A parameter correction method and system of a fixed-wing unmanned aerial vehicle semi-physical simulation system

Abstract

The application belongs to the technical field of unmanned aerial vehicle simulation, and particularly relates to a parameter correction method and system of a fixed-wing unmanned aerial vehicle semi-physical simulation system, which comprises the following steps: calculating the state information of a flight vehicle through a kinematics model and a dynamics model and sending the state information to a unmanned aerial vehicle guidance control system, the unmanned aerial vehicle guidance control system outputting rudder instructions according to the received state information of the flight vehicle to control the flight vehicle; constructing a deep learning network model, the input of the deep learning network model being the state information of the flight vehicle, and the output of the deep learning network model being the correction amount of aerodynamic coefficients; updating the kinematics model and the dynamics model by using the corrected aerodynamic coefficients, recalculating the state information of the flight vehicle and transmitting the state information to the unmanned aerial vehicle guidance control system to control the flight vehicle. The application improves the modeling accuracy of the 6DOF model, and further makes the semi-physical simulation system better simulate the real flight condition.

Description

Technical Field

[0001] This invention relates to the field of unmanned aerial vehicle (UAV) simulation technology, and specifically to a parameter correction method and system for a fixed-wing UAV hardware-in-the-loop simulation system. Background Technology

[0002] With the rapid development of aerospace technology, the design and development of aircraft have become increasingly complex. During the development process, it is necessary to fully verify and optimize the aircraft's performance, the stability of its control system, and its reliability before actual flight testing. However, conducting full-scale flight tests directly is costly, risky, and difficult to modify and adjust if problems arise. Traditional pure digital simulation methods have limitations in verifying aircraft performance and control systems, failing to fully and realistically reflect various situations encountered in actual flight. To more effectively design, develop, and test aircraft, hardware-in-the-loop (HIL) simulation technology has emerged.

[0003] A hardware-in-the-loop (HIL) simulation system mainly consists of a simulator (real-time simulation system), a guidance and control system, and a ground control station. The real-time simulation system comprises hardware and software: the hardware is an industrial control computer, and the software includes a real-time operating system, UAV kinematic and dynamic models, and a serial communication module. Traditional HIL simulation systems only use aerodynamic data for kinematic and dynamic modeling, without using actual flight data to refine the model.

[0004] The data used for kinematic and dynamic modeling of UAVs comes from software calculations or wind-blown experiments, which differ from the actual forces acting on the UAV during its motion. The forces acting on a UAV during flight change continuously with its motion (speed, altitude, etc.) and ambient temperature, but calculations or wind-blown experiments only collect discrete data points, leading to inaccuracies. Therefore, the resulting dynamic and kinematic model (6DOF model) deviates from reality, affecting the accuracy of the UAV hardware-in-the-loop simulation system. Summary of the Invention

[0005] Purpose of the invention: The purpose of this invention is to address the shortcomings of existing technologies by providing a parameter correction method and system for a fixed-wing unmanned aerial vehicle (UAV) hardware-in-the-loop simulation system, thereby improving the modeling accuracy of the 6DOF model and enabling the hardware-in-the-loop simulation system to better simulate real flight conditions.

[0006] Technical solution: The parameter correction method for the fixed-wing unmanned aerial vehicle (UAV) hardware-in-the-loop simulation system of the present invention includes the following steps:

[0007] S1: Construct a hardware-in-the-loop simulation system, including a real-time simulation system, an UAV guidance and control system, a UAV ground operation station, and a serial communication module. The real-time simulation system interacts with the UAV guidance and control system via the serial communication module, and the UAV guidance and control system interacts with the UAV ground operation station via a data link. The real-time simulation system includes a real-time operating system and real-time simulation software. The real-time operating system is mounted on a hardware platform, and the real-time simulation software includes a hardware driver module and an aircraft simulation model. The aircraft simulation model includes an aerodynamic model, a dynamic model, a kinematic model, and a sensor model. The dynamic model and kinematic model are used to calculate the aircraft's state information and, according to the data protocol of the real sensors, package the state information and send it to the UAV guidance and control system via the serial communication module. The UAV guidance and control system outputs rudder commands to control the aircraft based on the received aircraft state information.

[0008] S2: Construct a deep learning network model. The input of the deep learning network model is the state information of the aircraft, and the output of the deep learning network model is the correction amount of the aerodynamic coefficient. The gradient descent method is used to train the network model so that the output of the network model approximates the difference between the real aerodynamic coefficient and the original aerodynamic coefficient.

[0009] S3: The kinematic and dynamic models are updated using the corrected aerodynamic coefficients. The updated kinematic and dynamic models are then used to calculate the aircraft's state information and transmit it to the UAV guidance and control system for aircraft control.

[0010] To further improve the above technical solution, the dynamic model and kinematic model calculate the state information of the aircraft through the center of mass dynamic equation, center of mass kinematic equation, rotational dynamic equation, rotational kinematic equation, force and torque equations acting on the aircraft, and latitude and longitude calculation equation. The state information includes position (x, y, z), attitude (θ, ψ, φ), velocity (u, v, w), angular velocity (p, q, r), and acceleration (ax, ay, az).

[0011] Furthermore, the deep learning network model adopts a fully connected neural network, including a first channel (channel1) and a second channel (channel2). The first channel (channel1) consists of one input layer and six hidden layers, with the hidden layers having 48-dimensional state variables. The second channel (channel2) consists of one input layer and three hidden layers, with the hidden layers having 16-dimensional state variables.

[0012] Furthermore, the input to the deep learning network model is the UAV flight state information, including position (x, y, z), attitude (θ, ψ, φ), velocity (u, v, w), angular velocity (p, q, r), and acceleration (ax, ay, az), where θ, ψ, and φ are the pitch, yaw, and roll angles of the x, y, and z axes in the body coordinate system; u, v, and w represent the velocity components of the x, y, and z axes in the body coordinate system; p, q, and r are the angular velocity components of the x, y, and z axes in the body coordinate system; and ax, ay, and az are the acceleration components of the x, y, and z axes in the body coordinate system. The output of the deep learning network model is the correction amount for aerodynamic coefficients, including lift coefficient, drag coefficient, side force coefficient, pitch moment coefficient, yaw moment coefficient, and roll moment coefficient.

[0013] Furthermore, for different aerodynamic coefficients, the vector dimension of the input to the deep learning network model is different:

[0014] When correcting the lift coefficient, the first channel input is 12-dimensional state information, namely x, y, θ, ψ, φ, u, v, p, q, r, ax, ay, and the second channel input is 3-dimensional state information, namely z, w, az;

[0015] When correcting the drag coefficient, the first channel input is 12-dimensional state information, namely x, z, θ, ψ, φ, u, w, p, q, r, ax, and az; the second channel input is 3-dimensional state information, namely y, v, and ay.

[0016] When correcting the lateral force coefficient, the first channel input is 12-dimensional state information, namely y, z, θ, ψ, φ, v, w, p, q, r, ay, az, and the second channel input is 3-dimensional state information, namely x, u, ax;

[0017] When correcting the pitch moment coefficient, the first channel input is 13-dimensional state information, namely x, y, z, ψ, φ, u, v, w, p, r, ax, ay, az, and the second channel input is 2-dimensional state information, namely θ, q.

[0018] When correcting the roll moment coefficient and yaw moment coefficient, the first channel input is 11-dimensional state information, namely x, y, z, θ, u, v, w, q, ax, ay, az, and the second channel input is 4-dimensional state information, namely φ, ψ, p, r.

[0019] Furthermore, the corrected aerodynamic coefficient consists of two parts: the first part is the theoretical calculation value or wind tunnel experimental value; the second part is the output value of the deep learning network model.

[0020] Furthermore, the deep learning network model takes the state variables of the aircraft at the previous moment as input, performs iterative calculations, and outputs updated correction coefficients; the kinematic and dynamic models calculated using the updated correction coefficients are in a dynamic update process.

[0021] A hardware-in-the-loop simulation system for implementing the above method includes:

[0022] A real-time simulation system includes a real-time operating system and real-time simulation software. The real-time operating system is mounted on a hardware platform, and the real-time simulation software includes a hardware driver module and an aircraft simulation model. The aircraft simulation model includes an aerodynamic model, a dynamic model, a kinematic model, and a sensor model. The aerodynamic data model is used to provide basic aerodynamic parameters, the sensor model is used to generate data that conforms to the output format of real sensors, and the dynamic model and kinematic model are used to calculate the state information of the aircraft.

[0023] The deep learning network model takes as input the aircraft's state information, including position (x, y, z), attitude (θ, ψ, φ), velocity (u, v, w), angular velocity (p, q, r), and acceleration (ax, ay, az). The output of the deep learning network model is the correction amount for aerodynamic coefficients, including lift coefficient, drag coefficient, side force coefficient, pitch moment coefficient, yaw moment coefficient, and roll moment coefficient.

[0024] The UAV guidance and control system is used to obtain the state information of the aircraft by calculating the corrected aerodynamic coefficients from the dynamic model and kinematic model, and then control the aircraft.

[0025] Beneficial effects: Compared with the prior art, the advantages of this invention are as follows: This invention proposes a deep learning-based aerodynamic parameter correction method and designs a method for flight data processing and network training using flight data. It corrects the lift coefficient, drag coefficient, side force coefficient, pitch moment coefficient, yaw moment coefficient, and roll moment coefficient of the UAV, respectively, thereby improving the modeling accuracy of the kinematic and dynamic models. This enables the hardware-in-the-loop simulation system to better simulate real flight conditions, helps design and optimize the aircraft's control system, and allows for rapid testing and evaluation of different design schemes and control strategies. Attached Figure Description

[0026] Figure 1 This is a schematic diagram of the hardware-in-the-loop simulation system in this invention;

[0027] Figure 2 This is a network diagram of the deep learning model of the present invention. Detailed Implementation

[0028] The technical solution of the present invention will be described in detail below with reference to the accompanying drawings, but the scope of protection of the present invention is not limited to the embodiments described.

[0029] Example 1: This invention proposes a parameter correction method based on deep learning. By using deep learning networks and data from actual flight records, the lift coefficient, drag coefficient, side force coefficient, pitch moment coefficient, yaw moment coefficient, and roll moment coefficient of the UAV are corrected respectively, thereby improving the modeling accuracy of the 6DOF model and enabling the hardware-in-the-loop simulation system to better simulate real flight conditions.

[0030] I. Building a hardware-in-the-loop simulation system

[0031] like Figure 1 As shown, the hardware-in-the-loop simulation system mainly includes a real-time simulation system, a UAV guidance and control system, and a UAV ground operation station. Its system components include: a real-time simulation system, a UAV guidance and control system, a UAV ground operation station, and a serial communication module. The real-time simulation system comprises hardware and software; the hardware is an industrial control computer, and the software is a real-time operating system and real-time simulation software. The real-time simulation software includes aerodynamic models, dynamic models, kinematic models, and sensor models. Among these modules, the accuracy of the UAV dynamics and kinematic models directly determines the accuracy of the hardware-in-the-loop simulation system.

[0032] The first step is to establish kinematic and dynamic models:

[0033] (1) Equations of motion of the center of mass of the aircraft

[0034] The following formula is used to calculate the aircraft velocity projection in the computer body axis. , , .

[0035]

[0036] in, This is the projection of the aircraft velocity onto the body coordinate system. This is the projection of the aircraft's angular velocity onto the body coordinate system. This is the projection of the net external force acting on the aircraft in the velocity coordinate system; The mass of the aircraft.

[0037] (2) Kinematic equations of the center of mass of the aircraft

[0038] The velocity of the aircraft in the ground coordinate system is calculated based on the transformation matrix from the body coordinate system to the ground coordinate system. , , And positions x, y, z:

[0039]

[0040] in, To represent the body coordinate system The roll angle, pitch angle, and yaw angle of the shaft.

[0041] (3) Rotational dynamics equations of the aircraft

[0042]

[0043] in, The moments of inertia are the moments of inertia on each axis of the aircraft's body coordinate system; This is the projection of the resultant external torque of the aircraft onto the body coordinate system.

[0044] (4) Equations of rotational kinematics of the aircraft

[0045] To describe the flight attitude of an aircraft in atmospheric space, the equations for the rigid body's rotation around its center of mass need to be derived in a ground coordinate system. Established in the middle. The three angles describing its attitude are pitch angles. Yaw angle Roll angle Angular velocity is pitch angular velocity yaw rate Roll angular velocity .

[0046]

[0047] (5) Equations of forces and moments acting on the aircraft

[0048] The forces acting on an aircraft body are divided into aerodynamic forces, thrust, and gravity:

[0049]

[0050] Projection of aerodynamic forces onto the three axes of the aircraft's body coordinate system:

[0051]

[0052] in, This is the projection of aerodynamic forces onto the aircraft's body coordinate system; These are drag, lift, and lateral force acting on the aircraft, respectively. These are the angle of attack and sideslip angle of the aircraft.

[0053] Projection of gravity onto the three axes of the aircraft's body coordinate system:

[0054]

[0055] in, This is the projection of gravity onto the aircraft's body coordinate system; This refers to the gravity of the aircraft.

[0056] Three aerodynamic forces defined in the velocity coordinate system The calculation formula is:

[0057]

[0058] in, For dynamic pressure; Wing area; These are three aerodynamic coefficients.

[0059] Three aerodynamic moments defined in the body coordinate system The calculation formula is

[0060]

[0061] in, The average aerodynamic chord length; For wingspan; These are three aerodynamic moment coefficients.

[0062] The aerodynamic forces and torques acting on the aircraft are calculated using the formulas above. The force and torque coefficients are obtained by two-dimensional or three-dimensional interpolation of the original aerodynamic data.

[0063] The aerodynamic coefficient consists of the aerodynamic coefficient generated by the wing and the aerodynamic coefficient of the control surfaces:

[0064]

[0065] In the formula, the subscript Indicates wing, Indicates elevator, Indicates aileron, Indicates the rudder.

[0066] The aerodynamic moment coefficient is composed of the moment generated by the wing, the moment coefficient generated by the control surfaces, and the damping moment coefficient.

[0067]

[0068] (6) Latitude and longitude calculation equations

[0069] Since the Earth is a two-axis rotating ellipsoid, the distance from any point on the Earth's surface to the Earth's center is related to the latitude of that point at the Earth's center.

[0070] Earth's latitude The calculation formula is as follows:

[0071]

[0072] longitude The calculation formula is as follows:

[0073]

[0074] in, These are the position coordinates of the aircraft in the ground coordinate system; These are the coordinate components of the origin of the ground coordinate system in the geocentric coordinate system. It represents the projection components of the Earth's rotational angular velocity vector onto the three axes of the corresponding coordinate system (usually the ground system); Let be the modulus of the Earth's rotational angular velocity;

[0075] Real-time altitude on the flight path is ,in, The distance from any point on the flight path to the Earth's center. The geocentric radius of the real-time waypoint; The semi-major axis of the Earth's ellipsoid; It is the minor semi-axis of the Earth's ellipsoid.

[0076] (7) Equations for calculating aircraft speed, angle of attack, and sideslip angle

[0077] The formula for calculating aircraft speed is:

[0078]

[0079] The formulas for calculating the angle of attack and sideslip angle of an aircraft are:

[0080]

[0081]

[0082] (8) The position, attitude, speed, acceleration and other state information of the aircraft calculated according to the equations provided in (1) to (7) are packaged according to the data protocol of the real sensor used in the flight control system. Finally, the packaged data is sent to the UAV guidance and control system through the serial port, and the rudder command sent by the UAV guidance and control system is received through the serial port. The force and torque equations acting on the aircraft in (5) are input for calculation to complete the closed-loop control.

[0083] II. Optimization and Correction of Aerodynamic Parameters

[0084] Both aerodynamic parameters obtained from CFD calculations and wind tunnel experiments contain errors compared to their true values, which can lead to inaccurate simulations of actual flight conditions in the established kinematic and dynamic models. Therefore, it is necessary to optimize and correct the aerodynamic parameters using actual flight data. Since the correction methods for lift coefficient, drag coefficient, side force coefficient, pitch moment coefficient, yaw moment coefficient, and roll moment coefficient are consistent, we will focus on explaining the correction for the lift coefficient; the correction methods for the other coefficients are the same.

[0085] (1) Calculation of lift coefficient: Combine equations (1), (2), (3), (4), (5), (6), (7), (8), and (9) with the state data of the UAV (position, attitude, velocity, angular velocity, acceleration) to calculate the actual lift coefficient. .

[0086] (2) Correct the aerodynamic parameters in equation (10):

[0087]

[0088] Transform the above equation into:

[0089]

[0090] in, For the true lift coefficient, The original aerodynamic data is obtained from equation (10). Since the real aerodynamic parameters are variables that change continuously with the state of the UAV (position, speed, etc.), and the output of the deep learning network also changes according to the changes in the network input state, this invention proposes to use the deep learning network model DPnet to identify the real-time correction amount of the aircraft's aerodynamic parameters. Since the original aerodynamic data obtained from CFD calculations or wind tunnel experiments generally does not differ much from the true values, that is, it is generally near the optimal value, there is no need to worry about the network getting stuck in a local optimum during training. Therefore, a fully connected deep network is selected, and its network diagram is shown in Figure 10. Figure 2 As shown, the network consists of two channels: channel1 and channel2. Channel1 consists of one input layer and six hidden layers, with the hidden layers containing 48-dimensional state variables; channel2 consists of one input layer and three hidden layers, with the hidden layers containing 16-dimensional state variables.

[0091] The input vector X of the entire network consists of 15-dimensional state variables: position (x, y, z), attitude (θ, ψ, φ), velocity (u, v, w), angular velocity (p, q, r), and acceleration (ax, ay, az). Assuming a northeast-to-central coordinate system (x to the right, y forward, z upward), the state variables along the z-axis are closely related to lift, thus requiring fewer abstraction layers (hidden layers) between the input and output. To accelerate network training, channel 2 is used. Other state variables have indirect influences on lift, with lower correlation; therefore, channel 1, with its more abstraction layers, is used. Thus, when training the lift coefficient bias, the input vector x1 of channel 1 is 12-dimensional (x, y, θ, ψ, φ, u, v, p, q, r, ax, ay), and the input vector x2 of channel 2 is 3-dimensional (z, w, az). The output layer is 1-dimensional. The target value is The activation function is ReLU, and the loss function is... .

[0092] Gradient descent is used to train the network parameters using flight data, resulting in a trained parameter-corrected deep network model DPnet.

[0093] This leads to the corrected lift coefficient:

[0094]

[0095] (3) Using the same method, the corrected drag coefficient, side force coefficient, pitching moment coefficient, yaw moment coefficient, and rolling moment coefficient were obtained respectively.

[0096] The overall structure of the deep learning network model remains unchanged, but the input vectors need to be adjusted according to the correlation between the state variables and the coefficients to be corrected. When training the drag coefficient correction, the input vector x1 of channel 1 is 12-dimensional (x, z, θ, ψ, φ, u, w, p, q, r, ax, az), and the input vector x2 of channel 2 is 3-dimensional (y, v, ay). When training the lateral force coefficient correction, the input vector x1 of channel 1 is 12-dimensional (y, z, θ, ψ, φ, v, w, p, q, r, ay, az), and the input vector x2 of channel 2 is 3-dimensional (x, u, ax). When training the pitch moment coefficient correction, the input vector x1 of channel 1 is 13-dimensional (x, y, z, ψ, φ, u, v, w, p, r, ax, ay, az), and the input vector x2 of channel 2 is 2-dimensional (θ, q). When training the roll moment coefficient correction, the input vector x1 of channel 1 in the deep network is 11-dimensional (x, y, z, θ, u, v, w, q, ax, ay, az), and the input vector x2 of channel 2 is 4-dimensional (φ, ψ, p, r). When training the yaw moment coefficient correction, the input vector x1 of channel 1 in the deep network is 11-dimensional (x, y, z, θ, u, v, w, q, ax, ay, az), and the input vector x2 of channel 2 is 4-dimensional (φ, ψ, p, r).

[0097] Replace the lift coefficient, drag coefficient, side force coefficient, pitching moment coefficient, yaw moment coefficient, and roll moment coefficient in equations (1) and (3) with the corrected coefficients, such as the lift coefficient in the corrected equation (19). Rebuild the kinematic and dynamic models according to steps (1)-(8) of the hardware-in-the-loop simulation system, and replace the models in the original hardware-in-the-loop simulation system to optimize the simulation accuracy of the hardware-in-the-loop simulation system.

[0098] The revised kinematic and dynamic model formulas, aerodynamic parameters (such as lift coefficient) The model consists of two parts: one part is the theoretical calculation value or wind tunnel test value, and the other part is the deep network estimate DPnet. The DPnet value can be regarded as a function of the real-time state data (position, attitude, etc.) of the aircraft. The real-time state data used in the calculation process is the data of the previous frame, and the calculation process is iterative. Formulas (5) and (8) are updated again, and the updated kinematic and dynamic models are obtained.

[0099] Based on the characteristics of flight data and aerodynamic parameters—namely, aerodynamic parameters change in real time according to the aircraft's state, the initial values ​​of aerodynamic parameters are near their optimal values, and network training will not get stuck in local extrema—this invention proposes a parameter correction method based on deep learning. A deep learning network suitable for aerodynamic parameter correction is designed, using the deep learning network to identify the real-time correction amount of the aircraft's aerodynamic parameters. Furthermore, a flight data processing method and a method for training the network using flight data are designed to correct the aircraft's lift coefficient, drag coefficient, side force coefficient, pitch moment coefficient, yaw moment coefficient, and roll moment coefficient, respectively. This improves the modeling accuracy of kinematic and dynamic models, enabling the hardware-in-the-loop simulation system to better simulate real flight conditions. This aids in the design and optimization of the aircraft's control system and allows for rapid testing and evaluation of different design schemes and control strategies.

[0100] As described above, although the invention has been shown and described with reference to specific preferred embodiments, it should not be construed as limiting the invention itself. Various changes in form and detail may be made without departing from the spirit and scope of the invention as defined in the appended claims.

Claims

1. A parameter correction method for a hardware-in-the-loop simulation system of a fixed-wing unmanned aerial vehicle (UAV), characterized in that, Includes the following steps: S1: Construct a hardware-in-the-loop simulation system, including a real-time simulation system, an UAV guidance and control system, a UAV ground operation station, and a serial communication module. The real-time simulation system interacts with the UAV guidance and control system via the serial communication module, and the UAV guidance and control system interacts with the UAV ground operation station via a data link. The real-time simulation system includes a real-time operating system and real-time simulation software. The real-time operating system is mounted on a hardware platform, and the real-time simulation software includes a hardware driver module and an aircraft simulation model. The aircraft simulation model includes an aerodynamic model, a dynamic model, a kinematic model, and a sensor model. The dynamic model and kinematic model are used to calculate the aircraft's state information and, according to the data protocol of the real sensors, package the state information and send it to the UAV guidance and control system via the serial communication module. The UAV guidance and control system outputs rudder commands to control the aircraft based on the received aircraft state information. S2: Construct a deep learning network model. The input of the deep learning network model is the state information of the aircraft, and the output of the deep learning network model is the correction amount of the aerodynamic coefficient. The gradient descent method is used to train the network model so that the output of the network model approximates the difference between the true aerodynamic coefficient and the original aerodynamic coefficient. The deep learning network model adopts a fully connected neural network, including a first channel (channel1) and a second channel (channel2). The first channel (channel1) consists of one input layer and six hidden layers, and the hidden layers are 48-dimensional state variables. The second channel, channel2, consists of one input layer and three hidden layers, with the hidden layers containing 16-dimensional state variables. The input vectors of the first channel (channel 1) and the second channel (channel 2) are adjusted according to the correlation between the state variables and the aerodynamic coefficient to be corrected. For different aerodynamic coefficients, the dimensions of the input vectors of the first and second channels in the deep learning network model are different. When correcting the lift coefficient, the first channel input is 12-dimensional state information, namely x, y, θ, ψ, φ, u, v, p, q, r, ax, ay, and the second channel input is 3-dimensional state information, namely z, w, az; When correcting the drag coefficient, the first channel input is 12-dimensional state information, namely x, z, θ, ψ, φ, u, w, p, q, r, ax, and az; the second channel input is 3-dimensional state information, namely y, v, and ay. When correcting the lateral force coefficient, the first channel input is 12-dimensional state information, namely y, z, θ, ψ, φ, v, w, p, q, r, ay, az, and the second channel input is 3-dimensional state information, namely x, u, ax; When correcting the pitch moment coefficient, the first channel input is 13-dimensional state information, namely x, y, z, ψ, φ, u, v, w, p, r, ax, ay, az, and the second channel input is 2-dimensional state information, namely θ, q. When correcting the roll moment coefficient and yaw moment coefficient, the first channel input is 11-dimensional state information, namely x, y, z, θ, u, v, w, q, ax, ay, az, and the second channel input is 4-dimensional state information, namely ψ, φ, p, r. S3: Update the kinematic and dynamic models using the corrected aerodynamic coefficients, calculate the aircraft's state information using the updated kinematic and dynamic models, and transmit it to the UAV guidance and control system for aircraft control, including: The deep learning network model takes the state variables of the aircraft at the previous time step as input, performs iterative calculations, and outputs updated correction coefficients. The corrected aerodynamic coefficients consist of two parts: the first part is the theoretical calculation value or the wind tunnel experimental value; the second part is the output value of the deep learning network model. The kinematic and dynamic models calculated using the updated correction coefficients are in a dynamic updating process. The updated model is used to calculate the aircraft's state information and control the aircraft.

2. The parameter correction method for the fixed-wing unmanned aerial vehicle (UAV) hardware-in-the-loop simulation system according to claim 1, characterized in that, The dynamic and kinematic models calculate the state information of the aircraft through the center of mass dynamic equation, center of mass kinematic equation, rotational dynamic equation, rotational kinematic equation, force and torque equations acting on the aircraft, and latitude and longitude calculation equations. The state information includes position (x, y, z), attitude (θ, ψ, φ), velocity (u, v, w), angular velocity (p, q, r), and acceleration (ax, ay, az).

3. The parameter correction method for the fixed-wing unmanned aerial vehicle (UAV) hardware-in-the-loop simulation system according to claim 2, characterized in that, The input to the deep learning network model is the UAV flight state information, including position (x, y, z), attitude (θ, ψ, φ), velocity (u, v, w), angular velocity (p, q, r), and acceleration (ax, ay, az). Here, θ, ψ, and φ are the pitch, yaw, and roll angles of the x, y, and z axes in the body coordinate system; u, v, and w represent the velocity components of the x, y, and z axes in the body coordinate system; p, q, and r are the angular velocity components of the x, y, and z axes in the body coordinate system; and ax, ay, and az are the acceleration components of the x, y, and z axes in the body coordinate system. The output of the deep learning network model is the correction amount for aerodynamic coefficients, including lift coefficient, drag coefficient, side force coefficient, pitch moment coefficient, yaw moment coefficient, and roll moment coefficient.

4. A hardware-in-the-loop simulation system for implementing the method of claim 1, characterized in that, The system includes: A real-time simulation system includes a real-time operating system and real-time simulation software. The real-time operating system is mounted on a hardware platform, and the real-time simulation software includes a hardware driver module and an aircraft simulation model. The aircraft simulation model includes an aerodynamic model, a dynamic model, a kinematic model, and a sensor model. The aerodynamic data model is used to provide basic aerodynamic parameters, the sensor model is used to generate data that conforms to the output format of real sensors, and the dynamic model and kinematic model are used to calculate the state information of the aircraft. The deep learning network model takes as input the state information of the aircraft, including position (x, y, z), attitude (θ, ψ, φ), velocity (u, v, w), angular velocity (p, q, r), and acceleration (ax, ay, az). The output of the deep learning network model is the correction amount of the aerodynamic coefficients. The aerodynamic parameters include lift coefficient, drag coefficient, side force coefficient, pitch moment coefficient, yaw moment coefficient, and roll moment coefficient. The deep learning network model is a fully connected dual-channel network, including a first channel (channel1) and a second channel (channel2). The first channel (channel1) consists of one input layer and six hidden layers, with the hidden layers being 48-dimensional state variables. The second channel (channel2) consists of one input layer and three hidden layers, with the hidden layers being 16-dimensional state variables. The first channel (channel1) and the second channel (channel2) input state information according to the state variable division method corresponding to each aerodynamic coefficient in claim 1. The UAV guidance and control system is used to obtain the state information of the aircraft by calculating the corrected aerodynamic coefficients from the dynamic model and kinematic model, and then control the aircraft.

Images