change rate does not become excessively rapid due to model errors when the lift distribution factor is too low (i.e., λ is less than a certain lower threshold, such as 0.2), a safety protection is implemented: when λ is below the lower threshold, the thrust vector angle change rate is limited to 30% of the baseline change rate until λ exceeds the lower threshold.

[0216] Traditionally, the attitude control laws during vertical takeoff and landing (Hovering mode) and cruise (Fixed-wing mode) have different structures and parameters: Hovering mode typically employs proportional-integral-derivative control of angular rate damping and angle error, relying on rotor torque to generate attitude response; cruise mode relies on aerodynamic control surfaces (ailerons, elevators, rudders) for attitude control. During transitional flight, the two control effects coexist, and simple switching can cause disturbances. This embodiment uses a lift distribution factor λ as a weighting coefficient to smoothly fuse the outputs of the two attitude control laws:

[0217] Specifically, in each control cycle, the controller simultaneously calculates the output of the hovering mode attitude control law (i.e., the desired rotor speed difference, collective pitch, etc., converted into torque commands) and the output of the cruise mode attitude control law (i.e., the desired control surface deflection angle). Then, the weight of the hovering mode output is set to 1-λ, and the weight of the cruise mode output is set to λ. The two are weighted and averaged to obtain the final attitude control command. When λ=0.2 (the wing lift is low, and the rotor still provides the main lift), the hovering mode output accounts for 80%, and the cruise mode output accounts for 20%, meeting the requirement that rotor control is dominant during the hovering phase. When λ=0.8, the hovering mode output accounts for 20%, and the cruise mode output accounts for 80%, indicating that control surface control is dominant during the cruise phase. The physical meaning of this weighting method is clear: the smaller λ is (closer to hovering), the larger the hovering mode weight 1-λ is; the larger λ is (closer to cruise), the larger the cruise mode weight λ is, ensuring a continuous and smooth transition of attitude control weights throughout the lift transfer process, without abrupt changes.

[0218] To make the integration more reasonable, the command units of the two control laws need to be normalized. In engineering, the torque command output by hovering mode can be converted into equivalent pitch / roll / yaw torque, and the control surface deflection angle output by cruise mode can be converted into the corresponding torque (through the aerodynamic derivative model). After unifying the two into torque commands, they are weighted and summed, and finally converted into actual commands for each actuator (rotor speed difference, servo deflection angle) by the underlying allocation algorithm.

[0219] In practical engineering implementation, the lift coefficient table of the UAV's wing at different angles of attack and airspeeds is first obtained through computational fluid dynamics simulation or wind tunnel experiments and pre-stored in the flight control memory. During flight, the flight control system uses real-time measured airspeed and angle of attack to look up the table and interpolate to obtain the current lift coefficient, and then calculates the lift allocation factor. The update frequency of the lift allocation factor is the same as the main control cycle (e.g., 50 Hz) to ensure real-time control. To prevent abnormal signals from causing the thrust vector angle change rate command to exceed the limit, absolute maximum and minimum thrust vector angle change rate limits (e.g., 30 degrees per second and 3 degrees per second) are set on the basis of adaptive scheduling. In addition, the weighted fused command after attitude control law also needs to be amplitude limited to ensure that the final torque allocation command is within the physical capability range of all actuators (motors, servos).

[0220] The effects of the above technical solution are as follows:

[0221] The thrust vector angle change rate is automatically adjusted according to the lift transfer progress. Initially, a rapid tilt is used to establish airspeed, which is then reduced later to avoid excessive drag pulses. Relying solely on the lift distribution factor, a continuously changing scheduling coefficient is automatically generated, significantly reducing the workload of debugging. By merging the hovering and cruise attitude control laws through weighted averaging, abrupt control changes during mode switching are avoided. The lift distribution factor directly reflects the relationship between current aerodynamic lift and weight; therefore, for different takeoff weights (e.g., full fuel load versus low fuel load) or different payloads, this strategy automatically adjusts the thrust vector angle change rate and attitude weights without requiring parameter readjustment. This greatly enhances the mission adaptability of the UAV.

[0222] All calculations involve basic arithmetic operations and table lookups, which can be easily implemented with existing flight control computing resources; and all parameters have clear physical meanings, facilitating troubleshooting and on-site adjustments.

[0223] In one possible implementation, the scheduling process of the lift allocation factor is also subject to real-time modulation of the angle of attack margin:

[0224] The difference between the real-time computerized wing stall angle of attack and the current angle of attack is used as the angle of attack margin;

[0225] When the angle of attack margin is less than the preset angle of attack margin safety threshold, a modulation factor is introduced to limit the rate of change of the propulsion vector angle. The modulation factor is positively correlated with the angle of attack margin and has a preset lower limit value. The actual rate of change of the propulsion vector angle is equal to the normal rate of change multiplied by the modulation factor, so that the smaller the angle of attack margin, the slower the propulsion vector angle changes.

[0226] Meanwhile, in the weighted average reconstruction of the attitude feedback control quantity, the weight of the attitude control law in the hovering mode is dynamically corrected according to the angle of attack margin, so that the weight increases as the angle of attack margin decreases, and a rebound compensation amount in the pitch direction is superimposed on the pitch angle command. This rebound compensation amount is inversely proportional to the angle of attack margin and has a maximum value limit.

[0227] Building upon the aforementioned adaptive gain scheduling strategy based on the lift allocation factor, a real-time angle-of-attack margin modulation mechanism is further introduced to prevent stall during transitional flight due to excessively rapid lift allocation or inappropriate attitude response. This modulation mechanism operates independently of the lift allocation factor λ, directly affecting the propulsion vector angle change rate and attitude commands.

[0228] Taking a 35kg-class vertical takeoff and landing fixed-wing UAV as an example, its wing stall angle of attack is determined to be 15° through wind tunnel experiments or computational fluid dynamics simulations. During flight, the current angle of attack α is measured in real time using a weather vane-type angle of attack sensor or an estimation algorithm based on inertial / navigation data, and the angle of attack margin is calculated. = -α; where, This represents the critical stall angle of attack for the wing (unit: degrees). For the airfoil of a small UAV, the stall angle of attack is typically between 12° and 16°. In this example, 15° is used; a preset safety threshold for the angle of attack margin is also included. =5°. When When the angle is >5°, it indicates that the system is far from the stall boundary, the modulation mechanism does not intervene, and the system operates according to normal adaptive scheduling; when When the angle of attack is ≤5°, the angle of attack protection mode is activated.

[0229] Define modulation factor Its value is positively correlated with the angle of attack margin, for example, by using a linear relationship:

[0230]

[0231] in To ensure the minimum maneuverability, for example, take 0.2 (i.e., maintain a minimum of 20% normal propulsion vector angle change rate).

[0232] Then, multiply by the maximum allowable change in thrust vector angle during the current control cycle (i.e., the thrust vector angle change rate command). That is: Actual propulsion vector angle change rate command = Original adaptive propulsion vector angle change rate command × .

[0233] Therefore, when When approaching 0, = The propulsion vector angle change rate is reduced to 20% of the original, significantly slowing down the lift transfer speed and avoiding a further increase in the angle of attack due to excessive acceleration.

[0234] The original attitude control law weighted is: Total torque command = λ × hovering mode torque + (1-λ) × cruise mode torque, where λ is the proportion of wing lift.

[0235] It should be noted that the above weighted fusion formula is only applied to the transitional flight mode. During the transitional flight, the lift allocation factor λ ranges from 0 to 1, excluding endpoint values ​​(e.g., λ changes from a minimum value close to 0 to a maximum value close to 1, but will not equal 0 or 1). Therefore, the attitude control laws for hovering mode and cruise mode always participate in the fusion with non-zero weights. For pure hovering state (λ=0) and pure cruise state (λ=1), the UAV uses its own dedicated control laws independently, and this weighted fusion formula is not used.

[0236] When the angle of attack margin is insufficient, the weight of the hovering mode (rotor control) should be temporarily increased because the rotor can more directly generate nose-down torque. Therefore, a weight correction variable w_corr is introduced:

[0237]

[0238] Where W_max is the maximum additional weight, for example, 0.5. Then the corrected hover mode weight is:

[0239] λ'=λ+w_corr

[0240] The cruise mode weight is changed to 1-λ'. At the same time, λ' is limited to not exceed 0.95 to ensure that the cruise mode still has the minimum weight.

[0241] when When the value is very small, w_corr is close to 0.5 and λ' is close to λ+0.5. That is, no matter what value λ is, the hovering mode weight increases significantly, and the system prioritizes using rotor torque to suppress the pitching trend.

[0242] An additional pitch offset δθ is superimposed on the pitch angle command (the desired pitch angle). δθ is calculated using a saturation inverse proportional form:

[0243]

[0244] Where ε is a small positive number (e.g., 0.5°) to prevent division by zero; K is the nose-down compensation proportional coefficient, taken as 10 degrees, when the angle of attack margin... When =1°, K / =10 degrees; when When =2°, the result is 5°; when When the angle is 5°, the result is 2°. This value is then subjected to a maximum limit θ_max (e.g., 5°) to obtain the actual pitch compensation amount. θ_max is the maximum pitch compensation amount, for example, taken as 5°; when... As the angle of attack gradually decreases from 5° to 0°, δθ smoothly increases from approximately 1.8° to 5° (saturation value). This compensation amount guides the UAV to actively lower its nose, reducing the angle of attack. Once the angle of attack margin is restored, δθ gradually decays to 0 according to the same rule.

[0245] In terms of engineering implementation, the angle of attack information can be directly measured using a wind vane-type angle of attack sensor on the nose pitot tube, or estimated in real time based on combined navigation data from an inertial measurement unit and a global positioning system. All modulation calculations are synchronized with the main control cycle (e.g., a 50 Hz update frequency) to ensure real-time control. Key parameters discussed in this paper include the angle of attack margin safety threshold. Minimum Modulation Factor Lower Limit The maximum additional weight W_max, the nose-down compensation ratio coefficient K, and the maximum nose-down compensation amount θ_max can all be obtained through ground simulation and flight test iteration calibration. The specific values ​​given in this embodiment are only examples, and can be adjusted accordingly based on the characteristics of different aircraft models in actual applications.

[0246] Therefore, the propulsion vector angle change rate is reduced, rotor control is enhanced, and nose-down compensation is superimposed before the angle of attack approaches the stall boundary. Compared with the traditional approach of warning or forcibly pushing the stick after entering stall, the intervention time is earlier. The modulation factor, weight correction, and nose-down compensation all use continuous functions, and there are no threshold-triggered step changes. Through real-time modulation of the angle of attack margin, the stall safety under the transition flight mode is effectively improved, and the engineering feasibility is high.

[0247] In one possible implementation, the modulation factor adopts a nonlinear function: when the angle of attack margin is greater than the angle of attack margin safety threshold, the modulation factor is always 1; as the angle of attack margin decreases from the angle of attack margin safety threshold to zero, the modulation factor decreases nonlinearly from 1 to the minimum value, and its decreasing curve is an exponential curve or a sigmoid curve, so that the rate of change of the angle of attack margin is strongly suppressed when it approaches zero; the preset minimum value ranges from 0.1 to 0.3, and is pre-calibrated according to the wing stall characteristics and tilt mechanism response speed of the UAV.

[0248] Based on the aforementioned modulation mechanism based on angle of attack margin, the modulation factor is further limited to a nonlinear function to provide a stronger suppression effect when the angle of attack approaches the stall boundary.

[0249] Continuing with the aforementioned parameter settings: stall angle of attack is 15°, and the safety threshold for angle of attack margin is 5°.

[0250] The modulation factor takes values ​​according to the following rules:

[0251] When the angle of attack margin is within the safety threshold, the modulation factor is 1, modulation is not engaged, and the propulsion vector angle change rate is executed according to the original adaptive command.

[0252] 0≤ When the angle of attack margin is less than or equal to the safety threshold, the modulation factor decreases non-linearly from 1 to the preset minimum value.

[0253] This embodiment provides two achievable curves for the modulation factor:

[0254] Exponential curve:

[0255]

[0256] in, This is the attenuation coefficient (for example, between 0.8 and 1.5).

[0257] Sigmoid curve:

[0258]

[0259] c is the curve slope parameter, with the unit being 1 / degree, and the value range is usually between 0.3 and 2.0; in practice, it can be adjusted through ground simulation or test flight.

[0260] Or first Mapped to the [-5, 5] interval, then compressed using the Sigmoid function. The curve is... The change is gradual when approaching the threshold, the rate of change is large in the middle section, and slows down again when approaching zero, forming an S-shaped decline.

[0261] Minimum Modulation Factor Lower Bound The values ​​are calibrated based on the wing stall characteristics and tilt mechanism response speed of the UAV, with a range of 0.1 to 0.3.

[0262] The effects of the above technical solution are as follows:

[0263] When the angle-of-attack margin is less than 1°, the modulation factor of the nonlinear function can rapidly decrease to near 1°. (For example, 0.2) This compresses the rate of change of the thrust vector angle to an extremely low level, preventing stalling due to any minor disturbance near the stall critical point. The exponential or sigmoid curve is smooth and continuous, with no abrupt changes in the derivative, and will not cause jumps in the thrust vector angle change rate command, thus reducing servo wear.

[0264] Besides the minimum value, only one attenuation coefficient β or curve slope parameter is needed, and this parameter is not sensitive to wing stall characteristics, so the default value can be used. It can be quickly determined through ground simulation, reducing the number of test flights.

[0265] This nonlinear modulation, together with the aforementioned virtual airspeed estimation, attitude weight correction, and nose-down compensation, constitutes a complete angle-of-attack protection chain. When the angle-of-attack margin is extremely small, not only does tilting almost stop, but the hovering mode weight is increased and the nose-down compensation is amplified. The three work together to ultimately ensure the safety of the UAV.

[0266] In one possible implementation, the method further includes thermal management and power limiting protection strategies:

[0267] Real-time monitoring of engine cylinder temperature, generator winding temperature, motor controller temperature, and battery temperature;

[0268] The power limit is dynamically adjusted according to the thermal state: when any of the temperatures exceeds the preset first temperature threshold T1, power degradation is initiated; when any of the temperatures exceeds the preset second temperature threshold T2, protective power degradation is triggered, where T2>T1.

[0269] It also includes engine idling warm-up logic: during cold start, the engine is first run at low power until the lubricating oil temperature reaches the preset lower operating limit temperature before entering range-extending mode.

[0270] Based on the above-mentioned hierarchical coordinated control method, a specific implementation plan for thermal management and power limiting protection strategy is further given to prevent the performance of the power system from deteriorating or being damaged due to overheating, while ensuring reliable engine operation during cold start.

[0271] Taking a 35kg-class VTOL fixed-wing UAV as an example, its power system includes: a 120cc two-stroke gasoline engine (with cylinder temperature sensor), a permanent magnet synchronous generator (with winding temperature sensor), four VTOL motors and their controllers (with temperature sensors), and a 6S lithium-ion battery (with cell temperature sensor). Thermal management is jointly implemented by the mission decision-making layer and the energy management layer.

[0272] The following temperatures are collected in real time: engine cylinder head temperature (degrees Celsius), generator winding temperature (degrees Celsius), power transistor temperature of each motor controller (degrees Celsius), and highest cell temperature of the battery pack (degrees Celsius).

[0273] Two preset temperature thresholds:

[0274] First temperature threshold T1: Warning temperature, such as engine cylinder temperature 100℃, generator winding temperature 110℃, motor controller temperature 85℃, and battery temperature 50℃. When any temperature exceeds T1, it indicates that the system has entered a state of heat accumulation and requires active power reduction.

[0275] The second temperature threshold T2 is the ultimate protection temperature, such as 125°C for engine cylinders, 130°C for generator windings, 100°C for motor controllers, and 60°C for the battery. When any temperature exceeds T2, it indicates a risk of damage, and power must be reduced significantly and immediately.

[0276] The temperature difference between T1 and T2 is typically 15–25°C to ensure sufficient buffer space for power reduction operation. Specific values ​​are determined according to the specifications of each component and the required heat dissipation conditions.

[0277] When any monitored temperature exceeds T1, the energy management layer initiates power descent: gradually reducing the target engine power and the upper limit of available motor power at a linear rate (e.g., reducing current output power by 5% to 10% per second). During power descent, charging power is reduced first (if battery SOC is above the normal range), then non-essential loads (such as non-core modules in avionics) are reduced sequentially, and finally the main propulsion power is reduced. Descent continues until all temperatures fall below T1 (with a 2°C hysteresis allowance), or any part of the temperature exceeds T2.

[0278] When any temperature exceeds T2, protective power degradation is triggered:

[0279] Immediately force the total power limit down to 50% (or lower, such as 40%) of the rated power and lock this limit.

[0280] At the same time, the mission decision-making level will switch the flight mode to emergency return mode. If the altitude is sufficient, it will prioritize low-power return in fixed-wing mode; if the altitude is insufficient, it will land in vertical mode at the nearest landing site.

[0281] If the battery temperature exceeds T2, charging is also prohibited, and internal battery equalization or active cooling (such as a fan) is activated.

[0282] All degraded states persist until manual reset or the temperature drops below T1 for more than 30 seconds.

[0283] To prevent frequent entry and exit from degradation, a debounce timer is set: after the temperature exceeds the threshold, it must remain for 0.5 seconds before triggering a response; after the temperature falls back below the threshold, it must remain for 2 seconds before degrading is lifted.

[0284] When the engine is cold-started (shut down for more than 2 hours and the ambient temperature is below 10℃, or the cylinder temperature is below 40℃), it should not be directly put into range-extender mode for operation under load; otherwise, insufficient lubrication by the lubricating oil may cause cylinder scoring. The warm-up logic is as follows:

[0285] Starting at idle speed: The electric starter motor drives the engine to idle speed (e.g., 2500 rpm), and maintains idle speed after successful ignition.

[0286] Monitoring lubricating oil temperature: Since small two-stroke engines typically do not have independent oil temperature sensors, cylinder temperature can be used as an approximation (when the cylinder temperature reaches 60°C, the lubricating oil is considered to have sufficient fluidity). If an independent oil temperature sensor is available, it can be used directly.

[0287] Low-power warm-up: During idling, the engine only drives the generator to run idle (excitation current is zero) and does not supply power to the DC bus. If it is necessary to maintain power for onboard equipment, it is powered solely by the battery.

[0288] Range-extending mode enabled: Once the cylinder temperature (or oil temperature) reaches the preset lower operating limit (e.g., 60°C), the engine warm-up is complete. At this point, the energy management layer can allow the generator to output power, entering range-extending mode or combined power supply mode.

[0289] Warm-up Overtime Protection: If the warm-up period exceeds 3 minutes and the temperature still does not reach the lower limit, it is determined that there may be a sensor malfunction or extremely low ambient temperature. The system will issue an alarm and limit the engine power to no more than 120% of the idle power, and recommend returning to the starting point.

[0290] The effects of the above technical solution are as follows:

[0291] Through a two-stage protection system, power is proactively limited before the temperature approaches its limit, preventing crashes caused by sudden shutdowns or component burnout. The idling warm-up logic forces the engine to only carry a load after it reaches lubrication requirements, significantly reducing the risk of cylinder scoring or piston seizure when the engine is cold. Power reduction uses linear or proportional adjustments, preventing sudden thrust changes, allowing the drone to continue its mission or return safely. Temperature thresholds, power reduction rates, and target warm-up temperatures can all be calibrated for different drone models and seasons without code modification. This strategy has been adapted to multiple vertical takeoff and landing (VTOL) drones, effectively protecting their power systems. Combined with anti-shake timing and timeout protection, false triggering caused by sensor noise is avoided, and even under extreme conditions (such as heating failure), alarms are still issued and power is limited to prevent complete loss of power.

[0292] In one possible implementation, the method further includes fault detection and safety redundancy control strategies:

[0293] Generator fault handling: If an abnormal generator output voltage or communication loss is detected, the control clutch will automatically disengage, switch to direct battery supply mode, and execute the nearest landing procedure.

[0294] Engine in-flight shutdown handling: The electric starter will automatically attempt to restart the engine, with no more than 3 attempts. If the restart fails, the engine will switch to pure electric mode, assess the remaining flight time based on the remaining battery power, and plan an emergency route.

[0295] Severe battery undervoltage handling: Force the engine to start and enter range-extending mode, while limiting power supply to non-critical equipment to ensure that the power supply to the flight control system has the highest priority.

[0296] Taking a 35kg-class vertical takeoff and landing fixed-wing UAV as an example, its power system includes: an internal combustion engine, a permanent magnet synchronous generator, an electromagnetic clutch, a power battery, a main propulsion motor, and four vertical takeoff and landing motors. The mission decision-making layer continuously monitors the status of each component, and the energy management layer and the underlying execution layer work together to perform fault isolation and redundancy switching.

[0297] Generator malfunctions are mainly manifested as abnormal output voltage (e.g., voltage drop below 70% of the rated value, or severe ripple exceeding 20% ​​of the rated voltage) or communication interruption with the flight controller lasting for more than a preset time (e.g., 0.5 seconds). In this case, the generator cannot supply power to the DC bus normally and may even become a bus load (rectifier short circuit, etc.).

[0298] The mission decision-making layer performs real-time diagnostics based on voltage detection and communication heartbeats. Once a generator fault is confirmed, it immediately instructs the electromagnetic clutch to disengage, mechanically decoupling the engine from the generator. The clutch actuation time is approximately 0.1 seconds, and simultaneously the engine throttle is shut off, stopping fuel supply.

[0299] The energy management system switches the power supply mode to direct battery supply mode, where the power battery supplies power to the entire aircraft independently. Since the battery's rated discharge capacity is 10 times (instantaneous 15 times), when the maximum power requirement of the 35kg drone is about 8kW, the battery discharge current is about 360A, which is still within the instantaneous capacity range of a 6S 32Ah battery (15 times corresponds to 480A).

[0300] The mission decision-making level simultaneously switches the flight mode to emergency return mode and plans the nearest landing procedure based on the distance between the current position and the nearest safe landing point. Vertical landing is preferred as the landing method (because the range extender is unavailable after generator failure, the remaining power is limited, vertical landing consumes more power but has lower landing site requirements). If the distance is far, fixed-wing gliding can be selected to reduce power consumption.

[0301] During landing, the lower control layer reduces the power of non-essential motors (such as reducing the vertical motor speed to the minimum value just to maintain attitude) and shuts down all non-critical avionics (such as gimbal cameras and image transmission transmitters with reduced power) to extend endurance.

[0302] If the malfunction occurs during the takeoff climb phase and the altitude is below the safe vertical landing altitude (e.g., 10 meters), the system will immediately perform a vertical emergency landing without attempting to turn back.

[0303] Engine stalling in mid-air can be caused by fuel exhaustion, spark plug carbon buildup, or intake blockage. The mission decision-making layer identifies the stall state by detecting engine speed signals (crankshaft position sensor) or generator output frequency: when the engine speed drops from the normal operating speed (e.g., 5000 rpm) to below idle speed (e.g., 1000 rpm) within 0.2 seconds, and there is no deceleration command, it is determined to be an in-air stall.

[0304] The energy management system instructs the electric starter to engage, pulling the engine to idle speed (2500 rpm) while simultaneously restoring ignition and fuel supply. Each start attempt lasts no more than 3 seconds. If the engine speed does not rise above idle speed after 3 seconds, the process stops and waits 2 seconds before attempting again. A maximum of 3 attempts are allowed.

[0305] During restart, the engine does not output power, and the entire aircraft is powered solely by the battery. Since the starter motor consumes approximately 500-800W, plus the power consumption of flight control and motor standby, the total demand is about 1-2kW. The battery can support several restart attempts. If the restart is successful, the energy management system reverts to range-extended mode, the generator reconnects to the grid, and the drone continues its original mission or returns as needed.

[0306] If three restart attempts fail, the engine is deemed unrecoverable. The mission decision-making layer switches the system to pure electric mode, keeps the clutch disengaged, and shuts down the engine. Flight then relies solely on the remaining battery power. Based on the current battery state of charge (SOC) and total power requirements, the remaining flight time is estimated: for example, with an SOC of 60%, the total battery energy is approximately 22.2V × 32Ah ≈ 710Wh. Subtracting the safe discharge limit of 20% (approximately 284Wh), the effective energy can be approximately 8.5 minutes of flight at an average power of 2kW (fixed-wing cruise); or approximately 11 minutes at 1.5kW (low-speed cruise). The flight control system then plans an emergency route, prioritizing the nearest feasible landing point for a straight return flight.

[0307] In pure electric mode, the mission decision-making layer limits the maximum power to no more than the battery's continuous discharge capacity (e.g., approximately 3.5kW at 10x rate) and disables all non-essential functions (such as de-icing heating and secondary mission equipment). If the remaining battery power is lower than the expected return-to-base requirement, an immediate vertical landing will be performed, and no further attempts to continue flight will be made.

[0308] Severe undervoltage typically occurs during prolonged electric flight, when generator power is insufficient, or when batteries are aging. Severe undervoltage protection is triggered when the Battery Management System (BMS) detects that the voltage of any cell is below a preset low-voltage threshold (e.g., 3.0V per cell, 18V total voltage for a 6S battery). If the engine is not yet running, the mission decision layer instructs the engine to start (regardless of the current flight mode). After engine start, the energy management layer drives the generator to supply power to the DC bus at maximum permissible power (not exceeding the engine's economic limit).

[0309] Limit power supply to non-critical equipment: The energy management layer should immediately disconnect or reduce the following loads:

[0310] Power off the onboard gimbal, camera, lighting equipment, and other mission equipment;

[0311] Reduce the image transmission power from the default value to the lowest sustain level (e.g., from 1W to 100mW).

[0312] Turn off all switchable functions such as sensor heating and pitot tube heating.

[0313] Ensure that the power supply to the flight control computer, inertial measurement unit, GPS receiver, and basic actuators (servos, motor controllers) is prioritized and that their voltage is kept stable.

[0314] Simultaneously adjust the generator output: Set the generator output voltage above the battery's equalization voltage (e.g., 24V) so that the battery can perform emergency charging with a small current (e.g., 2-5A) to prevent the voltage from continuing to drop. If the engine power is sufficient, it can also charge the battery with a small current and constant voltage, gradually increasing the cell voltage to a safe range (e.g., 3.2V / cell).

[0315] The mission decision-making team assesses the flight path: If undervoltage occurs before takeoff or during hovering, immediately cancel takeoff or perform a vertical landing; if it occurs during cruise, assess the remaining fuel and battery recovery trend. If the engine can provide stable power and the battery voltage recovers to above 3.3V / cell, the flight can continue as planned or return to base; if the battery voltage continues to drop (e.g., the battery cell is damaged), immediately perform an emergency landing at the nearest airport.

[0316] Once the battery voltage returns to normal (e.g., >3.4V / cell for 10 seconds), the power supply restrictions for non-critical equipment are lifted, and the original mission configuration is restored. If the undervoltage condition recurs, the battery pack is considered severely aged, and the system issues an alarm and recommends replacing the battery after landing.

[0317] By automatically isolating, redundancy switching, and energy redistribution for different fault mechanisms, the internal combustion engine extended-range electric vertical take-off and landing fixed-wing UAV is equipped with the ability to cope with common power system failures, greatly reducing the risk of crash.

[0318] This invention also provides a hierarchical coordination control system for an internal combustion engine-extended electric vertical takeoff and landing fixed-wing unmanned aerial vehicle, comprising:

[0319] One or more processors; and

[0320] A memory storing a computer program that, when executed by the one or more processors, causes the system to perform the steps of the method described in Embodiment 1.

[0321] This invention also provides an electronic device, which includes a memory and a processor. The memory stores a computer program, and the processor executes the computer program to implement the steps of the method described in Embodiment 1 of this invention. The specific implementation method is consistent with the implementation method and the technical effects achieved in the above method embodiments, and some contents will not be repeated.

[0322] This invention also provides a computer-readable storage medium for storing a computer program. When the computer program is executed, it implements the steps of the method in Embodiment 1 of this invention. The specific implementation method is consistent with the implementation method and the technical effects achieved in the above method embodiments, and some contents will not be repeated.

[0323] In this invention, a readable storage medium can be any tangible medium containing or storing a program that can be used by or in conjunction with an instruction execution system, apparatus, or device. The program product can take the form of any combination of one or more readable media. A readable medium can be a readable signal medium or a readable storage medium. A readable storage medium can be, for example, but not limited to, an electrical, magnetic, optical, electromagnetic, infrared, or semiconductor system, apparatus, or device, or any combination thereof. More specific examples of readable storage media (a non-exhaustive list) include: an electrical connection having one or more wires, a portable disk, a hard disk, random access memory (RAM), read-only memory (ROM), erasable programmable read-only memory (EPROM or flash memory), optical fiber, portable compact disk read-only memory (CD-ROM), optical storage device, magnetic storage device, or any suitable combination thereof.

[0324] Although embodiments of the present invention have been shown and described above, it is understood that the above embodiments are exemplary and should not be construed as limiting the present invention. Those skilled in the art can make changes, modifications, substitutions and variations to the above embodiments within the scope of the invention without departing from the principles and spirit of the invention, and all such changes should fall within the protection scope of the claims of the present invention.

Claims

1. A hierarchical coordinated control method for unmanned aerial vehicles (UAVs), characterized in that, Applied to internal combustion engine-powered extended-range electric vertical takeoff and landing fixed-wing unmanned aerial vehicles (UAVs), the method is executed under a hierarchical control architecture, and the method includes: Real-time monitoring of flight status parameters; division of flight missions into multiple flight modes based on finite state machines; introduction of forward prediction windows; prediction of mode switching requirements based on flight mission profile and system response delay; and output of advance warning signals. An improved power follower strategy is adopted to dynamically allocate engine output power and battery charging and discharging power according to the relationship between total power demand and the engine's optimal operating range, and to limit the rate of change of engine power to suppress power surges. In the transitional flight mode, a gradual tilt control law is adopted, which makes the propulsion vector angle change with the airspeed according to the Sigmoid curve, introduces attitude angle error feedback correction, and dynamically adjusts the gain coefficient of the feedback correction according to the ratio of the current airspeed to the wing stall speed.

2. The method according to claim 1, characterized in that, It also includes steps for controlling the power flow to smoothly transition between multiple operating modes based on flight mode and battery state of charge: The various operating modes include pure electric mode, range-extended mode, combined power supply mode, charging mode, and energy recovery mode; A power-weighted transition filter is used during mode switching, where the first derivative of the transition function is continuous. When exiting the combined power supply mode or entering the charging mode, active bus voltage balancing is first performed: the difference between the DC bus voltage output by the generator after rectification and the battery pack voltage is detected. If the difference exceeds a preset first voltage threshold, the voltage difference is gradually reduced to below a preset second voltage threshold by adjusting the generator excitation current or the DC converter, and then the power switch is closed or opened, wherein the second voltage threshold is less than or equal to the first voltage threshold. Before the voltage difference is reduced to within the second voltage threshold, the mode switching is in a waiting state, and the power switch is prohibited from operating.

3. The method according to claim 1, characterized in that, The forward prediction window is implemented in the following ways: The system records the time of issuance of the mode switching command and the actual time when the corresponding power response reaches the target in real time, calculates the time delay between the two, and uses the recursive least squares method to identify the time delay sequence online, so as to obtain the equivalent delay time constant between the issuance of the mode switching command and the actual power response. Set a lead time for the prediction, which is equal to the equivalent delay time constant plus a safety margin. Read the pre-loaded flight mission profile and extract the target airspeed sequence and target altitude sequence within a time period equal to at least the predicted time lead. Based on the current flight status parameters and the extracted target sequence, the trend of state change within the predicted time lead time is predicted; when it is predicted that from the current moment, within the predicted time lead time, any flight status parameter will first reach or exceed the preset mode switching threshold, a mode switching warning signal is immediately output; the mode switching warning signal includes at least the predicted target mode type and the estimated remaining time.

4. The method according to claim 1, characterized in that, The improved power follower strategy includes: Set the total power demand, battery state of charge, and optimal operating range for the engine, which includes a lower limit, an economic value, and an upper limit. When the total power demand is within the optimal operating range, the engine output power is adjusted to be equal to the total power demand plus the charging compensation amount. This charging compensation amount is used to maintain the battery state of charge within a preset normal range. The charging compensation amount is calculated by a proportional-integral controller based on the deviation between the battery state of charge and the target midpoint value of the normal range, and its maximum amplitude is limited. When the total power demand is lower than the lower limit of the optimal operating range, if the battery state of charge is higher than the preset high charge threshold, the engine will be shut off and the battery will supply power alone; otherwise, the engine will operate at the lower limit of the optimal operating range, and the excess power will be used to charge the battery. When the total power demand exceeds the upper limit of the optimal operating range, the engine operates at the upper limit of the optimal operating range, and the difference in power is instantly replenished by the battery; at the same time, the rate of change of engine power is limited to not exceeding the preset rate limit.

5. The method according to claim 1, characterized in that, The gradual tilt control law includes: The relationship between the thrust vector angle and airspeed is determined using a Sigmoid curve; the corresponding base thrust vector angle is obtained by mapping the current airspeed from this Sigmoid curve. The pitch angle and pitch rate of the UAV are acquired in real time, as well as the current pitch angle command. When the deviation of the pitch angle from the pitch angle command exceeds a preset deviation threshold, or the pitch rate exceeds a preset rate threshold, a correction increment is superimposed on the base thrust vector angle. The correction increment is linearly related to the pitch angle deviation and pitch rate. The correction increment is obtained by multiplying the pitch angle deviation and pitch rate by a first gain coefficient and a second gain coefficient, respectively, summing the results, and then taking the negative of the sum. The first gain coefficient and the second gain coefficient are dynamically adjusted according to the ratio of the current airspeed to the wing stall speed: as the ratio decreases, the first gain coefficient gradually decreases and the second gain coefficient gradually increases, so that when the airspeed approaches the stall boundary, the control law mainly suppresses the pitch rate; at the same time, anti-integral saturation logic is introduced into the control loop of the tilt mechanism to prevent the integral depth caused by the saturation of the actuator.

6. The method according to claim 5, characterized in that, The transitional flight mode also includes fault-tolerant handling steps for airspeed anomalies: When the airspeed signal mutation rate is detected to exceed the preset mutation rate threshold, or when the airspeed signal is invalid, the tilt control law automatically enters the hold mode: the current thrust vector angle is locked and unchanged, while the thrust response gain of the electric rotor is increased, and the rotor directly generates the anti-disturbance torque. If the airspeed signal is invalid, the virtual airspeed estimator is activated to calculate the current airspeed estimate based on the ground speed and wind field estimation model. If the difference between the estimated value and the effective airspeed value at the previous moment does not exceed the preset difference threshold, then the Sigmoid tilt will continue to be performed based on the estimated value and the rate of change of the propulsion vector angle will be reduced to below the preset rate of change upper limit; otherwise, the hold mode will be maintained and an emergency landing will be triggered.

7. The method according to claim 6, characterized in that, The virtual airspeed estimator has a built-in fusion estimation module based on extended Kalman filtering. It inputs the acceleration signal from the inertial measurement unit, the ground speed measured by the global positioning system, the altitude change rate obtained by the barometric altimeter, and the fuselage attitude angular rate into the filtering module to generate an airspeed estimate and its estimation error covariance matrix. When the trace of the estimation error covariance is less than a set threshold, tilt control is allowed to resume; otherwise, an alarm is continuously issued and the hold mode is maintained.

8. The method according to claim 1, characterized in that, The control logic under the transition flight mode adopts an adaptive gain scheduling strategy based on the lift allocation factor: The lift distribution factor is defined as the ratio of estimated lift to UAV weight, where the estimated lift is obtained through real-time airspeed, angle of attack, and pre-stored aerodynamic data; Based on the calculated lift distribution factor, the rate of change command of the thrust vector angle and the attitude feedback control quantity are determined as follows: The rate of change of the thrust vector angle is equal to the reference rate of change multiplied by a coefficient, which is equal to the value obtained by mapping the lift distribution factor through a monotonically increasing Sigmoid function, so that the rate of change of the thrust vector angle increases with the increase of the lift distribution factor; The attitude feedback control quantity is reconstructed by weighted averaging of the pre-stored hovering mode attitude control law and cruise mode attitude control law, where the weight of the hovering mode attitude control law is equal to 1-λ and the weight of the cruise mode attitude control law is equal to λ, where λ is the lift distribution factor.

9. The method according to claim 8, characterized in that, The scheduling process of the lift allocation factor is also subject to real-time modulation of the angle of attack margin: The difference between the real-time computerized wing stall angle of attack and the current angle of attack is used as the angle of attack margin; When the angle of attack margin is less than the preset angle of attack margin safety threshold, a modulation factor is introduced to limit the rate of change of the propulsion vector angle. The modulation factor is positively correlated with the angle of attack margin and has a preset lower limit value. The actual rate of change of the propulsion vector angle is equal to the normal rate of change multiplied by the modulation factor, so that the smaller the angle of attack margin, the slower the propulsion vector angle changes. Meanwhile, in the weighted average reconstruction of the attitude feedback control quantity, the weight of the attitude control law in the hovering mode is dynamically corrected according to the angle of attack margin, so that the weight increases as the angle of attack margin decreases, and a rebound compensation amount in the pitch direction is superimposed on the pitch angle command. This rebound compensation amount is inversely proportional to the angle of attack margin and has a maximum value limit.

10. The method according to claim 9, characterized in that, The modulation factor adopts a nonlinear function: when the angle of attack margin is greater than the safe threshold of the angle of attack margin, the modulation factor is always 1; when the angle of attack margin decreases from the safe threshold of the angle of attack margin to zero, the modulation factor decreases nonlinearly from 1 to the minimum value, and its decreasing curve is an exponential curve or a sigmoid curve, so that the rate of change of the angle of attack margin is strongly suppressed when it approaches zero; the preset minimum value ranges from 0.1 to 0.3, and is pre-calibrated according to the wing stall characteristics and tilt mechanism response speed of the UAV.

Images