An aircraft geometric feature parameter inverse derivation modeling method based on deep learning

Abstract

This invention provides a deep learning-based method for reverse modeling of aircraft geometric feature parameters, comprising: Step 1, classifying the aircraft; Step 2, analyzing the shape characteristics and aerodynamic features of different types of aircraft, and extracting a set of geometric feature parameters; Step 3, constructing an aerodynamic database for training a neural network model; Step 4, establishing a deep neural network model; Step 5, training the deep neural network model, with the network output being the aircraft's geometric feature parameters; Step 6, inputting the designed aerodynamic characteristic information into the trained deep neural network model to obtain the aircraft's geometric feature parameters that meet the design requirements. The neural network model established by this invention, which transforms aerodynamic parameters into geometric shape feature parameters, outputs not the geometric design parameters used in parametric modeling, but rather a set of geometric feature parameters that can generalize to a certain type of aircraft shape, thus giving the network model shape generalization capabilities.

Description

Technical Field

[0001] This invention relates to a deep learning-based method for reverse modeling of aircraft geometric feature parameters. Background Technology

[0002] The demands for flight quality, aerodynamic performance, and flight efficiency of aircraft are increasing, often resulting in various constraints during the aerodynamic design phase. Aircraft aerodynamic design is mainly divided into forward design and inverse design methods. Forward design involves repeatedly modifying the aircraft's shape and performing CFD calculations to obtain an aerodynamic shape that meets design specifications, which undoubtedly consumes significant computational resources and time. Inverse design, on the other hand, directly solves for the shape that meets design requirements using pre-designed aerodynamic characteristic information, directly establishing a mapping from aerodynamic parameters to geometric shape, greatly improving design efficiency. Machine learning-based inverse design methods have become a current research hotspot. However, most current methods are applicable to airfoil inverse design, wing inverse design, or optimization problems of a single shape. The input to the machine learning model is the design parameters used in the parametric modeling of the shape, rather than geometric feature parameters, thus lacking applicability to inverse design problems of new shapes. Therefore, it is necessary to provide a deep learning-based inverse modeling method for aircraft geometric feature parameters, establishing a mapping model from aerodynamic characteristic information to geometric feature parameters applicable to a certain type of aircraft, rather than being limited to a single shape; that is, the constructed model has shape generalization capabilities. Summary of the Invention

[0003] Objective of the Invention: The technical problem to be solved by this invention is to address the shortcomings of existing technologies by providing a deep learning-based method for reverse modeling of aircraft geometric feature parameters. This invention directly establishes a mapping model from the aerodynamic characteristics of a three-dimensional aircraft to its geometric shape feature parameters using deep learning. This model can predict the geometric shape feature parameters of the aircraft that meet the design requirements based on the designed aerodynamic characteristic information, thereby shortening the overall aircraft shape design cycle.

[0004] The method of the present invention includes the following steps:

[0005] Step 1: Classify the aircraft based on its shape and aerodynamic characteristics. Currently, there are various ways to classify aircraft, such as by purpose (civilian vs. military), by speed (subsonic vs. supersonic), and by engine type (propeller vs. jet). Since the goal of this invention is to perform inverse modeling of the geometric features of aircraft, and different aerodynamic layouts result in significant differences in shape, classification based on both geometric and aerodynamic characteristics is necessary to ensure that the geometric features extracted in Step 2 are applicable to the specific type of aircraft. Current classification methods are mostly arbitrary; for example, hypersonic aircraft are classified according to their aerodynamic layout as lifting bodies, wing-body combinations, spinning bodies, and waveriders.

[0006] Step 2: For different types of aircraft, analyze their shape characteristics and aerodynamic characteristics, and extract a set of geometric feature parameters;

[0007] Step 3: Construct an aerodynamic database for training neural network models;

[0008] Step 4: Build a deep neural network model;

[0009] Step 5, Deep Neural Network Training and Testing: The deep neural network model established in Step 4 is trained using the aerodynamic database constructed in Step 3. The network input is the aerodynamic characteristic information of the aircraft, and the network output is the geometric feature parameters of the aircraft.

[0010] Step 6: Input the designed aerodynamic characteristics information into the trained deep neural network model to predict the geometric characteristic parameters of the aircraft that meet the design requirements.

[0011] In step 2,

[0012] The key feature of geometric characteristic parameters lies in combining the geometric shape and aerodynamic characteristics of an aircraft. For example, in a wing-body combination, the fuselage often adopts a slender body shape, and the wing adopts a large-sweep trapezoidal wing. For the trapezoidal wing, its relative planar shape can be determined by the aspect ratio, root-to-tip ratio, and leading-edge sweep angle. In order to meet the requirements of high speed and high lift-to-drag ratio and low speed and high lift, the wing is accompanied by a leading-edge extension design, from which geometric characteristic parameters such as wing aspect ratio, wing sweep angle, fuselage slenderness ratio, and leading-edge extension sweep angle can be extracted.

[0013] A spin-type fuselage is formed by rotating a generatrix (a smooth curve or a broken line) around a certain axis, which is called the body axis of the spin-type fuselage. Any cross-section of a spin-type fuselage is circular, and its shape is characterized by a pointed nose, high aspect ratio, large wing sweep angle, and low aspect ratio. It often adopts a tailless aerodynamic configuration. Furthermore, the wing position has a significant impact on the aerodynamic characteristics of the spin-type fuselage; therefore, geometric characteristic parameters such as aspect ratio, taper ratio, taper ratio, ratio of the maximum cross-sectional area of ​​the fuselage to the wing area, and ratio of the length from the leading edge of the wing root chord to the tip of the fuselage to the average aerodynamic chord length of the wing can be extracted.

[0014] For waveriders, their shape design is characterized by solving their geometry in reverse by using a known supersonic flow field. Therefore, for waverider configurations, it is necessary to combine flow field design and geometric shape to extract geometric feature parameters, such as aspect ratio, height-to-width ratio of the symmetry plane, and volume ratio.

[0015] In step 2, for lifting body aircraft, the Cuziman coefficient τ and volumetric factor I are extracted. vol Projected area S in the Y direction Y With reference area S ref The ratio of S Y / S ref Z-direction projected area S Z The ratio of the area to the reference area S Z / S ref , wetted area S wet With plane area S plan The ratio of S wet / S plan As a geometric characteristic parameter; the projected area of ​​the aircraft in the X direction is used as the reference area S. ref The projected area in the Z direction is taken as the plane area S. plan The coordinate system is defined as follows: the origin is the center of mass O of the aircraft, the positive direction of the X-axis is from the center of mass to the nose, the Z-axis is located in the longitudinal plane of symmetry of the aircraft, perpendicular to the X-axis, and the positive direction of the Z-axis is upward, and the Y-axis is perpendicular to the OXZ plane. The positive direction of the Y-axis is determined by the right-hand rule.

[0016] Step 2 also includes: using single-parameter sensitivity analysis to analyze the extracted geometric feature parameter A, plotting the aerodynamic characteristics of the aircraft as a function of parameter A, with parameter A as the abscissa and the aerodynamic characteristics as the ordinate, analyzing the impact of parameter A on the aerodynamic characteristics of the aircraft, such as whether it will increase or decrease the lift coefficient, or whether it will have no effect on the lift coefficient. If parameter A will increase or decrease the lift coefficient, then parameter A will be extracted into the geometric feature parameter set; if it has no effect, then it will not be extracted.

[0017] Step 3 includes:

[0018] Step 3-1, Establish a geometric shape library: Use parametric modeling to generate two or more aircraft shapes, that is, use two or more parameters to describe the shape when designing the aircraft geometry, such as length, width, height, etc., and then modify the aircraft shape by changing the value of the parameter, thereby generating two or more aircraft shapes; and extract the geometric feature parameters of each aircraft according to the geometric feature parameter set extracted in Step 2.

[0019] Step 3-2, Establish an aerodynamic database: Obtain aerodynamic characteristic data for each aircraft, including lift coefficient, drag coefficient, and moment coefficient, through engineering estimation methods, CFD simulation, or wind tunnel testing methods.

[0020] Step 3-1 includes: First, establishing a baseline model based on the geometric design parameters of the aircraft's parametric modeling. The modeling method adopts the control section shape modeling method, dividing the aircraft from the head to the tail into two or more sections, which are called control sections. For each section, the height, aspect ratio, and cross-sectional geometry are designed first to obtain a series of baseline shapes. The control parameters of each control section are the design variables for parametric modeling.

[0021] Secondly, since multiple control sections are used to parameterize the aircraft, multiple design variables are generated. Different aircraft shapes are produced by changing the values ​​of these design variables. Using all permutations and combinations of the design variables would result in an excessively large sample size. Therefore, orthogonal experimental design is used to generate the design space for the samples. The core of orthogonal experiments is the orthogonal array, using L... m (q n ) represents an orthogonal array, L m (q n The expression () indicates that for an orthogonal experiment with n design variables and q levels for each design variable, a total of m samples were generated. The level refers to the number of possible values ​​for a design variable. For example, the aspect ratio of section 1 could take values ​​of [0.3, 0.4, 0.5, 0.6, 0.7, 0.8], making this factor six levels. Furthermore, since it is impossible to guarantee that the number of levels for each design variable is the same, a mixed-level orthogonal array is used to design the sample space. The expression for the mixed-level orthogonal array is Lm(q1n) 1 ×q2n 2 ), This indicates that there are n1 design variables with q1 levels. This indicates that there are n² design variables with q² levels, denoted by L. 49 (4 6 ×7 1For example, this means there are 6 design variables with 4 levels and 1 design variable with 7 levels, generating a total of 49 samples. The parameter values ​​for different shaped aircraft are recorded in an orthogonal table, which is called a parameter design table.

[0022] The generation of aircraft with different shapes is achieved using OpenVSP software. Based on the obtained parameter design table, a design file (in *.des format) is generated to control the cross-sectional shape. This design file is used to adjust the geometry of the baseline model and includes the number of design parameters and their specific values. A script file (in *.vspscript format) is then written, and OpenVSP software is used to call the script file to generate two or more aircraft shapes in batches.

[0023] In step 3-2, for the lifting body aircraft, the head section shape, head section height, head section aspect ratio, tail section shape, tail section height, and tail section aspect ratio are used as design parameters. The section shape includes ellipse, triangle, and rhombus. Two or more lifting body models are generated by changing the design parameters.

[0024] The aerodynamic characteristics of each lifting body are calculated using engineering estimation methods: the aircraft shape is divided into two or more surface elements (usually many surfaces), and the pressure coefficient C of each surface element is solved. p Then, the aerodynamic characteristics of the entire aircraft are obtained by integration. When solving for the pressure of the surface element, the surface element is first divided into a windward side and a leeward side according to the sign of the impact angle. The formula for calculating the impact angle is:

[0025]

[0026] in It is the outward normal vector of the surface element; The velocity vector of the incoming flow; δ greater than 0 represents the impact angle on the windward side, and less than 0 represents the impact angle on the leeward side;

[0027] For the windward side, the selected pressure coefficient C p The calculation formula is based on the modified Newtonian theory:

[0028]

[0029] Where γ is the specific heat ratio, Ma ∞ It is the Mach number;

[0030] For the leeward side, the selected pressure coefficient C p The Prandtl-Meyer formula is used for calculation.

[0031]

[0032] Then, the pressure coefficient C of each element was used. p Unit normal vector Area ΔA and aircraft reference area S ref The aerodynamic coefficients of the aircraft in the X, Y, and Z directions are obtained by integration;

[0033] Unit normal vector The calculation formula is:

[0034]

[0035] Where |N| is The length of the mold, N x for The components of a vector in the X direction, N y for The components of the vector in the Y direction, N z for The component of the vector in the Z direction; for The component in the X direction, for ni The component in the Y direction, for The component in the Z direction;

[0036] The integral formula is:

[0037] F x =∑C p n x ΔA / S ref

[0038] F y =∑C p n y ΔA / S ref

[0039] F z =∑C p n z ΔA / S ref

[0040] Where, n x for The value of n y for The value of n z for The value of F; x Let F be the aerodynamic coefficient in the X direction. y Let F be the aerodynamic coefficient in the Y direction. z denoted as the aerodynamic coefficient in the Z direction.

[0041] Finally, the lift coefficient C is obtained. L Drag coefficient C D Torque coefficient C M :

[0042] C L =-F x sinα+F z cosα

[0043] C D =F x cosαcosβ+F y sinβ+F z sinαcosβ

[0044]

[0045]

[0046] Where L ref The characteristic length is represented by the subscript cg, which indicates the center of gravity position, α is the angle of attack, and β is the sideslip angle; C Mx C My C Mz These represent the moment coefficients of the aircraft about the X-axis, Y-axis, and Z-axis, respectively; x, y, and z are the coordinate values ​​of the surface element's location, x... cg Let y be the coordinate of the aircraft's center of gravity on the X-axis. cg Let z be the coordinate of the aircraft's center of gravity on the Y-axis. cg This represents the coordinates of the aircraft's center of gravity on the Z-axis.

[0047] Based on engineering estimation methods, the lift coefficient, drag coefficient, and pitch moment coefficient of each aircraft in the geometric shape library are calculated at different Mach numbers and angles of attack, and all the calculated aerodynamic data are used as an aerodynamic database.

[0048] In step 4, a one-dimensional convolutional neural network is used as the deep neural network model. The one-dimensional convolutional neural network includes convolutional layers, pooling layers, and fully connected layers. The convolutional layers can extract features from the input data; the pooling layers are used to reduce the data size, extract information after convolution, improve the computation speed, reduce the number of parameters, and prevent overfitting; the convolutional layers and pooling layers appear alternately to continuously learn the features of the input data, and finally the features are passed to the fully connected layer to obtain the network output.

[0049] In step 5, the aerodynamic database constructed in step 2 is randomly divided into a training set, a validation set, and a test set. The lift coefficient, drag coefficient, lift-to-drag ratio, pitch moment coefficient, and angle of attack of each shape are used as inputs to a one-dimensional convolutional neural network, and the geometric feature parameters of the corresponding shape are used as the network outputs. During network training, the learning rate is set to 2×10.-4 The optimizer chosen is Adam, and the loss function is the mean squared error (MSE).

[0050]

[0051] Where n is the sample size, Y i For the true value, This is the network prediction value.

[0052] The present invention also provides an electronic device, characterized in that it includes a processor and a memory, the memory storing program code that, when executed by the processor, causes the processor to perform the steps of the method.

[0053] Beneficial Effects: The neural network model established by this invention, which transforms aerodynamic parameters into geometric shape feature parameters, outputs not the geometric design parameters used in parametric modeling, but rather a set of geometric feature parameters that can generalize to a certain type of aircraft shape. This gives the network model shape generalization capabilities. Taking lifting bodies as an example, after obtaining the geometric feature parameters of a lifting body aircraft that meet aerodynamic design requirements through neural network prediction, and provided that the aerodynamic shape characteristics of this type of aircraft are satisfied, designers can establish a 3D model of the lifting body aircraft using different parametric methods. The design variables can be determined by the design requirements, as long as the geometric feature parameter values ​​of the model are consistent with the predicted geometric feature parameter values. Attached Figure Description

[0054] The present invention will be further described in detail below with reference to the accompanying drawings and specific embodiments, and the advantages of the present invention in the above and / or other aspects will become clearer.

[0055] Figure 1 This is a schematic diagram of the coordinate system definition.

[0056] Figure 2 This is a schematic diagram of a lifting body model generated using a parametric method.

[0057] Figure 3 This is a schematic diagram of the surface pressure distribution of X-37.

[0058] Figure 4a , Figure 4b , Figure 4c This is a schematic diagram comparing the aerodynamic characteristics estimated by the engineering team with the CDF results.

[0059] Figure 5 This is a diagram of a one-dimensional convolutional neural network structure.

[0060] Figure 6a , Figure 6b This is a diagram illustrating the network training loss.

[0061] Figure 7a , Figure 7b , Figure 7c , Figure 7d , Figure 7e This is a schematic diagram illustrating the network's prediction results of the geometric feature parameters of the training and test samples.

[0062] Figure 8 This is a flowchart of the aircraft shape library construction process. Detailed Implementation

[0063] This invention provides a deep learning-based method for reverse modeling of aircraft geometric feature parameters, comprising the following steps:

[0064] Step 1: Classify the aircraft based on its shape and aerodynamic characteristics. Currently, there are various ways to classify aircraft, such as by purpose (civilian vs. military), by speed (subsonic vs. supersonic), and by engine type (propeller vs. jet). Since the goal of this invention is to perform inverse modeling of the geometric features of aircraft, and different aerodynamic layouts result in significant differences in shape, classification based on both geometric and aerodynamic characteristics is necessary to ensure that the geometric features extracted in Step 2 are applicable to the specific type of aircraft. Current classification methods are mostly arbitrary; for example, hypersonic aircraft are classified according to their aerodynamic layout as lifting bodies, wing-body combinations, spinning bodies, and waveriders.

[0065] Step 2: For different types of aircraft, analyze their shape characteristics and aerodynamic characteristics, and extract a set of geometric feature parameters;

[0066] Step 3: Construct an aerodynamic database for training neural network models;

[0067] Step 4: Build a deep neural network model;

[0068] Step 5, Deep Neural Network Training and Testing: The deep neural network model established in Step 4 is trained using the aerodynamic database constructed in Step 3. The network input is the aerodynamic characteristic information of the aircraft, and the network output is the geometric feature parameters of the aircraft.

[0069] Step 6: Input the designed aerodynamic characteristics information into the trained deep neural network model to predict the geometric characteristic parameters of the aircraft that meet the design requirements.

[0070] In step 2,

[0071] The key feature of geometric characteristic parameters lies in combining the geometric shape and aerodynamic characteristics of an aircraft. For example, in a wing-body combination, the fuselage often adopts a slender body shape, and the wing adopts a large-sweep trapezoidal wing. For the trapezoidal wing, its relative planar shape can be determined by the aspect ratio, root-to-tip ratio, and leading-edge sweep angle. In order to meet the requirements of high speed and high lift-to-drag ratio and low speed and high lift, the wing is accompanied by a leading-edge extension design, from which geometric characteristic parameters such as wing aspect ratio, wing sweep angle, fuselage slenderness ratio, and leading-edge extension sweep angle can be extracted.

[0072] A spin-type fuselage is formed by rotating a generatrix (a smooth curve or a broken line) around a certain axis, which is called the body axis of the spin-type fuselage. Any cross-section of a spin-type fuselage is circular, and its shape is characterized by a pointed nose, high aspect ratio, large wing sweep angle, and low aspect ratio. It often adopts a tailless aerodynamic configuration. Furthermore, the wing position has a significant impact on the aerodynamic characteristics of the spin-type fuselage; therefore, geometric characteristic parameters such as aspect ratio, taper ratio, taper ratio, ratio of the maximum cross-sectional area of ​​the fuselage to the wing area, and ratio of the length from the leading edge of the wing root chord to the tip of the fuselage to the average aerodynamic chord length of the wing can be extracted.

[0073] For waveriders, their shape design is characterized by solving their geometry in reverse by using a known supersonic flow field. Therefore, for waverider configurations, it is necessary to combine flow field design and geometric shape to extract geometric feature parameters, such as aspect ratio, height-to-width ratio of the symmetry plane, and volume ratio.

[0074] In step 2, for lifting body aircraft, the Cuziman coefficient τ and volumetric factor I are extracted. vol Projected area S in the Y direction Y With reference area S ref The ratio of S Y / S ref Z-direction projected area S Z The ratio of the area to the reference area S Z / S ref , wetted area S wet With plane area S plan The ratio of S wet / S plan As a geometric characteristic parameter; the projected area of ​​the aircraft in the X direction is used as the reference area S. ref The projected area in the Z direction is taken as the plane area S. plan The coordinate system is defined as follows: the origin is the center of mass O of the aircraft, the positive direction of the X-axis is from the center of mass to the nose, the Z-axis is located in the longitudinal plane of symmetry of the aircraft, perpendicular to the X-axis, and the positive direction of the Z-axis is upward, and the Y-axis is perpendicular to the OXZ plane. The positive direction of the Y-axis is determined by the right-hand rule.

[0075] Step 2 also includes: using single-parameter sensitivity analysis to analyze the extracted geometric feature parameter A, plotting the aerodynamic characteristics of the aircraft as a function of parameter A, with parameter A as the abscissa and the aerodynamic characteristics as the ordinate, analyzing the impact of parameter A on the aerodynamic characteristics of the aircraft, such as whether it will increase or decrease the lift coefficient, or whether it will have no effect on the lift coefficient. If parameter A will increase or decrease the lift coefficient, then parameter A will be extracted into the geometric feature parameter set; if it has no effect, then it will not be extracted.

[0076] Step 3 includes:

[0077] Step 3-1, Establish a geometric shape library: Use parametric modeling to generate two or more aircraft shapes, that is, use two or more parameters to describe the shape when designing the aircraft geometry, such as length, width, height, etc., and then modify the aircraft shape by changing the value of the parameter, thereby generating two or more aircraft shapes; and extract the geometric feature parameters of each aircraft according to the geometric feature parameter set extracted in Step 2.

[0078] Step 3-2, Establish an aerodynamic database: Obtain aerodynamic characteristic data for each aircraft, including lift coefficient, drag coefficient, and moment coefficient, through engineering estimation methods, CFD simulation, or wind tunnel testing methods.

[0079] Step 3-1 includes: First, establishing a baseline model based on the geometric design parameters of the aircraft's parametric modeling. The modeling method adopts the control section shape modeling method, dividing the aircraft from the head to the tail into two or more sections, which are called control sections. For each section, the height, aspect ratio, and cross-sectional geometry are designed first to obtain a series of baseline shapes. The control parameters of each control section are the design variables for parametric modeling.

[0080] Secondly, since multiple control sections are used to parameterize the aircraft, multiple design variables are generated. Different aircraft shapes are produced by changing the values ​​of these design variables. Using all permutations and combinations of the design variables would result in an excessively large sample size. Therefore, orthogonal experimental design is used to generate the design space for the samples. The core of orthogonal experiments is the orthogonal array, using L... m (q n ) represents an orthogonal array, L m (q n The expression () indicates that for an orthogonal experiment with n design variables and q levels for each design variable, a total of m samples were generated. The level refers to the number of possible values ​​for a design variable. For example, the aspect ratio of section 1 could take values ​​of [0.3, 0.4, 0.5, 0.6, 0.7, 0.8], making this factor six levels. Furthermore, since it is impossible to guarantee that the number of levels for each design variable is the same, a mixed-level orthogonal array is used to design the sample space. The expression for the mixed-level orthogonal array is: This indicates that there are n1 design variables with q1 levels. This indicates that there are n² design variables with q² levels, denoted by L. 49 (4 6 ×7 1 For example, this means there are 6 design variables with 4 levels and 1 design variable with 7 levels, generating a total of 49 samples. The parameter values ​​for different shaped aircraft are recorded in an orthogonal table, which is called a parameter design table.

[0081] The generation of aircraft with different shapes is achieved using OpenVSP software. Based on the obtained parameter design table, a design file (in *.des format) is generated to control the cross-sectional shape. This design file is used to adjust the geometry of the baseline model and includes the number of design parameters and their specific values. A script file (in *.vspscript format) is written, and OpenVSP software calls this script file to generate two or more aircraft shapes in batches. Figure 8 The diagram shows the process of building an aircraft shape library.

[0082] In step 3-2, for the lifting body aircraft, the head section shape, head section height, head section aspect ratio, tail section shape, tail section height, and tail section aspect ratio are used as design parameters. The section shape includes ellipse, triangle, and rhombus. Two or more lifting body models are generated by changing the design parameters.

[0083] The aerodynamic characteristics of each lifting body are calculated using engineering estimation methods: the aircraft shape is divided into two or more surface elements (usually many surfaces), and the pressure coefficient C of each surface element is solved. p Then, the aerodynamic characteristics of the entire aircraft are obtained by integration. When solving for the pressure of the surface element, the surface element is first divided into a windward side and a leeward side according to the sign of the impact angle. The formula for calculating the impact angle is:

[0084]

[0085] in It is the outward normal vector of the surface element; The velocity vector of the incoming flow; δ greater than 0 represents the impact angle on the windward side, and less than 0 represents the impact angle on the leeward side;

[0086] For the windward side, the selected pressure coefficient C p The calculation formula is based on the modified Newtonian theory:

[0087]

[0088] Where γ is the specific heat ratio, Ma ∞ It is the Mach number;

[0089] For the leeward side, the selected pressure coefficient C p The Prandtl-Meyer formula is used for calculation.

[0090]

[0091] Then, the pressure coefficient C of each element was used. p Unit normal vector Area ΔA and aircraft reference area S ref The aerodynamic coefficients of the aircraft in the X, Y, and Z directions are obtained by integration;

[0092] Unit normal vector The calculation formula is:

[0093]

[0094] Where |N| is The length of the mold, N x for The components of a vector in the X direction, N y for The components of the vector in the Y direction, N z for The component of the vector in the Z direction; for The component in the X direction, for The component in the Y direction, for The component in the Z direction;

[0095] The integral formula is:

[0096] F x =∑C p n x ΔA / S ref

[0097] F y =∑C p n y ΔA / S ref

[0098] F z =∑C p n z ΔA / S ref

[0099] Where, n x for The value of n y for The value of n z for The value of F; x Let F be the aerodynamic coefficient in the X direction. y Let F be the aerodynamic coefficient in the Y direction. z denoted as the aerodynamic coefficient in the Z direction.

[0100] Finally, the lift coefficient C is obtained. L Drag coefficient C D Torque coefficient C M :

[0101] C L =-F x sinα+F z cosα

[0102] C D =F x cosαcosβ+F y sinβ+F z sinαcosβ

[0103]

[0104] Where L ref The characteristic length is represented by the subscript cg, which indicates the center of gravity position, α is the angle of attack, and β is the sideslip angle; C Mx C My C Mz These represent the moment coefficients of the aircraft about the X-axis, Y-axis, and Z-axis, respectively; x, y, and z are the coordinate values ​​of the surface element's location, x... cg Let y be the coordinate of the aircraft's center of gravity on the X-axis. cg Let z be the coordinate of the aircraft's center of gravity on the Y-axis. cg This represents the coordinates of the aircraft's center of gravity on the Z-axis.

[0105] Based on engineering estimation methods, the lift coefficient, drag coefficient, and pitch moment coefficient of each aircraft in the geometric shape library are calculated at different Mach numbers and angles of attack, and all the calculated aerodynamic data are used as an aerodynamic database.

[0106] In step 4, a one-dimensional convolutional neural network is used as the deep neural network model. The one-dimensional convolutional neural network includes convolutional layers, pooling layers, and fully connected layers. The convolutional layers can extract features from the input data; the pooling layers are used to reduce the data size, extract information after convolution, improve the computation speed, reduce the number of parameters, and prevent overfitting; the convolutional layers and pooling layers appear alternately to continuously learn the features of the input data, and finally the features are passed to the fully connected layer to obtain the network output.

[0107] In step 5, the aerodynamic database constructed in step 2 is randomly divided into a training set, a validation set, and a test set. The lift coefficient, drag coefficient, lift-to-drag ratio, pitch moment coefficient, and angle of attack of each shape are used as inputs to a one-dimensional convolutional neural network, and the geometric feature parameters of the corresponding shape are used as the network outputs. During network training, the learning rate is set to 2×10. -4 The optimizer chosen is Adam, and the loss function is the mean squared error (MSE).

[0108]

[0109] Where n is the sample size, Y i For the true value, This is the network prediction value.

[0110] In a specific embodiment of the present invention, a deep learning-based method for reverse modeling of aircraft geometric feature parameters is provided, comprising the following steps:

[0111] Step 1: Taking a lifting body aircraft as an example, extract the Cuziman coefficient τ and the volumetric ratio I. vol The ratio of the projected area in the Y direction to the reference area, S Y / S ref The ratio of the projected area in the Z direction to the reference area, S Z / S ref The ratio S of wetted area to planar area wet / S plan As a set of geometric feature parameters, the projected area of ​​the aircraft in the X direction is used as the reference area S. ref The projected area in the Z direction is taken as the plane area S. plan . Figure 1 This is the coordinate system definition method used in this embodiment.

[0112] Step 2: Establish a lifting body shape database using parametric methods. In this embodiment, the head section shape, head section height, head section aspect ratio, tail section shape, tail section height, and tail section aspect ratio are used as design parameters. The section shapes include ellipses, triangles, rhombuses, etc. Multiple lifting body models are generated by changing the design parameters. Figure 2 The generated partial lifting body model is shown.

[0113] This embodiment employs an engineering estimation method to calculate the aerodynamic characteristics of each lifting body. Specifically, the aircraft's shape is divided into several surfaces, and the pressure coefficients of these local surfaces are calculated, followed by integration to obtain the overall aerodynamic characteristics of the aircraft. For the windward surface, this embodiment uses a modified Newtonian theory formula for calculating the pressure coefficient:

[0114]

[0115] Where γ is the specific heat ratio, Ma ∞ Where is the Mach number and δ is the impact angle. The leeward side is chosen using the Prandtl-Meyer formula:

[0116]

[0117] Figure 3 The pressure distribution on the X-37 surface is calculated using the above formula. In the figure, Cp_ep represents the pressure coefficient obtained by engineering estimation method. Figure 4a , Figure 4b , Figure 4c This figure compares the engineering estimation results and CFD results for X-37. In the figure, EEM represents the engineering estimation results, and C... L C D L and L / D represent the lift coefficient, drag coefficient, and lift-to-drag ratio, respectively, and AOA is the angle of attack. It is evident that the engineering estimation and CFD results agree well; therefore, this paper employs the engineering estimation method to calculate the aerodynamic characteristics of the lifting body.

[0118] In this embodiment, a total of 450 lifting body shapes were generated, and the lift coefficient, drag coefficient, lift-to-drag ratio and pitching moment coefficient of each lifting body shape were calculated. The Mach number for the calculation condition was 3, the angle of attack range was -10° to 20°, and the step size was 2°.

[0119] Step 3, Deep Neural Network Model Construction. This embodiment uses a one-dimensional convolutional neural network as the deep neural network model. Figure 5 This describes the network framework. A one-dimensional convolutional neural network (CNN) consists of convolutional layers, pooling layers, and fully connected layers. Convolutional layers extract features from the input data. Pooling layers reduce the data size, extract the main information after convolution, improve computational speed, reduce the number of parameters, and prevent overfitting. In CNNs, convolutional and pooling layers typically alternate, continuously learning the features of the input data, and finally passing these features to the fully connected layers to obtain the network output.

[0120] Step 4, Neural Network Training. The aerodynamic database constructed in Step 2 is divided into a training set, a validation set, and a test set in an 8:1:1 ratio, i.e., 360 shapes for training, 45 shapes for validation, and 45 shapes for test. The lift coefficient, drag coefficient, lift-to-drag ratio, pitch moment coefficient, and angle of attack of each shape are used as inputs to a one-dimensional convolutional neural network, and the geometric feature parameters of the corresponding shape are used as the network output. During network training, the learning rate is set to 2×10⁻⁶. -4 The optimizer chosen is Adam, and the loss function is Mean Square Error (MSE).

[0121]

[0122] Where n is the sample size, Y i For the true value, This is the network prediction value. Figure 6a , Figure 6b The training set loss and validation set loss are used for network training. In the figure, Epoch represents the number of training iterations.

[0123] Step 5, Network Training Results and Testing. The network trained in Step 4 has converged and can therefore be used for testing. The network is tested using the test set data divided in Step 4, with the lift coefficient, drag coefficient, lift-to-drag ratio, pitch moment coefficient, and angle of attack information for each shape inputted. Figure 7a , Figure 7b , Figure 7c , Figure 7d , Figure 7e The diagram shows the network's prediction results for the geometric feature parameters of the lifting body shape on the training and test sets. The left side shows the network's prediction results for the training samples, and the right side shows the network's prediction results for the test samples. Figure 7a , Figure 7b , Figure 7c , Figure 7d , Figure 7e The horizontal axis represents the true value, and the vertical axis represents the predicted value. Each point represents a sample. The closer a point is to the 45° line, the closer the predicted value is to the true value. Kuchemann's tau is the Kuchemann coefficient. For the training set data, all sample points are located on the 45° line. For the test data, some points deviate from the 45° line. Among all test samples, the average relative errors for each geometric parameter prediction are 0.889%, 0.489%, 2.322%, 0.821%, and 0.754%, respectively. The largest relative errors are 5.341%, 3.301%, 8.093%, 6.011%, and 2.414%, respectively, all of which are less than 10%, indicating that the network has high prediction accuracy for the test samples.

[0124] In another embodiment of the present invention, in step 3, an aerodynamic database is constructed from historical aerodynamic data. When constructing the shape database, different shapes can also be generated using the Free-Form Deformation (FFD) method, and CFD methods or wind tunnel testing methods can be used to obtain aerodynamic characteristic data.

[0125] In another embodiment of the present invention, in step 4, other surrogate models are used to replace the deep neural network model, such as the multinomial response surface model, the Kriging model, the radial basis function, the support vector machine, etc.

[0126] This invention provides a deep learning-based method for reverse modeling of aircraft geometric feature parameters. Many methods and approaches exist for implementing this technical solution; the above description is merely a preferred embodiment. It should be noted that those skilled in the art can make various improvements and modifications without departing from the principles of this invention, and these improvements and modifications should also be considered within the scope of protection of this invention. All components not explicitly stated in this embodiment can be implemented using existing technologies.

Claims

1. A deep learning-based method for reverse modeling of aircraft geometric feature parameters, characterized in that, Includes the following steps: Step 1: Classify aircraft based on their shape and aerodynamic characteristics; Step 2: For different types of aircraft, analyze their shape characteristics and aerodynamic characteristics, and extract a set of geometric feature parameters; Step 3: Construct an aerodynamic database for training neural network models; Step 4: Build a deep neural network model; Step 5, Deep Neural Network Training and Testing: The deep neural network model established in Step 4 is trained using the aerodynamic database constructed in Step 3. The network input is the aerodynamic characteristic information of the aircraft, and the network output is the geometric feature parameters of the aircraft. Step 6: Input the designed aerodynamic characteristic information into the trained deep neural network model to predict the geometric characteristic parameters of the aircraft that meet the design requirements. In step 2, for the wing-body combination, the wing aspect ratio, wing sweep angle, fuselage slenderness ratio, and strake sweep angle are extracted as geometric feature parameters. For a rotating body, the following geometric feature parameters are extracted: slenderness ratio, contraction ratio, taper ratio, ratio of the maximum cross-sectional area of ​​the fuselage to the area of ​​the wing, and ratio of the length from the leading edge of the wing root chord to the tip of the fuselage to the average aerodynamic chord length of the wing. For waveriders, the aspect ratio, the height-to-width ratio of the symmetry plane, and the volume ratio are extracted as geometric feature parameters. In step 2, for lifting body aircraft, the Cuziman coefficient is extracted. Floor area ratio Projected area in the Y direction Compared with reference area ratio Z-direction projected area Ratio to reference area wetted area With plane area ratio As a geometric feature parameter; The projected area of ​​the aircraft in the X direction is used as the reference area. The projected area in the Z direction is used as the plane area. For the definition of the coordinate system, the origin of the coordinate system is the center of mass O of the aircraft. The positive direction of the X-axis is from the center of mass to the nose of the aircraft. The Z-axis is located in the longitudinal symmetry plane of the aircraft, perpendicular to the X-axis, and upward is the positive direction of the Z-axis. The Y-axis is perpendicular to the OXZ plane, and the positive direction of the Y-axis is determined according to the right-hand rule. Step 2 also includes: using single-parameter sensitivity analysis to analyze the extracted geometric feature parameter A, plotting the aerodynamic characteristics of the aircraft as a function of parameter A, using parameter A as the abscissa and aerodynamic characteristics as the ordinate, analyzing the impact of parameter A on the aerodynamic characteristics of the aircraft, if parameter A increases or decreases the lift coefficient, then parameter A is extracted into the geometric feature parameter set, otherwise it is not extracted; Step 3 includes: Step 3-1, Establish a geometric shape library: Use parametric modeling to generate two or more aircraft shapes, that is, use two or more parameters to describe the shape when designing the aircraft geometry, and then modify the aircraft shape by changing the value of the parameter, thereby generating two or more aircraft shapes; and extract the geometric feature parameters of each aircraft according to the geometric feature parameter set extracted in Step 2. Step 3-2, Establish an aerodynamic database: Obtain aerodynamic characteristic data for each aircraft, including lift coefficient, drag coefficient, and moment coefficient; Step 3-1 includes: establishing a baseline model based on the geometric design parameters of the aircraft's parametric modeling; the modeling method adopts the control section shape modeling method, dividing the aircraft from the head to the tail into two or more sections, and referring to these sections as control sections; for each section, first designing the height, aspect ratio, and cross-sectional geometry to obtain a series of baseline shapes; the control parameters of each control section are the design variables for parametric modeling. Different aircraft shapes are generated by changing the values ​​of various design variables: an orthogonal experimental design method is used to generate the design space of the samples. Represents an orthogonal array. Indicates that for those who have There are 1 design variables, and each design variable contains 1 design variable. A total of m samples were generated from the orthogonal experiment at each level; A mixed-level orthogonal array was used to design the sample space. The expression for the mixed-level orthogonal array is: , Indicates that there is The design variables are A level, Indicates that there is The design variables are: At each level, the parameter values ​​of different shaped aircraft are recorded in an orthogonal table, which is called a parameter design table. OpenVSP software is used to generate aircraft with different shapes. Based on the obtained parameter design table, a parameter design file is generated to control the cross-sectional shape. The parameter design file is used to adjust the geometry of the reference model. The content includes the number of design parameters and the specific value of each design parameter. A script file is written and OpenVSP software is used to call the script file to generate more than two aircraft shapes in batches. In step 3-2, for the lifting body aircraft, the head section shape, head section height, head section aspect ratio, tail section shape, tail section height, and tail section aspect ratio are used as design parameters. The section shape includes ellipse, triangle, and rhombus. Two or more lifting body models are generated by changing the design parameters. The aerodynamic characteristics of each lifting body are calculated using engineering estimation methods: the aircraft shape is divided into two or more surface elements, and the pressure coefficient of each surface element is solved. Then, the aerodynamic characteristics of the entire aircraft are obtained by integration. When solving for the pressure of the surface element, the surface element is first divided into the windward and leeward sides according to the sign of the impact angle. The formula for calculating the impact angle is: ， in It is the outward normal vector of the surface element; The velocity vector of the incoming flow; A value greater than 0 indicates the angle of impact on the windward side, and a value less than 0 indicates the angle of impact on the leeward side. For the windward side, the selected pressure coefficient The calculation formula is based on the modified Newtonian theory: ， in For specific heat ratio, It is the Mach number; For the leeward side, the selected pressure coefficient The Prandtl-Meyer formula is used for calculation. ， Utilizing the pressure coefficient of each surface element Unit normal vector ,area and aircraft reference area Integrating the components yields the aerodynamic coefficients of the aircraft in the X, Y, and Z directions. Unit normal vector The calculation formula is: ， in for The length of the mold, , for The components of the vector in the X direction, for The components of the vector in the Y direction, for The component of the vector in the Z direction; for The component in the X direction, for The component in the Y direction, for The component in the Z direction; The integral formula is: ， ， ， in, for The value, for The value, for The value; Let X be the aerodynamic coefficient in the X direction. Let Y be the aerodynamic coefficient. The aerodynamic coefficient in the Z direction; Finally, the lift coefficient is obtained. drag coefficient Torque coefficient : ， ， ， ， ， Where L ref The characteristic length is represented by the subscript cg, which indicates the position of the center of gravity, α is the angle of attack, and β is the sideslip angle. , , These represent the moment coefficients of the aircraft about the X-axis, Y-axis, and Z-axis, respectively; x, y, and z are the coordinate values ​​of the surface element's location. This represents the coordinates of the aircraft's center of gravity on the X-axis. This represents the coordinates of the aircraft's center of gravity on the Y-axis. This represents the coordinates of the aircraft's center of gravity on the Z-axis. Based on engineering estimation methods, the lift coefficient, drag coefficient, and pitch moment coefficient of each aircraft in the geometric shape library are calculated at different Mach numbers and angles of attack, and all the calculated aerodynamic data are used as an aerodynamic database. In step 4, a one-dimensional convolutional neural network is used as the deep neural network model. The one-dimensional convolutional neural network includes convolutional layers, pooling layers, and fully connected layers. The convolutional layers can extract features from the input data; the pooling layers are used to reduce the data size, extract information after convolution, improve the computation speed, reduce the number of parameters, and prevent overfitting; the convolutional layers and pooling layers appear alternately to continuously learn the features of the input data, and finally the features are passed to the fully connected layer to obtain the network output. In step 5, the aerodynamic database constructed in step 2 is randomly divided into a training set, a validation set, and a test set. The lift coefficient, drag coefficient, lift-to-drag ratio, pitch moment coefficient, and angle of attack of each shape are used as inputs to a one-dimensional convolutional neural network, and the geometric feature parameters of the corresponding shape are used as the network outputs. During network training, the learning rate is set to 2×10. -4 The optimizer chosen is Adam, and the loss function is the mean squared error (MSE). ， in For sample size, For the true value, This is the network prediction value.

2. An electronic device, characterized in that, It includes a processor and a memory, the memory storing program code that, when executed by the processor, causes the processor to perform the steps of the method as described in claim 1.

Images