An active identification method for multi-surface effectiveness loss of flying wing unmanned aerial vehicle

Abstract

This invention belongs to the field of flight control and fault diagnosis technology for flying-wing unmanned aerial vehicles (UAVs), and relates to an active identification method for the performance loss of multiple control surfaces in flying-wing UAVs. The method superimposes a single, mutually exclusive set of orthogonal multi-sine excitation signals with distinct frequencies onto the closed-loop control commands of the six trailing-edge control surfaces of the flying-wing UAV; it collects three-axis angular velocities, obtains the three-axis torque coefficients through numerical differentiation and inverse kinematics of rigid body rotation; constructs three-channel regression models for roll, pitch, and yaw; and uses a recursive least squares equation error method with a forgetting factor to identify the control derivatives of each control surface online; normalizes the identified control derivative values ​​of each control surface in the three torque channels to obtain channel-by-channel performance coefficients; pre-determines the set of torque channels participating in the voting according to the physical main control action of each control surface; weights the selected channels within a sliding window using the reciprocal of the coefficient of variation as the weight; and uses a dual criterion of coefficient of variation and drift criterion for convergence determination, outputting the final performance coefficients.

Description

Technical Field

[0001] This invention belongs to the field of flight control and fault diagnosis technology for flying-wing unmanned aerial vehicles (UAVs), and relates to an active identification method for multi-control surface performance loss in flying-wing UAVs. Background Technology

[0002] Flying-wing UAVs integrate the wings and fuselage into one unit, eliminating conventional horizontal and vertical stabilizers. Their three-axis control authority is distributed across multiple trailing-edge control surfaces, each participating in two to three torque channels simultaneously, forming an overdriven, strongly coupled control system. This system relies on control allocation to distribute three-axis torque commands to each control surface, and the establishment of the control allocation matrix is ​​predicated on the nominal performance of the control surfaces. Once a control surface experiences sustained performance degradation, this precondition is broken, the closed-loop response quality deteriorates, and in severe cases, it can lead to loss of aircraft control.

[0003] Control surface effectiveness loss is a multiplicative fault: the control surface position feedback is always consistent with the command, and the deviation only occurs between the commanded deflection amount and its actual aerodynamic effect. Therefore, it cannot be detected by hardware redundancy at the actuator level and must be identified by analytical redundancy.

[0004] The patent "A Fault Diagnosis Method for Control Surfaces of Flying-Wing UAVs Based on Neural Network Adaptive Observer" (CN115659502A) addresses the combined characteristics of strong coupling and strong nonlinearity in the control surfaces of flying-wing UAVs. It establishes an affine nonlinear model and a control surface fault model for the flying-wing UAV and proposes a control surface fault diagnosis method based on an RBF neural network adaptive observer. This method uses a neural network to estimate fault terms in the observer online, forming an adaptive adjustment mechanism. When the system output error exceeds an alarm threshold, fault diagnosis is triggered, achieving real-time detection and estimation of control surface faults. However, this method only detects and qualitatively estimates faults in a single control surface, failing to provide quantitative identification results of the degree of performance loss for each control surface. Furthermore, it does not consider the mirror coupling characteristics of aerodynamically symmetrical control surfaces in the moment equation in a flying-wing configuration, making it difficult to distinguish the fault sources of each control surface when multiple control surfaces experience simultaneous performance loss.

[0005] The patent "Aircraft Fault Diagnosis Method, Apparatus, and Electronic Equipment" (CN113221719A) addresses the poor adaptability of aircraft sensor faults by proposing a fault diagnosis method based on an extended state observer and frequency domain analysis. This method utilizes an extended state observer to extract angle-of-attack residual signals, analyzes the energy characteristics of the residual signals across multiple preset frequency bands, and identifies fault types using a preset fault classification model. By replacing time-domain residuals with frequency domain features, it reduces sensitivity to sensor measurement noise and improves the adaptability of the diagnostic method. However, this method only identifies faults in a single sensor channel and cannot be extended to scenarios where multiple control surfaces experience simultaneous performance losses. Furthermore, its fixed frequency band division and classification model struggle to adaptively integrate multi-channel identification information, lacking the ability to dynamically adjust channel weights under different fault combinations.

[0006] For flying wing configurations, existing methods for identifying performance loss have two shortcomings. First, flying wing configurations contain aerodynamically symmetrical control surfaces, which enter the moment equation in a mirror manner. Relying solely on residual signals makes it difficult to distinguish which control surface the fault originates from. Second, there is a lack of specific research on situations where multiple control surfaces experience performance loss simultaneously. When using fixed inter-channel weights to fuse multi-channel identification results, it is difficult to adapt to different fault combinations. The control derivative of some control surfaces in a certain moment channel is very small, resulting in a low signal-to-noise ratio for that channel. If fixed weights are still used for fusion, the overall identification accuracy will be reduced.

[0007] To overcome the above shortcomings, a method is needed that can simultaneously identify multiple control surfaces and adaptively fuse the multi-channel identification results according to the physical characteristics of each control surface. Summary of the Invention

[0008] To address the challenge of accurately quantifying the performance loss of each control surface when multiple control surfaces of a flying-wing UAV simultaneously experience performance degradation, and to overcome the difficulties in distinguishing fault sources due to the aerodynamic symmetry of the flying-wing configuration and the inability of fixed channel weights to adapt to different fault combinations, this invention proposes an active identification method for multi-control surface performance loss in flying-wing UAVs. This invention can accurately quantify the performance loss of each control surface even under conditions of simultaneous performance degradation and sensor noise, providing reliable per-control-surface performance coefficients for downstream reconfigurable control allocation.

[0009] The technical concept of this invention is as follows: After confirming a malfunction of the control surface, an orthogonal multi-sine excitation signal is superimposed on the command of each control surface to ensure sufficient excitation for each control surface; based on rigid body rotational dynamics, the three-axis angular rate measurement values ​​are numerically differentiated and inversely solved to obtain the three-axis torque coefficients; the aerodynamic derivative vector, including all control surface control derivatives, is identified online using the recursive least squares equation error method with a forgetting factor; the identified control derivative values ​​of each control surface in the three torque channels are normalized to obtain the channel-by-channel efficiency coefficients; according to the physical main control action of each control surface, a set of torque channels participating in the voting is pre-assigned for each control surface; the mean and coefficient of variation of the efficiency coefficients of each channel are calculated within a sliding window, and the selected channels are weighted and fused using the reciprocal of the coefficient of variation as the weight; finally, the coefficient of variation and drift of the fused sequence are used as dual criteria to determine the convergence of the identification, and the final efficiency coefficients of each control surface are output.

[0010] The technical solution of the present invention is as follows: An active identification method for multi-control surface performance loss of a flying-wing unmanned aerial vehicle includes the following steps: Step 1: Superimpose orthogonal multi-sine excitation signals onto each control surface The flying-wing UAV is equipped with six trailing-edge control surfaces, symmetrically distributed along the wingspan. From the inside to the outside, they are a pitch flap (PF), an elevon (ELE), and an all-moving wingtip (AMT). Specifically, they include a left elevon, a right elevon, a left pitch flap, a right pitch flap, a left wingtip all-moving control surface, and a right wingtip all-moving control surface. After confirming that the control surfaces have suffered performance loss, an orthogonal multi-sine excitation signal is superimposed on the closed-loop control command of each control surface. The excitation signal for each control surface channel is: (1) In the formula, For the first Excitation signal for each control surface channel , For time; For the first The amplitude of the excitation signal for each control surface channel; For the first The number of harmonics contained in each control surface channel; , , The first The first control surface channel The frequency, phase, and power allocation coefficients of each harmonic are determined, and the frequency sets allocated to each control surface channel are mutually exclusive, thus ensuring that the excitation signals between channels are orthogonal. The excitation signal is superimposed on the original closed-loop command, and the amplitude of the excitation signal is kept small to provide sufficient excitation identification without significantly disturbing the closed-loop flight state.

[0011] The excitation signal is an orthogonal multi-sine excitation signal. The excitation signal for each control surface channel is a weighted sine superposition of several harmonics within the frequency set allocated to that channel. The excitation signal is superimposed on the original closed-loop control command.

[0012] Step 2: Inversely solve the triaxial torque coefficients from the angular rate measurements. The triaxial angular velocities are collected, and their numerical differentiation yields estimates of the triaxial angular accelerations. These estimates are then substituted into the rigid body rotational dynamics equations to inversely solve for the triaxial moments. Finally, the roll, pitch, and yaw moment coefficients are obtained by dimensionless dynamic pressure calculation. , , As an identification observation; specifically as follows: The equation of motion for the rigid body rotation of a flying-wing UAV about its center of mass is as follows: (2) In the formula, , , These are the roll, pitch, and yaw rates of the airframe shaft system, respectively. , , These are the first derivatives of the corresponding angular velocities; , , These are the roll, pitch, and yaw moments of the airframe axis system, respectively. The inertia coupling coefficient is determined by the aircraft's rotational inertia.

[0013] The triaxial angular velocities are measured by inertial navigation, and the central difference numerical differential is performed to obtain estimates of the triaxial angular accelerations. The rigid body rotational dynamics equations are then rewritten in a form with triaxial moments as unknowns: the roll and yaw channels are combined to form a equation regarding... , A system of two linear equations, with the pitch channel given separately. Solving for the problem yields: (3) in: (4) In the formula, , These are the remaining values ​​after subtracting the inertial coupling terms unrelated to control from the estimated angular acceleration values ​​in the roll and yaw channels, respectively.

[0014] Then, the triaxial torque is dimensionless, yielding the triaxial torque coefficients: (5) In the formula, For dynamic pressure; This refers to the wing reference area. For wingspan; The mean aerodynamic chord length is given.

[0015] Step 3: Construct a three-channel regression model The three-axis moment coefficients are linearly parameterized with respect to flight state variables and six control surface deflections, constructing regression models for three moment channels: roll, pitch, and yaw; the regression model is expressed as follows: ,in For torque channel labeling, For discrete time indices. For channel The regression vector, The aerodynamic derivative vector to be identified for the channel; The three-axis moment coefficients are linearly parameterized with respect to flight state quantities and six control surface deflections, expressed as: (6) In the formula, , , These are the reference values ​​for the roll, pitch, and yaw moment coefficients, respectively. , , These are the three-axis torque coefficients relative to the angle of attack. The derivative; , These are the roll and yaw moment coefficients relative to the sideslip angle, respectively. The derivative; , , These are the damping derivatives of the roll moment coefficient with respect to the roll rate, the pitch moment coefficient with respect to the pitch rate, and the yaw moment coefficient with respect to the yaw rate, respectively. , , The first Control derivatives of each control surface in the roll, pitch, and yaw channels; For the first The deflection of each control surface; This refers to flight speed.

[0016] The aerodynamic derivatives to be identified in the three torque channels are stacked into an aerodynamic derivative vector: (7) In the formula, , , Let be the aerodynamic derivative vectors to be identified for the roll, pitch, and yaw channels, respectively, with dimensions of 10, 9, and 10. Correspondingly, the ... The regression vectors for the three channels at each moment are composed of the flight state quantity and the control surface deflection quantity at the current moment: (8) In the formula, , , The first Regression vectors for roll, pitch, and yaw channels at all times; subscript Indicates the discrete time sequence number; , , , , , The first Angle of attack, sideslip angle, flight speed, roll rate, pitch rate, and yaw rate at any given moment; For the first The deflection of the six control surfaces at any given time. The regression model for the three channels is uniformly written as: (9) In the formula, For the first Time Channel The observed torque coefficient is obtained from the inverse solution in step 2.

[0017] Step 4: Online identification of errors in recursive least squares equations with forgetting factors. For each of the three channels, a recursive least squares equation error method with a forgetting factor is used to identify the aerodynamic derivative vector, including the six control surface derivatives, online. For each channel... , No. The recursive update law for each moment is: (10) In the formula, For the first Time Channel The estimated value of the aerodynamic derivative; For the first Time Channel The estimated value of the aerodynamic derivative; For the first Time Channel The gain vector; For the first Time Channel The covariance matrix; For the first Time Channel The covariance matrix; For the first Time Channel The regression vector; The forgetting factor has a value range of 1. This is used to reduce the weight of historical data and suppress the data saturation phenomenon caused by the decay of the positive definiteness of the information matrix as data accumulates, so that the identification can track the changes in aerodynamic derivatives after the fault. For the channel An identity matrix with the same dimension as the aerodynamic derivative vector. Initial values ​​for aerodynamic derivative estimation. The aerodynamic derivatives of the healthy nominal aerodynamic model at the balance point are solved using the small perturbation method.

[0018] Initial values ​​of the covariance matrix Take as: (11) The recursion is carried out step by step from the injection time of the orthogonal multi-sine excitation signal, and the aerodynamic derivative estimates of the three channels are updated according to Equation (10) at each time.

[0019] Step 5: Extract the three-channel efficiency coefficients per rudder surface Divide the control derivative identification value of each control surface in the three torque channels by the corresponding healthy nominal control derivative to obtain the channel-by-channel effectiveness coefficient of the control surface in the three channels; as follows: At any given moment , will the Dividing the control derivative identification value of each control surface in the three torque channels by its healthy nominal control derivative, we obtain the channel-by-channel effectiveness coefficient of the control surface in the three channels: (12) In the formula, For the first Time of the first Each rudder surface in the channel The efficiency coefficient; For the first The first time to identify Each rudder surface in the channel The estimated value of the control derivative on; For the first Each rudder surface in the channel The aerodynamic derivatives are obtained using the small perturbation method. An efficiency coefficient of 1 indicates no energy loss in the channel, less than 1 indicates performance degradation, and 0 indicates complete failure.

[0020] Step 6: Specify the set of torque channels participating in the voting on a per-control-face basis according to the physical main control action. Each control surface is pre-assigned a set of torque channels to participate in the voting. ,in For the first The set of moment channels for each control surface to participate in the voting. Among them, the voting channels for the left and right elevons and the left and right pitch flaps are the roll and pitch channels, while the voting channels for the left and right wingtip all-moving control surfaces are the roll, pitch, and yaw channels.

[0021] Specifically, the left and right elevons generate roll moment through differential deflection and serve as the primary roll control surfaces, while the left and right pitch flaps generate pitch moment through symmetrical deflection and serve as the primary pitch control surfaces. The set of channels for both to participate in the voting is taken as follows: (13) The left and right wingtip all-moving control surfaces generate yaw moment through differential deflection and serve as the primary yaw control surfaces. The set of channels participating in the voting is: (14) Step 7: Sliding window variation coefficient weighted fusion Enter the sliding window, for each control surface, the set of channels. For each channel within the range, calculate the window mean of the channel-by-channel performance coefficient. With window coefficient of variation The fusion efficiency coefficient of the control surface is obtained by weighting the window mean of each channel in the channel set with the reciprocal of the coefficient of variation as the weight; as follows: The introduced length is Seconds, corresponding A sliding window for each sampling step. At that moment, for the first For each channel participating in the voting for each control surface, calculate the mean and coefficient of variation of its channel-by-channel effectiveness coefficient within the window: (15) In the formula, For the first Time of the first individual control surface channels The window mean of the efficiency coefficient; The corresponding window standard deviation; This represents the corresponding window variation coefficient; For the first Time of the first individual control surface channels The efficiency coefficient; To prevent small positive numbers with a denominator of zero, it is generally advisable to take the value as 1e-7; This is the sampling time sequence number within the window.

[0022] Using the reciprocal of the coefficient of variation of each channel as the weight, for the th The fused efficiency coefficient of a control surface is obtained by weighted fusion of the average efficiency coefficient window values ​​of the selected channels for each control surface. (16) In the formula, For the first Time of the first The fusion efficiency coefficient of each control surface; For the first Time of the first individual control surface channels The fusion weight; For the first Time of the first individual control surface channels The window mean of the efficiency coefficient. This weighting automatically assigns higher weights to channels with low coefficients of variation, i.e., good convergence and low volatility, while correspondingly reducing the weights of channels with high coefficients of variation, i.e., low identification reliability.

[0023] Step 8: Dual-criteria convergence determination and final performance coefficient output For each control surface, the coefficient of variation criterion and drift criterion are calculated based on the fusion performance coefficient sequence. When both criteria are simultaneously below their respective thresholds and remain below the set confirmation step count, the control surface is considered to have converged. The average value of the fusion performance coefficients within the window preceding the convergence time is taken as the final performance coefficient of the control surface. Specifically: For the first The sequence of fusion efficiency coefficients for each control surface Simultaneously calculate the coefficient of variation criterion and the drift criterion within the sliding window: (17) In the formula, The coefficient of variation criterion for the fusion sequence within the window; The criterion for determining the drift of the fused sequence within the window; , These are the fusion efficiency coefficients. Mean and standard deviation within the window; The slope is obtained by performing a linear fit on the fused sequence within the window; The window duration is... Coefficient of variation criterion Constraints on the fluctuation amplitude and drift criterion of the fusion sequence The relative drift over the window length is used to measure whether the fused sequence has a monotonic trend. Combining the two can eliminate slow drift cases that cannot be identified by the simple fluctuation criterion.

[0024] when and At the same time, it remains below its respective set threshold and continuously maintains that level. At the sampling step, determine the . The recognition convergence of individual rudder surfaces, among which To confirm the number of steps, the convergence time is denoted as . The final efficiency coefficient of the control surface is taken as the average of the fused efficiency coefficients within a window before the convergence time: (18) In the formula, For the first The final efficiency coefficient of each control surface; For the first The convergence moment of each rudder surface; The number of sampling steps contained in the window; For the first Time of the first The fusion efficiency coefficient of each control surface. A value less than 1 indicates that the control surface has suffered an effectiveness loss.

[0025] The beneficial effects of this invention are: First, by simultaneously superimposing orthogonal multi-sine excitation signals on six control surfaces and jointly identifying them, quantitative performance coefficients of each control surface can be given simultaneously when multiple control surfaces experience performance loss. This overcomes the shortcomings of existing methods, which are mostly designed for single control surface failures and lack the ability to identify multiple control surfaces simultaneously.

[0026] Second, the set of torque channels participating in the voting is pre-specified according to the physical main control function of each control surface, and channels with too small a control derivative and low identification signal-to-noise ratio are excluded from the voting, so as to avoid low signal-to-noise ratio channels from lowering the overall identification accuracy.

[0027] Third, the selected channels are weighted and fused using the inverse of the coefficient of variation as the weight, so that channels with good convergence and small fluctuations automatically receive higher weights. The fusion result does not depend on fixed inter-channel weights and can adapt to different fault combinations.

[0028] Fourth, the convergence determination is based on the dual criteria of the coefficient of variation and the amount of drift. This not only constrains the fluctuation range of the fused sequence, but also eliminates the slow drift that cannot be identified by the simple fluctuation criterion, thus improving the reliability of the convergence determination under sensor noise. Attached Figure Description

[0029] Figure 1 This is a flowchart of a method for actively identifying the performance loss of multiple control surfaces in flying-wing UAVs; Figure 2 It is a time-domain waveform diagram of the orthogonal multi-sine excitation signals of the six control surfaces; Figure 3 It is a spectrum diagram of the orthogonal multi-sine excitation signals of the six control surfaces; Figure 4 These are the channel-by-channel identification curves of the effectiveness coefficients of the two faulty control surfaces; Figure 5It is a channel-by-channel weighted fusion curve of the efficiency coefficients of two faulty control surfaces and a convergence time diagram. Detailed Implementation

[0030] The specific embodiments of the present invention will be further described below with reference to the accompanying drawings and technical solutions.

[0031] The overall process of the method of the present invention is as follows: Figure 1 As shown.

[0032] The diagnostic subject in this embodiment is a flying-wing unmanned aerial vehicle (UAV) equipped with six trailing-edge control surfaces. The aircraft performs trim cruise at an altitude of 1200 meters and a speed of 200 meters per second, with a discrete sampling period of 0.01 seconds. The nominal aerodynamic derivatives required for recursive identification, as well as the nominal control derivatives of the control surfaces used for efficiency coefficient normalization, are all taken from the calculated values ​​of the aircraft's nominal aerodynamic model at the aforementioned trim points.

[0033] This method is initiated after the upstream fault detection process confirms that the control surfaces have suffered performance loss. First, step 1 is executed, superimposing one orthogonal multi-sine excitation signal onto each of the six control surfaces' closed-loop control commands. The frequency sets of each channel are divided between 0.1 and 2.0 Hz and do not overlap to ensure orthogonality between the excitation signals; the amplitude of the excitation signals is kept small, ensuring that the required excitation intensity for identification does not significantly disturb the trim cruise state. The excitation signals are continuously injected from the start time until all control surfaces have been identified or the excitation interval ends. In this embodiment, the time-domain waveforms and spectra of the six orthogonal multi-sine excitation signals are as follows: Figure 2 and Figure 3 As shown, the frequencies of each channel do not overlap, and the amplitudes of the excitation signals are all small.

[0034] Then proceed to step 2. Acquire the three-axis angular rates output by the inertial navigation system, perform a center difference to obtain estimated three-axis angular acceleration values, substitute these values ​​into the rigid body rotation dynamics equations to solve for the three-axis torques, and then perform dimensionless dynamic pressure to obtain the roll, pitch, and yaw torque coefficients, which are used as identification observations. When the sensor measurement noise is high, a low-pass filter can be applied to the angular rate sequence before numerical differentiation to suppress the amplification of high-frequency noise during differentiation.

[0035] Perform steps 3 and 4. Following step 3, linearly parameterize the three-axis moment coefficients with respect to flight state variables and six control surface deflections, constructing regression models for roll, pitch, and yaw channels. The dimensions of the aerodynamic derivative vectors to be identified for the three channels are 10, 9, and 10, respectively. Following step 4, run the recursive least squares equation error method with a forgetting factor on each of the three channels, identifying the aerodynamic derivative vectors of each channel online from the injection time of the orthogonal multisine excitation signal. In this embodiment, the forgetting factor is 0.998, which can be adjusted within the range (0,1] depending on the time-varying rate of the aerodynamic derivative. The initial value of the aerodynamic derivative estimation is taken as the corresponding value of the healthy nominal aerodynamic model at the trim point, and the initial value of the covariance matrix is ​​the product of a sufficiently large positive number and the identity matrix.

[0036] Perform steps 5 and 6. In step 5, at each moment of the identification process, divide the identified control derivative values ​​of each control surface in the three torque channels by the corresponding health nominal control derivative to obtain the channel-by-channel efficiency coefficient of the control surface. Figure 4 The control derivative identification convergence curves for two faulty control surfaces (right elevon ELE_R and left pitch flap PF_L) on three moment channels are presented. The solid blue line represents the least-squares identification value, and the dashed red line represents the simulation true value. In the figure, (a) and (b) show the control derivatives of ELE_R and PF_L on the roll channel, respectively. The identification curves and identification values ​​converge to near the true value relatively quickly after the excitation is applied, with good convergence accuracy; (c) and (d) are the control derivatives of ELE_R and PF_L on the pitch channel, respectively. The identification curve shows convergence behavior similar to the roll channel, with a smaller steady-state estimation error; (e) and (f) are the control derivatives of ELE_R and PF_L on the yaw channel, respectively. The identification curves show that the identification values ​​of the yaw channel fluctuate greatly and the steady-state error is significantly higher. This is consistent with the characteristics that the contribution of the elevons and pitch flaps to the yaw moment is relatively small and the signal-to-noise ratio is low.

[0037] Perform steps 7 and 8. Following step 7, introduce a sliding window of 5.0 seconds. For each selected channel on each control surface, calculate the window mean and coefficient of variation of the channel-by-channel effectiveness coefficient, and use the reciprocal of the coefficient of variation (weighted smaller positive constants are taken as 1×10⁻⁶). -6 The selected channels are weighted and fused to obtain the fusion efficiency coefficient of the control surface. According to step 8, the coefficient of variation criterion and drift criterion are calculated simultaneously for the fusion efficiency coefficient sequence of each control surface. When both criteria are lower than the threshold (0.02 in this embodiment) and remain so for 200 consecutive steps, convergence is determined, and the average value of the fusion efficiency coefficient in the window before the convergence time is taken as the final efficiency coefficient.

[0038] In the typical simulation scenario of this embodiment, a 20% performance loss (actual performance coefficient) is applied simultaneously to the right elevon (ELE_R) and the left pitch flap (PF_L). The other four control surfaces remain healthy. ). Figure 5 The graph shows the weighted fusion efficiency coefficient curves and convergence times for the six control surfaces. In each subgraph, the thin lines represent the channel-by-channel efficiency coefficient curves participating in the voting, the thick lines represent the weighted voting fusion results, and the vertical dotted lines indicate the convergence time. Specifically, (a) shows the fusion efficiency coefficient curve for ELE_R, which converged at 54.3 seconds, with a final identified value of 0.803 and an identification error of 0.32% compared to the true value of 0.8; (b) shows the fusion efficiency coefficient curve for ELE_L, which converged at 52.5 seconds, with a final identified value of 1.002 and an identification error of 0.24%; and (c) shows the fusion efficiency coefficient curve for PF_R, which converged at 52.5 seconds, with a final identified value of 1.016. The error is 1.59%; (d) is the fusion efficiency coefficient curve of PF_L, which converges in 72.7 seconds, with a final identified value of 0.781 and an identification error of 2.36%; (e) is the fusion efficiency coefficient curve of AMT_R, which converges in 55.3 seconds, with a final identified value of 0.993 and an identification error of 0.66%; (f) is the fusion efficiency coefficient curve of AMT_L, which converges in 57.3 seconds, with a final identified value of 1.010 and an identification error of 0.99%. The identification errors of the two faulty control surfaces are both within 2.5%, and the identification values ​​of the four healthy control surfaces are all around 1.0 with an error of no more than 2%. This verifies the quantitative identification accuracy of this method under conditions of simultaneous failure of multiple control surfaces and the presence of sensor noise, and can provide reliable per-control-surface efficiency coefficients for downstream reconfigurable control allocation.

Claims

1. An active identification method for the performance loss of multiple control surfaces in a flying-wing unmanned aerial vehicle, characterized in that, Includes the following steps: Step 1: Superimpose orthogonal multi-sine excitation signals onto each control surface The flying-wing UAV is equipped with six trailing edge control surfaces, which are symmetrically distributed along the wingspan direction, including the left elevon, right elevon, left pitch flap, right pitch flap, left wingtip all-moving control surface, and right wingtip all-moving control surface; after confirming that the control surface has lost its effectiveness, an orthogonal multi-sine excitation signal is superimposed on the closed-loop control command of each control surface. Step 2: Inversely solve the triaxial torque coefficients from the angular rate measurements. The triaxial angular velocities are collected, and their numerical differentiation yields estimates of the triaxial angular accelerations. These estimates are then substituted into the rigid body rotational dynamics equations to inversely solve for the triaxial moments. Finally, the roll, pitch, and yaw moment coefficients are obtained by dimensionless dynamic pressure calculation. , , , as an identification observation; Step 3: Construct a three-channel regression model The three-axis moment coefficients are linearly parameterized with respect to flight state variables and six control surface deflections, constructing regression models for three moment channels: roll, pitch, and yaw; the regression model is expressed as follows: ,in For torque channel labeling, For discrete time indices. For channel The regression vector, Let be the aerodynamic derivative vector to be identified for the channel. For the first Time Channel The observed values ​​of the torque coefficients are obtained by inverse solution from step 2; Step 4: Online identification of errors in recursive least squares equations with forgetting factors. The recursive least squares equation error method with forgetting factor is used for each of the three channels to identify the aerodynamic derivative vectors of each channel, including the six control surface derivatives. Step 5: Extract the three-channel efficiency coefficients per rudder surface Divide the control derivative identification value of each control surface in the three torque channels by the corresponding health nominal control derivative to obtain the channel-by-channel efficiency coefficient of the control surface in the three channels. Step 6: Specify the set of torque channels participating in the voting on a per-control-face basis according to the physical main control action. Each control surface is pre-assigned a set of torque channels to participate in the voting. ,in For the first The set of torque channels for each control surface to participate in the voting; Step 7: Sliding window variation coefficient weighted fusion Enter the sliding window, for each control surface, the set of channels. For each channel within the range, calculate the window mean of the channel-by-channel performance coefficient. With window coefficient of variation The window mean values ​​of each channel in the channel set are weighted and fused using the reciprocal of the coefficient of variation as the weight to obtain the fusion efficiency coefficient of the control surface. Step 8: Dual-criteria convergence determination and final performance coefficient output For each control surface, the coefficient of variation criterion and drift criterion are calculated for the fusion efficiency coefficient sequence. When both criteria are simultaneously below their respective thresholds and the set number of confirmation steps is maintained, the control surface is determined to have converged. The average value of the fusion efficiency coefficients within the window before the convergence time is taken as the final efficiency coefficient of the control surface.

2. The active identification method for multi-control surface performance loss of a flying-wing UAV according to claim 1, characterized in that, In step 1, the first The excitation signal for each control surface channel is: (1) In the formula, For the first Excitation signal for each control surface channel , For time; For the first The amplitude of the excitation signal for each control surface channel; For the first The number of harmonics contained in each control surface channel; , , The first The first control surface channel The frequency, phase, and power distribution coefficients of each harmonic, and the frequency sets allocated to each control surface channel are mutually exclusive, thereby ensuring that the excitation signals between channels are mutually orthogonal.

3. The active identification method for multi-control surface performance loss of a flying-wing UAV according to claim 1, characterized in that, Step 2 is as follows: The equation of motion for the rigid body rotation of a flying-wing UAV about its center of mass is as follows: (2) In the formula, , , These are the roll, pitch, and yaw rates of the airframe shaft system, respectively. , , These are the first derivatives of the corresponding angular velocities; , , These are the roll, pitch, and yaw moments of the airframe axis system, respectively. The inertia coupling coefficient is determined by the aircraft's moment of inertia. The triaxial angular velocities are measured by inertial navigation, and the central difference numerical differential is performed to obtain the estimated triaxial angular accelerations. The rigid body rotational dynamics equations are rewritten in the form where the triaxial torques are unknowns: the roll channel and yaw channel equations are combined to form the equations regarding... , A system of two linear equations, with the pitch channel given separately. Solving for the problem yields: (3) in: (4) In the formula, , These are the remaining values ​​after subtracting the control-inertial coupling terms from the estimated angular acceleration values ​​in the roll and yaw channels, respectively. Then, the triaxial torque is dimensionless, yielding the triaxial torque coefficients: (5) In the formula, For dynamic pressure; This refers to the wing reference area. For wingspan; The mean aerodynamic chord length is given.

4. The active identification method for multi-control surface performance loss of a flying-wing UAV according to claim 1, characterized in that, Step 3 is as follows: The three-axis moment coefficients are linearly parameterized with respect to flight state quantities and six control surface deflections, expressed as: (6) In the formula, , , These are the reference values ​​for the roll, pitch, and yaw moment coefficients, respectively. , , These are the three-axis torque coefficients relative to the angle of attack. The derivative; , These are the roll and yaw moment coefficients relative to the sideslip angle, respectively. The derivative; , , These are the damping derivatives of the roll moment coefficient with respect to the roll rate, the pitch moment coefficient with respect to the pitch rate, and the yaw moment coefficient with respect to the yaw rate, respectively. , , The first Control derivatives of each control surface in the roll, pitch, and yaw channels; For the first The deflection of each control surface; For flight speed; The aerodynamic derivatives to be identified in the three torque channels are stacked into an aerodynamic derivative vector: (7) In the formula, , , These are the aerodynamic derivative vectors to be identified for the roll, pitch, and yaw channels, respectively; The regression vectors for the three channels at each moment are composed of the flight state quantity and the control surface deflection quantity at the current moment: (8) In the formula, , , The first Regression vectors for roll, pitch, and yaw channels at all times; subscript Indicates the discrete time sequence number; , , , , , The first Angle of attack, sideslip angle, flight speed, roll rate, pitch rate, and yaw rate at any given moment; For the first The deflection of the six control surfaces at any given time; the regression model for the three channels is uniformly written as: (9)。 5. The active identification method for multi-control surface performance loss of a flying-wing UAV according to claim 1, characterized in that, Step 4 is as follows: For the channel , No. The recursive update law for each moment is: (10) In the formula, For the first Time Channel The estimated value of the aerodynamic derivative; For the first Time Channel The estimated value of the aerodynamic derivative; For the first Time Channel The gain vector; For the first Time Channel The covariance matrix; For the first Time Channel The covariance matrix; For the first Time Channel The regression vector; The forgetting factor has a value range of 1. ; For the channel An identity matrix with the same dimension as the aerodynamic derivative vector; Initial values ​​for aerodynamic derivative estimation The aerodynamic derivatives of the healthy nominal aerodynamic model at the balance point are obtained by using the small perturbation method. Initial values ​​of the covariance matrix Take as: (11) The recursion is carried out step by step from the injection time of the orthogonal multi-sine excitation signal, and the aerodynamic derivative estimates of the three channels are updated according to Equation (10) at each time.

6. The active identification method for multi-control surface performance loss of a flying-wing UAV according to claim 1, characterized in that, Step 5 is as follows: At any given moment , will the Dividing the control derivative identification value of each control surface in the three torque channels by its healthy nominal control derivative, we obtain the channel-by-channel effectiveness coefficient of the control surface in the three channels: (12) In the formula, For the first Time of the first Each rudder surface in the channel The efficiency coefficient; For the first The first time to identify Each rudder surface in the channel The estimated value of the control derivative on; For the first Each rudder surface in the channel The aerodynamic derivatives are obtained using the small perturbation method; an efficiency coefficient of 1 indicates that the channel has no energy loss, less than 1 indicates that there is efficiency degradation, and 0 indicates complete failure.

7. The active identification method for multi-control surface performance loss of a flying-wing UAV according to claim 1, characterized in that, Step 6 is as follows: The left and right elevons generate roll moment through differential deflection and serve as the primary roll control surfaces; the left and right pitch flaps generate pitch moment through symmetrical deflection and serve as the primary pitch control surfaces. The set of channels for both to participate in the voting is taken as: (13) The left and right wingtip all-moving control surfaces generate yaw moment through differential deflection and serve as the primary yaw control surfaces. The set of channels participating in the voting is: (14)。 8. The active identification method for multi-control surface performance loss of a flying-wing UAV according to claim 1, characterized in that, Step 7 is as follows: The introduced length is Seconds, corresponding The sliding window of the sampling step; in the first sampling step At that moment, for the first For each channel participating in the voting for each control surface, calculate the mean and coefficient of variation of its channel-by-channel effectiveness coefficient within the window: (15) In the formula, For the first Time of the first individual control surface channels The window mean of the efficiency coefficient; The corresponding window standard deviation; This represents the corresponding window variation coefficient; For the first Time of the first individual control surface channels The efficiency coefficient; To prevent positive numbers with a denominator of zero; This refers to the sampling time sequence number within the window; Using the reciprocal of the coefficient of variation of each channel as the weight, for the th The fused efficiency coefficient of a control surface is obtained by weighted fusion of the average efficiency coefficient window values ​​of the selected channels for each control surface. (16) In the formula, For the first Time of the first The fusion efficiency coefficient of each control surface; For the first Time of the first individual control surface channels The fusion weight; For the first Time of the first individual control surface channels The window mean of the efficiency coefficient.

9. The active identification method for multi-control surface performance loss of a flying-wing unmanned aerial vehicle according to claim 1, characterized in that, Step 8 is as follows: For the first The sequence of fusion efficiency coefficients for each control surface Simultaneously calculate the coefficient of variation criterion and the drift criterion within the sliding window: (17) In the formula, The coefficient of variation criterion for the fusion sequence within the window; The criterion for determining the drift of the fused sequence within the window; , These are the fusion efficiency coefficients. Mean and standard deviation within the window; The slope is obtained by performing a linear fit on the fused sequence within the window; The window duration is... ; when and At the same time, it remains below its respective set threshold and continuously maintains that level. At the sampling step, determine the . The recognition convergence of individual rudder surfaces, among which To confirm the number of steps; the convergence time is denoted as . The final efficiency coefficient of the control surface is taken as the average of the fused efficiency coefficients within a window before the convergence time: (18) In the formula, For the first The final efficiency coefficient of each control surface; For the first The convergence moment of each rudder surface; The number of sampling steps contained in the window; For the first Time of the first The fusion efficiency coefficient of each control surface; A value less than 1 indicates that the control surface has suffered an effectiveness loss.

Images