An adaptive control method for an aircraft wing and an aircraft wing

Abstract

An adaptive control method and an aircraft wing of an aircraft, wherein the adaptive control method comprises: constructing a prediction model, inputting real-time external aerodynamic disturbance parameters into the prediction model, predicting a prediction disturbance error at the next moment, realizing feedforward prediction compensation of the external aerodynamic disturbance, and eliminating the influence of predictable disturbance on the airfoil in advance; then monitoring the wing pose to obtain a tracking airfoil parameter, and calculating the deviation between the tracking airfoil parameter and a target airfoil parameter to obtain an actual tracking error, so as to provide a real-time airfoil pose feedback signal. Finally, a control component generates a control instruction based on the prediction disturbance error, the actual tracking error and the target airfoil parameter, and acts on a mechanical neural network to realize adaptive adjustment of the wing airfoil, so that the wing airfoil continuously approximates to the target airfoil neighborhood through the composite control mechanism of feedforward and feedback.

Description

Technical Field

[0001] This invention relates to the field of aircraft control technology, and in particular to an adaptive control method for an aircraft wing and an aircraft wing. Background Technology

[0002] The core mission of civil aircraft is to transport personnel and goods safely, efficiently, economically, and comfortably by air. As the core lift component of a civil aircraft, the wing is a concentrated embodiment of aerodynamic characteristics, structural performance, and functional integration. Its design directly determines the aircraft's fuel efficiency, takeoff and landing performance, cruise capability, operating costs, and environmental performance. The core functions of a civil aircraft wing span the entire lifecycle of the aircraft, both in flight and on the ground. Specifically, these include: generating all the lift required for flight to support the fuselage weight; reducing air resistance during cruise and takeoff / landing phases by optimizing the aerodynamic shape; improving low-speed takeoff and landing performance in conjunction with high-lift devices such as flaps and slats; and providing a mounting platform and structural support for engines, fuel tanks, and avionics. From a technical perspective, the modern civil aircraft wing system is an integration of aerodynamic design, structural engineering, materials science and control technology. It mainly consists of airfoil structure (main wing, winglets), high-lift system (slats, flaps, actuators), load-bearing structure (spars, ribs, skin), fuel storage system and auxiliary control devices. The coordinated performance of each subsystem directly determines the overall aerodynamic efficiency and structural reliability of the wing.

[0003] In the design system of civil aircraft, the wing typically accounts for 20%-30% of the weight, and its aerodynamic efficiency directly affects 30%-40% of the aircraft's fuel consumption. Therefore, optimizing its performance plays a crucial role in reducing aircraft weight and increasing efficiency, lowering fuel consumption, and controlling operating costs. Existing fixed airfoil designs only achieve optimal aerodynamic performance under specific flight conditions. When encountering external aerodynamic disturbances such as atmospheric turbulence, crosswind loads, temperature gradients, and changes in flight speed, the wing shape undergoes uncontrollable deformation, leading to a significant decrease in lift-to-drag ratio. Furthermore, traditional actively deformable wings often employ discrete actuators (such as shape memory alloy sheets and piezoelectric ceramic stacks), which have coarse adjustment granularity, large response delays, and rely on centralized electronic control systems for closed-loop calculations, resulting in computational bottlenecks and communication delays, making it difficult to achieve continuous, fine-grained, and real-time collaborative optimization of airfoil parameters. How to combine algorithms and control theory to achieve a highly adaptable and efficient flight control method is a key challenge in the current development of aircraft technology. Summary of the Invention

[0004] This invention provides an adaptive control method for an aircraft wing and an aircraft wing, aiming to solve the technical bottlenecks of insufficient adaptability and difficulty in balancing control accuracy and response speed in traditional wing control methods.

[0005] To achieve the above objectives, the present invention provides the following technical solution:

[0006] An adaptive control method for an aircraft wing, the aircraft wing comprising a frame and a skin covering the frame, wherein a plurality of sequentially connected mechanical neural networks are provided between the frame and the skin, and the end nodes of each mechanical neural network are connected to the skin, such that the output pose of each mechanical neural network constitutes the airfoil of the aircraft wing, characterized in that the adaptive control method includes:

[0007] S1. Construct a prediction model, and then input the collected real-time external aerodynamic disturbance parameters into the prediction model to predict the wing shape error caused by the external aerodynamic disturbance at the next moment. The wing shape error is the predicted disturbance error.

[0008] S2. Based on the real-time output pose of the outer contour nodes of each of the mechanical neural networks, the tracking airfoil parameters are obtained, and the deviation between the tracking airfoil parameters and the target airfoil parameters is calculated to obtain the actual tracking error; the target airfoil parameters are preset in the control component.

[0009] S3. The control component generates control commands based on the predicted disturbance error, the actual tracking error, and the target airfoil parameters. The control commands are applied to the mechanical neural network to achieve adaptive adjustment of the airfoil.

[0010] Furthermore, the prediction model in step S1 includes a cascaded aerodynamic condition prediction model and an error prediction model. The aerodynamic condition prediction model predicts the external aerodynamic disturbance parameters at the next moment based on the real-time external aerodynamic disturbance parameters. The error prediction model then maps the predicted external aerodynamic disturbance parameters to obtain the predicted disturbance error. The external aerodynamic disturbance parameters include atmospheric turbulence intensity, relative speed of the aircraft, aerodynamic load distribution, ambient temperature, and crosswind load.

[0011] Furthermore, in step S1, historical monitoring data is used as the training dataset for the prediction model. The prediction model extracts features from the historical monitoring data, learns from them, and outputs the prediction perturbation error for the next moment. The historical monitoring data includes the external aerodynamic perturbation parameters and the corresponding actual tracking error.

[0012] Furthermore, the prediction error of the prediction model satisfies: ,in, and The poses of the feature points are respectively the predicted and the actual measured poses of the next time step under the influence of the external aerodynamic disturbance parameters.

[0013] Furthermore, in step S2, the acquisition component acquires the output pose of the mechanical neural network in real time and obtains the tracking airfoil parameters. The control component calculates the deviation between the tracking airfoil parameters and the target airfoil parameters dimension by dimension to obtain the actual tracking error. ;in, for The target airfoil parameters at time 1, for The tracking airfoil parameters at time t, and the actual tracking error includes displacement error and angle error.

[0014] Furthermore, in step S3, the control command generated by the control component is: ;in, The actual tracking error is... This represents the feedforward control gain coefficient. The pose of the feature points under the influence of the external aerodynamic disturbance parameters at the next predicted moment. For proportional gain, For differential gain, The actual rate of change of the tracking error is given.

[0015] Furthermore, the mechanical neural network includes an adjustable beam, and the control component controls the output pose of the mechanical neural network by controlling the extension and retraction of the adjustable beam; step S3 includes:

[0016] S31, The control component generates the control command based on the predicted disturbance error, the actual tracking error, and the target airfoil parameters;

[0017] S32. Using a genetic algorithm, the optimal stiffness parameter combination corresponding to the control command is output. The control component controls the extension and retraction of the adjustable beam based on the optimal stiffness parameter combination to achieve adaptive adjustment of the wing airfoil.

[0018] Furthermore, step S32 includes:

[0019] S321. Within a preset stiffness value range, several sets of stiffness parameter combinations are randomly generated to form an initial population; each set of stiffness parameter combinations defines the axial stiffness value of each adjustable beam in the mechanical neural network.

[0020] S322. For each set of stiffness parameter combinations in the population, calculate the MSE value between the output pose of the mechanical neural network and the target airfoil parameters under the preset external aerodynamic disturbance parameters, and obtain the fitness of each set of stiffness parameter combinations.

[0021] S323. Based on the fitness, select excellent individuals, generate a new generation of stiffness parameter combinations through crossover and mutation operations to update the population, and repeat step S322 until the preset iteration termination condition is reached.

[0022] S324. After the iteration terminates, output the stiffness parameter combination with the smallest MSE value in the population throughout the generations, as the optimal stiffness parameter combination for controlling the mechanical neural network.

[0023] An aircraft wing, used to implement an adaptive control method for an aircraft wing as described above, includes a load-bearing component for connection to the aircraft cabin, the load-bearing component including a wing frame and a skin covering the wing frame; it also includes a plurality of mechanical neural networks disposed between the frame and the skin, each mechanical neural network consisting of a plurality of neuron motion branches, the end nodes of each neuron motion branch being connected to the skin; it further includes a data acquisition component for acquiring external aerodynamic disturbance parameters and the output pose of the mechanical neural networks; and it further includes a control component connected to the data acquisition component, the control component controlling the displacement of each neuron motion branch based on the external aerodynamic disturbance parameters and the output pose of the mechanical neural networks to achieve adaptive adjustment of the wing airfoil formed by the skin.

[0024] Furthermore, the neuronal motor branch is an RPR structure motor branch, including a first rotational joint, a prismatic joint, and a second rotational joint connected in sequence; adjacent mechanical neural networks are connected by a UPS structure motor branch.

[0025] The beneficial effects of this invention are:

[0026] 1. This invention proposes an adaptive control method for an aircraft wing. The aircraft wing includes a frame and a skin covering the frame. Several sequentially connected mechanical neural networks are provided between the frame and the skin. The adaptive control method includes: constructing a prediction model; inputting real-time external aerodynamic disturbance parameters into the prediction model; predicting the predicted disturbance error at the next moment; achieving feedforward prediction compensation for external aerodynamic disturbances; and eliminating the influence of predictable disturbances on the airfoil in advance. Then, based on the real-time output pose of the outer contour nodes of each mechanical neural network, the tracking airfoil parameters are obtained, and the deviation between the tracking airfoil parameters and the target airfoil parameters is calculated to obtain the actual tracking error, providing a real-time airfoil pose feedback signal. Finally, the control component generates control commands based on the predicted disturbance error, the actual tracking error, and the target airfoil parameters, which are applied to the mechanical neural networks to achieve adaptive adjustment of the wing airfoil. Through a composite control mechanism of feedforward and feedback, the wing airfoil continuously approaches the neighborhood of the target airfoil.

[0027] 2. This invention proposes an adaptive control method for an aircraft wing. The prediction model includes a cascaded aerodynamic condition prediction model and an error prediction model. The aerodynamic condition prediction model predicts the external aerodynamic disturbance parameters at the next moment based on real-time external aerodynamic disturbance parameters. The error prediction model then maps the predicted external aerodynamic disturbance parameters to obtain the predicted disturbance error. This hierarchical design helps improve the prediction accuracy of the disturbance error, thereby enhancing the effectiveness of the composite control strategy. The external aerodynamic disturbance parameters include atmospheric turbulence intensity, relative flight speed of the aircraft, aerodynamic load distribution, ambient temperature, and crosswind load. These multi-dimensional and high-precision external aerodynamic disturbance parameters together constitute the input basis of the prediction model, ensuring that the aerodynamic condition prediction model can fully perceive the dynamic changes of the complex flight environment, providing rich feature information for the error prediction model, thereby improving the matching degree between the predicted disturbance error and the actual disturbance error, and laying a data foundation for the accurate generation of subsequent control commands.

[0028] 3. The adaptive control method for an aircraft wing proposed in this invention uses historical monitoring data as the training dataset for a prediction model in step S1. The prediction model extracts features from the historical monitoring data, learns from them, and outputs the predicted disturbance error for the next moment. The historical monitoring data includes external aerodynamic disturbance parameters and the corresponding actual tracking errors. Furthermore, the prediction error of the prediction model must satisfy the following: During model training, the network parameters are continuously optimized through iterative processes, so that the prediction error continues to converge to the threshold range, thereby ensuring that the prediction model can maintain high prediction accuracy in complex and ever-changing flight environments.

[0029] 4. The adaptive control method for an aircraft wing proposed in this invention involves a data acquisition component that acquires the output pose of a mechanical neural network in real time to obtain the tracking airfoil parameters. The control component then calculates the actual tracking error by performing a dimensional deviation calculation between the tracking airfoil parameters and the target airfoil parameters. The actual tracking error includes displacement error and angle error, providing a precise control basis for subsequent adaptive adjustment of the wing airfoil.

[0030] 5. An adaptive control method for an aircraft wing is proposed in the invention. Step S3 includes step S32: A genetic algorithm outputs an optimal stiffness parameter combination corresponding to real-time external aerodynamic disturbance parameters. The control component then controls the extension and retraction of the adjustable beam based on the optimal stiffness parameter combination, thereby changing the bending and torsional stiffness of different regions of the wing. For example, when encountering strong crosswind disturbances, the genetic algorithm outputs a parameter combination that increases the stiffness of the outer region of the wing. The control component then controls the extension of the outer adjustable beam to enhance the structural strength of this region, reduce unexpected deformation of the wing, ensure that the wing can stably maintain the target airfoil, and guarantee the flight attitude stability and handling performance of the aircraft.

[0031] 6. This invention proposes an aircraft wing, comprising a load-bearing component for connection to the aircraft cabin. The load-bearing component includes a wing frame and a skin covering the wing frame. It also includes several mechanical neural networks disposed between the frame and the skin. Each mechanical neural network consists of several neuron motion branches, with the terminal nodes of each neuron motion branch connected to the skin. Utilizing the distributed redundancy and fault tolerance, real-time physical layer response, and hierarchical node pose adjustment characteristics of the mechanical neural network, continuous fine-grained collaborative optimization of airfoil parameters is achieved. The invention further includes a data acquisition component for acquiring external aerodynamic disturbance parameters and the output pose of the mechanical neural network. It also includes a control component that, based on the external aerodynamic disturbance parameters and the output pose of the mechanical neural network, controls the displacement of each neuron motion branch to achieve adaptive adjustment of the wing airfoil formed by the skin, ensuring that the wing maintains ideal aerodynamic performance under complex aerodynamic environments. Attached Figure Description

[0032] To more clearly illustrate the technical solutions in the embodiments of the invention or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0033] Figure 1 A schematic diagram of an aircraft incorporating the wing of this invention;

[0034] Figure 2 This is one of the schematic diagrams of an aircraft wing according to the present invention;

[0035] Figure 3 This is a second schematic diagram of an aircraft wing according to the present invention;

[0036] Figure 4 This is a schematic diagram of the mechanical neural network of an aircraft wing operating under airflow disturbance according to the present invention;

[0037] Figure 5 This is a two-dimensional schematic diagram of a mechanical neural network for an aircraft wing according to the present invention;

[0038] Figure 6 This is a three-dimensional schematic diagram of a mechanical neural network for an aircraft wing according to the present invention;

[0039] Figure 7 This is a front view of a mechanical neural network of an aircraft wing according to the present invention;

[0040] Figure 8 This is a side view of a mechanical neural network of an aircraft wing according to the present invention;

[0041] Figure 9 This is a flowchart of an adaptive control method for an aircraft wing according to the present invention;

[0042] Figure 10 This is a flowchart of step S32 in an adaptive control method for an aircraft wing according to the present invention;

[0043] In the diagram, 10 is the aircraft; 20 is the aircraft wing; 30 is the wing frame; 40 is the skin; 50 is the mechanical neural network; 501 is the RPR kinematic branch; and 502 is the UPS kinematic branch. Detailed Implementation

[0044] The following is combined Figure 1-10 The present invention will be described in detail below.

[0045] Example 1

[0046] This embodiment provides an aircraft wing 20, such as Figures 1 to 4 As shown, the system includes a load-bearing component for connection to the cabin of the aircraft 10. The load-bearing component includes a wing frame 30 and a skin 40 covering the wing frame 30. It also includes several mechanical neural networks 50 disposed between the wing frame 30 and the skin 40. Each mechanical neural network 50 is composed of several neuron motion branches, and the end nodes of each neuron motion branch are connected to the skin 40. The system also includes a data acquisition component for acquiring external aerodynamic disturbance parameters and the output pose of the mechanical neural networks 50. Furthermore, the system includes a control component connected to the data acquisition component. Based on the external aerodynamic disturbance parameters and the output pose of the mechanical neural networks 50, the control component controls the displacement of each neuron motion branch to achieve adaptive adjustment of the airfoil formed by the skin 40.

[0047] like Figures 5 to 8 As shown, the neuron's motor branches are RPR structure motor branches. RPR motor branches 501 include a first revolute joint, a prismatic joint, and a second revolute joint connected in sequence. Adjacent mechanical neural networks 50 are connected by UPS structure motor branches. UPS motor branches 502 include a universal joint, a prismatic joint, and a ball joint connected in sequence. Each motor branch is composed of a servo motor-driven ball screw and an elastic hinge structure, possessing high rigidity and high-precision motion adjustment capabilities, enabling independent or coordinated adjustment of wing camber, torsion angle, and sweep angle.

[0048] The RPR motion chain 501 can achieve high-precision, multi-degree-of-freedom active control of the pose of 50 nodes in a single-layer mechanical neural network. Meanwhile, the UPS motion chain 502 provides large-angle deflection tolerance and axial extension buffering capability between adjacent layers. The two have clear division of labor and hierarchical decoupling. While ensuring the accuracy of airfoil dynamic reconstruction, they effectively suppress interlayer stress concentration and motion interference, thereby supporting the stable configuration of the aircraft 10 wing in the neighborhood of the target airfoil under complex airflow disturbances.

[0049] Example 2

[0050] This embodiment provides an adaptive control method for an aircraft wing 20, applicable to the aforementioned aircraft wing 20. The aircraft wing 20 includes a frame and a skin 40 covering the frame. A plurality of sequentially connected mechanical neural networks 50 are provided between the frame and the skin 40. The end nodes of each mechanical neural network 50 are connected to the skin 40, such that the output pose of each mechanical neural network 50 constitutes the wing airfoil of the aircraft 10. Figure 9 As shown, the adaptive control method includes:

[0051] S1. Construct a prediction model, and then input the collected real-time external aerodynamic disturbance parameters into the prediction model to predict the wing shape error caused by the external aerodynamic disturbance at the next moment. The wing shape error is the predicted disturbance error.

[0052] The prediction model is a mathematical model used to establish the mapping relationship between external aerodynamic disturbance parameters and wing shape errors. Its input includes one or more external aerodynamic disturbance parameters collected in real time, such as atmospheric turbulence intensity, relative flight speed of the aircraft 10, aerodynamic load distribution, ambient temperature, and crosswind load. The output is the set of pose offsets of the outer contour nodes of each mechanical neural network 50 of the wing in three-dimensional space at the next moment. The wing shape error generated by the predicted external disturbance is quantified using a three-dimensional pose deviation vector in Cartesian coordinates and is directly related to the core aerodynamic parameters of the wing. For the entire mechanism, it specifically manifests as a combination of three types of geometric deviations and one type of aerodynamic performance deviation: the geometric deviations include the prediction error of the twist angle in the spanwise direction (Y-axis). camber error in the chord direction (X-axis) And the sweep angle error about the vertical axis (Z-axis). From the above-mentioned pose deviations, the lift-to-drag ratio prediction error can also be derived. .

[0053] The core of operational economy for civil aircraft lies in achieving a high lift-to-drag ratio and low fuel consumption. External aerodynamic disturbances can cause the wing's real-time attitude to deviate from its design state, ultimately resulting in a reduced lift-to-drag ratio and deteriorated flight performance. Therefore, the core control objective of this application is to maintain the wing attitude within the optimal or suboptimal range that sustains a high lift-to-drag ratio through real-time adaptive adjustment in complex dynamic flight environments. To achieve this objective, the target airfoil parameters preset by the control system correspond to the ideal geometric configuration that maximizes the theoretical lift-to-drag ratio under specific cruise conditions.

[0054] The data acquisition components include an air density monitoring device and a flight speed monitoring device to monitor air density and relative flight speed at flight altitude. The air density monitoring device is based on the International Standard Atmosphere (ISA) model and combines measured parameters from built-in sensors to indirectly calculate the air density at the current flight altitude. Its core formula is: ,in This is the static air pressure measured by the built-in pressure sensor, in Pa. Let be the gas constant for dry air, and take the value 287.05. . The corrected temperature based on the influence of water vapor is calculated using the following formula:

[0055] ,

[0056] in The temperature is measured in degrees Celsius, obtained directly through a built-in temperature sensor. The relative humidity is measured in kg / kg (the mass of water vapor contained in one kilogram of dry air), and is obtained by monitoring the humidity using a built-in humidity sensor.

[0057] The flight speed monitoring device is based on Bernoulli's principle, calculating the dynamic and static pressure of the airflow around the aircraft. This device mainly consists of a pitot tube, a static pressure orifice, and related accessories. The pitot tube is used to monitor the total pressure. The static pressure orifice is used to monitor static pressure. To avoid airflow disturbance. The formula for calculating flight speed is:

[0058] .

[0059] The pressure distribution and friction generated by airflow over the wing surface constitute the aerodynamic load on the wing. This aerodynamic load simultaneously generates lift L and drag D, which are calculated using the following formulas:

[0060] ,

[0061] Where S is the wing planar area, that is, the two-dimensional wing planar area composed of a single layer of mechanical neural network 50. Lift coefficient and drag coefficient All are wing shape functions, determined by the wing profile and angle of attack.

[0062] In the 50-layer mechanical neural network structure The calculation formula is: , The zero lift coefficient is for a two-dimensional airfoil; due to the use of a symmetrical airfoil, its value is 0. Let be the slope of the lift line of the two-dimensional airfoil, and take a value of . , This is the angle of attack, determined by the shape of the wing profile. The angle of attack is zero lift, and due to the use of a symmetrical airfoil, its value is 0. In the single-layer mechanical neural network 50 organizational structure, The calculation formula is:

[0063]

[0064] Induced drag coefficient Caused by wingtip vortices, this is a "drag that accompanies" lift and is directly related to the wing's planar shape. It is the Otto efficiency factor. Let A be the aspect ratio. Minimal drag produces sufficient lift, i.e., maintaining a high lift-to-drag ratio. The formula for calculating the lift-to-drag ratio is: .

[0065] In this embodiment, the prediction model in step S1 includes a cascaded aerodynamic condition prediction model and an error prediction model. The two are cascaded through a data interface. The aerodynamic condition prediction model predicts the external aerodynamic disturbance parameters at the next moment based on the real-time external aerodynamic disturbance parameters. The error prediction model then obtains the predicted disturbance error based on the predicted external aerodynamic disturbance parameters.

[0066] External aerodynamic disturbance parameters refer to external aerodynamic loads and environmental parameters that affect wing airfoil stability and aerodynamic efficiency, including but not limited to: atmospheric turbulence intensity (turbulence scale, pulsation velocity), relative speed of passenger aircraft (mainly cruise speed), aerodynamic load distribution (wing surface pressure gradient, lift and drag torque), ambient temperature (affecting atmospheric density and wing material mechanical properties), and crosswind loads (wind direction angle, etc.). The acquisition unit can collect historical monitoring data of the above-mentioned external aerodynamic disturbance parameters through an integrated aerodynamic environment detection unit. This unit includes a miniature turbulence sensor for measuring atmospheric turbulence intensity, an airspeed sensor for measuring atmospheric turbulence intensity, a distributed pressure sensor array for monitoring the distribution of aerodynamic loads on the wing surface, a temperature sensor deployed between the wing skin 40 and the wing frame 30, and an upper-air meteorological sensor for collecting wind direction angle and atmospheric density parameters. All monitored external aerodynamic disturbance parameter data are transmitted to the control unit in real time, forming a time-series of historical monitoring data for a specific flight path.

[0067] In some embodiments, external aerodynamic disturbance parameters can be further defined into three core aerodynamic parameters. Differential monitoring methods are used to scan and collect the flight airspace environment and aerodynamic parameters in real time, ensuring the comprehensiveness and accuracy of historical monitoring data. The three core aerodynamic parameters include: atmospheric flow field and aerodynamic environment parameters, flight airspace wind field parameters, and airspace temperature and humidity. Specifically, atmospheric flow field and aerodynamic environment parameters encompass atmospheric turbulence intensity parameters and atmospheric density ρ (0.3~1.3 kg / m³). Correspondingly, for monitoring atmospheric flow field and aerodynamic environment parameters, an airborne miniature turbulence detection radar is used to acquire three-dimensional distribution data of atmospheric turbulence. Frequency domain analysis algorithms are used to extract turbulence scale and fluctuating velocity. A high-precision pitot tube and an atmospheric data computer work together to calculate atmospheric density in real time. The flight airspace wind field parameters include high-altitude wind speed v (0~40m / s) and wind direction angle θ (based on the aircraft's heading). Correspondingly, for monitoring these parameters, an integrated fiber optic anemometer is deployed near the static pressure port on the leading edge of the wing to collect high-altitude wind field data in real time. Airspace environmental temperature and humidity include ambient temperature T (-55℃~70℃) and relative humidity RH (5%~100%). These parameters are used to correct for the influence of temperature and humidity on atmospheric density and the mechanical properties of the wing material. Therefore, an integrated temperature and humidity sensor is used, deployed in a non-stressed area inside the wing to avoid the influence of aerodynamic loads and structural vibrations on the sensor.

[0068] Furthermore, in step S1, historical monitoring data is used as the training dataset for the prediction model. The prediction model extracts features from the historical monitoring data, learns from it, and outputs the predicted disturbance error for the next moment. The historical monitoring data includes external aerodynamic disturbance parameters and corresponding actual tracking errors. A prediction model is constructed based on a genetic learning algorithm, and the trained prediction model is pre-installed in the control component. By extracting features from the historical monitoring data, such as the peak values ​​of lift and drag coefficient parameters, and performing learning, the final output is completed. The historical monitoring data used for training includes pre-collected historical monitoring data. In some embodiments, the prediction model can also adopt an online incremental learning mechanism. After each flight mission, newly collected external aerodynamic disturbance parameters and corresponding measured actual tracking errors are added to the training set to continuously adapt to long-term degradation effects such as wing material aging and skin wear.

[0069] In this embodiment, the prediction error of the prediction model satisfies: ,in, and The feature point poses are defined as follows: under the influence of predicted and actually measured external aerodynamic disturbance parameters at the next moment. The feature point can be a node of a motion chain. In this embodiment, the feature point is specifically the center point of the wing's tip chord. and the aerodynamic center of the wing , The pose of the feature points is predicted to be influenced by external aerodynamic disturbance parameters at the next moment. Let be the pose of the feature point under the external aerodynamic disturbance parameters at the next moment, as measured in the actual experiment. For ease of error calculation, and are both expressed as the absolute coordinate difference between the two feature points, i.e. .

[0070] Specifically, the prediction model employs a nonlinear mapping architecture based on a radial basis function (RBF) neural network. The number of hidden layer neurons in the RBF neural network is determined using cross-validation, and a Gaussian activation function is used. The mean squared error (MSE) loss function between the prediction error and the measured error is minimized using gradient descent, iteratively optimizing the weight matrix W and the bias vector b. After training, the overall prediction error of the model satisfies: This ensures the compensation accuracy of the feedforward control. Furthermore, parameters such as the order of the aerodynamic condition prediction model and the number of neurons in the hidden layer of the RBF neural network can be dynamically adjusted according to the flight characteristics and airspace environment differences of different civil aircraft; they are not fixed parameters.

[0071] S2. Based on the real-time output pose of the outer contour nodes of each mechanical neural network 50, the tracking airfoil parameters are obtained, and the deviation between the tracking airfoil parameters and the target airfoil parameters is calculated to obtain the actual tracking error. By providing real-time feedback on the deviation between the current airfoil pose (tracking airfoil parameters) and the target pose (target airfoil parameters), the prediction deviation of the prediction model and the airfoil offset caused by unpredictable disturbances (such as sudden strong turbulence and extreme crosswinds) can be corrected in a timely manner. This complements the feedforward control, constructing a "prediction and feedback" dual closed-loop control foundation. The target airfoil parameters are preset in the control component.

[0072] The outer contour nodes refer to the end nodes of several motion branches that constitute the boundary contour of each mechanical neural network 50. These end nodes are connected to the skin 40, and their spatial coordinates and attitude angles jointly define the local airfoil of this layer of the wing. The target airfoil parameters are preset by the design specifications of the aircraft 10 and the flight condition requirements, and stored in the control component. Optionally, the target airfoil parameters can be set as the "optimal cruise airfoil attitude under corresponding loads", that is, the geometric parameters of the outer contour formed by the mechanical neural network 50 are completely matched with the target airfoil parameters (the geometric deviation of each dimension of the wing shape is ≤5mm). At this time, the aerodynamic structure of the wing platform is in the optimal stress state, which can maximize the cruise lift-to-drag ratio and avoid fatigue damage to key components (such as the wing frame 30 and adaptive skin 40) caused by sudden changes in aerodynamic loads.

[0073] Specifically, in step S2, the acquisition component acquires the output pose of the mechanical neural network 50 in real time and obtains the tracking airfoil parameters. The control component calculates the deviation between the tracking airfoil parameters and the target airfoil parameters dimension by dimension to obtain the actual tracking error. ;in, Let be the target airfoil parameters at time t. Let be the tracking airfoil parameters at time t, and the actual tracking error includes displacement error and angle error.

[0074] The acquisition components include a high-precision airfoil attitude monitoring unit that acquires the current pose of the mechanical neural network 50 in real time. This monitoring unit may include: a fiber optic grating sensor array (measuring the wing twist angle and sweep angle), a laser displacement sensor (range 0~300mm, measuring the wing camber change), and a three-dimensional scanning vision module (positioning accuracy ±0.05mm, assisting in verifying the airfoil attitude data).

[0075] S3. The control component generates control commands based on the predicted disturbance error, the actual tracking error, and the target airfoil parameters. These control commands are applied to the mechanical neural network 50 to achieve adaptive adjustment of the airfoil. The control commands generated by the control component are as follows: ;in, This represents the actual tracking error. This represents the feedforward control gain coefficient. To predict the pose of the feature points under the influence of external aerodynamic disturbance parameters at the next moment. For proportional gain, For differential gain, This represents the rate of change of the actual tracking error. The control component fuses the feedforward control signal and the feedback control signal through a "weight allocation algorithm" to generate control commands, which drive the parallel motion chain to adjust the wings of the aircraft 10. Ultimately, this ensures that the pose of each node of the mechanical neural network 50 at the next moment is maintained within the neighborhood of the target pose. The neighborhood is defined as the absolute value of the geometric deviation of each node ≤ 0.01 mm and the absolute value of the angular deviation ≤ 0.1°.

[0076] The control component converts control commands into servo motor drive signals for each motion chain. By adjusting the motor speed and direction, it controls the extension and retraction of the ball screw, thereby adjusting the wing's attitude. For example, when atmospheric turbulence is predicted to cause a +0.5° disturbance error in the wing's twist angle, the feedforward control command drives the wingspan direction chain to reverse the terminal attitude by 0.5° in advance to counteract the turbulence impact. When the rate of change of the current tracking error is detected to be large, the differential element suppresses high-frequency oscillations in the airfoil attitude, ensuring a smooth adjustment process and avoiding damage to the wing structure caused by sudden aerodynamic load changes.

[0077] Steps S1 to S3 establish a dual-error feedback control mechanism based on prediction error and current tracking error, thereby achieving coordinated control of the motion branches of each layer of the mechanical neural network 50. The final control command output under this mechanism determines the target output pose of the entire mechanical neural network 50. Since a single (single-layer) mechanical neural network 50 is composed of multiple RPR motion branches 501 connected in parallel, for the same target output pose, each RPR motion branch 501 and UPS motion branch 502 may have multiple different displacement combinations. Therefore, step S32 can be further introduced in step S3, such as... Figure 10 As shown, an optimal combination of branch displacements is determined from all feasible solutions through an optimization algorithm, thereby refining the macroscopic pose control commands and allocating them to each specific motion branch.

[0078] In some embodiments, step S3 includes:

[0079] S31. The control component generates control commands based on the predicted disturbance error, the actual tracking error, and the target airfoil parameters.

[0080] S32: The genetic algorithm outputs the optimal combination of stiffness parameters corresponding to the control commands. The control component controls the extension and retraction of the adjustable beam based on the optimal combination of stiffness parameters. The prismatic joint (P-joint) in each kinematic branch is integrated with the adjustable beam and ball screw. The axial stiffness of the adjustable beam is the core optimization parameter of the genetic algorithm. The neuron kinematic branch includes the adjustable beam. The control component controls the output pose of the mechanical neural network 50 by controlling the extension and retraction of the adjustable beam.

[0081] Specifically, step S32 includes:

[0082] S321. Within the preset stiffness value range, several sets of stiffness parameter combinations are randomly generated to form an initial population; each set of stiffness parameter combinations defines the axial stiffness value of each adjustable beam in the mechanical neural network 50.

[0083] The number of stiffness combinations in each generation of optimization is defined, and in this embodiment, it is set to 1000 per generation. During population initialization, the stiffness value of each beam in all individuals within the stiffness range is randomly generated, denoted as , which is the stiffness value of the beam in the individual. The engineering constraint range of the adjustable beam axial stiffness is limited to a stiffness suitable for actual working conditions based on the structural mechanical properties and material strength threshold of civil aircraft wings. The stiffness value range is limited to , the maximum number of iterations is 40, and the convergence tolerance is . is the number of adjustable beams; is the number of output nodes used to construct the wing profile.

[0084] S322. For each stiffness parameter combination in the population, calculate the MSE value between the output pose of the mechanical neural network 50 and the target airfoil parameters under preset external aerodynamic disturbance parameters, and obtain the fitness of each stiffness parameter combination.

[0085] A preset external load is applied to the outer contour nodes of the 50-organization structure of the 3D mechanical neural network, and this load is converted into the displacement of the driving elements on the mechanical neurons based on the stiffness equation. The displacement is mapped by the parameters of the 50-organization structure of the 3D mechanical neural network, and the actual pose of each node is obtained after parameter combination. Then, the mean squared error (MSE) is used as the fitness function to quantify the deviation between the actual displacement and the target displacement of each mechanical neuron connection node. The current best MSE value and the corresponding combination are recorded as the parent stiffness combination.

[0086] Regarding the calculation of expansion and contraction (axial displacement): The axial stiffness of a single adjustable beam is defined as the sum of its axial driving force and its axial displacement. The ratio, the core conversion formula is:

[0087] ,

[0088] Overall axial displacement Due to the axial deformation of each adjustable beam The coupling effect determines its calculation formula:

[0089] ,

[0090] For the first The displacement weighting coefficient of the adjustable beam is determined by its topological position and force transmission path in the mechanical neural network 50 organization, and is calculated by the finite element method or the direct stiffness method.

[0091] Regarding the calculation of mean square error: For each layer (single) of the mechanical neural network 50, the expression is as follows:

[0092] ,

[0093] For the mechanical neurons that connect the layers of the mechanical neural network 50, the MSE value expression is:

[0094] .

[0095] in, This represents the number of output nodes; For the first The actual displacement of each output node in the corresponding direction; For the first The target displacement of each output node.

[0096] S323. Select excellent individuals based on fitness, generate a new generation of stiffness parameter combinations through crossover and mutation operations to update the population, and repeat step S322 until the preset iteration termination condition is reached.

[0097] Specifically, after the fitness of all individuals in the population has been calculated, a convergence test is performed: the current optimal parameter combination is compared with the historical optimal combination, and the one with better performance (elite) is retained as the new historical optimal combination. If no better combination is produced in five consecutive iterations, that is, the historical optimal combination remains unchanged for five consecutive iterations, then the algorithm is considered to have converged.

[0098] Elite individuals are selected using an ascending sorting strategy, with the top 50% forming the elite population. These elite individuals are then randomly paired and crossover is performed to generate offspring, which are then replenished to the initial population size before proceeding to the next round of iterative computation.

[0099] S324. After the iteration terminates, output the stiffness parameter combination with the smallest MSE value in the population of all generations, as the optimal stiffness parameter combination for controlling the mechanical neural network 50.

[0100] The data recorded in steps S1 to S2 can be used as the training set for the genetic algorithm in step S3, thereby determining the optimal combination of stiffness parameters to be adopted under specific external aerodynamic disturbance parameters. Step S32 deeply couples the optimal combination of stiffness parameters output by the genetic algorithm with the adjustable beam extension control, enabling each adjustable beam to obtain differentiated stiffness configurations under different disturbance conditions. This ensures both rapid response capability under strong disturbances and maintains compliant and energy-saving characteristics under weak disturbances, achieving adaptive adjustment of the airfoil.

[0101] The present invention provides an adaptive control method for the wing of an aircraft 10, comprising:

[0102] S1. Predicted Disturbance Error: Based on real-time collected external aerodynamic disturbance parameters, the predicted wing shape error caused by external disturbance at the next moment is calculated through a pre-constructed disturbance error prediction model.

[0103] S2. Obtain the actual tracking error: In real time, obtain the output pose of the outer contour node of the mechanical neural network 50, obtain the current tracking airfoil parameters, and compare them with the target airfoil parameters pre-stored in the control component to calculate the actual tracking error.

[0104] S3. Generate control commands, then determine and apply the optimal stiffness strategy: The control component integrates the predicted disturbance error with the actual tracking error, and combines it with the target airfoil parameters to generate control commands; The control component calls the corresponding optimal stiffness parameter combination for the obtained control commands, and then adjusts the equivalent axial stiffness of each adjustable beam in the mechanical neural network 50 based on the optimal stiffness parameter combination to optimize the overall mechanical properties of the wing structure and realize real-time adaptive adjustment of the wing airfoil.

[0105] The above embodiments are only for illustrating the technical concept and features of the present invention, and are intended to enable those skilled in the art to understand and implement the present invention. They should not be construed as limiting the scope of protection of the present invention. All equivalent changes or modifications made in accordance with the spirit and essence of the present invention should be covered within the scope of protection of the present invention.

Claims

1. An adaptive control method for an aircraft wing, the aircraft wing comprising a frame and a skin covering the frame, wherein a plurality of sequentially connected mechanical neural networks are provided between the frame and the skin, the end nodes of each mechanical neural network being connected to the skin, such that the output pose of each mechanical neural network constitutes the airfoil of the aircraft, the mechanical neural network comprising an adjustable beam, and a control component controlling the output pose of the mechanical neural network by controlling the extension and retraction of the adjustable beam; characterized in that, The adaptive control method includes: S1. Construct a prediction model, and then input the collected real-time external aerodynamic disturbance parameters into the prediction model to predict the wing shape error caused by the external aerodynamic disturbance at the next moment. The wing shape error is the predicted disturbance error. S2. Based on the real-time output pose of the outer contour nodes of each of the mechanical neural networks, the tracking airfoil parameters are obtained, and the deviation between the tracking airfoil parameters and the target airfoil parameters is calculated to obtain the actual tracking error; the target airfoil parameters are preset in the control component. S3. The control component generates control commands based on the predicted disturbance error, the actual tracking error, and the target airfoil parameters. The control commands act on the mechanical neural network to achieve adaptive adjustment of the airfoil. Step S3 includes: S31, The control component generates the control command based on the predicted disturbance error, the actual tracking error, and the target airfoil parameters; S32. Using a genetic algorithm, the optimal stiffness parameter combination corresponding to the control command is output. The control component controls the extension and retraction of the adjustable beam based on the optimal stiffness parameter combination to achieve adaptive adjustment of the wing airfoil. Step S32 includes: S321. Within a preset stiffness value range, several sets of stiffness parameter combinations are randomly generated to form an initial population; each set of stiffness parameter combinations defines the axial stiffness value of each adjustable beam in the mechanical neural network. S322. For each set of stiffness parameter combinations in the population, calculate the MSE value between the output pose of the mechanical neural network and the target airfoil parameters under the preset external aerodynamic disturbance parameters, and obtain the fitness of each set of stiffness parameter combinations. S323. Based on the fitness, select excellent individuals, generate a new generation of stiffness parameter combinations through crossover and mutation operations to update the population, and repeat step S322 until the preset iteration termination condition is reached. S324. After the iteration terminates, output the stiffness parameter combination with the smallest MSE value in the population throughout the generations, as the optimal stiffness parameter combination for controlling the mechanical neural network.

2. The adaptive control method for an aircraft wing as described in claim 1, characterized in that, The prediction model in step S1 includes a cascaded aerodynamic condition prediction model and an error prediction model. The aerodynamic condition prediction model predicts the external aerodynamic disturbance parameters at the next moment based on the real-time external aerodynamic disturbance parameters. The error prediction model then obtains the predicted disturbance error based on the predicted external aerodynamic disturbance parameters. The external aerodynamic disturbance parameters include atmospheric turbulence intensity, relative speed of the aircraft, aerodynamic load distribution, ambient temperature, and crosswind load.

3. The adaptive control method for an aircraft wing as described in claim 2, characterized in that, In step S1, historical monitoring data is used as the training dataset for the prediction model. The prediction model extracts features from the historical monitoring data, learns from them, and outputs the prediction perturbation error for the next moment. The historical monitoring data includes the external aerodynamic perturbation parameters and the corresponding actual tracking error.

4. The adaptive control method for an aircraft wing as described in claim 3, characterized in that, The prediction error of the prediction model satisfies: ,in, and The poses of the feature points under the influence of the external aerodynamic disturbance parameters at the next moment, as predicted and actually measured, are respectively.

5. The adaptive control method for an aircraft wing as described in claim 1, characterized in that, In step S2, the acquisition component acquires the output pose of the mechanical neural network in real time and obtains the tracking airfoil parameters. The control component calculates the deviation between the tracking airfoil parameters and the target airfoil parameters dimension by dimension to obtain the actual tracking error. ;in, Let be the target airfoil parameters at time t. Let be the tracking airfoil parameters at time t, and let be the actual tracking error including displacement error and angle error.

6. The adaptive control method for an aircraft wing as described in claim 1, characterized in that, In step S3, the control command generated by the control component is: ;in, The actual tracking error is... This represents the feedforward control gain coefficient. The pose of the feature points under the influence of the external aerodynamic disturbance parameters at the next predicted moment. For proportional gain, For differential gain, The actual rate of change of the tracking error is given.

7. An aircraft wing for implementing the adaptive control method for an aircraft wing as described in any one of claims 1-6, comprising a load-bearing component for connection to an aircraft cabin, the load-bearing component comprising a wing frame and a skin covering the wing frame; characterized in that, It also includes a plurality of mechanical neural networks disposed between the skeleton and the skin, each mechanical neural network consisting of a plurality of neuronal motion branches, the end nodes of the execution ends of each neuronal motion branch being connected to the skin; it also includes a data acquisition component for acquiring the external aerodynamic disturbance parameters and the output pose of the mechanical neural networks; and it also includes a control component connected to the data acquisition component, the control component controlling the displacement of each neuronal motion branch based on the external aerodynamic disturbance parameters and the output pose of the mechanical neural networks to achieve adaptive adjustment of the airfoil formed by the skin.

8. An aircraft wing as described in claim 7, characterized in that, The neuronal motor branch is an RPR structure motor branch, including a first rotational joint, a prismatic joint and a second rotational joint connected in sequence; adjacent mechanical neural networks are connected by a UPS structure motor branch.

Images