An aircraft control distribution method that combines failure detection with anti-windup redistribution

Abstract

The application discloses an aircraft control distribution method combining failure detection and anti-saturation redistribution, which firstly collects motor phase current, rotating speed, vibration and temperature and other multi-source sensor data, extracts features and carries out fusion processing, utilizes principal component analysis and Euclidean distance determination to identify motor fault state in real time, and generates a motor state matrix. Subsequently, in the control distribution stage, weights are set according to the priority of flight attitude control, and zero space projection technology is introduced, and under the premise of meeting motor physical constraints, optimal control signals are solved through quadratic programming and gradient descent method, and hierarchical anti-saturation redistribution of control commands is realized. The application effectively solves the problem that the control performance of the traditional method sharply decreases when the motor fails or is saturated, and significantly improves the flight stability and safety of the multi-rotor aircraft under the conditions of partial power failure and high load.

Description

Technical Field

[0001] This invention belongs to the field of aircraft control technology, and in particular relates to an aircraft control allocation method that combines failure detection and anti-saturation redistribution. Background Technology

[0002] With the aviation industry's urgent need for energy conservation and emission reduction, electric aircraft have become an important development direction for future green aviation. Flight control of electric aircraft is crucial for flight safety. Motor failure not only leads to mission failure but also poses a serious threat to safety and stability; therefore, timely detection of motor anomalies or failures is essential. However, in practical applications, there are delays in the transmission of failure information, or inaccuracies due to limitations in sensors and algorithms. Therefore, secondary safeguards are needed to ensure stable system operation even in the event of motor failure. Existing aircraft motor fault detection methods typically rely on data from a single sensor, which suffers from sensitivity to noise, high false negative rates, and insufficient real-time performance.

[0003] During flight, the ideal control input calculated by the flight control system is typically distributed to each motor to achieve attitude stabilization and trajectory tracking. However, due to the inherent physical characteristics and power limitations of the motors, their rotational speeds have maximum and minimum constraints. When the control signal output by the flight control system far exceeds the motor's actuation range, motor control saturation occurs. Motor saturation not only introduces control errors but may also trigger attitude instability. Especially under conditions of partial motor failure or high load, traditional linear limiting or simple shaving methods are insufficient to guarantee safe flight, exhibiting problems of insufficient real-time performance and poor robustness.

[0004] Existing patent CN113741549A discloses a method for allocating control inputs to a multi-rotor unmanned aerial vehicle (UAV). This method aims to optimize motor control signals to ensure stable flight of the aircraft under motor saturation or limitation conditions. By setting the maximum and minimum speeds of each motor and combining this with the aircraft's control tasks, a control matrix is ​​constructed. Optimization methods are then used to adjust the control signals, ensuring that all motor control signals are within physical limits. The optimization process involves phased corrections: first, altitude control inputs are adjusted, then roll, pitch, and yaw control inputs are optimized, gradually reducing the impact of motor saturation and fully utilizing unsaturated motors. However, this method only performs allocation / limiting and lacks a fault / anomaly detection mechanism, making it unable to promptly develop safer and more reliable control schemes for faults or anomalies.

[0005] Existing patent CN115610680A discloses a method and system for fault detection and redundant control of unmanned aerial vehicles (UAVs). This method acquires the pitch, roll, and yaw control variables of the UAV, calculates their average values, and compares them with preset conditions to determine whether a fault has occurred. If a fault is detected, the system reallocates the control signals of the normal power subsystem using a pre-designed control allocation matrix to compensate for the absence of the faulty subsystem. However, this method relies excessively on fixed thresholds and triggering conditions such as maximum throttle, is insensitive to early performance degradation (speed drop, efficiency decrease), and requires a long time window for detection lag. In addition, this method uses a pre-designed offline control allocation matrix that switches by "number," making it difficult to cover scenarios such as partial degradation (incomplete failure), multiple failures / combined failures, and configuration changes (tilt / load changes); if it does not match reality, it may produce excessive control errors or oscillations.

[0006] Therefore, it is necessary to provide a new aircraft control allocation method that combines failure detection and anti-saturation redistribution to solve the above-mentioned technical problems. Summary of the Invention

[0007] This invention is a method for aircraft control allocation that combines failure detection and anti-saturation redistribution. By using failure detection based on multi-source sensor fusion and anti-saturation control allocation technology based on priority null space projection, it solves the control stability problem of multi-rotor aircraft under motor failure and output saturation conditions, and achieves the technical effect of maintaining the stability of the core attitude of the aircraft and safe flight even when some power fails.

[0008] This invention provides an aircraft control allocation method that combines failure detection and anti-saturation reallocation, comprising the following steps: S1. Use multiple sensors to comprehensively collect the operating data of each motor, including motor current data, speed data, vibration data and temperature data; S2. Preprocess the collected operational data to eliminate noise interference and obtain preprocessed data. Then extract key features from the preprocessed data to obtain the feature vectors of each sensor. S3. Concatenate the feature vectors of each sensor into a high-dimensional feature vector. Principal component analysis was used to analyze this high-dimensional eigenvector. Dimensionality reduction and fusion are performed to generate the core feature matrix. ; S4. Calculate the Euclidean distance between the current characteristic data and the normal reference data for each motor. and the distance from the set threshold By comparing the states of each motor, we can determine whether the state is normal and obtain the motor state matrix. ; S5. Construct an effective control allocation matrix based on the motor state matrix FM and the control allocation matrix B. ; S6. Assign priorities to different flight control channels, with attitude control as the highest priority, altitude control as the second highest, and yaw control as the lowest priority, and construct the priority matrix W. S7. Combining the priority matrix W with the effective control allocation matrix Obtain the weighted effective allocation matrix ; S8. The motor control signal allocation problem is constructed as a constrained quadratic programming problem. The goal is to minimize the control error and obtain the optimal control signal while satisfying all physical output constraints of the motor. ; S9. Based on the priority constraints of attitude control, altitude control, and yaw control, select the optimal control signal. By performing hierarchical null space projection, priority constraints for the control signals are obtained. , ; S10. Combining the gradient descent method and the priority constraints of the control signals, the optimization problem is solved, and the final actual control command for each motor is calculated. .

[0009] Furthermore, in step S1, the various sensors include a motor phase current sensor for collecting load and stall information, a motor speed sensor for detecting motor operating speed and abnormal fluctuations, a vibration sensor for detecting vibrations caused by mechanical imbalance, bearing wear, or propeller loss, and a motor temperature sensor for monitoring changes in winding and housing temperature.

[0010] Furthermore, in step S2, the key feature includes the average current in the current data. Peak current The instantaneous speed in the speed data Speed ​​fluctuation range The root mean square of vibration in the vibration data Vibration peak value and the instantaneous temperature in the temperature data. heating rate .

[0011] Furthermore, in step S2, the preprocessing includes filtering, denoising, and normalizing the key features.

[0012] Furthermore, step S3 includes the following steps: S31. Unify the dimensions of the feature vectors from each sensor to form a unified high-dimensional vector, and then concatenate them into a high-dimensional feature vector. : , in, This is the feature vector of the current sensor. This is the feature vector of the speed sensor. This is the feature vector of the vibration sensor. This is the feature vector of the temperature sensor; S32. Using principal component analysis, high-dimensional eigenvectors are transformed... Mapping to a unified feature space yields a fused low-dimensional core feature matrix. : ; in, It is the centered feature matrix; It is a matrix composed of the first d principal components.

[0013] Furthermore, step S4 includes the following steps: S41. Calculate the Euclidean distance between the current characteristic data and the normal reference data for each motor. : , , , in, It is the feature vector of the i-th motor. It is a reference point. It is the number of samples in the training set. It is the set threshold distance; S42. Based on the Euclidean distance of each motor Distance from threshold Determine the status of each motor. : ; S43. Combine the state values ​​of all motors to obtain the motor state matrix. .

[0014] Furthermore, in step S6, the expression for the priority matrix W is: W , in, ; These are the roll attitude control weights, pitch attitude control weights, altitude control weights, and yaw control weights, respectively.

[0015] Furthermore, step S8 includes the following steps: S81. Define the control objective function as follows: , The constraints are: , in, It is a control signal vector. To minimize the control signal amount, To maximize the control signal, Q is a positive definite weighting matrix. It is the bias vector of the target command; S82. Calculate the gradient of the objective function in step S81. and Hessian matrix They are respectively: ; ; in, It is a given control command. It is a regularization parameter. It is the identity matrix; S83. Solve the objective function using the quadratic programming method to obtain the optimal control signal. for: .

[0016] Furthermore, step S9 includes: For the attitude control channel, use a high-priority constraint matrix. Perform null space projection to obtain the high-priority constraint matrix. null space orthogonal projection matrix : ; For the altitude control channel, while keeping the attitude control channel control signal unchanged, a medium-priority constraint matrix is ​​used. Perform null space projection to obtain the medium priority constraint matrix. null space orthogonal projection matrix : .

[0017] Furthermore, step S10 includes: S101, retrieve the optimal control signal from the previous cycle. As the control signal of the current cycle Based on the Hessian matrix H and the control signal of the current period The current objective function is calculated. And update the gradient of the objective function. : ; ; S102. Update the control signal command based on the calculated update gradient to obtain the updated control signal command. : , in, The learning rate; S103. Further adjust and update the control signal command based on the priority control signal obtained from the null space projection. To obtain the final control signal : .

[0018] The beneficial effects of this invention's aircraft control allocation method, which combines failure detection and anti-saturation redistribution, are as follows: it achieves high-precision and high-reliability monitoring of motor states, thereby improving the aircraft's flight safety and fault tolerance, and providing reliable input information for subsequent redistribution. Based on the detected motor fault states, the flight control system can adjust the control allocation strategy in real time, transferring the load of the affected motors to the normal motors, ensuring that the aircraft can maintain stable flight even in the event of partial motor failure. Specifically: (1) Considering that the actuator level is susceptible to noise and operating condition fluctuations, this invention constructs a multi-source fusion state assessment link. Signals such as current, speed, vibration, and temperature are filtered, normalized, and feature-extracted. After dimensionality reduction via principal component analysis (PCA), anomaly detection is achieved based on distance metrics and adaptive thresholds, outputting a motor health matrix. This mechanism maintains stable discrimination sensitivity even in the presence of external disturbances and measurement noise, significantly reducing missed detections, false detections, and detection delays. It provides a reliable prior for "health weights" in subsequent control allocation, thereby improving overall fault tolerance.

[0019] (2) To address the saturation propagation and critical channel degradation problems caused by traditional pseudo-inversion and amplitude limiting, this invention introduces a task priority matrix and an effective allocation matrix to construct a constrained quadratic programming (QP) model. A "hierarchical solution + null space projection" strategy is adopted, first obtaining feasible optimal solutions in high-priority tasks, and then performing constrained optimization on low- and medium-priority targets in its null space. This method effectively suppresses packet and diffusion when channels tend to saturate or partially fail, maintains the stability of roll / pitch channels, and significantly reduces tracking errors.

[0020] (3) This patent constructs an effective allocation matrix in real time based on the motor health matrix, priority weighting matrix, and allocation matrix, thereby reconstructing the feasible region online and completing the adaptive load transfer. When a failure occurs, the channel can be automatically eliminated and degraded operation can be achieved, ensuring that core tasks such as attitude and lift are prioritized under resource-constrained conditions.

[0021] (4) Due to the computational bottleneck and convergence uncertainty of online optimization, this invention maintains the convexity and sparse structure of the problem during the modeling stage, provides the analytical gradient and Hessian matrix of the objective function, and combines a lightweight solver for projected gradients to achieve rapid online updates. This design can operate stably under resource constraints, taking into account control smoothness, actuator lifespan and energy efficiency, significantly improving computational efficiency and engineering feasibility. Attached Figure Description

[0022] Figure 1 This is a structural block diagram of the present invention. Detailed Implementation

[0023] Example 1: Please refer to Figure 1 This invention provides an aircraft control allocation method that combines failure detection and anti-saturation reallocation, comprising the following steps: S1, Motor Failure Detection.

[0024] S11. Multi-source sensor data acquisition: The system uses multiple sensors to collect key status parameters in real time, obtaining multi-source sensor data. These sensors specifically include a motor phase current sensor (for acquiring load and stall information), a motor speed sensor (for detecting motor operating speed and abnormal fluctuations), a vibration sensor (for detecting vibrations caused by mechanical imbalance, bearing wear, or propeller loss), and a motor temperature sensor (for monitoring winding and casing temperature changes). The multi-source sensor data includes motor current data, speed data, vibration data, and temperature data.

[0025] S12. Preprocessing and Feature Extraction: The initial data collected by each sensor is filtered, denoised, and normalized to reduce the impact of environmental interference and measurement errors on feature calculation, resulting in preprocessed data; the mean current is extracted from the preprocessed data. Peak current As a current feature, instantaneous rotational speed is extracted. Speed ​​fluctuation range As a rotational speed feature, the root mean square of vibration is extracted. Vibration peak value As a vibration feature, instantaneous temperature is extracted. heating rate As a temperature characteristic, feature vectors for each sensor are obtained, including the feature vector of the current sensor. Speed ​​sensor feature vector Vibration sensor feature vector Temperature sensor feature vector .

[0026] S13. Unified Feature Representation and Fusion: S131. Unify the dimensions of the feature vectors extracted from each sensor in step S12 to form a unified high-dimensional vector; concatenate the feature vectors of each sensor in sequence as follows: ; in, It is a high-dimensional feature vector formed by concatenating the feature vectors of each sensor, including the feature vector of the current sensor. Speed ​​sensor feature vector Vibration sensor feature vector Temperature sensor feature vector .

[0027] S132. Using Principal Component Analysis (PCA), high-dimensional feature vectors with multi-source features are transformed. Mapping to a unified feature space yields a centered feature matrix. To optimize information retention and computational efficiency; then the centered feature matrix After fusion using the PCA method, a low-dimensional core feature matrix is ​​obtained. : ; in, It is a matrix composed of the first d principal components.

[0028] S133. Based on Euclidean distance, calculate the feature vector of each sample point (i.e., the i-th object to be judged) in the fusion space, which is equal to... The distance between the i-th row and the reference point (i.e., the center point of the normally operating sample in the same fusion space) This is used to subsequently determine whether the sample is an anomaly; distance The calculation is as follows: , , , in, It is the feature vector of the i-th sample, i.e., the core feature matrix. The i-th row; It is a reference point. It is the number of samples in the training set. It is the set threshold distance.

[0029] S134, Based on sample distance Distance from threshold Determine the status of each motor: ; in, This represents the state of the i-th motor, indicating whether the motor is functioning correctly; the state values ​​of all motors are combined to form the motor state matrix. : .

[0030] Motor state matrix Used for subsequent control allocation decisions.

[0031] S2, Anti-saturation redistribution: By introducing priority control and zero-space projection technology, the motor saturation problem is fully considered in the multi-motor control distribution, ensuring that the system can still maintain normal operation under high load and dynamic changes.

[0032] The flight control system of a multi-rotor aircraft typically outputs a desired flight control command. It consists of force and torque: , in, It is total lift. It is the rolling torque. It is the pitching moment. It is the yaw moment.

[0033] Motor output and flight control desired commands The relationship between them is: , Where B is the control allocation matrix, It is the input signal for the motor.

[0034] Assuming the aircraft has N motors, the control objective is to allocate control inputs. , The motor speed allows the system to optimize control performance while meeting constraints. Based on this idea, step S2, anti-saturation redistribution, specifically includes the following steps: S21. Obtain the effective control allocation matrix based on the motor state matrix FM and control allocation matrix B obtained in step S1. : ; The constraints for the motor control signals are: .

[0035] S22. Set the priority of the pitch, roll, altitude, and yaw channels.

[0036] During flight, due to motor saturation or resource constraints, it is impossible to simultaneously meet all control tasks. Therefore, it is necessary to rationally allocate control signals based on the priority of each control channel. Attitude control (pitch and roll channels) directly affects the aircraft's maneuverability and stability, and therefore needs to be set as the highest priority. The aircraft's vertical stability is also important, but it is usually not as critical as attitude control during flight; therefore, the altitude channel is set as the second highest priority. The yaw channel is set as the lowest priority, as it mainly affects heading control, and the need for yaw adjustment during flight is relatively low.

[0037] This step uses a priority matrix W to represent the priority of each channel: ; in ; These are the roll attitude control weight, pitch attitude control weight, altitude control weight, and yaw control weight. These four weights form a priority matrix W. For example, the weights can be set to 100, 100, 30, and 5 respectively, thus ensuring that the pitch and roll channels are assigned the highest priority, the throttle channel the second highest priority, and the yaw channel the lowest priority.

[0038] S23. Combining the priority matrix W with the effective control allocation matrix Obtain the weighted effective allocation matrix : .

[0039] S24. Use the quadratic programming (QP) method to optimize the control signal to control the error of the control signal.

[0040] The QP method minimizes control error under constraints, ensuring that the control signal meets both the dynamic requirements of the aircraft and the limitations and priority requirements of the motors. Specifically, it includes: S241. Define the control objective function as follows: , in, It is a control signal vector. To minimize the control signal amount, To maximize the control signal, Q is a positive definite weighting matrix. It is the bias vector of the target command.

[0041] S242. Calculate the gradient of the objective function in step S241. and Hessian matrix They are respectively: ; ; in, It is a given control command. It is a regularization parameter. It is an identity matrix.

[0042] S243. Solve the objective function using the quadratic programming method to obtain the optimal control signal. for: .

[0043] S25. Based on priority constraints, select the optimal control signal. Perform hierarchical null-space projection. The core idea of ​​null-space projection is to further optimize the control signals of low-priority tasks while satisfying the constraints of high-priority tasks.

[0044] For roll and pitch channels, use high-priority constraint matrices. By performing null-space projection while keeping the two control tasks unchanged, the high-priority constraint matrix is ​​obtained. null space orthogonal projection matrix : ; For the altitude channel, while keeping the roll and pitch channel control signals unchanged, a medium-priority constraint matrix is ​​used. Null-space projection is performed to ensure that the optimization process satisfies all priority constraints, resulting in the medium-priority constraint matrix. null space orthogonal projection matrix : .

[0045] For the lowest priority yaw channels, they can be completely discarded or optimized within zero space according to specific needs.

[0046] S26. By combining priority and null space projection, high-priority and medium-priority control signal constraints are obtained, and the gradient descent method is used to solve the optimization problem.

[0047] S261. Take the optimal control signal from the previous cycle. As the control signal of the current cycle Based on the Hessian matrix H and the control signal of the current period The current objective function is calculated. And update the gradient of the objective function. : ; ; S262. Update the control signal command based on the calculated update gradient to obtain the updated control signal command. : , in, The learning rate; Solving for the results Then, based on the constraints, iterations are performed to make the objective function... The process is minimized until convergence, thus obtaining the optimal control signal for the current cycle, and then repeating the calculation for the next cycle. S263. Further adjust and update the control signal command based on the priority control signal obtained from the null space projection. To obtain the final control signal : .

[0048] This process ensures that the control signals of each motor do not exceed their physical limits, and optimizes low-priority control tasks to the maximum extent while satisfying high-priority tasks.

[0049] The core technical solution of this invention proposes an efficient control allocation method by combining failure detection and anti-saturation redistribution algorithms. This method can automatically adjust control inputs and redistribute tasks when a motor fails, by detecting the operating status of the failed motor in real time, thus ensuring stable system operation. By comprehensively considering motor speed constraints, motor health status, and flight control priorities, the flight control output signals are dynamically allocated in real time, ensuring a reasonable distribution of control commands among the motors, thereby avoiding saturation of individual motors and guaranteeing the control accuracy of the aircraft's attitude and torque.

[0050] By comprehensively utilizing data from multiple sensors, high-precision and high-reliability monitoring of motor status is achieved, thereby improving the aircraft's flight safety and fault tolerance, and providing reliable input information for subsequent reallocation. Based on the detected motor fault status, the flight control system can adjust the control allocation strategy in real time, transferring the load of the affected motors to the normal motors, ensuring that the aircraft can maintain stable flight even in the event of partial motor failure.

[0051] The advantages of this embodiment are: (1) This embodiment directly constructs a constrained anti-saturation redistribution problem in the actuator instruction space, introduces priority and null space projection, and completes the allocation with optimization quantity. It provides stronger protection for global consistency and priority reachability in boundary conditions. It avoids the sequential dependency and channel coupling residue caused by the "channel-by-channel / stage-by-stage pruning" in the prior art patent CN113741549A. It can still obtain the globally consistent minimum deviation solution when approaching the boundary or even entering saturation, and has stronger priority reachability and stability.

[0052] (2) In this embodiment, the heading channel is incorporated into a low-weight hierarchical system and solved in a high-priority null space to ensure orderly degradation when resources are limited. Unlike the stage-based independent correction adopted in the prior art patent CN113741549A, the heading adjustment in this embodiment will not destroy the high-priority effects such as altitude, roll, and pitch, significantly reducing jumps and oscillations, making the transition process smoother and the convergence faster.

[0053] (3) This embodiment handles saturation and resource allocation in the priority weight unified objective, and suppresses saturation diffusion through null space projection. Compared with the existing technology patent CN113741549A, which requires multiple heuristic adjustments and amplitude limiting, this solution only needs to solve once to meet multiple constraints, minimize deviation, has fewer parameters, better real-time performance, and has a clear mathematical optimality guarantee.

[0054] (4) This embodiment is based on the fusion of multiple sensors such as current and speed, and is uniformly mapped to a low-dimensional feature space by PCA. Distance threshold is used to judge and construct the motor state matrix for subsequent control allocation. Compared with the prior art patent CN115610680A, which only judges faults based on the average threshold of pitch / roll / yaw control quantities, is easily disturbed and is not sensitive to early degradation, this solution is more robust in noise resistance, detects earlier, has lower false alarms and false misses, and can produce health quantification results for motor-by-motor use by the control layer.

[0055] (5) This embodiment solves the problem under the quadratic objective of priority weighted error and uses null space projection to ensure high priority subtasks. It can still give a unique and provably convergent minimum deviation control quantity when saturation and multiple constraints coexist. It is numerically stable and easy to perform Cholesky pre-decomposition to meet real-time requirements.

[0056] (6) In this embodiment, the allocation model is reconstructed and downgraded online using the motor state matrix as input, and the reallocation and priority maintenance are completed within the same optimization framework. Compared with the prior art patent CN115610680A, which only switches according to the predefined matrix and is difficult to cover complex situations such as partial degradation, combined faults, and configuration changes, this solution can continuously adapt to the actual health state, reduce switching abrupt changes, and improve redundancy utilization, task completion rate and safety margin.

[0057] For those skilled in the art, various modifications and improvements can be made without departing from the inventive concept of this invention, and these all fall within the protection scope of this invention.

Claims

1. A method for aircraft control allocation combining failure detection and anti-saturation reallocation, characterized in that, Includes the following steps: S1. Use multiple sensors to comprehensively collect the operating data of each motor, including motor current data, speed data, vibration data and temperature data; S2. Preprocess the collected operational data to eliminate noise interference and obtain preprocessed data. Then extract key features from the preprocessed data to obtain the feature vectors of each sensor. S3. Concatenate the feature vectors of each sensor into a high-dimensional feature vector. Principal component analysis was used to analyze this high-dimensional eigenvector. Dimensionality reduction and fusion are performed to generate the core feature matrix. ; S4. Calculate the Euclidean distance between the current characteristic data and the normal reference data for each motor. and the distance from the set threshold By comparing the states, we can determine whether each motor is in a normal state and obtain the motor state matrix. ; S5. Construct an effective control allocation matrix based on the motor state matrix FM and the control allocation matrix B. ; S6. Assign priorities to different flight control channels, with attitude control as the highest priority, altitude control as the second highest, and yaw control as the lowest priority, and construct the priority matrix W. S7. Combining the priority matrix W with the effective control allocation matrix Obtain the weighted effective allocation matrix ; S8. The motor control signal allocation problem is constructed as a constrained quadratic programming problem. The goal is to minimize the control error and obtain the optimal control signal while satisfying all physical output constraints of the motor. ; S9. Based on the priority constraints of attitude control, altitude control, and yaw control, select the optimal control signal. By performing hierarchical null space projection, priority constraints for the control signals are obtained. , ; S10. Combining the gradient descent method and the priority constraints of the control signals, the optimization problem is solved, and the final actual control command for each motor is calculated. .

2. The aircraft control allocation method combining failure detection and anti-saturation reallocation as described in claim 1, characterized in that, In step S1, the various sensors include a motor phase current sensor for collecting load and stall information, a motor speed sensor for detecting motor operating speed and abnormal fluctuations, a vibration sensor for detecting vibrations caused by mechanical imbalance, bearing wear or propeller loss, and a motor temperature sensor for monitoring changes in winding and housing temperature.

3. The aircraft control allocation method combining failure detection and anti-saturation reallocation as described in claim 1, characterized in that, In step S2, the key feature includes the average current in the current data. Peak current The instantaneous speed in the speed data Speed ​​fluctuation range The root mean square of vibration in the vibration data Vibration peak value and the instantaneous temperature in the temperature data. heating rate .

4. The aircraft control allocation method combining failure detection and anti-saturation reallocation as described in claim 1, characterized in that, In step S2, the preprocessing includes filtering, denoising, and normalizing the key features.

5. The aircraft control allocation method combining failure detection and anti-saturation reallocation as described in claim 1, characterized in that, Step S3 includes the following steps: S31. Unify the dimensions of the feature vectors from each sensor to form a unified high-dimensional vector, and then concatenate them into a high-dimensional feature vector. : ， in, This is the feature vector of the current sensor. This is the feature vector of the speed sensor. This is the feature vector of the vibration sensor. This is the feature vector of the temperature sensor; S32. Using principal component analysis, high-dimensional eigenvectors are transformed... Mapping to a unified feature space yields a fused low-dimensional core feature matrix. : ； in, It is the centered feature matrix; It is a matrix composed of the first d principal components.

6. The aircraft control allocation method combining failure detection and anti-saturation reallocation as described in claim 1, characterized in that, Step S4 includes the following steps: S41. Calculate the Euclidean distance between the current characteristic data and the normal reference data for each motor. : ， ， ， in, It is the feature vector of the i-th motor. It is a reference point. It is the number of samples in the training set. It is the set threshold distance; S42. Based on the Euclidean distance of each motor Distance from threshold Determine the status of each motor. : ； S43. Combine the state values ​​of all motors to obtain the motor state matrix. .

7. The aircraft control allocation method combining failure detection and anti-saturation reallocation as described in claim 1, characterized in that, In step S6, the expression for the priority matrix W is: W ， in, ; These are the roll attitude control weights, pitch attitude control weights, altitude control weights, and yaw control weights, respectively.

8. The aircraft control allocation method combining failure detection and anti-saturation reallocation as described in claim 1, characterized in that, Step S8 includes the following steps: S81. Define the control objective function as follows: ， The constraints are: , in, It is a control signal vector. To minimize the control signal amount, To maximize the control signal, Q is a positive definite weighting matrix. It is the bias vector of the target command; S82. Calculate the objective function in step S81. gradient and Hessian matrix They are respectively: ； ； in, It is a given control command. It is a regularization parameter. It is the identity matrix; S83. Solve the objective function using the quadratic programming method to obtain the optimal control signal. for: 。 9. The aircraft control allocation method combining failure detection and anti-saturation reallocation as described in claim 1, characterized in that, Step S9 includes: For the attitude control channel, use a high-priority constraint matrix. Perform null space projection to obtain the high-priority constraint matrix. null space orthogonal projection matrix : ； For the altitude control channel, while keeping the attitude control channel control signal unchanged, a medium-priority constraint matrix is ​​used. Perform null space projection to obtain the medium priority constraint matrix. null space orthogonal projection matrix : 。 10. The aircraft control allocation method combining failure detection and anti-saturation reallocation as described in claim 8, characterized in that, Step S10 includes: S101, retrieve the optimal control signal from the previous cycle. As the control signal of the current cycle Based on the Hessian matrix H and the control signal of the current period The current objective function is calculated. And update the gradient of the objective function. : ； ； S102. Update the control signal command based on the calculated update gradient to obtain the updated control signal command. : ， in, The learning rate; S103. Further adjust and update the control signal command based on the priority control signal obtained from the null space projection. To obtain the final control signal : 。

Images