A bionic flapping wing control system and method for realizing trajectory tracking and vibration suppression

Abstract

The present invention discloses a bionic flapping-wing control system and method for realizing trajectory tracking and vibration suppression, and relates to the field of automatic control technology. The present invention builds a virtual bionic flapping-wing model and performs analysis, establishes a model equation of the bionic flapping-wing control system, designs a bionic flapping-wing control algorithm in combination with the model equation, uses a neural network to approximate the uncertain information of the system, constructs a bionic flapping-wing control algorithm execution module, and then designs a bionic flapping-wing control algorithm evaluation module to evaluate the control algorithm, thereby continuously adjusting the bionic flapping-wing control algorithm online, and at the same time simulating the motion process of the system, adjusting the control parameters, so that the system can suppress system vibration while realizing trajectory tracking.

Description

Technical Field

[0001] The present invention belongs to the field of automatic control technology, and more specifically, is a bionic flapping wing control system and method for realizing trajectory tracking and vibration suppression. Background Art

[0002] The inspiration of bionic flapping-wing flying robots comes from the flying postures of seagulls, bats, insects and other flying creatures. They are lightweight and have low energy consumption. Compared with rotary and fixed wings, the flight efficiency of bionic flapping-wing flying robots has more obvious advantages, and it has broad application prospects in both military and civilian fields. In the military, bionic flapping-wing UAVs can be used for enemy reconnaissance, target tracking, electronic interference, damage assessment, nuclear, biological and chemical sampling, deployment of sensors, relay communication, and even active attack and defense. In civilian use, bionic flapping-wing UAVs can be used for various types of monitoring, surveillance, patrol, search and rescue, photography, surveying and mapping investigations and surveys, such as environmental monitoring, disaster monitoring, traffic road monitoring, border patrol and control, drug embargo, agricultural survey, large ranch patrol, urban surveillance and aerial photography. However, the current research in the field of bionic flapping-wing flying robots is still in the development stage, and its theoretical modeling and analysis are quite difficult. The research on control theory and method technology is also in the early stages. People still have a long way to go to fully understand the mechanism of flapping-wing flight. More importantly, even in the rare research on bionic flapping-wing flying robots, most of the research only focuses on the single-stage flapping-wing system, and there are few studies considering the two-stage rigid-flexible coupling bionic flapping-wing system. Whether from the perspective of its potential application value or from the perspective of promoting bionic technology, control research and platform building for the bionic rigid-flexible coupling flapping-wing system are imperative research work.

[0003] In the existing bionic flapping wing flying robot control technology, the performance of most control algorithms is directly related to the flapping wing structural parameters. However, the bionic flapping wing system usually works in a complex environment, and the nonlinearity, time-varying and uncertainty factors in the actual system make it difficult to accurately obtain its dynamic parameters. The current bionic flapping wing control technology (such as distributed control, boundary control, etc.) can no longer meet the control requirements of the actual system. In addition, it should be noted that many studies on the control methods of bionic flapping wing systems have ignored vibration signals. Unlike traditional rigid structures, the use of flexible materials in bionic flapping wings brings about elastic vibration problems that cannot be ignored. Only by successfully solving the vibration problem of bionic rigid-flexible coupling flapping wings can the stability of the system be guaranteed. The current bionic flapping wing control technology (such as iterative learning control, fuzzy logic control, etc.) can no longer adapt to the challenges brought to control design by flexible characteristics and structural coupling. There are also a small number of control methods that can consider elastic vibration, most of which require the installation of actuators at the end of the flapping wing, which will cause the mechanical design of the bionic structure to be too complicated, making it difficult to achieve the ideal state. Therefore, the present invention will propose a trajectory tracking and vibration suppression method for a bionic rigid-flexible coupling flapping wing system that comprehensively considers the above factors.

[0004] In the existing bionic flapping wing flying robot model building work, the rigid-flexible coupling characteristics make the system model of the bionic flapping wing flying robot have the characteristics of infinite dimensions. Most research works usually use the Hamiltonian principle to analyze its energy and derive a series of partial differential equations and ordinary differential equations to describe the rigid-flexible coupling flapping wing system. Although this modeling method has high accuracy, the complicated derivation and tedious calculations that follow make the system model building work very difficult. In order to improve the modeling efficiency, some research works (such as the assumed modal method, the lumped mass method, etc.) use the mathematical method of dimensionality reduction, ignore the high-order modes of the system, and transform the infinite-dimensional model of the system into a finite-dimensional one. However, this method still does not improve the difficulty of large calculation amount, the model dimensionality reduction process is also very tedious, and the accuracy of the model is also reduced. In response to the above problems, many new modeling software and methods have emerged (such as the numerical simulation method based on XFlow software, the collaborative simulation method based on ADAMS, etc.). However, the scalability of these software modeling methods is very limited. Once the model is determined, it will be difficult to modify it, which makes the complex and changeable bionic system modeling unable to meet the real-time requirements. Therefore, the present invention proposes an efficient and highly simulated bionic rigid-flexible coupling flapping wing system virtual experimental device that comprehensively considers the above factors to meet the needs of model analysis and control algorithm verification of the bionic flapping wing system. Summary of the invention

[0005] The technology of the present invention solves the following problems: overcomes the shortcomings of the prior art, builds a high-efficiency, high-simulation, and visualized bionic flapping-wing model, and provides a bionic flapping-wing control system that solves system uncertainty, the flexible characteristics of bionic flapping wings, and the rationality of actuator installation, so that the system can accurately track the desired trajectory, while suppressing the undesirable elastic vibrations generated during the system movement, and realizing online learning and optimal control.

[0006] The technical solution of the present invention is: a bionic flapping-wing control system for realizing trajectory tracking and vibration suppression, which is characterized by comprising: a virtual bionic flapping-wing model module, a bionic flapping-wing control algorithm execution module and a bionic flapping-wing control algorithm evaluation module;

[0007] The virtual bionic flapping wing model module constructs a virtual bionic flapping wing model composed of rigid joints, rigid wings and flexible wings, and determines the model equation of the bionic flapping wing control system by the virtual bionic flapping wing model. The model equation is a second-order linear model equation.

[0008] The bionic flapping wing control algorithm execution module constructs a forward three-layer execution neural network based on the second-order linear model equation to approximate the unknown dynamic information in the model equation. Given the expected trajectory of the flapping angle of the bionic flapping wing, the control execution algorithm is designed to control the flapping angle and vibration of the bionic flapping wing according to the state error of the bionic flapping wing control system to achieve trajectory tracking and vibration suppression. At the same time, the bionic flapping wing control algorithm execution module will adjust the control execution algorithm according to the evaluation results, and apply the adjusted control execution algorithm to the virtual bionic flapping wing model module to obtain the real-time output of the virtual bionic flapping wing model module. The effectiveness of the control execution algorithm of the current bionic flapping wing control algorithm execution module is verified according to the real-time output of the virtual bionic flapping wing model module.

[0009] The bionic flapping-wing control algorithm evaluation module designs the control algorithm objective function according to the control objectives of trajectory tracking and vibration suppression, constructs a radial basis evaluation neural network, approximates the objective function of the control algorithm, evaluates whether the control algorithm in the current bionic flapping-wing control algorithm execution module can reach the optimal state, and feeds back the evaluation results to the bionic flapping-wing control algorithm execution module.

[0010] A bionic flapping-wing control algorithm execution module is designed based on the virtual bionic flapping-wing model module. The bionic flapping-wing control algorithm evaluation module evaluates whether the control algorithm of the current bionic flapping-wing control algorithm execution module can reach the optimal level, and feeds back the evaluation result to the bionic flapping-wing control algorithm execution module. The bionic flapping-wing control algorithm execution module will adjust the control execution algorithm according to the evaluation result. The control execution algorithm output by the bionic flapping-wing control algorithm execution module acts on the virtual bionic flapping-wing model module. The effectiveness of the control execution algorithm of the current bionic flapping-wing control algorithm execution module can be verified according to the real-time output of the virtual bionic flapping-wing model module.

[0011] The virtual bionic flapping wing model module is specifically implemented as follows: a wing portion of the bionic flapping wing close to the fuselage is constructed by a rigid material component, and a wing portion of the bionic flapping wing away from the fuselage is constructed by a flexible material component, and the two wing sections are connected by a rigid joint component, and the component is a visualization component tool in MapleSim software;

[0012] Constructing 10 rigid body segment strips of the same width, each rigid body segment strip is connected to each other by a flexible spring damper assembly, thereby defining a flexible wing portion;

[0013] A drive motor assembly is installed at the fuselage and joints as an actuator, and an angle sensor assembly is installed to measure the flapping angle of the wing. A displacement sensor assembly is installed on the flexible wing to measure the vibration amount.

[0014] Collecting the flapping angle of the rigid wing of the bionic flapping wing system , flexible wing flapping angle And the vibration of the 10 rigid body segments on the flexible wing Information is used as the state quantity and output quantity of the bionic flapping-wing system to determine the model equation of the bionic flapping-wing system. The model equation is a second-order linear model equation, which is as follows:

[0015] ,

[0016] In the formula, is the state matrix of the bionic flapping wing system, and are the derivative and second-order derivative of the state matrix, respectively. Represents the dynamic information of the bionic flapping wing system, Represents the control input of the bionic flapping wing system.

[0017] Flexible systems are usually described by infinite-dimensional partial differential equations, which have large computational complexity and complicated derivation. The above method realizes the description of flexible systems as finite-dimensional ordinary differential equations without determining specific dynamic information, and only needs to obtain the form of the model equation.

[0018] In the bionic flapping wing control algorithm execution module,

[0019] The expected trajectory of a given bionic flapping wing flapping angle (including rigid wing flapping angle and flexible wing flapping angle) , according to the state matrix of the bionic flapping wing system and expected trajectory By subtracting, we can get the state error and the derivative of the error , design control execution algorithm to achieve trajectory tracking and vibration suppression;

[0020] The specific implementation is as follows:

[0021] ,

[0022] In the formula, represents the model-based control algorithm, represents the control gain; and The expected trajectory The derivative and second-order derivative of Dynamic information representing the biomimetic flapping wing control system (usually not accurately available);

[0023] Construct a forward three-layer executive neural network to approximate the dynamic information of the bionic flapping wing system that cannot be accurately obtained The structure of the forward three-layer execution neural network includes a single input layer, a single hidden layer and a single output layer. The state error and error derivative of the bionic flapping wing system are input into the input layer as test data. The input data is mapped to the hidden layer space based on the Gaussian function, and the output of the hidden layer is obtained as follows: , the corresponding weight is , and get the predicted output ;

[0024] According to the output of the bionic flapping wing control algorithm evaluation module, the weights of the forward three-layer execution neural network are adaptively adjusted to obtain a new control execution algorithm;

[0025] The adaptive law for adjusting the weight is: , represents the learning rate, is a constant gain, Represent the hidden layer output and predicted output of the i-th neural network respectively, represents the prediction output of the hidden layer of the RBF neural network for the i-th time;

[0026] When the output of the bionic flapping wing control algorithm evaluation module satisfies the minimum value of the objective function, the output of the hidden layer of the constructed neural network is , the predicted output is (i.e. the final output data) is the approximate value of the dynamic information ;

[0027] The control execution algorithm is obtained as follows:

[0028] ,

[0029] In the formula, represents the control execution algorithm, represents the control gain, An approximation representing kinetic information.

[0030] The bionic flapping wing control algorithm evaluation module designs the control algorithm objective function as follows:

[0031] ,

[0032] In the formula, are constants, is the objective function, and the smaller its value is, the better the bionic flapping wing control system is.

[0033] A radial basis evaluation neural network is constructed to approximate the objective function of the control algorithm with arbitrarily small accuracy. The state error and error derivative of the bionic flapping wing system are input into the input layer as test data. The input data is mapped to the hidden layer space based on the Gaussian function, and the output of the hidden layer is obtained as follows: , predict output ;

[0034] The update rate of the neural network weights is designed to be ,in is the learning rate, is a constant gain, Respectively represent the hidden layer output and predicted output of the i-th neural network;

[0035] The output data after n updates is , which is the approximate value of the objective function.

[0036] A bionic flapping wing control method for achieving trajectory tracking and vibration suppression of the present invention comprises the following steps:

[0037] Step 1: Establish the bionic flapping wing model equation:

[0038] ,

[0039] In the formula, is the state matrix of the bionic flapping wing system, Represents the dynamic information of the bionic flapping wing system; represents control input;

[0040] Step 2: Based on the bionic flapping wing model in step 1, a model-based control algorithm is designed to achieve trajectory tracking and vibration suppression, as follows:

[0041] ,

[0042] In the formula, is the system state error, is the derivative of the error, represents the model-based control algorithm, represents the control gain; Represents the dynamic information of the bionic flapping wing system;

[0043] Step 3: Construct a forward three-layer executive neural network to approximate the dynamic information of the bionic flapping wing system that cannot be accurately obtained above The structure of the forward three-layer execution neural network includes a single input layer, a single hidden layer and a single output layer. The state error and the error derivative are input into the input layer as test data. The input data is mapped to the hidden layer space based on the Gaussian function, and the output of the hidden layer is obtained as follows: , the corresponding weight is , and get the predicted output ;

[0044] Design the control algorithm objective function as follows:

[0045] ,

[0046] In the formula, are constants, is the objective function, and the smaller its value is, the better the bionic flapping wing control system is.

[0047] Construct a radial basis evaluation neural network to approximate the control algorithm objective function with arbitrarily small accuracy; input the system state error and error derivative as test data into the input layer, map the input data to the hidden layer space based on the Gaussian function, and obtain the output of the hidden layer as , predict output ;

[0048] The update rate of the neural network weights is designed to be ,in is the learning rate, is a constant gain, Respectively represent the hidden layer output and predicted output of the i-th neural network;

[0049] The output data after n updates is , which is the approximate value of the objective function;

[0050] According to the value of the above objective function, the weights of the forward three-layer execution neural network are adjusted according to the weight adjustment adaptive law. The weight adjustment adaptive law is: , represents the learning rate, is a constant gain, Represent the hidden layer output and predicted output of the i-th neural network respectively, represents the prediction output of the hidden layer of the RBF neural network for the i-th time;

[0051] When the objective function value is minimized, the output of the constructed neural network hidden layer is , the predicted output is (i.e. the final output data) is the approximate value of the dynamic information ;

[0052] Thus, a bionic flapping wing control execution algorithm for trajectory tracking and vibration suppression is obtained, which is as follows:

[0053] ,

[0054] In the formula, represents the control execution algorithm, represents the control gain, Approximation value representing kinetic information;

[0055] Step 4: Given the desired flapping angle trajectory, parameters The data measured by the sensor in the virtual bionic flapping wing model in step 1 is subtracted from the given expected trajectory; Obtained by executing the output of the neural network in three forward layers; The pre-training is obtained by co-simulating a bionic flapping wing system by MapleSim and MATLAB software, and performing simulation training by using the control execution algorithm in step 3 to ensure that the flapping angle can accurately track the expected trajectory and that the terminal vibration amount converges to zero;

[0056] Step 5: Substitute the real-time parameters obtained in step 4 into the bionic flapping wing control execution algorithm of step 3 to obtain the control input at the current moment, and then apply force to the bionic flapping wing system according to the control input to make the bionic flapping wing flap according to the desired flapping angle trajectory while suppressing the vibration of the bionic flapping wing.

[0057] The beneficial effects of the present invention compared with the prior art are as follows:

[0058] (1) The present invention discloses a bionic flapping-wing control system and method for realizing trajectory tracking and vibration suppression, and relates to the field of automatic control technology. The present invention establishes a model equation of a bionic flapping-wing control system by building a virtual bionic flapping-wing model and performing analysis, and builds a visualized virtual bionic flapping-wing model with high simulation degree and low modeling difficulty, thereby overcoming the problems of large amount of calculation, cumbersome derivation process and poor scalability of traditional modeling methods, and can be used for model analysis and control strategy verification, and obtains the model equation of a bionic flapping-wing system.

[0059] (2) The present invention designs a bionic flapping-wing control algorithm in combination with the model equation, adopts a neural network to approximate the uncertain information of the system, constructs a bionic flapping-wing control algorithm execution module, and then designs a bionic flapping-wing control algorithm evaluation module to evaluate the control algorithm, thereby continuously adjusting the bionic flapping-wing control algorithm online, and at the same time simulating the motion process of the system, adjusting the control parameters, so that the bionic flapping-wing system can accurately track the desired trajectory, while suppressing the adverse elastic vibration generated during the motion process, thereby realizing online learning and optimal control. BRIEF DESCRIPTION OF THE DRAWINGS

[0060] Figure 1 A block diagram of the bionic flapping-wing control system for achieving trajectory tracking and vibration suppression according to the present invention;

[0061] Figure 2 A flow chart of a bionic flapping wing control method for achieving trajectory tracking and vibration suppression according to the present invention;

[0062] Figure 3 is a schematic diagram of a virtual bionic flapping wing model;

[0063] Figure 4 It is a graph of chord versus span;

[0064] Figure 5 Simulation diagram of bionic flapping wing flapping angle trajectory and tracking error under traditional PD control;

[0065] Figure 6 Simulation diagram of the terminal vibration of the bionic flapping wing system under traditional PD control;

[0066] Figure 7 Simulation diagram of the flapping angle trajectory and tracking error of the bionic flapping wing under the proposed control;

[0067] Figure 8 Simulation diagram of the terminal vibration of the bionic flapping-wing system under the proposed control. DETAILED DESCRIPTION

[0068] The present invention is described in detail below with reference to the accompanying drawings and embodiments.

[0069] like Figure 1 As shown, a bionic flapping-wing control system for realizing trajectory tracking and vibration suppression of the present invention comprises: a virtual bionic flapping-wing model module, a bionic flapping-wing control algorithm execution module and a bionic flapping-wing control algorithm evaluation module;

[0070] The virtual bionic flapping wing model module, inspired by the flying mode of seagulls in bionics, constructs a virtual bionic flapping wing model consisting of rigid joints, rigid wings and flexible wings, and determines the bionic flapping wing control system as a second-order linear model equation based on the virtual bionic flapping wing model;

[0071] The virtual bionic flapping wing model module is specifically implemented as follows: a wing portion of the bionic flapping wing close to the fuselage is constructed by a rigid material component, and a wing portion of the bionic flapping wing away from the fuselage is constructed by a flexible material component, and the two wing sections are connected by a rigid joint component, and the component is a visualization component tool in MapleSim software;

[0072] Constructing 10 rigid body segment strips of the same width, each rigid body segment strip is connected to each other by a flexible spring damper assembly, thereby defining a flexible wing portion;

[0073] A drive motor assembly is installed at the fuselage and joints as an actuator, and an angle sensor assembly is installed to measure the flapping angle of the wing. A displacement sensor assembly is installed on the flexible wing to measure the vibration amount.

[0074] Collecting the flapping angle of the rigid wing of the bionic flapping wing system , flexible wing flapping angle And the vibration of the 10 rigid body segments on the flexible wing Information is used as the state quantity and output quantity of the bionic flapping-wing system to determine the model equation of the bionic flapping-wing system. The model equation is a second-order linear model equation, as follows:

[0075] ,

[0076] In the formula, is the state matrix of the bionic flapping wing system, Represents the dynamic information of the bionic flapping wing system; Represents the control input of the bionic flapping wing system.

[0077] Flexible systems are usually described by infinite-dimensional partial differential equations, which have large computational complexity and complicated derivation. The above method realizes the description of flexible systems as finite-dimensional ordinary differential equations without determining specific dynamic information, and only needs to obtain the form of the model equation.

[0078] The bionic flapping wing control algorithm execution module constructs a forward three-layer execution neural network based on the second-order linear model equation of the bionic flapping wing control system determined by the virtual bionic flapping wing model module to approximate the unknown dynamic information in the model equation, and then gives the desired flapping angle trajectory, designs the control execution algorithm to control the flapping angle and vibration of the bionic flapping wing, and realizes trajectory tracking and vibration suppression;

[0079] In the bionic flapping wing control algorithm execution module,

[0080] The expected trajectory of a given bionic flapping wing flapping angle (including rigid wing flapping angle and flexible wing flapping angle) , according to the state matrix of the bionic flapping wing system and expected trajectory By subtracting, we can get the state error and the derivative of the error Design control execution algorithms to achieve trajectory tracking and vibration suppression;

[0081] The specific implementation is as follows:

[0082] ,

[0083] In the formula, represents the model-based control algorithm, represents the control gain; Represents the dynamic information of the bionic flapping wing system;

[0084] Construct a forward three-layer executive neural network to approximate the dynamic information of the bionic flapping wing system that cannot be accurately obtained The structure of the forward three-layer execution neural network includes a single input layer, a single hidden layer and a single output layer. The state error and error derivative of the bionic flapping wing system are input into the input layer as test data. The input data is mapped to the hidden layer space based on the Gaussian function, and the output of the hidden layer is obtained as follows: , the corresponding weight is , and get the predicted output ;

[0085] According to the output of the bionic flapping wing control algorithm evaluation module, the weights of the forward three-layer execution neural network are adaptively adjusted to obtain a new control execution algorithm;

[0086] The adaptive law for adjusting the weight is: , represents the learning rate, Represent the hidden layer output and predicted output of the i-th neural network respectively, represents the prediction output of the hidden layer of the RBF neural network for the i-th time;

[0087] When the output of the bionic flapping wing control algorithm evaluation module satisfies the minimum value of the objective function, the output of the hidden layer of the constructed neural network is , the predicted output is (i.e. the final output data) is the approximate value of the dynamic information ;

[0088] The control execution algorithm is obtained as follows:

[0089] ,

[0090] In the formula, represents the control execution algorithm, represents the control gain, An approximation representing kinetic information.

[0091] The bionic flapping wing control algorithm evaluation module designs the control algorithm objective function based on the control objectives of trajectory tracking and vibration suppression, constructs a radial basis evaluation neural network, and approximates the control algorithm objective function;

[0092] The bionic flapping wing control algorithm evaluation module designs the control algorithm objective function as follows:

[0093] ,

[0094] In the formula, the smaller the value of the objective function is, the better the bionic flapping wing control system is.

[0095] A radial basis evaluation neural network is constructed to approximate the objective function of the control algorithm with arbitrarily small accuracy. The state error and error derivative of the bionic flapping wing system are input into the input layer as test data. The input data is mapped to the hidden layer space based on the Gaussian function, and the output of the hidden layer is obtained as follows: , predict output ;

[0096] The update rate of the neural network weights is designed to be ,in is the learning rate, Respectively represent the hidden layer output and predicted output of the i-th neural network;

[0097] The output data after n updates is , which is the approximate value of the objective function.

[0098] Based on the virtual bionic flapping-wing model module, a bionic flapping-wing control algorithm execution module can be designed. The bionic flapping-wing control algorithm evaluation module can evaluate whether the control execution algorithm of the current bionic flapping-wing control algorithm execution module can reach the optimal state, and feed back the evaluation result to the bionic flapping-wing control algorithm execution module. The bionic flapping-wing control algorithm execution module will adjust the control execution algorithm according to the evaluation result. The control execution algorithm output by the bionic flapping-wing control algorithm execution module acts on the virtual bionic flapping-wing model module. According to the real-time output of the virtual bionic flapping-wing model module, the effectiveness of the control execution algorithm of the current bionic flapping-wing control algorithm execution module can be verified.

[0099] like Figure 2 As shown, a bionic flapping wing control method for realizing trajectory tracking and vibration suppression of the present invention is implemented as follows:

[0100] Step 1: Based on the inspiration of the seagull flying mode in bionics, a virtual bionic flapping wing model consisting of rigid joints, rigid wings and flexible wings is constructed, and the bionic flapping wing control system is determined by the virtual bionic flapping wing model as a second-order linear model equation:

[0101] ,

[0102] In the formula, is the system state matrix, Represents the dynamic information of the bionic flapping wing system, represents control input;

[0103] Step 2: Based on the above model, design a model-based control algorithm to achieve trajectory tracking and vibration suppression, as follows:

[0104] ,

[0105] In the formula, is the system state error, is the derivative of the error, represents the model-based control algorithm, represents the control gain; Represents the dynamic information of the biomimetic flapping wing control system, which is usually not accurately available;

[0106] Step 3: Consider the dynamic information in the above model-based control algorithm Usually it is impossible to obtain it accurately. A forward three-layer executive neural network is constructed to approximate the above dynamic information that cannot be obtained accurately. Its structure includes a single input layer, a single hidden layer and a single output layer. The state error and error derivative of the bionic flapping wing control system are input into the input layer as test data. The input data is mapped to the hidden layer space based on the Gaussian function, and the output of the hidden layer is obtained as follows: , the corresponding weight is , and get the predicted output ;

[0107] Design the control algorithm objective function as follows:

[0108] ,

[0109] In the formula, the smaller the objective function value is, the smaller the error of the bionic flapping-wing control system is, and the control input is as small as possible. This is the optimal state of the bionic flapping-wing control system.

[0110] Construct a radial basis evaluation neural network to approximate the control algorithm objective function with arbitrarily small accuracy; input the system state error and error derivative as test data into the input layer, map the input data to the hidden layer space based on the Gaussian function, and obtain the output of the hidden layer as , predict output ;

[0111] The update rate of the neural network weights is designed to be ,in is the learning rate, Respectively represent the hidden layer output and predicted output of the i-th neural network;

[0112] The output data after n updates is , which is the approximate value of the objective function.

[0113] According to the value of the above objective function, the weights of the forward three-layer execution neural network are adjusted according to the weight adjustment adaptive law. The weight adjustment adaptive law is: , represents the learning rate, Respectively represent the hidden layer output and predicted output of the i-th neural network;

[0114] When the objective function value is minimized, the output of the constructed neural network hidden layer is , the predicted output is (i.e. the final output data) is the approximate value of the dynamic information ;

[0115] Thus, a bionic flapping wing control method for achieving trajectory tracking and vibration suppression is obtained, which is as follows:

[0116] ,

[0117] In the formula, represents the control execution algorithm, represents the control gain, An approximation representing kinetic information.

[0118] like Figure 3 As shown, a virtual bionic flapping wing model module is built, which includes three major parts: a rigid joint, a rigid wing and a flexible wing. The wing part of the bionic flapping wing close to the fuselage is constructed by a rigid material component, and the wing part of the bionic flapping wing away from the fuselage is constructed by a flexible material component. The two wing sections are connected by a rigid joint component.

[0119] The rigid wing is constructed directly from rigid material components, and the angle between the rigid wing and the horizontal plane is the flapping angle of the rigid wing.

[0120] Construct 10 rigid body segments of the same width, and connect each rigid body segment with a flexible spring damper assembly to determine the flexible wing part and the angle between the flexible wing and the horizontal plane. is the flapping angle of the flexible wing. Elastic deformation at for The vibration amount at the flexible wing is Elastic deformation at is the end vibration amount.

[0121] A driving motor assembly is installed at the fuselage and the joints as an actuator, and an angle sensor assembly is installed at the fuselage and the joints to measure the flapping angle of the wing. A displacement sensor assembly is installed on the flexible wing to measure the vibration amount. The assembly is a visualization component tool in the MapleSim software.

[0122] The relationship between chord and wingspan is as follows Figure 4 As shown, according to bionics and statistical data, the length of the rigid wing is set to 18 cm, and the total mass is 288 g, of which the weight of the rigid wing leading edge beam is 201.6 g, and the weight of the rigid wing surface is 86.4 g; the length of the flexible wing is 32 cm, and the total mass is 171 g, of which the weight of the flexible wing leading edge beam is 136.8 g, and the weight of the flexible wing surface is 34.2 g.

[0123] In order to explain the technical content of the present invention in a more organized manner, in the embodiment of the present invention, an implementation verification is carried out on the designed bionic flapping wing control system. In order to simulate the flight of a seagull, the desired flapping angle is set to:

[0124] ,

[0125] Among them, the first 10 seconds simulated flapping flight mode, and the last 5 seconds simulated gliding flight mode.

[0126] Firstly, the traditional proportional differential controller (PD) is applied to the virtual experimental device, and the control gain is set to , . Observe the trajectory tracking and vibration suppression performance of the bionic rigid-flexible coupling flapping wing system, such as Figure 5 and Figure 6 As shown. Figure 5 It can be seen that under traditional PD control, although the bionic rigid-flexible coupling flapping wing system can track the desired trajectory, the peak tracking errors of the rigid wing flapping angle and the flexible wing flapping angle reach 0.58cm and 0.28cm respectively, and the tracking error fluctuates greatly within 1 second after the start of the movement. Figure 6 It can be seen that under traditional PD control, the terminal vibration peak of the bionic rigid-flexible coupled flapping wing system reached 0.13 cm, and it took 1.6 seconds for the terminal vibration to reach a stable range. Oscillation also occurred when switching from flapping wing flight mode to sliding flight mode, and it took 0.8 seconds to reach stability.

[0127] The method of the present invention is applied to the bionic rigid-flexible coupling flapping wing virtual experimental device, and the control gains are set to , and . Observe the trajectory tracking and vibration suppression performance of the bionic rigid-flexible coupling flapping wing system, such as Figure 7 and Figure 8 As shown. Figure 7It can be seen that under the designed vibration control, the bionic rigid-flexible coupling flapping wing system can accurately track the desired trajectory, and the tracking error peaks of the rigid wing flapping angle and the flexible wing flapping angle are reduced to 0.34cm and 0.17cm, respectively, which are reduced by 41.4% and 39.3% compared with traditional PD control. Figure 8 It can be seen that under the designed vibration control, the terminal vibration peak value of the bionic rigid-flexible coupling flapping wing system is reduced to 0.07cm, and it takes 0.3 seconds for the terminal vibration to reach the stable range. When switching from the flapping wing flight mode to the sliding flight mode, the oscillation phenomenon is almost eliminated, and it takes only 0.1 seconds to reach stability. Therefore, it can be concluded that the present invention has better effects in tracking control and vibration suppression.

Claims

1. A bionic flapping wing control system for achieving trajectory tracking and vibration suppression, characterized in that: include: Virtual bionic flapping-wing model module, bionic flapping-wing control algorithm execution module and bionic flapping-wing control algorithm evaluation module; The virtual bionic flapping wing model module constructs a virtual bionic flapping wing model composed of rigid joints, rigid wings and flexible wings, and determines the model equation of the bionic flapping wing control system by the virtual bionic flapping wing model. The model equation is a second-order linear model equation. The bionic flapping wing control algorithm execution module constructs a forward three-layer execution neural network based on the second-order linear model equation to approximate the unknown dynamic information in the model equation. Given the expected trajectory of the flapping angle of the bionic flapping wing, the control execution algorithm is designed to control the flapping angle and vibration of the bionic flapping wing according to the state error of the bionic flapping wing control system to achieve trajectory tracking and vibration suppression. At the same time, the bionic flapping wing control algorithm execution module will adjust the control execution algorithm according to the evaluation results, and apply the adjusted control execution algorithm to the virtual bionic flapping wing model module to obtain the real-time output of the virtual bionic flapping wing model module. The effectiveness of the control execution algorithm of the current bionic flapping wing control algorithm execution module is verified according to the real-time output of the virtual bionic flapping wing model module. The bionic flapping-wing control algorithm evaluation module designs the control algorithm objective function according to the control objectives of trajectory tracking and vibration suppression, constructs a radial basis evaluation neural network, approximates the objective function of the control algorithm, evaluates whether the control algorithm in the current bionic flapping-wing control algorithm execution module can achieve the optimal control algorithm, and feeds back the evaluation results to the bionic flapping-wing control algorithm execution module; A bionic flapping-wing control algorithm execution module is designed based on the virtual bionic flapping-wing model module. The bionic flapping-wing control algorithm evaluation module evaluates whether the control algorithm of the current bionic flapping-wing control algorithm execution module can achieve the optimal control algorithm, and feeds back the evaluation result to the bionic flapping-wing control algorithm execution module. The bionic flapping-wing control algorithm execution module adjusts the control execution algorithm according to the evaluation result. The control execution algorithm output by the bionic flapping-wing control algorithm execution module acts on the virtual bionic flapping-wing model module. The effectiveness of the control execution algorithm of the current bionic flapping-wing control algorithm execution module can be verified according to the real-time output of the virtual bionic flapping-wing model module. The virtual bionic flapping wing model module is specifically implemented as follows: a wing portion of the bionic flapping wing close to the fuselage is constructed by a rigid material component, and a wing portion of the bionic flapping wing away from the fuselage is constructed by a flexible material component, and the two wing sections are connected by a rigid joint component, and the component is a visualization component tool in MapleSim software; Constructing 10 rigid body segment strips of the same width, each rigid body segment strip is connected to each other by a flexible spring damper assembly, thereby defining a flexible wing section; A drive motor assembly is installed at the fuselage and joints as an actuator, and an angle sensor assembly is installed to measure the flapping angle of the wing. A displacement sensor assembly is installed on the flexible wing to measure the vibration amount. Collecting the flapping angle of the rigid wing of the bionic flapping wing system , flexible wing flapping angle And the vibration of the 10 rigid body segments on the flexible wing Information is used as the state quantity and output quantity of the bionic flapping-wing system to determine the model equation of the bionic flapping-wing system. The model equation is a second-order linear model equation, as follows: ， In the formula, is the state matrix of the bionic flapping wing system, and are the derivative and second-order derivative of the state matrix, respectively. Represents the dynamic information of the bionic flapping wing system; represents the control input of the bionic flapping wing system; In the bionic flapping wing control algorithm execution module, Given the flapping angle of the bionic flapping wing, including the expected trajectory of the rigid wing flapping angle and the flexible wing flapping angle , according to the state matrix of the bionic flapping wing system and expected trajectory Subtract and get the state error and the derivative of the error Design control execution algorithms to achieve trajectory tracking and vibration suppression; The specific implementation is as follows: ， In the formula, represents the model-based control algorithm, represents the control gain; and The expected trajectory The derivative and second-order derivative of Represents the dynamic information of the bionic flapping wing system; Construct a forward three-layer executive neural network to approximate the dynamic information of the bionic flapping wing system that cannot be accurately obtained The structure of the forward three-layer execution neural network includes a single input layer, a single hidden layer and a single output layer. The state error and error derivative of the bionic flapping wing system are input into the input layer as test data. The input data is mapped to the hidden layer space based on the Gaussian function, and the output of the hidden layer is obtained as follows: , the corresponding weight is , and get the predicted output ; According to the output of the bionic flapping wing control algorithm evaluation module, the weights of the forward three-layer execution neural network are adaptively adjusted to obtain a new control execution algorithm; The adaptive law for adjusting the weight is: , represents the learning rate, is a constant gain, Represent the hidden layer output and predicted output of the i-th neural network respectively, represents the prediction output of the hidden layer of the RBF neural network for the i-th time; When the output of the bionic flapping wing control algorithm evaluation module satisfies the minimum value of the objective function, the output of the hidden layer of the constructed neural network is , the predicted output is This is the approximation of the dynamic information ; The control execution algorithm is obtained as follows: ， In the formula, represents the control execution algorithm, represents the control gain, Approximation value representing kinetic information; The bionic flapping wing control algorithm evaluation module designs the control algorithm objective function as follows: ， In the formula, are constants, is the objective function, and the smaller its value is, the better the bionic flapping wing control system is. A radial basis evaluation neural network is constructed to approximate the objective function of the control algorithm with arbitrarily small accuracy. The state error and error derivative of the bionic flapping wing system are input into the input layer as test data. The input data is mapped to the hidden layer space based on the Gaussian function, and the output of the hidden layer is obtained as follows: , predict output ; The update rate of the neural network weights is designed to be ,in is the learning rate, is a constant gain, Respectively represent the hidden layer output and predicted output of the i-th neural network; The output data after n updates is , which is the approximate value of the objective function.

2. A bionic flapping wing control method for achieving trajectory tracking and vibration suppression, characterized in that: The following steps are involved: Step 1: Establish the bionic flapping wing model equation: ， In the formula, is the state matrix of the bionic flapping wing system, Represents the dynamic information of the bionic flapping wing system; represents control input; Step 2: Based on the bionic flapping wing model in step 1, a model-based control algorithm is designed to achieve trajectory tracking and vibration suppression, as follows: ， In the formula, is the system state error, is the derivative of the error, represents the model-based control algorithm, represents the control gain; Represents the dynamic information of the bionic flapping wing control system; Step 3: Construct a forward three-layer execution neural network to approximate the dynamic information of the bionic flapping wing system that cannot be accurately obtained above The structure of the forward three-layer execution neural network includes a single input layer, a single hidden layer and a single output layer. The state error and the error derivative are input into the input layer as test data. The input data is mapped to the hidden layer space based on the Gaussian function, and the output of the hidden layer is obtained as follows: , the corresponding weight is , and get the predicted output ; Design the control algorithm objective function as follows: ， In the formula, the smaller the value of the objective function is, the better the bionic flapping wing control system is. Construct a radial basis evaluation neural network to approximate the control algorithm objective function with arbitrarily small accuracy; input the system state error and error derivative as test data into the input layer, map the input data to the hidden layer space based on the Gaussian function, and obtain the output of the hidden layer as , predict output ; The update rate of the neural network weights is designed to be ,in is the learning rate, is a constant gain, Respectively represent the hidden layer output and predicted output of the i-th neural network; The output data after n updates is , which is the approximate value of the objective function; According to the value of the above objective function, the weights of the forward three-layer execution neural network are adjusted according to the weight adjustment adaptive law. The weight adjustment adaptive law is: , represents the learning rate, is a constant gain, Represent the hidden layer output and predicted output of the i-th neural network respectively, represents the prediction output of the hidden layer of the RBF neural network for the i-th time; When the objective function value is minimized, the output of the constructed neural network hidden layer is , the predicted output is This is the approximation of the dynamic information ; Thus, a bionic flapping wing control execution algorithm for trajectory tracking and vibration suppression is obtained, which is as follows: ， In the formula, represents the control execution algorithm, represents the control gain, Approximation value representing kinetic information; Step 4: Given the desired flapping angle trajectory, parameters The data measured by the sensor in the virtual bionic flapping wing model in step 1 is subtracted from the given expected trajectory; The output of the forward three-layer neural network is obtained; real-time parameters The pre-training is obtained by co-simulating a bionic flapping wing system by MapleSim and MATLAB software, and performing simulation training by using the control execution algorithm in step 3 to ensure that the flapping angle can accurately track the expected trajectory and that the terminal vibration amount converges to zero; Step 5: Substitute the real-time parameters obtained in step 4 into the bionic flapping wing control execution algorithm of step 3 to obtain the control input at the current moment, and then apply force to the bionic flapping wing system according to the control input to make the bionic flapping wing flap according to the desired flapping angle trajectory while suppressing the vibration of the bionic flapping wing.

Images