Method for identifying aerodynamic parameters of aircraft with uncertain shape
Through the "offline learning + online correction" method of neural network, the problem of insufficient real-time identification of aerodynamic parameters of uncertain appearance aircraft is solved, and the rapid and accurate identification of aerodynamic parameters of uncertain appearance aircraft is achieved, and the aerodynamic parameters of uncertain appearance aircraft is improved. Real-time and accuracy.
Patent Information
- Application Number
- CN202410915967.6
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2024-07-09
- Publication Date
- 2024-11-12
- Estimated Expiration
- 2044-07-09
AI Technical Summary
The existing online identification methods for aerodynamic parameters of uncertain aircraft are insufficient in real-time and adaptability, making it difficult to meet the needs of fast and accurate identification of aerodynamic parameters.
Using the 'offline learning + online correction' aerodynamic parameters online identification method based on neural networks, aerodynamic model learning sample library is constructed by collecting wind tunnel tests and historical flight data, and using neural network training to build a basic aerodynamic identification network model, and in flight In the process, incremental adjustment is performed through the stochastic gradient descent method to achieve real-time identification of aerodynamic parameters.
It improves the real-time and accuracy of online identification of aerodynamic parameters of uncertain appearance aircraft, realizes the rapid and accurate identification of aerodynamic parameters of uncertain appearance aircraft, and avoids the need to build an explicit aerodynamic model.
Smart Images

Figure CN118940393A_ABST
Abstract
Description
Technical Field
[0001] The invention relates to an aerodynamic parameter identification method for an aircraft, and in particular to an aerodynamic parameter identification method for an aircraft with an uncertain shape. Background Art
[0002] Uncertain shape aircraft can change aerodynamic characteristics through deformation, and have stronger environmental and mission adaptability, and are being studied by countries around the world. However, compared with traditional determined shape aircraft, uncertain shape aircraft have wider flight airspace and speed range, and the aerodynamic effects they are subjected to during flight are more complex and the mechanism is unknown. In addition, the coupling of uncertain deformation and aerodynamics makes it impossible to build an accurate uncertain shape aircraft aerodynamic model covering the entire deformation process offline, and a fast and accurate online aerodynamic parameter identification method is urgently needed.
[0003] At present, there are two main methods for online identification of aerodynamic parameters: frequency / time domain online identification of aerodynamic parameters and artificial intelligence-based online identification of aerodynamic parameters. Frequency / time domain online identification of aerodynamic parameters requires human experience, and the identification process is essentially an estimation based on the optimal criterion, which is difficult to meet the real-time requirements of online identification of aerodynamic parameters of aircraft with uncertain shapes; artificial intelligence-based aerodynamic parameter identification methods require a large number of samples for training, which requires a large amount of samples and computing resources, and has limited generalization performance.
[0004] Therefore, it is urgent to develop a method that can realize online rapid and accurate identification of aerodynamic parameters of aircraft with uncertain shapes. Summary of the invention
[0005] In order to overcome the problems of insufficient real-time and adaptability of existing online aerodynamic parameter identification methods, the present invention provides an aerodynamic parameter identification method for an aircraft with uncertain shape. The method adopts an online aerodynamic parameter identification method of "offline learning + online correction", fully utilizing the advantages of aerodynamic parameter identification based on artificial intelligence, and realizing online rapid and accurate identification of aerodynamic parameters of an aircraft with uncertain shape.
[0006] The objective of the present invention is achieved through the following technical solutions:
[0007] A method for identifying aerodynamic parameters of an aircraft with uncertain shape comprises the following steps:
[0008] Step 1: Collect wind tunnel test and historical flight data of uncertain shape aircraft, obtain flight status and aerodynamic force and moment coefficients, and build a sample library for aerodynamic model learning;
[0009] Step 2: Based on the aerodynamic model learning sample library constructed in step 1, a basic aerodynamic identification network model from flight state to aerodynamic force / torque coefficient is constructed through neural network learning training;
[0010] Step 3: During online flight, collect airspeed and overload measurement information, and use flight mechanics principles to build an aerodynamic identification network sample;
[0011] Step 4: Using the aerodynamic identification network sample constructed in step 3, the stochastic gradient descent method is used to incrementally adjust the aerodynamic identification network at each interval T, so as to obtain a more accurate aerodynamic identification network model;
[0012] Step 5: According to the real-time flight status and the aerodynamic identification network model adjusted in step 4, the aerodynamic coefficients of the uncertain shape aircraft are obtained, and the aerodynamic parameters are obtained by numerically differentiating the aerodynamic coefficients.
[0013] Compared with the prior art, the present invention has the following advantages:
[0014] The present invention adopts an online identification method of aerodynamic parameters based on neural network "offline learning + online correction". There is no need to build an explicit aerodynamic model. It can make full use of the online and offline data of uncertain shape aircraft, improve the real-time and accuracy of online identification of aerodynamic parameters, and realize rapid and accurate identification of aerodynamic parameters of uncertain shape aircraft. BRIEF DESCRIPTION OF THE DRAWINGS
[0015] Figure 1 It is a block diagram of the aerodynamic parameter identification scheme of the uncertain shape aircraft of the present invention;
[0016] Figure 2 The pitch moment coefficient identification result of the basic aerodynamic model;
[0017] Figure 3 It is the identification result of the pitching moment coefficient of the online correction model. DETAILED DESCRIPTION
[0018] The technical solution of the present invention is further described below in conjunction with the accompanying drawings, but is not limited thereto. Any modification or equivalent replacement of the technical solution of the present invention without departing from the spirit and scope of the technical solution of the present invention should be included in the protection scope of the present invention.
[0019] The present invention provides a method for identifying aerodynamic parameters of an aircraft with an uncertain shape. The method comprises two parts: offline basic aerodynamic identification model construction and online aerodynamic identification model correction. The specific steps are as follows:
[0020] Construction of offline basic aerodynamic identification model:
[0021] Step 1: Collect wind tunnel test and historical flight data of uncertain shape aircraft, obtain flight status (Mach number, angle of attack, sideslip angle, rudder angle, deformation) and aerodynamic force and moment coefficient, and build an aerodynamic model learning sample library.
[0022] Step 2: Based on the aerodynamic model learning sample library constructed in step 1, a basic aerodynamic identification network model from flight state to aerodynamic force / torque coefficient is constructed through neural network learning training. The specific steps are as follows:
[0023] The flight state of the uncertain shape aircraft is used as the neural network input, the aerodynamic force / torque coefficient is used as the neural network output, the hyperbolic tangent function f(x)=tanh(x) is used as the activation function of the hidden layer, the mean square error is selected as the loss function, and the Adam algorithm is selected for neural network training optimization.
[0024] Online aerodynamic identification model correction:
[0025] Step 3: During online flight, collect measurement information: airspeed, overload, attitude angular rate, rudder deflection, deformation, and use flight mechanics principles to build an aerodynamic identification network sample. The specific steps are as follows:
[0026] Step 3.1: Online measurable data collection: Online measurable data includes the three-channel attitude angular rate ω of the aircraft x ,ω y ,ω z 、Axial overload of the elastic system n x , normal overload of the projectile system n y , lateral overload of the missile system n z , dynamic pressure q, pitch rudder deflection angle Rolling rudder angle δ γ 、Yaw rudder deflection angle δ ψ , deformation ξ.
[0027] Step 3.2: Unmeasurable sample features and label construction: The angle of attack and sideslip angle are obtained by the components of the velocity in the projectile system. is the component of airspeed in the missile system, the angle of attack can be given by:
[0028]
[0029] Sample label Axial force coefficient C A , Normal force coefficient C N , lateral force coefficient C Z , rolling moment coefficient C mx , Yaw moment coefficient C my , pitch moment coefficient C mz It can be obtained through the following simplified relationship:
[0030]
[0031] Among them, attitude angular acceleration The attitude angular velocity ω x ,ω y ,ωz The numerical difference is obtained, m is the mass, S is the aerodynamic reference area, L is the aerodynamic reference length, J x ,J y ,J z is the moment of inertia.
[0032] Step 3.3: Sample normalization: Use the maximum-minimum normalization method to normalize the samples. The normalization formula is as follows:
[0033]
[0034] Among them, Y represents the original sample sequence of a single variable, is the normalized data sequence, n is the number of samples, Up is the normalized upper bound, and Low is the normalized lower bound.
[0035] Step 4: Using the aerodynamic identification network samples constructed in step 3, the stochastic gradient descent method is used to incrementally adjust the aerodynamic identification network constructed in step 2 at every time interval T, so as to obtain a more accurate aerodynamic identification network model.
[0036] Step 5: According to the real-time flight status and the aerodynamic identification network model after step 4, the aerodynamic coefficients of the uncertain shape aircraft are obtained, and the aerodynamic parameters are obtained by numerically differentiating the aerodynamic coefficients. The specific steps are as follows:
[0037] Step 5.1: Obtaining aerodynamic coefficients: Input the aerodynamic coefficients into the aerodynamic identification network model adjusted in step 4 to quickly obtain accurate aerodynamic coefficients.
[0038] Step 5.2: Obtaining aerodynamic parameters: Use numerical differentiation method to numerically differentiate the aerodynamic coefficients to obtain aerodynamic parameters.
[0039] Example:
[0040] Taking a deformable glider as an example, the specific steps of offline basic aerodynamic identification model construction and online aerodynamic identification model correction in the process of aerodynamic parameter identification are as follows:
[0041] ① Construction of offline basic aerodynamic identification model
[0042] Step 1: Collect the wind tunnel test and historical flight data of the deformable glider, obtain the flight state (Mach number, angle of attack, sideslip angle, pitch rudder angle, yaw rudder angle, roll rudder angle, deformation) and the corresponding aerodynamic force and moment coefficient, and build an aerodynamic model learning sample library.
[0043] Step 2: Establish a fully connected neural network and use the gradient descent method to train the glider aerodynamic identification model network using the sample library.
[0044] Taking the pitch moment coefficient identification network construction as an example, the network input is the Mach number, angle of attack, sideslip angle, pitch rudder deflection angle, yaw rudder deflection angle, roll rudder deflection angle, and deformation of the glider, and the output is the pitch moment coefficient. The neural network training optimization method is as follows:
[0045] Activation function: In order to avoid the gradient vanishing phenomenon when using the gradient descent algorithm to optimize the neural network, the hyperbolic tangent function f(x) = tanh(x) is used as the activation function of the hidden layer to improve the optimization efficiency of the neural network.
[0046] Loss function: Select mean square error (MSE) as the loss function:
[0047]
[0048] In the formula, Represents the network output value, θ represents the sample label value, and N represents the number of samples.
[0049] Optimization function: Select the Adam algorithm with adaptive learning rate to optimize the training of neural network.
[0050] ②Online aerodynamic identification model correction
[0051] Step 3: During online flight, collect measurement information such as airspeed and overload, and use the principles of flight mechanics to build an aerodynamic identification network sample.
[0052] Step 3.1: Online measurable data collection: The measurable flight state variables actually obtained by the glider include the three-channel attitude angular rate ω of the aircraft x ,ω y ,ω z 、Axial overload of the elastic system n x , normal overload of the projectile system n y , lateral overload of the missile system n z , dynamic pressure q, pitch rudder deflection angle Rolling rudder angle δ γ 、Yaw rudder deflection angle δ ψ And the deformation ξ1 is given by the flight control system.
[0053] Step 3.2: Unmeasurable sample features and label construction: The angle of attack and sideslip angle cannot be measured directly, but can be obtained through the component of the velocity in the missile system. Since the flight speed of the glider is relatively high, the influence of wind can be ignored. is the component of airspeed in the missile system, the angle of attack can be given by:
[0054]
[0055] Sample label Axial force coefficient C A, Normal force coefficient C N , lateral force coefficient C Z , rolling moment coefficient C mx , Yaw moment coefficient C my , pitch moment coefficient C mz It can be obtained through the following simplified relationship:
[0056]
[0057] Among them, attitude angular acceleration The attitude angular velocity ω x ,ω y ,ω z The numerical difference is obtained, m is the mass, S is the aerodynamic reference area, L is the aerodynamic reference length, J x ,J y ,J z is the moment of inertia.
[0058] Step 3.3: Sample normalization: Since the numerical intervals of different raw data distributions are different, the samples need to be normalized before training to improve the training speed and the generalization performance of the neural network. The present invention adopts the maximum-minimum normalization method, and the normalization formula is as follows:
[0059]
[0060] Among them, Y represents the original sample sequence of a single variable, Y is the normalized data sequence, and n is the number of samples. Up is the normalized upper bound, and Low is the normalized lower bound. Usually Low=-1,Up=1, that is, all variables in the sample are placed in the interval [-1,1] after normalization.
[0061] Step 4: Using the aerodynamic identification network samples constructed in step 3, at every time interval T, a network with a loss value within an acceptable range that does not require all training samples and has a small amount of calculation is used to train and adjust the aerodynamic identification network constructed in step 2, so as to obtain a more accurate aerodynamic identification network model.
[0062] Step 5: Obtain the aerodynamic coefficients of the glider according to the real-time flight status and the aerodynamic model, and perform numerical differentiation on the aerodynamic coefficients to obtain the aerodynamic parameters.
[0063] Step 5.1: Obtaining aerodynamic coefficients: Input the Mach number, angle of attack, sideslip angle, pitch rudder deflection angle, yaw rudder deflection angle, roll rudder deflection angle, and deformation of the glider into the aerodynamic identification network model adjusted in step 4 to quickly obtain accurate aerodynamic coefficients.
[0064] Step 5.2: Aerodynamic parameter calculation: After the aerodynamic model is constructed, the numerical differentiation method is used to calculate The basic steps of numerical differentiation are as follows:
[0065] a. Select an input variable and add a small disturbance to the input of the aerodynamic identification network model while keeping other input variables unchanged;
[0066] b. The corresponding aerodynamic derivative is equal to the ratio of the change in the model output to the disturbance at the input;
[0067] c. To ensure the robustness of the numerical method, the above steps a and b are performed at each sample point for the same aerodynamic derivative.
[0068] d. The aerodynamic derivative to be calculated is equal to the average of the estimated values of the aerodynamic derivatives obtained at different sample points.
[0069] The identification results of the pitch moment coefficient of the basic aerodynamic model are as follows: Figure 2 As shown in the figure, the pitch moment coefficient identification results of the online correction model are as follows: Figure 3 shown.
Claims
1. A method for identifying aerodynamic parameters of an aircraft with uncertain shape, characterized in that The method comprises the following steps: Step 1: Collect wind tunnel test and historical flight data of uncertain shape aircraft, obtain flight status and aerodynamic force and moment coefficients, and build a sample library for aerodynamic model learning; Step 2: Based on the aerodynamic model learning sample library constructed in step 1, a basic aerodynamic identification network model from flight state to aerodynamic force / torque coefficient is constructed through neural network learning training; Step 3: During online flight, collect airspeed and overload measurement information, and use flight mechanics principles to build an aerodynamic identification network sample; Step 4: Using the aerodynamic identification network sample constructed in step 3, the stochastic gradient descent method is used to incrementally adjust the aerodynamic identification network at each interval T, so as to obtain a more accurate aerodynamic identification network model; Step 5: According to the real-time flight status and the aerodynamic identification network model adjusted in step 4, the aerodynamic coefficients of the uncertain shape aircraft are obtained, and the aerodynamic parameters are obtained by numerically differentiating the aerodynamic coefficients.
2. The method for identifying aerodynamic parameters of an uncertain shape aircraft according to claim 1, characterized in that The flight status includes Mach number, angle of attack, sideslip angle, rudder angle, and deformation.
3. The method for identifying aerodynamic parameters of an uncertain shape aircraft according to claim 1, characterized in that The specific steps of step 2 are as follows: The flight state of the uncertain shape aircraft is used as the neural network input, the aerodynamic force / torque coefficient is used as the neural network output, the hyperbolic tangent function f(x)=tanh(x) is used as the activation function of the hidden layer, the mean square error is selected as the loss function, and the Adam algorithm is selected for neural network training optimization.
4. The method for identifying aerodynamic parameters of an uncertain shape aircraft according to claim 1, characterized in that The specific steps of step 3 are as follows: Step 3.1: Online measurable data collection: Online measurable data includes the three-channel attitude angular rate ω of the aircraft x ,ω y ,ω z 、Axial overload of the elastic system n x , normal overload of the projectile system n y , lateral overload of the missile system n z , dynamic pressure q, pitch rudder deflection angle Rolling rudder angle δ γ 、Yaw rudder deflection angle δ ψ , deformation ξ; Step 3.2: Unmeasurable sample features and label construction: The angle of attack and sideslip angle are obtained by the components of the velocity in the projectile system. is the component of airspeed in the missile system, the angle of attack is given by: Sample label Axial force coefficient C A , Normal force coefficient C N , lateral force coefficient C Z , rolling moment coefficient C mx , Yaw moment coefficient C my , pitch moment coefficient C mz The following simplified relationship is obtained: Among them, attitude angular acceleration The attitude angular velocity ω x ,ω y ,ω z The numerical difference is obtained, m is the mass, S is the aerodynamic reference area, L is the aerodynamic reference length, J x ,J y ,J z is the moment of inertia; Step 3.3: Sample normalization: Use the maximum-minimum normalization method to normalize the samples. The normalization formula is as follows: Among them, Y represents the original sample sequence of a single variable, is the normalized data sequence, n is the number of samples, Up is the normalized upper bound, and Low is the normalized lower bound.
5. The method for identifying aerodynamic parameters of an uncertain shape aircraft according to claim 1, characterized in that The specific steps of step 5 are as follows: Step 5.1: Obtaining aerodynamic coefficients: Input the aerodynamic coefficients into the aerodynamic identification network model adjusted in step 4 to quickly obtain accurate aerodynamic coefficients; Step 5.2: Obtaining aerodynamic parameters: Use numerical differentiation method to numerically differentiate the aerodynamic coefficients to obtain aerodynamic parameters.
Citation Information
Patent Citations
Online identification method for aerodynamic parameters of hypersonic aircraft
CN112668104A
Aircraft aerodynamic parameter identification method based on pre-training and incremental learning
CN114462149A
Pitching moment coefficient and center-of-pressure coefficient correction method for rocket projectile, and storage medium
WO2021098886A1
