Disturbance-resistant vector control method and device for UAV servo motors adapted to harsh near-space conditions
By identifying the stator resistance and permanent magnet flux of the permanent magnet synchronous motor in a hierarchical and time-division online manner, and combining the dynamic feedforward decoupling technology of adaptive law and extended state observer, the problem of deep coupling between internal parameter drift and external wind shear disturbance of UAV servo in near space is solved, and the servo can achieve rapid and stable recovery and high-precision control under strong disturbance.
Patent Information
- Application Number
- CN202611092158.5
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2026-07-22
- Publication Date
- 2026-11-17
AI Technical Summary
In harsh near-space conditions, the nonlinear drift of the internal parameters of the UAV servo motor is deeply coupled with the external wind shear disturbance. Traditional vector control and active disturbance rejection control cannot simultaneously take into account parameter adaptation and disturbance rejection response, resulting in the dynamic stability of the servo motor under strong disturbances.
A step-by-step model reference adaptive identification method based on Popov's superstability theory is adopted to identify the stator resistance and permanent magnet flux of the permanent magnet synchronous motor in a hierarchical and time-division online manner. The control input gain is updated in real time through an adaptive law. Combined with an extended state observer and dynamic feedforward decoupling technology, external wind load disturbances are isolated and internal parameter drift is compensated, so as to achieve a synergistic improvement in parameter adaptation and disturbance resistance.
Under strong wind shear conditions, the servo motor speed recovery time is shortened to 0.1 seconds, the torque pulsation and current harmonic suppression effects are significant, the observer load is reduced, and the problems of phase lag and observation error in traditional methods are avoided.
Smart Images

Figure CN123113994A_ABST
Abstract
Description
Technical Field
[0001] This invention relates to the field of unmanned aerial vehicle (UAV) servo control technology, and in particular to a method and device for UAV servo anti-disturbance vector control adapted to harsh near-space conditions. Background Technology
[0002] Near space typically refers to the airspace 20 to 100 kilometers above the Earth's surface, situated between the maximum flight altitude of traditional aircraft and the minimum operational orbit of low-Earth orbit satellites. This region has an extremely thin atmosphere, extremely low air pressure, background temperatures that can drop to approximately 233 K, and is characterized by high concentrations of ozone, intense ultraviolet and solar radiation, and severe and unpredictable wind shear induced by gravitational and planetary waves. In recent years, with the increasingly widespread application of high-altitude, low-speed flight platforms such as stratospheric airships, high-altitude scientific balloons, and high-altitude long-endurance unmanned aerial vehicles (UAVs) in missions such as Earth observation, communication relay, and environmental monitoring, the ability of these platforms to remain stationary in near space for extended periods, precisely maintain their flight attitude, and track predetermined trajectories increasingly relies on the performance of their onboard servo systems.
[0003] As the core actuator driving the deflection of control surfaces, UAV servo systems often use permanent magnet synchronous motors (PMSMs) as their power source, combined with field-oriented vector control technology to achieve independent excitation and torque control characteristics close to those of a DC motor. This results in millisecond-level torque dynamic response and large torque output. However, when the aforementioned mature and reliable control schemes are transplanted to near space, their performance is difficult to maintain. The extremely low ambient temperature in near space, coupled with the difficulty of heat dissipation for the motor under low air pressure, and the Joule heating and aerodynamic heating generated by prolonged high-load operation, create a severe cycle of thermal stress. This causes a large-scale and nonlinear drift in the stator resistance, stator inductance, and especially the permanent magnet flux linkage of the PMSM. The rate of change in stator resistance can even exceed 50% across the entire operating temperature range. The permanent magnet flux linkage is also prone to flux density decay and even irreversible demagnetization under alternating high and low temperatures. Traditional vector control current loops typically rely on fixed parameters obtained from nameplates or offline identification for feedforward decoupling. When actual parameters deviate from preset values, the decoupling conditions of the stator voltage equation are disrupted, leading to severe cross-coupling between the d-axis and q-axis currents. This results in increased torque ripple, sluggish dynamic response, and even system instability under extreme conditions. Meanwhile, near-space often experiences strong wind shear exceeding 100 m / s. These sudden airflow changes apply instantaneous, large-gradient aerodynamic torque to the control surfaces. However, dual-loop control based on proportional-integral regulation has limited bandwidth and insufficient ability to suppress nonlinear transient disturbances.
[0004] To enhance disturbance rejection capabilities, existing technologies often incorporate Active Disturbance Rejection Control (ADRC), which uses an extended state observer to treat external load disturbances and internal model uncertainties as a unified total disturbance for real-time estimation and compensation. However, in near-space conditions, the aforementioned drastic drift of internal parameters and abrupt changes in external wind loads often occur simultaneously. In this case, ADRC indiscriminately combines the two, requiring the observer to simultaneously observe both internal parameter perturbations and external aerodynamic disturbances, significantly increasing its workload. Especially when the control input gain mismatches with flux linkage changes, the observer exhibits significant phase lag and observation errors, easily causing a sharp drop in speed or even step loss at the moment of sudden wind shear increase, making it difficult to meet the requirements of rapid recovery and stable operation of such servos under transient strong disturbances. Fundamentally, internal parameter drift and external wind load disturbances are not isolated from each other in near-space conditions, but are deeply coupled: internal parameter drift weakens the execution accuracy of the controller's inner loop, and the loss of inner loop accuracy, in turn, amplifies the damage of external wind load disturbances to the dynamic stability of the outer loop. How to accurately identify and compensate for internal parameter drift while separating external wind shear disturbances from the total disturbances observed by the system, so as to synergistically improve parameter adaptation capability and disturbance resistance capability, has become an urgent technical problem to be solved in this field. Summary of the Invention
[0005] The purpose of this invention is to provide a method and device for anti-disturbance vector control of UAV servo motors that adapts to the harsh working conditions of near space. This method addresses the problem that the nonlinear drift of internal motor parameters caused by near-space thermal stress cycles is deeply coupled with external strong wind shear disturbances, and that traditional vector control and active disturbance rejection control schemes cannot simultaneously take into account parameter adaptation and disturbance rejection response. While accurately identifying and compensating for internal parameter drift, this method separates external wind load disturbances from the total disturbances observed by the system, thereby synergistically improving parameter adaptation capability and disturbance rejection capability.
[0006] To achieve the above objectives, the present invention adopts the following technical solution.
[0007] A disturbance rejection vector control method for UAV servo motors adapted to harsh near-space conditions is disclosed. The servo motor includes a permanent magnet synchronous motor (PMSM) driving the control surface deflection and a three-phase power inverter driving the PMSM. The method executes the following steps in each control cycle: 1) Acquire the three-phase stator current and rotor position of the PMSM, and obtain the d-axis current, q-axis current, electrical angular velocity, and mechanical angular velocity in the dq coordinate system through coordinate transformation; 2) Using the PMSM as a reference model and an adjustable model with the same structure and containing parameters to be identified, constructed in software, as the adjustment object, the difference in stator current between the reference model and the adjustable model is used as the identification error; 3) Based on the adaptive law derived from Popov's hyperstability theory, the stator resistance and permanent magnet flux linkage of the PMSM are identified online; 4) The online identified permanent magnet flux linkage is mapped in real-time to the control input gain of the speed loop active disturbance rejection control, and the control input gain is dynamically adjusted. The identification value of the permanent magnet flux linkage is updated periodically. In the speed loop, an active disturbance rejection control system with an extended state observer is constructed using the control input gain. The extended state observer estimates the mechanical speed and total disturbance, and generates a q-axis current command based on the target speed, estimated speed, and estimated total disturbance. The control input gain perturbation caused by the change in permanent magnet flux linkage is compensated by the real-time mapping and is not included in the total disturbance estimated by the extended state observer, thereby concentrating the observation bandwidth of the extended state observer for estimating external aerodynamic disturbances. In the current loop, the stator resistance and permanent magnet flux linkage obtained online are used to perform dynamic feedforward decoupling on the dq-axis voltage equation. D-axis voltage commands and q-axis voltage commands are generated based on the q-axis current command and d-axis current command. The d-axis voltage command and q-axis voltage command are then space vector modulated to drive the three-phase power inverter, which outputs three-phase voltage to the permanent magnet synchronous motor. Thus, the same online identification result is used simultaneously for voltage decoupling of the current loop and gain mapping of the speed loop, which not only compensates for the damage to the inner loop decoupling accuracy caused by parameter drift, but also separates the internal perturbation from the outer loop observation target.
[0008] As a preferred technical solution of the present invention, the online identification is performed in a step-by-step, time-division manner: First, the stator inductance is taken as a slowly varying parameter. Under the condition that the permanent magnet synchronous motor is running steadily at a speed not lower than the set speed and a limited amplitude d-axis excitation current is injected, the differential term of the stator current is made to approach zero, thereby reducing the stator current state equation to a steady-state algebraic equation. Based on this, the d-axis inductance and q-axis inductance are calculated and treated as known constants in subsequent identification cycles. Then, under the premise that the stator inductance is known, the stator resistance and permanent magnet flux linkage are identified, thereby avoiding the coupling error introduced by the rank reduction of the voltage equation when multiple parameters are identified simultaneously.
[0009] In a preferred embodiment of the present invention, the adaptive laws for identifying the stator resistance and the permanent magnet flux linkage are both proportional-integral (PI) type. Specifically, the identification value of the stator resistance is obtained by a first PI circuit calculating a first error signal, where the first error signal is the algebraic sum of the product of the d-axis identification error and the d-axis current of the adjustable model, and the product of the q-axis identification error and the q-axis current of the adjustable model. The identification value of the permanent magnet flux linkage is obtained by a second PI circuit calculating a second error signal, where the second error signal is the product of the electric angular velocity and the q-axis identification error. The output directions of both the first and second PI circuits are set to negative feedback directions that reduce the identification error, and their integrator initial values are taken from the nameplate values of the stator resistance and the permanent magnet flux linkage, or the most recent convergence value, respectively, to ensure asymptotic convergence of the identification error.
[0010] As a preferred embodiment of the present invention, when the electric angular velocity is lower than a set low-speed threshold, the identification value of the permanent magnet flux linkage is frozen to maintain the most recent valid identification result, and only the identification of the stator resistance is performed; when the electric angular velocity is higher than the low-speed threshold, the stator resistance and the permanent magnet flux linkage are identified simultaneously; and a hysteresis interval is set on both sides of the low-speed threshold to avoid flux linkage identification divergence and suppress frequent switching of identification channels when the back electromotive force signal-to-noise ratio is insufficient.
[0011] As a preferred technical solution of the present invention, within the interval where the stator resistance and permanent magnet flux are simultaneously identified, time-division excitation and alternating updates are adopted: short-term d-axis excitation windows are inserted at set time intervals, a limited amplitude d-axis current is injected into the d-axis excitation window and the stator resistance is identified in a concentrated manner, while the identification value of the permanent magnet flux is frozen, and the identification value of the stator resistance is latched at the end of the d-axis excitation window; outside the d-axis excitation window, the d-axis current command is set to zero, and the permanent magnet flux is identified separately with the identification value of the stator resistance known, thereby separating the identification excitation of the two parameters in the time dimension and ensuring the unique identifiability of the two parameters.
[0012] As a preferred embodiment of the present invention, the extended state observer is a second-order linear extended state observer with the control input gain as the input gain. Its two states are used to estimate the mechanical speed and the total disturbance, respectively. The state gain and disturbance gain are determined by the observer bandwidth by configuring the dual poles of the observer's characteristic polynomial to the same negative real pole. The q-axis current command is obtained by subtracting the estimated total disturbance from the virtual control quantity obtained by proportional control of the difference between the target speed and the estimated speed, dividing by the control input gain, and then limiting the amplitude. Thus, based on proportional feedback, equal feedforward compensation is implemented for external wind load disturbance.
[0013] As a preferred technical solution of the present invention, when performing dead-time compensation on the three-phase power inverter in the space vector modulation, the freewheeling direction during the dead-time period is determined according to the polarity of the current in each phase, and an equivalent correction amount to the duty cycle of the corresponding bridge arm is applied to the dead-time. Furthermore, the equivalent winding temperature is derived from the stator resistance obtained online based on the conductor resistance temperature characteristics, and the preset switching delay compensation table is checked and corrected using the equivalent winding temperature as an index. Thus, the voltage deviation caused by the switching characteristic drift of power devices in low-temperature environments can be compensated without the need to add a temperature sensor.
[0014] This invention also provides a disturbance-resistant vector control device for UAV servo motors adapted to harsh near-space conditions, comprising a permanent magnet synchronous motor driving servo surface deflection, a three-phase power inverter driving the permanent magnet synchronous motor, a current sensor for acquiring three-phase stator current, a rotary encoder for detecting rotor position, a bus voltage detection circuit for detecting DC bus voltage, and a microprocessor; the output terminals of the current sensor, rotary encoder, and bus voltage detection circuit are connected to the acquisition terminal of the microprocessor, and the drive signal output terminal of the microprocessor is connected to the gate drive circuit of the three-phase power inverter; the microprocessor is configured to: transmit the signals from the current sensor and rotary encoder via a... The current, electrical angular velocity, and mechanical angular velocity in the dq coordinate system are obtained by standard transformation. Using the permanent magnet synchronous motor as a reference model and an adjustable model containing the parameters to be identified constructed by software as the adjustment object, the stator resistance and permanent magnet flux of the permanent magnet synchronous motor are identified online based on the adaptive law derived from Popov's superstability theory. The permanent magnet flux obtained online is mapped in real time to the control input gain of the speed loop active disturbance rejection control, so that the control input gain perturbation caused by the change of permanent magnet flux is not included in the total disturbance estimated by the speed loop expanded state observer. Dynamic feedforward decoupling is performed on the dq axis voltage equation of the current loop using the stator resistance and permanent magnet flux obtained online.
[0015] As a preferred embodiment of the present invention, the microprocessor includes a main control module, a sensor data acquisition and coordinate transformation module, an online parameter identification module, an adaptive active disturbance rejection speed control module, a dynamic feedforward decoupling current control module, and a space vector modulation and dead-zone compensation module. Each module runs periodically in the form of an interrupt service subroutine and exchanges data through a global memory pool. The main control module generates a timed interrupt and schedules the sensor data acquisition and coordinate transformation module, the online parameter identification module, the adaptive active disturbance rejection speed control module, the dynamic feedforward decoupling current control module, and the space vector modulation and dead-zone compensation module in sequence according to the timing order within each interrupt cycle.
[0016] As a preferred embodiment of the present invention, the DC side of the three-phase power inverter is connected in parallel with an energy dissipation branch consisting of a braking resistor and a discharge switch connected in series. The microprocessor controls the discharge switch to turn on when the energy is fed back to the DC bus and the bus voltage exceeds the limit due to the rudder surface returning to center or reverse braking, based on the sampling value of the bus voltage detection circuit, so as to dissipate the feedback energy through the braking resistor.
[0017] This invention employs a stepped model reference adaptive identification based on Popov's superstability theory. First, under steady-state conditions satisfying the speed and excitation requirements, the stator inductance is estimated. Then, the stator resistance is identified online separately in the low-speed range, and in the medium-to-high-speed range with sufficient back EMF, the stator resistance and permanent magnet flux linkage are alternately identified using short-time d-axis excitation applied in a time-division manner. This avoids coupling and compensation errors introduced by the rank reduction of the voltage equation when identifying multiple parameters simultaneously. It can obtain the current values of stator resistance and permanent magnet flux linkage online much faster than the temperature drift process, with the flux linkage in the low-speed range maintaining the most recent valid identification result. These identification values are fed back in real-time to the dynamic feedforward decoupling stage of the current loop, thereby eliminating parameter mismatch caused by alternating heating and cooling, ensuring stable decoupling accuracy of the d-axis and q-axis currents over a wide temperature range, and suppressing torque pulsation and current harmonics that occur under fixed parameter schemes.
[0018] Based on this, the present invention maps the identified dynamic flux linkage in real time to the control input gain of the system state equation in active disturbance rejection speed control, instead of taking it as a fixed empirical constant. This removes the measurable internal parameter perturbations caused by temperature changes from the total disturbances that the extended state observer needs to observe, allowing the observer's observation bandwidth and computational resources to be concentrated on estimating unpredictable external aerodynamic wind loads. The observer's observation burden is thus reduced, phase lag is decreased, and the drop in motor speed when strong wind shear increases is reduced, allowing it to recover to steady state within approximately 0.1 s. All of these effects are achieved at the control software logic level, and the temperature information required for power device dead-zone compensation is also obtained by inversion from the identified stator resistance, eliminating the need for additional sensors or power hardware. Attached Figure Description
[0019] Figure 1 A schematic diagram of the hardware structure of a UAV servo anti-disturbance vector control device adapted to harsh near-space conditions provided in an embodiment of the present invention;
[0020] Figure 2 This is a block diagram showing the module composition of the microprocessor's internal control software logic in an embodiment of the present invention;
[0021] Figure 3 This is a block diagram illustrating the principle of the online parameter identification module in an embodiment of the present invention.
[0022] Figure 4This is a block diagram illustrating the principle of the adaptive active disturbance rejection speed control module in an embodiment of the present invention.
[0023] Figure 5 A flowchart of an anti-disturbance vector control method for UAV servo motors adapted to harsh near-space conditions provided in an embodiment of the present invention;
[0024] Figure 6 This is a comparison curve of the rotational speed response of the embodiment of the present invention and the traditional fixed parameter scheme under the condition of sudden wind load change.
[0025] In the diagram: 1. Permanent magnet synchronous motor; 2. Three-phase power inverter; 3. Microprocessor; 4. Current sensor; 5. Rotary encoder; 6. Bus voltage detection circuit; 7. Control surface; 8. Reduction gear transmission mechanism; 9. Flight control computer; 100. Main control module; 101. Sensor data acquisition and coordinate transformation module; 102. Online parameter identification module; 103. Adaptive and active disturbance rejection speed control module; 104. Dynamic feedforward decoupling current control module; 105. Space vector modulation and dead zone compensation module. Detailed Implementation
[0026] To make the objectives, technical solutions, and advantages of this invention clearer, the invention will be further described in detail below with reference to the accompanying drawings and specific embodiments. It should be understood that the specific embodiments described herein are merely illustrative and not intended to limit the invention. All other embodiments obtained by those skilled in the art based on the embodiments of this invention without inventive effort are within the scope of protection of this invention.
[0027] This embodiment focuses on a high-altitude, long-endurance unmanned aerial vehicle (UAV) capable of sustained cruising at altitudes between 20 and 30 kilometers. In this airspace, the atmosphere is extremely thin, convective heat dissipation is almost nonexistent, and the onboard electronic equipment and motors primarily rely on thermal radiation and conduction for heat dissipation. Background temperatures can be as low as approximately 233 K, accompanied by high concentrations of ozone and strong ultraviolet radiation. Furthermore, strong wind shear, often reaching speeds of 100 m / s, frequently occurs, causing abrupt changes in aerodynamic torque acting on the control surfaces within milliseconds. Simultaneously, the Joule heat accumulated by the motor windings during prolonged high-load operation under low air pressure interacts with the external cryogenic environment, creating a severe cycle of thermal stress. Taking the surface-mount permanent magnet synchronous motor with a rated power of approximately 800 W, a pole pair count of 4, and a rated speed of approximately 3000 r / min used in this embodiment as an example, its stator resistance is approximately 0.55 Ω at room temperature of approximately 293 K. Due to the characteristic that the resistivity of copper conductors changes approximately linearly with temperature, it drops to approximately 0.42 Ω when cryogenically immersed at approximately 233 K. However, when the winding temperature is raised to approximately 370 K due to pneumatic heating and Joule heating, it rises again to approximately 0.72 Ω. The resistance ratio between the high-temperature end and the cryogenic end is close to 1.7, and the resistance change rate over the entire operating temperature range exceeds 50%. The permanent magnet flux linkage is approximately 0.15 Wb at room temperature. Due to the reversible temperature coefficient of approximately -1.1 per Kelvin for neodymium iron boron permanent magnet materials, the drift amount under alternating cryogenic and high-temperature conditions can reach more than 8%, and there is even a risk of irreversible demagnetization under extreme high temperatures. It is precisely this working condition, in which the severe nonlinear drift of internal parameters and the external strong wind load disturbance highly coincide in time and are deeply coupled in physical mechanism, that constitutes the technical obstacle that the anti-disturbance vector control device and corresponding control method provided in this embodiment need to overcome.
[0028] See Figure 1 The UAV servo anti-disturbance vector control device adapted to the harsh conditions of near space provided in this embodiment includes a permanent magnet synchronous motor 1, a three-phase power inverter 2, a microprocessor 3, a current sensor 4, a rotary encoder 5, a bus voltage detection circuit 6, and control surfaces 7 and a reduction gear transmission mechanism 8 as the driven objects. The entire device establishes a two-way interactive channel of command and status with the host flight control computer 9 through an airborne data bus. The permanent magnet synchronous motor 1 adopts a surface-mount or built-in structure customized for aerospace applications. The winding insulation material is selected to be ozone-resistant and UV-resistant. The bearings are filled with wide-temperature-range low-temperature grease to adapt to the cryogenic environment. Its stator three-phase windings are electrically connected to the AC output terminal of the three-phase power inverter 2. The output shaft of the permanent magnet synchronous motor 1 is connected to the rotating shaft of the rudder surface 7 via the reduction transmission mechanism 8. While amplifying the output torque and matching the deflection rate of the rudder surface, the reduction transmission mechanism 8 also pushes the transient aerodynamic disturbance torque caused by wind shear on the rudder surface 7 back to the motor rotor side according to the transmission ratio, which constitutes the main source of external disturbance that this device aims to suppress.
[0029] The rotary encoder 5 is a high-resolution absolute photoelectric encoder or magnetoelectric encoder, coaxially mounted with the shaft of the permanent magnet synchronous motor 1. It is used to detect the rotor's mechanical angle in real time and process the signals to obtain electrical angle and speed signals. In this embodiment, its resolution is no less than 17 bits per revolution to ensure a sufficient signal-to-noise ratio when calculating the speed using angle differentiation in the low-speed range. The current sensor 4 is used to collect the three-phase stator current. , , The current sensor can be a magnetic balance Hall current sensor installed on the phase line between the three-phase power inverter 2 and the permanent magnet synchronous motor 1, or a sampling resistor can be connected in series in the lower tube branch of the bridge arm of the three-phase power inverter 2. In the latter case, the complete three-phase current is restored by the phase current reconstruction stage described later. The bus voltage detection circuit 6 is connected in parallel across the DC bus of the three-phase power inverter 2. The DC bus voltage is obtained by resistor voltage division and sampling by isolation operational amplifier. The detection result is used for the normalization calculation of the duty cycle in the space vector modulation described later, and is also used as a criterion for the bus overvoltage and undervoltage protection, as well as the activation and deactivation of the energy dissipation branch. The three-phase power inverter 2 adopts a radiation-hardened design. Its DC side is connected in parallel with a large-capacity thin-film support capacitor to absorb bus ripple. It also has an energy dissipation branch consisting of a braking resistor and a discharge switch in series. When the steering surface 7 returns to the center or reverse braking, causing energy to be fed back to the bus and the bus voltage rises beyond the limit, the microprocessor 3 controls the discharge switch to turn on according to the sampling value of the bus voltage detection circuit 6, and dissipates the feedback energy as heat through the braking resistor. The six power switching devices are selected from insulated gate bipolar transistors or silicon carbide metal oxide semiconductor field-effect transistors and are configured with anti-parallel freewheeling diodes. The gate is driven by a dedicated driver chip with isolation and undervoltage lockout functions. The AC side outputs a three-phase pulse width modulation voltage to drive the permanent magnet synchronous motor 1. Microprocessor 3 is a high-frequency digital signal processor with a hardware floating-point unit, such as a TMS320F28335 chip with a clock frequency of not less than 150 MHz, or a dedicated system-on-a-chip for motor control. Its analog-to-digital conversion peripheral's sample-and-hold channel is electrically connected to the outputs of current sensor 4 and bus voltage detection circuit 6, respectively, and is synchronously triggered by a pulse-width modulation carrier. Its synchronous serial communication interface is connected to rotary encoder 5 to read absolute angle data cycle by cycle. Its enhanced pulse-width modulation peripheral outputs six complementary drive signals with dead time to the gate drive circuit of three-phase power inverter 2. Thus, at the hardware level, a complete closed loop is formed from sensing, computation, power drive to mechanical execution.
[0030] See Figure 2The focus of this invention is the control software logic running inside the microprocessor 3. This software logic is functionally divided into a main control module 100, a sensor data acquisition and coordinate transformation module 101, an online parameter identification module 102, an adaptive active disturbance rejection speed control module 103, a dynamic feedforward decoupling current control module 104, and a space vector modulation and dead-zone compensation module 105. Each module runs periodically as an interrupt service routine and interacts with the global memory pool via a software bus. A dedicated shared data area is allocated in the global memory pool to store the physical parameters obtained from online identification, observer state variables, instructions at all levels, and excitation request flags. Each module reads and writes to this data area according to a fixed sequence, thereby avoiding data contention caused by concurrent access from multiple modules. The main control module 100 is responsible for configuring the system clock tree, initializing peripherals, clearing global variables, and managing fault flags. It generates timed interrupts at an execution frequency of 10 kHz to 20 kHz. Within each interrupt cycle, the module performs scheduling sequentially according to the following order: sensor data acquisition and coordinate transformation module 101, online parameter identification module 102, adaptive active disturbance rejection speed control module 103, dynamic feedforward decoupling current control module 104, and space vector modulation and dead-time compensation module 105. In this embodiment, the interrupt frequency is 10 kHz, corresponding to the control cycle. The value is 100 μs. Considering that the dynamic bandwidth of the outer speed loop is much lower than that of the inner current loop, in order to save computing resources, the adaptive self-disturbance rejection speed control module 103 can also adopt a frequency division execution strategy, that is, the speed loop operation is performed once every two or four current loop cycles, while sensor sampling, coordinate transformation, current decoupling and space vector modulation always run cycle by cycle; when using frequency division execution, the control cycle in equation (10) described later is... The actual execution cycle of the adaptive active disturbance rejection speed control module 103 should be taken accordingly, and the product of the bandwidth of the extended state observer (described later) and the actual execution cycle should be ensured to be within the allowable range of stability and accuracy of the forward Euler discretization, for example, not greater than 0.3. If necessary, the observer bandwidth should be adjusted downward accordingly. This arrangement reduces the processor load without sacrificing the outer loop stability margin and does not affect the implementation effect of this scheme.
[0031] After an interrupt is triggered, the sensor data acquisition and coordinate transformation module 101 reads the raw signals from the current sensor 4 and the rotary encoder 5 at high speed. It first performs zero-point calibration and range restoration on the sampled values, then performs digital low-pass filtering to suppress switching noise and quantization glitches. When using a dual-resistor or single-resistor bridge arm sampling scheme, the sensor data acquisition and coordinate transformation module 101 selects an effective sampling window based on the modulation sector of the current pulse width modulation and reconstructs the missing phase current according to Kirchhoff's current law to obtain the complete three-phase stator current. Based on this, the current in the three-phase stationary coordinate system is mapped to the two-phase stationary α-β coordinate system via Clarke transform, and then rotated to the dq coordinate system synchronized with the rotor magnetic field via Park transform to obtain the excitation component current. With torque component current Simultaneously, the absolute angle signal is subjected to continuous unwinding and differential filtering to calculate the rotor electrical angle θ and electrical angular velocity. With mechanical angular velocity The aforementioned state variables, along with their previous cycle values, are written into the global memory pool for subsequent modules to read.
[0032] See Figure 3 The online parameter identification module 102 is constructed based on a model reference adaptive mechanism, where the reference model is the physical motor body, and its actual current response is directly obtained through sensor sampling. The stator current state equation of the permanent magnet synchronous motor 1 in the dq coordinate system is affected by the alternating extreme temperatures in the near space and can be expressed as equation (1):
[0033]
[0034] In the formula, , These are the stator voltages along the d-axis and q-axis. , These are the stator currents along the d-axis and q-axis. , For the d-axis and q-axis stator inductance, For stator resistance, It is a permanent magnet flux chain. Let be the electric angular velocity. For the surface-mount motor used in this embodiment, it can be approximated as... and Equal to, with a value of approximately 3.2 mH. An adjustable model with the same structure as equation (1) is iteratively constructed in software within the memory of microprocessor 3, containing the unknown physical parameters to be identified, with superscript " The variable " represents the estimated value of the system at the current moment, as shown in equation (2):
[0035]
[0036] The voltage input of the adjustable model is taken from the actual voltage command issued in the previous control cycle. , This ensures that the reference model and the adjustable model are subjected to strictly the same physical excitation, thereby eliminating identification bias introduced by inconsistent inputs. Define arbitrary discrete time intervals. The state current identification error matrix is This error reflects both the degree of mismatch between the adjustable model parameters and the true parameters, and is also the signal source that drives the adaptive law to continuously correct the estimated value.
[0037] It is worth noting that Equation (1) can only provide two independent scalar constraints at any given moment. If the three types of parameters, namely resistance, inductance, and flux linkage, are included in the online identification at the same time, the number of unknowns will exceed the number of independent equations. Due to the rank reduction of the voltage equation, the solution will lose its uniqueness. In actual operation, the estimated parameters will be coupled and compensate for each other, resulting in continuous oscillations or even divergence. To avoid this rank deficiency dilemma, the online parameter identification module 102 adopts a step-by-step identification and judgment logic to extract parameters with different temperature sensitivities in layers and time-sharing. Considering that the stator inductance mainly depends on the magnetic circuit structure and the saturation degree of the iron core, its sensitivity to temperature changes is much lower than that of the winding resistance and the flux linkage of the permanent magnet. Therefore, the inductance is placed as a slowly changing parameter in the first step of the identification process for separate estimation. After the estimation is completed, it is regarded as a known constant in the subsequent identification cycle and participates in the calculation. Inductance prediction is performed under the condition that the motor is operating in steady state at an electrical angular velocity not less than 20% of the rated value: During the power-on self-test phase, the main control module 100 drives the motor to the speed range according to the preset self-tuning process and maintains a steady state; during the steady-state cruise phase, the operating range that meets the speed condition is directly selected; during the prediction period, the parameter online identification module 102 sets the excitation request flag to the dynamic feedforward decoupling current control module 104 through the global memory pool, and the module then sends the excitation current command accordingly. The current temporarily switches from zero to a limited injection value, such as 10% to 20% of the rated current, and returns to zero after tens of milliseconds. For surface-mount motors, the d-axis current does not generate reluctance torque, and this injection process does not disturb the operation of the speed loop. During the injection period, the differential term of the current is approximately zero, and equation (1) degenerates into a steady-state algebraic equation, from which the two-axis inductance can be solved inversely, as shown in equation (3):
[0038]
[0039] In the formula, the subscript "0" represents the steady-state average value over several sampling periods during the prediction phase. , Retrieve the value from the motor nameplate or the stored value after the most recent identification convergence. Solving the second equation of equation (3), we get the denominator. Because of the aforementioned lower limit condition of rotational speed and the actively injected non-zero d-axis current, the numerical condition is always non-zero; Solving equation (3) from the first equation, we find that its denominator contains the q-axis current. Therefore, it is only executed in the load steady-state range when the q-axis current amplitude exceeds the set threshold. For the surface-mount motor with symmetrical d- and q-axis magnetic circuits in this embodiment, it can be directly used when under no-load or light-load conditions and when the q-axis current is below the threshold. equal Since the permeability of the iron core changes very little with temperature, this prediction remains valid over a long timescale and is only refreshed when the overall temperature of the machine undergoes a large step change.
[0040] After completing the inductance prediction, the online parameter identification module 102 enters the online identification ladder of stator resistance and permanent magnet flux linkage. To ensure that the identification error asymptotically converges under the complex noise and gusty wind load excitation in the near space, this embodiment adopts the Popov superstability theory to derive the adaptive law: compared with the approach of explicitly constructing the Lyapunov function for the error system, this framework decomposes the convergence condition into two conditions that can be verified separately: the passivity of the forward channel and the integral inequality of the nonlinear feedback link, which facilitates the direct acquisition of the proportional integral type adaptive law that is easy to implement in engineering. The error dynamic system obtained by subtracting equation (1) from equation (2) is equivalent to a feedback interconnection structure composed of a linear main channel and a nonlinear time-varying feedback channel, wherein the main channel is composed of the resistance damping term and the speed coupling term of the error state equation, and the nonlinear feedback channel contains a set of terms including the resistance damping term and the speed coupling term of the error state equation. , The undetermined adaptive expression. For the surface-mount motor with equal d-axis and q-axis inductances in this embodiment, the speed coupling term in the main channel coefficient matrix is antisymmetric and does no work on the quadratic energy of the error. The symmetric part of this matrix is the diagonal element equal to Because it is a negative fixed diagonal matrix, the main channel remains strictly passive for any operating speed and any positive stator resistance. For built-in motors with unequal d-axis and q-axis inductances, the same passive structure can be obtained through an equivalent transformation using the stator flux linkage components as state variables. Under the premise of a passive main channel, as long as the output of the nonlinear feedback loop... If the Popov integral inequality shown in equation (4) is satisfied at any time, then the interconnected system is asymptotically superstable, and the error matrix is... Asymptotically converges to zero:
[0041]
[0042] In the formula Let be any finite positive constant related to the initial state. Expanding and solving the integrand of equation (4) according to the parameter to be identified, we can obtain the proportional-integral type adaptive update law. The online update logic of stator resistance and permanent magnet flux linkage is shown in equations (5) and (6), respectively:
[0043]
[0044] (6)
[0045] In the formula, , The proportional and integral coefficients of the resistance adaptive law. , The proportional and integral coefficients of the magnetic flux adaptive law are both taken as positive values. For the Laplace operator, the initial values of the integrators for the two update laws are set to the nameplate resistance value and the nameplate flux linkage value, or the stored value after the most recent convergence, respectively. The negative signs on the right-hand side of equations (5) and (6) ensure that the parameter correction is carried out in the negative feedback direction that reduces the identification error: taking the resistor channel as an example, when When the resistance is less than the actual resistance, the adjustable model has insufficient damping and the estimated current is too high. When the value is negative, equation (5) is driven accordingly. Increase and approximate the true value, and vice versa; substitute this form into equation (4) to verify that the lower bound of the integral is determined by the initial parameter deviation and is always a finite value, thus Popov's integral inequality holds. In terms of signal composition, and These represent the regression directions of resistance mismatch and flux mismatch in the error equation, respectively. The proportional term in the above proportional-integral structure provides the identification with fast tracking capability, while the integral term eliminates steady-state deviation. The specific coefficients can be tuned by those skilled in the art based on motor parameters and sampling noise levels, while ensuring stability margin. Generally, the equivalent bandwidth of the resistance identification loop is slightly higher than that of the flux identification loop to accommodate the more drastic changes in resistance with temperature.
[0046] It should be noted that the method normally used in this embodiment is... Under operating conditions, the d-axis current is approximately zero, and the effects of resistance mismatch and flux mismatch converge on the q-axis error path, with their instantaneous return directions collinear. If two renewal laws are simultaneously activated in the steady-state region under constant speed and load, any law that causes the resistance deviation and flux deviation to revert to their respective values will result in a loss of d-axis current. The mutually canceling combination of estimates will cause the error to approach zero. and The two sides will compensate for each other and fail to converge uniquely to the true value, which is the same as the rank reduction problem of the three-parameter co-identification mentioned above. To this end, the online parameter identification module 102 further adopts a time-sharing excitation and alternating update mechanism in the medium- and high-speed joint identification ladder: it sets the excitation request flag at a set time interval, for example, once every 1 s to 5 s, and the dynamic feedforward decoupling current control module 104 inserts a short d-axis excitation window accordingly, the window lasts from 10 ms to 50 ms, and the injected amplitude is equivalent to that of the inductance prediction stage; during the window period The value is not zero, and flux mismatch does not directly enter the d-axis error equation, while resistance mismatch passes through a non-zero value. It is directly displayed on the d-axis channel, and equation (5) uses the error information within this window for centralized correction. And latch at the end of the window, while freezing equation (6); outside the normal range, equation (5) holds the latched value or operates with minimal gain, and equation (6) in Corrected separately via the q-axis channel, given the prior knowledge. Thus, the identification excitations of the two parameters are separated in the time dimension, and the two-parameter identification restores its unique identifiability. The time constants of resistance and flux drift caused by temperature changes are on the order of tens of seconds or even minutes, which is much slower than the second-level time-division identification cycle. Therefore, the above mechanism is sufficient to track the parameter evolution caused by the alternation of hot and cold in near space without significant lag.
[0047] In discrete implementation, the current differential equation of the adjustable model shown in equation (2) controls the period. Following the forward Euler method, the integral steps in equations (5) and (6) are discretized using the backward Euler or trapezoidal method, and a reasonable limit is applied to the identification results. and Each is constrained within a physically reliable range estimated based on the characteristics of the motor material; once a certain identification quantity exceeds the limit due to instantaneous strong noise, it is truncated at the boundary value and the identification holding of this cycle is triggered to prevent abnormal values from propagating to subsequent stages. Considering that flux linkage identification essentially relies on back electromotive force information, when the electric angular velocity is lower than the set low speed threshold, the back electromotive force amplitude is too small and the signal-to-noise ratio of the input signal of equation (6) deteriorates significantly. At this time, the parameter online identification module 102 freezes the flux linkage update channel, making The previous effective convergence value is maintained, and only equation (5) is retained for the identification of the stator resistance. If necessary, the signal-to-noise ratio is enhanced by using a short-time d-axis excitation window. When the electric angular velocity exceeds the threshold and enters the medium-to-high speed region, and the back electromotive force is fully manifested, the dual-parameter identification is carried out according to the aforementioned time-sharing alternation mechanism. In this embodiment, the low-speed threshold is taken as about 10% of the rated electric angular velocity, corresponding to about 126 rad / s, and a hysteresis interval of about 5% is set on both sides of the threshold to avoid frequent switching of the identification channel and the introduction of jitter when the rotational speed crosses back and forth near the threshold. With the help of the above-mentioned step-like mechanism, the parameter online identification module 102 can stably track the decrease in winding resistance caused by the cryogenic environment, the increase in resistance caused by Joule heat accumulation, and the flux drift caused by temperature stress at a speed much faster than the thermal time constant of the device. and The data is written to the global memory pool in real time, so that the adaptive self-disturbance rejection speed control module 103 and the dynamic feedforward decoupling current control module 104 can directly read and call it.
[0048] See Figure 4 The adaptive active disturbance rejection speed control module 103 constructs a speed outer loop based on a linear active disturbance rejection control framework. Compared with the traditional active disturbance rejection structure that relies on nonlinear functions and requires tuning of a large number of parameters, the linear framework can be tuned using only two physical quantities: the observer bandwidth and the controller bandwidth. It is easier to ensure stability when facing high-frequency sudden wind loads and is more convenient for engineering application. The mechanical motion equation of the permanent magnet synchronous motor 1 when running in near space is shown in equation (7):
[0049]
[0050] In the formula, The rotor's mechanical angular velocity, This represents the number of pole pairs of the motor. The rotational inertia of the entire servo drivetrain, including the reduced inertia of the reduction gear mechanism 8 and the control surface 7, is approximately 3 × 10⁻⁶ in this embodiment. -4 kg·m², The coefficient of viscous friction, This is the high-frequency external aerodynamic disturbance torque caused by wind shear, which is pushed back to the rotor side via the rudder surface 7 and the reduction gear mechanism 8. Equation (7) is rewritten in the standard single-input state-space form:
[0051] (8)
[0052] in Defined as the total system disturbance, it encompasses internal uncertainties such as internal viscous friction, unmodeled cogging torque, and transmission backlash, as well as external wind loads. The combined effect To control the input gain coefficient, conventional linear active disturbance rejection design will... Taking a fixed empirical constant, once the magnetic flux linkage drifts significantly due to high-temperature demagnetization or changes in magnetic flux density under cryogenic conditions, the deviation term between the actual physical gain and the mathematical model gain will be affected. Treating the entire extended state observer as part of an external disturbance not only encroaches on the observer's limited observation bandwidth but also induces low-frequency oscillations and phase lag in the observation output during rapid parameter perturbations, ultimately causing a significant drop in speed or even loss of synchronization during sudden increases in wind shear. The adaptive active disturbance rejection speed control module 103 changes the conventional design of isolating the identification and disturbance rejection control stages. In each control cycle, it reads the flux linkage estimate output by the online parameter identification module 102 from the global memory pool and dynamically maps and corrects the control gain.
[0053]
[0054] Substituting the room temperature parameters of this embodiment, It is approximately 3000, and when the magnetic flux drifts under cryogenic or high-temperature conditions, The model gain is then updated periodically, ensuring it closely follows the actual physical gain. Through this mapping, the measurable internal component of the magnetic flux perturbation caused by temperature—originally mixed in with the total disturbance—is separated from the observed target and directly compensated by the model. This reduces the dimensionality and workload of the extended state observer, allowing its observation bandwidth and computational resources to be concentrated on estimating unpredictable external aerodynamic wind loads. This is achieved through periodically updated dynamic gain. Based on this, the forward Euler method is used to discretize the second-order linear extended state observer, where the observer state... Used to track and estimate machine speed Expansion state Specifically designed for estimating the total disturbance after removing flux linkage fluctuations, the discrete recursive relationship is shown in equation (10):
[0055]
[0056] In the formula, For the observer velocity tracking error, , These represent the observer state gain and the disturbance gain, respectively. For ease of tuning, based on pole placement theory, the double poles of the observer characteristic polynomial are placed on the negative real axis of the complex plane. place, thereby obtaining , , Configure parameters for the observer bandwidth. In this embodiment, the observer bandwidth... Take approximately 1200 rad / s, corresponding to Approximately 2400 Approximately 1.44 × 10 6 Since the main part of the model gain bias contributed by flux drift has been offset at the source by equation (9), the remaining bias only comes from the nominal error of the moment of inertia and the flux identification residual, and the magnitude is small, thus expanding the state. The components were significantly purified, and the acceleration disturbance component caused by wind shear aerodynamic torque was mainly characterized. The remaining unmodeled terms are only viscous friction, cogging torque, and other relatively small internal terms. The effective bandwidth of the observer can be concentrated on the estimation of the symmetrical wind load, and the phase lag is significantly reduced. To further suppress the impact of the desired command step on the observer, the adaptive active disturbance rejection speed control module 103 sets a tracking differentiator or a first-order inertial smoothing element at the command entry point to soften the transition of the target speed issued by the flight control computer 9, and arranges a transition process curve that is both fast and without overshoot.
[0057] After obtaining the disturbance estimate, the adaptive active disturbance rejection speed control module 103 executes a linear error feedback control law. The target speed is then smoothed. With observed rotational speed After the difference is calculated, proportional control is applied, as shown in equation (11); then, equal feedforward compensation is applied to the disturbance estimate and dynamic gain normalization and limiting processing are performed to generate the final q-axis reference current command issued to the inner current loop, as shown in equation (12):
[0058]
[0059]
[0060] In the formula, This is a virtual control quantity without disturbance compensation, and its dimension is angular acceleration; For controller gain, and controller bandwidth Strong correlation, usually taken In this embodiment, the controller bandwidth Take approximately 300 rad / s, observer bandwidth Adjusted to To ensure that the observation speed is 3 to 5 times faster than the feedback speed and to avoid observation lag dominating the closed-loop dynamics; Represents the amplitude limiting function. The current limit value is set based on the allowable current of the three-phase power inverter 2 and the overcurrent capability of the permanent magnet synchronous motor 1. The numerator in equation (12) This embodies the core idea of active disturbance rejection control, which is to actively compensate and cancel out the observed wind load disturbances on the basis of proportional feedback, making the equivalent controlled object approximate as a pure integral element, thereby improving the stiffness of the servo motor against gust disturbances. This limiting element can prevent the current command from exceeding the safe operating range of the power device under extreme gust conditions, and, in conjunction with the integral anti-saturation strategy, avoids response hysteresis and overshoot when the command exits saturation.
[0061] The dynamic feedforward decoupling current control module 104 receives the output from the adaptive active disturbance rejection speed control module 103. and excitation current command In this embodiment, the normal operation is as follows: The control strategy aims to maximize the torque output per unit current and reduce copper loss; when the parameter online identification module 102 sets the excitation request flag, the dynamic feedforward decoupling current control module 104 will, within the corresponding window period, The injection amplitude is temporarily switched to the requested value and automatically returns to zero after the window ends. Due to the combined effects of the heat dissipation difficulties caused by the ultra-low air pressure in near-space and the high-speed requirements, the dq-axis voltage equation exhibits severe nonlinear cross-coupling caused by the rotating electromotive force. If feedforward decoupling is still performed using the nameplate-fixed parameters, the decoupling conditions will be disrupted as the parameters drift, causing cross-perturbations in the d-axis and q-axis currents, leading to integral saturation and overmodulation of the regulator. Therefore, the dynamic feedforward decoupling current control module 104 extracts the latest identified value from the global memory pool in each cycle. and The voltage command calculation with dynamic compensation including cross-coupling terms is performed according to equations (13) and (14):
[0062] (13)
[0063]
[0064] In the formula, , and , These are the proportional and integral coefficients of the d-axis and q-axis current regulators, respectively. , Take the predicted result of equation (3). The last term of equation (13) and the term in equation (14) contain... The terms respectively cancel out the coupling voltage induced on the d-axis by the q-axis current through rotation and the coupling voltage induced on the q-axis by the d-axis current, in equation (14). The feedforward compensation term compensates for the rotational back electromotive force that changes in real time with temperature. Equations (13) and (14) are written in continuous domain operators for ease of expression, and in implementation, both regulators are also based on the control cycle. Discrete, the integral term employs the backward Euler method and is configured to resist integral saturation. To ensure the closed-loop bandwidth of the current loop remains constant under parameter drift and to avoid dynamic degradation due to the disappearance of zero-pole pairs caused by resistance changes, the coefficients of each regulator are based on the desired current loop bandwidth. And based on the online identification results, adaptive retuning is performed according to the internal model control principle, as shown in equation (15):
[0065] (15)
[0066] In this embodiment, the current loop bandwidth is... Taking approximately 2500 rad / s, which is far higher than the bandwidth of the outer loop, ensures that the inner loop can approximate an ideal first-order inertial or even proportional element relative to the outer loop. Thanks to... The q-axis back EMF feedforward term is introduced without hysteresis, and the integral coefficient increases with... Even when the device is in a cryogenic environment of about 233 K and is subjected to intense alternating aerodynamic heating, and the permanent magnet characteristics and winding resistance change drastically at the same time, Equations (13) and (14) can still maintain accurate voltage feedforward decoupling, effectively suppressing torque pulsation and current harmonics induced by the cross coupling of d-axis and q-axis currents under the fixed parameter scheme, and preventing the inverter from entering the overmodulation region.
[0067] The space vector modulation and dead-time compensation module 105 will transmit voltage commands. , After mapping to a two-phase stationary coordinate system by the inverse Park transformation, we obtain , The reference voltage vector is normalized by combining the DC bus voltage collected by the bus voltage detection circuit 6. Based on this, the spatial sector in which it is located is determined, and the action time of two adjacent basic voltage vectors and the zero vector is calculated according to the volt-second balance principle. This generates a seven-segment pulse width modulation duty cycle waveform with saddle wave envelope characteristics, which improves the inverter bus voltage utilization rate by about 15% compared to sinusoidal pulse width modulation. Considering that the low temperature environment in the vicinity will cause the turn-on and turn-off delays, conduction voltage drops, and gate thresholds of power switching devices to drift, thus causing deviations between the actual output voltage and the command voltage during the dead time and aggravating current waveform distortion, the space vector modulation and dead time compensation module 105 determines the direction of the freewheeling path during the dead time based on the measured polarity of the current in each phase, applies a compensation correction amount equivalent to the dead time to the duty cycle of the corresponding bridge arm, and presets a switching delay compensation table indexed by temperature. The temperature information required for compensation does not require an additional temperature sensor, but is output by the parameter online identification module 102. The equivalent temperature of the winding is obtained by inverting the resistance-temperature characteristics of the copper conductor. This temperature is then used as a slow variable characterizing the internal heating and cooling levels of the device for lookup table interpolation correction. After the above processing, the space vector modulation and dead-zone compensation module 105 finally outputs six complementary drive signals to the gate drive circuit of the three-phase power inverter 2, driving each switching device of the inverter bridge to turn on and off according to the calculated timing sequence, outputting three-phase AC power to the stator of the permanent magnet synchronous motor 1.
[0068] The following is combined Figure 5 Taking a complete interrupt cycle as an example, the execution flow of the control method in this embodiment and the signal transmission path between modules are explained. When the UAV is cruising in the stratosphere at an altitude of 20 to 30 kilometers, and its yaw or pitch control surface 7 is simultaneously subjected to continuous deep cooling and occasional strong wind shear, the timer of the main control module 100 triggers an interrupt service subroutine every 100 μs. After the interrupt is entered, the sensor data acquisition and coordinate transformation module 101 reads the original signals from the current sensor 4 and the rotary encoder 5, completes calibration, filtering, phase current reconstruction, and Clarke and Parker transforms, and obtains... , , and And write it to the global memory pool. The processed state quantity is compared with the delayed voltage instruction of the previous cycle. , The data flows into the online parameter identification module 102, which substitutes the actual applied voltage excitation into the differential equation matrix of the adjustable model constructed in software, and calculates the difference in current output between the reference model and the adjustable model in parallel. , Based on the current speed range, the corresponding identification channel is activated, and the estimated value is corrected along the direction of error reduction by the adaptive law of equations (5) and (6), and the estimated resistance value reflecting the current temperature state of the stator winding is output. And the estimated flux linkage reflecting the current magnetic state of the permanent magnet. The data is then immediately updated to the shared data area of the global memory pool. If the current cycle is within the time-sharing excitation window, the online parameter identification module 102 also requests the dynamic feedforward decoupling current control module 104 to apply a short-term d-axis excitation current through the global memory pool to refresh the resistance identification results.
[0069] The data stream smoothly transitions to the adaptive disturbance rejection speed control module 103. At this time, the external wind shear aerodynamic torque has been pushed back to the motor rotor by the reduction gear transmission mechanism 8, at the mechanical speed... The upper part agitates the fluctuation. The adaptive self-disturbance rejection speed control module 103 first presses equation (9) with the latest Reconstruct control gain And then take the current , and the output current at the previous moment The observer recursion is performed to input equation (10). Since the deviation between the model gain and the actual gain is updated periodically... Significant compression, with the remaining deviation stemming solely from the nominal error of rotational inertia and the residual error in flux identification, in an expanded state. This allows for a relatively low hysteresis and a clean reflection of the aerodynamic disturbance torque amplitude applied to the control surface 7 by the current strong wind shear. After obtaining this disturbance estimate, the adaptive active disturbance rejection speed control module 103 fuses the target rotational speed according to equations (11) and (12). Observation speed With disturbance quantity The torque-current command is generated by algebraic feedforward subtraction to simultaneously compensate for the effects of sudden changes in wind shear load and low-temperature parameter drift. This instruction, along with the normal... Seamlessly transferred to the dynamic feedforward decoupling current control module 104, which uses the latest Reset the regulator gain according to equation (15), and utilize... The cross-coupled electromotive force term completes the feedforward decoupling operation of equations (13) and (14), and outputs a voltage control vector that achieves precise decoupling under severe parameter variations. , The voltage vector input to the space vector modulation and dead zone compensation module 105 completes vector modulation, dead zone, and cryogenic drift correction, then drives the three-phase power inverter 2. The electromagnetic torque generated by the three-phase AC power output from the inverter in the permanent magnet synchronous motor 1 is updated cycle by cycle, continuously offsetting the combined disturbances exerted on the servo system by cryogenic conditions, low air pressure, and strong wind shear, enabling the control surface 7 to accurately follow the attitude commands issued by the flight control computer 9. At this point, the complete business process of a single interruption cycle is completed, and the system waits for the next timed interruption to arrive, thus maintaining high-rigidity servo control of the control surface attitude in a continuous cycle.
[0070] To verify the actual effect of this embodiment, a comparative test was conducted on a semi-physical simulation test bench, comparing it with a traditional linear active disturbance rejection vector control scheme using fixed nameplate parameters. During the test, a temperature and low-pressure combined environment test chamber was used to simulate a deep cryogenic immersion environment of approximately 233 K and a low-pressure heat dissipation-limited environment. When the motor was running at a steady speed of half its rated speed, a step aerodynamic torque equivalent to 75% of the rated load was suddenly applied to simulate the sudden load change of the large-area control surface 7 under a strong wind shear on the order of 100 m / s. Figure 6 Comparison curves of the speed response of the two schemes under the same disturbance excitation are presented. The control scheme, due to its fixed control gain and severe mismatch with the actual flux linkage under cryogenic conditions, exhibits significant phase lag and steady-state observation errors in the extended state observer. The speed drops by approximately 18% at the instant of the disturbance, and only barely recovers after 2 to 3 overshoot cycles, with a recovery time exceeding 0.3 s. In contrast, the scheme in this embodiment reduces the speed drop to less than 5% at the instant of the same disturbance, and fully recovers to steady state within approximately 0.1 s. The adjustment process is smooth and monotonous, with virtually no overshoot cycle. In the full-temperature range scanning test from cryogenic to high temperature, thanks to the real-time online identification of resistance and flux linkage and its cross-module feedforward intervention to both the inner and outer loops, the decoupling accuracy of the dq-axis current remains at the design level throughout the entire temperature range. The total harmonic distortion rate of the steady-state phase current is reduced from over 6% in the control scheme to less than 2%, and the output torque ripple is also reduced from approximately 12% to less than 4%. The aforementioned performance improvements are entirely achieved at the software logic level of the microprocessor 3. The temperature information required for dead zone temperature compensation is also obtained by inversion of the stator resistance identified online. No additional temperature compensation sensors, observation hardware, or power devices are required, which has significant advantages in terms of lightweight and cost. It is particularly suitable for near-space vehicle servo systems that are sensitive to weight, size, and power consumption.
[0071] It should be noted that the specific model of microprocessor 3, interrupt frequency, and observer bandwidth in the above embodiments are as follows. Controller bandwidth With current loop bandwidth The values of , low-speed threshold and hysteresis range, interval and duration of time-sharing excitation window and injected current amplitude, adaptive laws and regulator coefficients, and current limit value. The parameters are all exemplary. Those skilled in the art can adjust them within a reasonable range based on the rated parameters of the specific motor, the inertia of the transmission chain, and the task profile without departing from the concept of the present invention. The rotary encoder 5 can also be replaced by a rotary transformer in conjunction with the decoding circuit, or, when conditions permit, a sensorless observation algorithm can be used to assist in the verification. Current sampling can be flexibly selected between phase line Hall detection and bridge arm lower tube resistance sampling. Power devices can be selected from insulated gate bipolar transistors and silicon carbide metal oxide semiconductor field-effect transistors according to the bus voltage level. All such variations, as long as they still use the motor physical parameters obtained from online identification to simultaneously drive the inner loop voltage decoupling and the outer loop active disturbance rejection gain mapping, and thereby separate the internal parameter perturbation from the total disturbance that the observer needs to estimate, all belong to the technical solutions of the present invention and do not affect the realization of the purpose of the present invention or the acquisition of beneficial effects.
[0072] The above description is only a preferred embodiment of the present invention and is not intended to limit the present invention. Any modifications, equivalent substitutions, and improvements made within the spirit and principles of the present invention should be included within the protection scope of the present invention.
Claims
1. A disturbance-resistant vector control method for a UAV servo motor adapted to harsh near-space conditions, wherein the servo motor includes a permanent magnet synchronous motor for driving control surface deflection and a three-phase power inverter for driving the permanent magnet synchronous motor, characterized in that, The method performs the following steps in each control cycle: The three-phase stator current and rotor position of the permanent magnet synchronous motor are collected, and the d-axis current, q-axis current, electrical angular velocity and mechanical angular velocity in the dq coordinate system are obtained through coordinate transformation. Using the permanent magnet synchronous motor as a reference model and an adjustable model with the same structure and containing parameters to be identified constructed by software as the adjustment object, the difference between the stator current of the reference model and the adjustable model is used as the identification error. Based on the adaptive law derived from Popov's superstability theory, the stator resistance and permanent magnet flux of the permanent magnet synchronous motor are identified online. The permanent magnet flux obtained by online identification is mapped in real time to the control input gain of the speed loop active disturbance rejection control, and the control input gain is updated periodically with the identification value of the permanent magnet flux. Active disturbance rejection control with an extended state observer is constructed in the speed loop using the control input gain. The extended state observer estimates the mechanical speed and total disturbance, and generates a q-axis current command based on the target speed, estimated speed, and estimated total disturbance. The control input gain perturbation caused by the change in permanent magnet flux linkage is compensated by the real-time mapping and is not included in the total disturbance estimated by the extended state observer. In the current loop, the stator resistance and permanent magnet flux obtained online are used to perform dynamic feedforward decoupling on the dq axis voltage equation, and the d-axis voltage command and q-axis voltage command are generated according to the q-axis current command and d-axis current command. The d-axis voltage command and q-axis voltage command are space vector modulated to drive the three-phase power inverter, which outputs three-phase voltage to the permanent magnet synchronous motor.
2. The UAV servo motor anti-disturbance vector control method for adapting to harsh near-space conditions according to claim 1, characterized in that, The online identification is performed in a step-by-step, time-division manner: First, the stator inductance is taken as a slowly changing parameter. Under the condition that the permanent magnet synchronous motor is running steadily at a speed not lower than the set speed and a limited amplitude d-axis excitation current is injected, the differential term of the stator current is made to approach zero, and the stator current state equation is degenerated into a steady-state algebraic equation. Based on this, the d-axis inductance and q-axis inductance are calculated and regarded as known constants in the subsequent identification cycle. Then, under the premise that the stator inductance is known, the stator resistance and permanent magnet flux linkage are identified.
3. The UAV servo motor anti-disturbance vector control method for adapting to harsh near-space conditions according to claim 2, characterized in that, The adaptive laws for both the stator resistance and the permanent magnet flux linkage are identified as proportional-integral type. The stator resistor identification value is obtained by the first proportional-integral circuit calculating the first error signal. The first error signal is the algebraic sum of the product of the d-axis identification error and the d-axis current of the adjustable model and the product of the q-axis identification error and the q-axis current of the adjustable model. The identification value of the permanent magnet flux linkage is obtained by the second proportional-integral link calculating the second error signal, which is the product of the electric angular velocity and the q-axis identification error. The output directions of the first proportional-integral circuit and the second proportional-integral circuit are both set to negative feedback directions that reduce the identification error, and the initial values of their integrators are respectively taken from the nameplate values of the stator resistance and the permanent magnet flux linkage or the most recent convergence value.
4. The UAV servo motor anti-disturbance vector control method for adapting to harsh near-space conditions according to claim 3, characterized in that, When the electric angular velocity is lower than the set low speed threshold, the identification value of the permanent magnet flux linkage is frozen to keep the most recent valid identification result, and only the identification of the stator resistance is performed; when the electric angular velocity is higher than the low speed threshold, the stator resistance and the permanent magnet flux linkage are identified at the same time; and hysteresis intervals are set on both sides of the low speed threshold.
5. The UAV servo motor anti-disturbance vector control method for adapting to harsh near-space conditions according to claim 3, characterized in that, Within the interval where the stator resistance and permanent magnet flux linkage are simultaneously identified, time-division excitation and alternating updates are employed: short d-axis excitation windows are inserted at set time intervals, a limited amplitude d-axis current is injected into the d-axis excitation window, and the stator resistance is identified in a concentrated manner, while the identification value of the permanent magnet flux linkage is frozen, and the identification value of the stator resistance is latched at the end of the d-axis excitation window; outside the d-axis excitation window, the d-axis current command is set to zero, and the permanent magnet flux linkage is identified separately given that the identification value of the stator resistance is known.
6. The UAV servo motor anti-disturbance vector control method for adapting to harsh near-space conditions according to claim 1, characterized in that, The extended state observer is a second-order linear extended state observer with the control input gain as the input gain. Its two states are used to estimate the mechanical speed and the total disturbance, respectively. The state gain and disturbance gain are determined by the observer bandwidth by placing the dual poles of the observer's characteristic polynomial at the same negative real pole. The q-axis current command is obtained by subtracting the estimated total disturbance from the virtual control quantity obtained by proportional control of the difference between the target speed and the estimated speed, dividing by the control input gain, and then limiting the amplitude.
7. The UAV servo motor anti-disturbance vector control method for adapting to harsh near-space conditions according to claim 1, characterized in that, When performing dead-time compensation on the three-phase power inverter in the space vector modulation, the freewheeling direction during the dead-time period is determined according to the polarity of the current in each phase, and an equivalent correction amount to the duty cycle of the corresponding bridge arm is applied to the dead-time. Furthermore, the equivalent winding temperature is derived from the stator resistance obtained online based on the conductor resistance temperature characteristics, and the preset switching delay compensation table is checked and corrected using the equivalent winding temperature as an index.
8. A disturbance-resistant vector control device for UAV servos adapted to harsh near-space conditions, characterized in that, It includes a permanent magnet synchronous motor that drives the deflection of the steering surface, a three-phase power inverter that drives the permanent magnet synchronous motor, a current sensor that collects the three-phase stator current, a rotary encoder that detects the rotor position, a bus voltage detection circuit that detects the DC bus voltage, and a microprocessor. The current sensor, rotary encoder, and the output terminal of the bus voltage detection circuit are connected to the acquisition terminal of the microprocessor, and the drive signal output terminal of the microprocessor is connected to the gate drive circuit of the three-phase power inverter. The microprocessor is configured to obtain the current, electrical angular velocity, and mechanical angular velocity in the dq coordinate system by transforming the signals from the current sensor and the rotary encoder. Using the permanent magnet synchronous motor as a reference model and an adjustable model containing parameters to be identified constructed by software as the adjustment object, the stator resistance and permanent magnet flux of the permanent magnet synchronous motor are identified online based on the adaptive law derived from Popov's superstability theory. The permanent magnet flux obtained by online identification is mapped in real time to the control input gain of the speed loop active disturbance rejection control, so that the control input gain perturbation caused by the change of permanent magnet flux is not included in the total disturbance estimated by the speed loop extended state observer; Dynamic feedforward decoupling is performed using the dq-axis voltage equations of the stator resistance and permanent magnet flux linkage obtained online for the current loop.
9. The UAV servo anti-disturbance vector control device adapted to harsh near-space conditions according to claim 8, characterized in that, The microprocessor contains a main control module, a sensor data acquisition and coordinate transformation module, an online parameter identification module, an adaptive active disturbance rejection speed control module, a dynamic feedforward decoupling current control module, and a space vector modulation and dead-time compensation module. Each module runs periodically in the form of interrupt service subroutines and exchanges data through a global memory pool. The main control module generates timed interrupts and schedules the sensor data acquisition and coordinate transformation module, the online parameter identification module, the adaptive active disturbance rejection speed control module, the dynamic feedforward decoupling current control module, and the space vector modulation and dead-time compensation module in sequence within each interrupt cycle.
10. The UAV servo anti-disturbance vector control device adapted to harsh near-space conditions according to claim 8, characterized in that, The DC side of the three-phase power inverter is connected in parallel with an energy dissipation branch consisting of a braking resistor and a discharge switch connected in series. The microprocessor controls the discharge switch to turn on when the energy is fed back to the DC bus and the bus voltage exceeds the limit due to the rudder surface returning to center or reverse braking, based on the sampling value of the bus voltage detection circuit, so as to dissipate the feedback energy through the braking resistor.
