Aircraft unsteady aerodynamic force prediction method, system, equipment and medium

Abstract

The invention discloses an aircraft unsteady aerodynamic force prediction method, system and device and a medium, and relates to the technical field of aircraft modeling and time sequence prediction. According to the method, an LSTM neural network prediction model is constructed, an aircraft unsteady aerodynamic force time sequence data set processed by a sliding window is used as a training data set, the LSTM neural network prediction model is trained, and the aircraft unsteady aerodynamic force time sequence data set is predicted by using the trained LSTM neural network prediction model. And the change of the unsteady aerodynamic performance of the aircraft under new incoming flow disturbance is predicted online in real time. According to the method, the prediction response to the attitude angle and the unsteady aerodynamic performance of the aircraft under the incoming flow disturbance is fast, the accuracy is high, and meanwhile reference can be provided for reducing the engineering problems of aircraft vibration, flight attitude control and the like.

Description

Technical Field

[0001] The present application relates to the technical field of aircraft modeling and time series prediction, and in particular to a method, system, device and medium for predicting unsteady aerodynamic force of an aircraft based on machine learning. Background Art

[0002] Under the condition of continuous change of incoming flow direction, quickly and accurately predicting the change of aircraft attitude is the premise of active control of aircraft, which is of great significance to flight safety and mission completion. Affected by the incoming flow disturbance, the flow field on the aircraft surface may show unsteady and highly nonlinear characteristics, affecting the stability of the aircraft aerodynamics and threatening flight safety.

[0003] The research methods of unsteady aerodynamic characteristics of aircraft include wind tunnel experiments, flight tests, numerical simulation and other methods. However, faced with complex flight environments, especially in real-time flight control application scenarios, neither experiments nor high-precision CFD simulations can meet the real-time prediction needs of engineering departments. Prediction methods based on mathematical and physical models include least squares method, extended Kalman filter method, maximum likelihood method, etc. Such methods with clear physical connotations can provide a solid foundation for unsteady aerodynamic modeling, but they often require accurate mathematical description of the power system and a priori estimation of the statistical characteristics of noise. However, the above two conditions are often difficult to meet in actual flight.

[0004] With the optimization of intelligent algorithms and the improvement of hardware performance, artificial intelligence-based modeling methods have become a powerful tool for dealing with complex nonlinear problems. The core advantages of these methods are that, on the one hand, they can bypass the complexity of traditional aerodynamic modeling and directly predict aerodynamic performance by learning nonlinear input-output mapping relationships; on the other hand, the trained intelligent models have a faster prediction speed than traditional aerodynamic research methods and meet the immediate flight control needs. Traditional machine learning methods often face challenges in nonlinear time series prediction tasks, and require specialized models to handle the time dependency and nonlinear characteristics of such tasks.

[0005] It can be seen that in the prior art, there is a lack of accurate and immediate prediction of the unsteady aerodynamic force of an aircraft under incoming flow disturbances, and the immediate and accurate prediction of the aerodynamic characteristics of an aircraft usually relies on a complete model of aircraft dynamics and immediate incoming flow information. In view of this, developing an effective unsteady aerodynamic force prediction method when there is incoming flow disturbance has important application value for engineering practice in the aviation field. Summary of the invention

[0006] Based on this, it is necessary to provide a method, system, equipment and medium based on the prediction of unsteady aerodynamic force of aircraft to address the above technical problems.

[0007] This manual adopts the following technical solutions:

[0008] This specification provides a method for predicting unsteady aerodynamic forces of an aircraft, including:

[0009] Obtain the time series data set of unsteady aerodynamic force of the aircraft under free stream disturbance;

[0010] The data set is processed by sliding window, and the long time series is processed into multiple short time series to extract the unsteady aerodynamic data set of the aircraft in each short time series;

[0011] Build an LSTM neural network prediction model, and use the aircraft unsteady aerodynamic time series data set processed by sliding window as a training data set to train the built LSTM neural network prediction model;

[0012] The trained LSTM neural network prediction model is used to predict the unsteady aerodynamic force of the aircraft under the predicted incoming flow disturbance.

[0013] Preferably, the building of the LSTM neural network prediction model specifically includes:

[0014] Designing the structure of the LSTM neural network model, including the number of layers and the number of hidden units;

[0015] Move the data set processed by the sliding window to two fully connected layers with 128 hidden units each to generate a new time series feature vector;

[0016] Pass the newly generated time series feature vector into an LSTM layer with 256 hidden units;

[0017] The operation results of the LSTM layer are passed to two fully connected layers, and finally the estimated value and the standard deviation of the estimated value of the predicted value of the neural network model are output;

[0018] Based on the estimated value of the predicted quantity of the neural network model and the standard deviation of the estimated value, the loss function and activation function for training the LSTM neural network are determined.

[0019] Preferably, the constructed LSTM neural network prediction model is trained, specifically including:

[0020] Based on the time series data set processed by sliding window, the input and output data of the LSTM neural network prediction model are set, where the input data is the incoming flow velocity vector and the historical flight attitude angle of the aircraft; the output data is the flight attitude angle and torque of the aircraft at the next moment;

[0021] Set the loss function for training the LSTM neural network and the activation function for the output standard deviation;

[0022] Train the LSTM neural network prediction model according to the set loss function, joint network layer number and activation function;

[0023] Output the trained LSTM neural network prediction model and parameters.

[0024] Preferably, the sliding window processing of the data set specifically includes: each data sample corresponds to a set of feature quantities selected by a specific s×τ sliding window, s represents the input feature quantity, τ is the time span of each sliding window, that is, the window size, which is set to 11 time steps, and one time unit is moved between adjacent sliding windows.

[0025] Preferably, the step of obtaining a time series data set of unsteady aerodynamic forces of an aircraft under free stream disturbance comprises:

[0026] Establish CFD calculation domain based on sliding moving grid technology;

[0027] Construct a single-degree-of-freedom aircraft pitch motion model and a three-degree-of-freedom aircraft pitch, yaw and roll motion model;

[0028] The unsteady aerodynamic time series data set of the aircraft under different incoming flow disturbances is collected, including the incoming flow velocity vector, the aircraft flight attitude angle and torque.

[0029] Preferably, the time series data set of unsteady aerodynamic force of the aircraft under free stream disturbance is obtained by numerical simulation using Ansys Fluent.

[0030] This specification provides an aircraft unsteady aerodynamic prediction system, comprising:

[0031] Data acquisition module: used to obtain the time series data set of unsteady aerodynamic force of the aircraft under free stream disturbance;

[0032] Data processing module: used to perform sliding window processing on the data set, processing the long time series into multiple short time series to extract the aircraft unsteady aerodynamic data set under each short time series;

[0033] Model building module: used to build an LSTM neural network prediction model, and use the aircraft unsteady aerodynamic time series data set processed by the sliding window as a training data set to train the built LSTM neural network prediction model;

[0034] Prediction module: Use the trained LSTM neural network prediction model to predict the unsteady aerodynamic force of the aircraft under the predicted incoming flow disturbance.

[0035] The present specification provides a computer-readable storage medium, characterized in that the storage medium stores a computer program, and when the computer program is executed by a processor, the above-mentioned method for predicting unsteady aerodynamic forces based on an aircraft is implemented.

[0036] The present specification provides a computer device, characterized in that it includes a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor implements the above-mentioned method for predicting unsteady aerodynamic forces of an aircraft when executing the program.

[0037] This specification provides a method, system, device and medium for predicting unsteady aerodynamic force of an aircraft. Compared with the prior art, the beneficial effects are as follows:

[0038] The present specification provides a method for predicting the unsteady aerodynamic force of an aircraft, which can construct a nonlinear mapping between flow disturbances and the unsteady aerodynamic response of the aircraft, and make instant and accurate predictions of the aircraft attitude angle and unsteady aerodynamic performance under incoming flow disturbances; and the constructed LSTM neural network prediction model uses the incoming flow information and flight attitude information that can be obtained in real time during the flight of the aircraft as input and output parameters, thereby improving its prediction response speed, while avoiding high-order quantities that are difficult to accurately calculate in actual flight, such as velocity gradient and rate of change of angle of attack, and has potential engineering application value. BRIEF DESCRIPTION OF THE DRAWINGS

[0039] The drawings described herein are used to provide a further understanding of the present application and constitute a part of the present application. The illustrative embodiments of the present application and their descriptions are used to explain the present application and do not constitute an improper limitation on the present application. In the drawings:

[0040] Figure 1 A schematic diagram of a flow chart of a method for predicting unsteady aerodynamic force of an aircraft provided by the present invention;

[0041] Figure 2 The present invention provides an aircraft unsteady aerodynamic prediction method for predicting the pitch angle and pitch moment of a NACA 0012 airfoil;

[0042] Figure 3 A diagram showing the prediction effect of the pitch angle and pitch moment of a B-2 flying wing using an aircraft unsteady aerodynamic prediction method provided by the present invention;

[0043] Figure 4 The pitch, yaw, roll angle and moment prediction effect diagram of a CHN-F1 flying wing using an aircraft unsteady aerodynamic prediction method provided by the present invention;

[0044] Figure 5This is a diagram showing the uncertainty prediction effect of the pitch, yaw and roll moments of the CHN-F1 flying wing using an aircraft unsteady aerodynamic prediction method provided by the present invention. DETAILED DESCRIPTION

[0045] In order to make the purpose, technical solutions and advantages of this specification more clear, the technical solutions of this application will be clearly and completely described below in combination with the specific embodiments of this specification and the corresponding drawings. Obviously, the described embodiments are only part of the embodiments of this application, not all of them. Based on the embodiments in the specification, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of this application.

[0046] The technical solutions provided by various embodiments of the present application are described in detail below in conjunction with the accompanying drawings.

[0047] Figure 1 The following is a flow chart of a method for predicting unsteady aerodynamic force of an aircraft in this specification, which specifically includes the following steps:

[0048] Step 1: Use Ansys Fluent numerical simulation to obtain the time series data set of the unsteady aerodynamic force of the aircraft under free stream disturbance, including the incoming flow velocity vector, aircraft attitude angle and torque, including:

[0049] Establish the CFD calculation domain based on sliding moving mesh technology.

[0050] Construct the pitch, yaw, and roll motion model of the aircraft.

[0051] Collect a time series dataset of unsteady aerodynamic forces on a vehicle under different incoming flows.

[0052] Step 2: Perform sliding window processing on the aircraft unsteady aerodynamic time series data set, and process the long time series into multiple short time series to extract the aircraft unsteady aerodynamic data set under each short time series, including:

[0053] Each data sample corresponds to a set of feature quantities selected by a specific s×τ sliding window, where s represents the input feature quantity and τ is the time span of each sliding window, i.e., the window size, which is set to 10 time steps. Adjacent sliding windows move one time unit.

[0054] Step 3: Build an LSTM neural network prediction model, and use the aircraft unsteady aerodynamic time series data set processed by the sliding window as a training data set to train the built LSTM neural network prediction model, including:

[0055] Build an LSTM neural network prediction model.

[0056] Set the input and output data of the neural network prediction model based on the time series data set processed by the sliding window.

[0057]

[0058]

[0059] Among them, X train Input the historical information matrix for historical information, Y train is the output vector, u is the incoming flow velocity vector, θ is the attitude angle, M is the torque, and the subscript t is the time step.

[0060] Set the loss function for training the LSTM neural network:

[0061]

[0062] Where y is the true value, f(x) is the neural network estimate, and σ0 is the standard deviation of the estimate.

[0063] Set the activation function for training the LSTM neural network output standard deviation:

[0064] soft plus(σ0)=log(1+exp(σ0))

[0065] Among them, σ o is the standard deviation of the predicted data.

[0066] Train the LSTM neural network prediction model according to the set loss function, number of joint network layers, and activation function;

[0067] Output the trained LSTM neural network prediction model and parameters.

[0068] Step 4: Use the trained LSTM neural network model to predict the unsteady aerodynamic force of the aircraft under the predicted incoming flow disturbance.

[0069] Furthermore, the input vector after sliding window processing is moved to two fully connected layers with 128 hidden units respectively to generate a new time series feature vector; then the newly generated time series feature vector is passed to an LSTM layer with 256 hidden units; finally, the result of LSTM is passed to two fully connected layers, and finally the estimated value and the standard deviation of the estimated value predicted by the neural network model are output, and the loss function and activation function of training the LSTM neural network are calculated based on the estimated value and the standard deviation of the estimated value.

[0070] Furthermore, 10% of the time series data set processed by the sliding window is randomly selected as the test set to test the neural network training effect. The examples selected as the test set do not participate in the subsequent sampling and training process to ensure the effectiveness of the test. The input and output vectors of the test set are consistent with the training process, but unlike the training set, the final neural network is directly used for prediction.

[0071] Embodiment 1:

[0072] Based on an aircraft unsteady aerodynamic prediction method provided in this specification, the future flight attitude angle and moment of a NACA 0012 two-dimensional airfoil are predicted.

[0073] Firstly, the SST k-ω turbulence model is used to solve the RANS equations in a closed manner. The solution method is COUPLED, and the second-order upwind scheme is selected as the discrete format. The CFD calculation domain based on the sliding moving mesh technology is established, which is a c-type mesh.

[0074] The center of gravity of the airfoil and flying wing model is located at the center of the semicircle, the chord length of the airfoil is 1m, the calculation medium of the inlet flow is set to ideal gas, the flow velocity is 35m / s, and the inlet is 8 times the reference length of the flying wing to ensure that the inlet flow is fully developed.

[0075] The computational grid is generated using the ANSYS Mesh tool, using an unstructured grid and a specific angle of attack as the initial state of the airfoil. In this embodiment, the angle of the airfoil is set to 6.5°. The pressure center of the wing at a specific angle of attack is solved by integrating the pressure on the airfoil surface to be 0.18m, and the equilibrium position of the wing is obtained by balancing with the center of gravity, and the pressure center is used as the rotation center of the airfoil.

[0076] Construct a single-degree-of-freedom pitch motion model, write UDF (User-Defined Functions) in Fluent to set the flow disturbance equation α whose angle changes with time, and simulate clear-air turbulence by adding disturbances to the initial flow. The flow velocity is,

[0077] u=(ucosα(t)i,usinα(t)j)

[0078] Among them, u=|u|=35(m / s) is the magnitude of the incoming flow velocity, and α(t) determines the direction of the incoming flow velocity vector, which is expressed as,

[0079] α(t)=α(t0)+a1sin(b1t)+a2cos(b2t)

[0080] +a3sin(b3t)sin(b4t)

[0081] In the formula, α(t0) is the initial flow direction, and the parameter value set is set to a∈[1,8], b∈[10,80]. Data is randomly generated within this range and randomly combined into different flow conditions to achieve continuous changes in the flow direction under the premise of constant flow velocity. 85 sets of NACA0012 two-dimensional airfoil aerodynamic time series data under different flows were collected, and 10% were randomly selected as the test set. Figure 2 Shown is the predicted effect diagram of pitch angle and pitching moment of NACA0012 airfoil.

[0082] Embodiment 2:

[0083] Based on a method for predicting unsteady aerodynamic force of an aircraft provided in this specification, the future flight attitude angle and moment of a B-2 flying wing model are predicted.

[0084] The longitudinal length of the B-2 flying wing model is about 1m, and the incoming flow Mach number Ma=0.9. Similar to the NACA 0012 airfoil embodiment, the computational domain grid is C-shaped, the flying wing is located at the center of the semicircular computational domain, the inlet condition of the flow field is the pressure far field, the outlet is the pressure outlet, and the inlet boundary is set at a distance of 10 times the length of the airfoil nose. An unstructured grid is used, the center of gravity is used as the rotation center, and the incoming flow disturbance is added in the flying wing level flight state.

[0085] For the B-2-like flying wing model, although three-dimensional calculations are carried out, the flying wing is limited to single-degree-of-freedom pitch motion. The setting of the incoming flow disturbance in the calculation is the same as that of the single-degree-of-freedom NACA 0012 airfoil. 85 sets of aerodynamic time series data of the B-2-like flying wing model under different incoming flows are collected, and 10% are randomly selected as the test set.

[0086] like Figure 3 Shown is a diagram showing the effect of predicting the pitch angle and pitch moment of a B-2 flying wing using an aircraft unsteady aerodynamic prediction method provided by the present invention.

[0087] Embodiment 3:

[0088] In order to face more realistic working conditions, a three-degree-of-freedom rotational motion model of the CHN-F1 flying wing is constructed.

[0089] A CFD calculation domain based on dynamic mesh technology is established, which is a C-type mesh similar to the B-2 flying wing. The center of gravity of the flying wing model is located at the center of the semicircle, and the inlet is 10 times the reference length of the flying wing to ensure that the inlet flow is fully developed. In order to achieve three-degree-of-freedom rotational motion, a UDF is written to set the moment of inertia and mass of the flying wing around the X, Y, and Z axes, and constrain the translation of the flying wing on the three axes, limiting the freedom of the flying wing in translation, and only retaining the rotational motion of the flying wing in three directions. At the same time, a UDF is written to set the future flow disturbance as a three-dimensional disturbance, and the three-dimensional velocity vector is expressed as,

[0090] u=(u1i,u2j,u3k)

[0091] The three components of the velocity vector are expressed as,

[0092]

[0093]

[0094] u3=usin(θ(t))

[0095] Where u = |u| is the velocity of the incoming flow, satisfying the Mach number Ma = 0.6, i, j, k are the unit vectors of the X, Y, and Z axes, ensuring that the velocity of the incoming flow remains unchanged. It is only necessary to determine θ(t) and The change of the two angle functions over time can give the flow disturbance with the continuously changing flow angle. The three-dimensional flow disturbance equation needs to add a transverse flow equation on the basis of the two-dimensional flow equation to determine the direction of the flow velocity vector θ(t) and Expressed as,

[0096] θ(t)=θ(t0)+a1 sin(b1t)+a2 cos(b2t)+a3 sin(b3t)sin(b4t)

[0097]

[0098] The parameter value range is a∈[1,8], b∈[500,1500]. Data is randomly generated within this range and randomly combined into different incoming flow conditions. The free flow disturbance corresponding to each set of parameter values ​​is regarded as a working condition. 300 sets of aerodynamic time series data of the CHN-F1 flying wing model under different incoming flows are collected, and 10% are randomly selected as the test set.

[0099] like Figure 4 The figure shows the effect of the pitch, yaw, roll angle and moment prediction of the CHN-F1 flying wing using the unsteady aerodynamic force prediction method provided by the present invention.

[0100] like Figure 5 The figure shows the effect of the pitch, yaw and roll moment uncertainty prediction of the CHN-F1 flying wing using the unsteady aerodynamic force prediction method provided by the present invention.

[0101] In summary, the present invention provides a method for predicting the unsteady aerodynamic force of an aircraft based on machine learning, develops a real-time online prediction model for the unsteady aerodynamic force of an aircraft, and can construct a nonlinear mapping between flow disturbances and the unsteady aerodynamic response of the aircraft. In the prediction of pitch angle and moment changes of the single-degree-of-freedom two-dimensional NACA 0012 airfoil and the three-dimensional B-2 flying wing model, the model prediction results are in good agreement with the CFD calculation results. At the same time, the method for predicting the unsteady aerodynamic force of an aircraft proposed in the present invention has certain potential for practical engineering applications. In the study of the three-degree-of-freedom unsteady embodiment of the CHN-F1 flying wing facing real flight scenarios, the LSTM neural network with the flow history information and the flight attitude history information as input can still accurately predict the aerodynamic force consistent with the numerical simulation results.

[0102] The aircraft unsteady aerodynamic prediction method proposed in the present invention not only has good prediction performance, but also has certain modeling capabilities, and can predict and model other aerodynamic forces such as torque when only the incoming flow information and the aircraft rotation angle are available. The model has achieved a breakthrough in the prediction time order of magnitude while ensuring the calculation accuracy, and is more suitable for engineering application scenarios that require rapid prediction and rapid response, and can provide a reference for engineering problems such as reducing aircraft vibration and flight attitude control.

[0103] The technical features of the above embodiments may be combined arbitrarily. To make the description concise, not all possible combinations of the technical features in the above embodiments are described. However, as long as there is no contradiction in the combination of these technical features, they should be considered to be within the scope of this specification.

Claims

1. A method for predicting unsteady aerodynamic force of an aircraft, characterized in that: include: Obtain the time series data set of unsteady aerodynamic force of the aircraft under free stream disturbance; The data set is processed by sliding window, and the long time series is processed into multiple short time series to extract the unsteady aerodynamic data set of the aircraft in each short time series; Build an LSTM neural network prediction model, and use the aircraft unsteady aerodynamic time series data set processed by sliding window as a training data set to train the built LSTM neural network prediction model; The trained LSTM neural network prediction model is used to predict the unsteady aerodynamic force of the aircraft under the predicted incoming flow disturbance.

2. The method for predicting unsteady aerodynamic force of an aircraft according to claim 1, characterized in that: The construction of the LSTM neural network prediction model specifically includes: Designing the structure of the LSTM neural network model, including the number of layers and the number of hidden units; Move the data set processed by the sliding window to two fully connected layers with 128 hidden units each to generate a new time series feature vector; The newly generated time series feature vector is passed into an LSTM layer with 256 hidden units; The operation results of the LSTM layer are passed to two fully connected layers, and finally the estimated value and the standard deviation of the estimated value of the predicted value of the neural network model are output; Based on the estimated value of the predicted quantity of the neural network model and the standard deviation of the estimated value, the loss function and activation function for training the LSTM neural network are determined.

3. The method for predicting unsteady aerodynamic force of an aircraft according to claim 1, characterized in that: The training of the built LSTM neural network prediction model includes: Based on the time series data set processed by sliding window, the input and output data of the LSTM neural network prediction model are set, where the input data is the incoming flow velocity vector and the historical flight attitude angle of the aircraft; the output data is the flight attitude angle and torque of the aircraft at the next moment; Set the loss function for training the LSTM neural network and the activation function for the output standard deviation; Train the LSTM neural network prediction model according to the set loss function, joint network layer number and activation function; Output the trained LSTM neural network prediction model and parameters.

4. The method for predicting unsteady aerodynamic force of an aircraft according to claim 1, characterized in that: The sliding window processing of the data set specifically includes: each data sample corresponds to a set of feature quantities selected by a specific s×τ sliding window, s represents the input feature quantity, τ is the time span of each sliding window, that is, the window size, which is set to 11 time steps, and one time unit is moved between adjacent sliding windows.

5. The method for predicting unsteady aerodynamic force of an aircraft according to claim 1, characterized in that: The specific steps of obtaining the time series data set of unsteady aerodynamic force of the aircraft under free stream disturbance include: Establish CFD calculation domain based on sliding moving grid technology; Construct a single-degree-of-freedom aircraft pitch motion model and a three-degree-of-freedom aircraft pitch, yaw and roll motion model; The unsteady aerodynamic time series data set of the aircraft under different incoming flow disturbances is collected, including the incoming flow velocity vector, the aircraft flight attitude angle and torque.

6. The method for predicting unsteady aerodynamic force of an aircraft according to claim 1, characterized in that: The time series data set of the unsteady aerodynamic force of the aircraft under the free stream disturbance is obtained by Ansys Fluent numerical simulation.

7. An aircraft unsteady aerodynamic prediction system, characterized in that: include: Data acquisition module: used to obtain the time series data set of unsteady aerodynamic force of the aircraft under free stream disturbance; Data processing module: used to perform sliding window processing on the data set, processing the long time series into multiple short time series to extract the aircraft unsteady aerodynamic data set under each short time series; Model building module: used to build an LSTM neural network prediction model, and use the aircraft unsteady aerodynamic time series data set processed by the sliding window as a training data set to train the built LSTM neural network prediction model; Prediction module: Use the trained LSTM neural network prediction model to predict the unsteady aerodynamic force of the aircraft under the predicted incoming flow disturbance.

8. A computer-readable storage medium, characterized in that: The storage medium stores a computer program, and when the computer program is executed by a processor, the method according to any one of claims 1 to 6 is implemented.

9. A computer device, characterized in that: The method comprises a memory, a processor and a computer program stored in the memory and executable on the processor, wherein when the processor executes the program, the method described in any one of claims 1 to 6 is implemented.

Images